From: Mark Anderson <mda@di...> - 2003-06-26 06:22:12
Having to deal with these large file system files on the host
side can be a little frustrating as far as maintainance from
the host side is concerned.
So I started to try to imagine how to use UML without once
With a network of "real" computers one way to address this is
with PXE boot, which typically gest a minimal system
which then mounts various file systems over NFS (though
obviously PXE clients by no means have to be diskless).
This can be combined with something like http://clusternfs.sourceforge.net/
so that each machine can be configured identically (except for ip
or mac) and yet still get different actual file systems for the
same mount, when those mounts are read-write.
One idea is that UML might really offer a real pxe implementation
(say, by integrating with pxelinux). The host would have to run dhcp, tftpd,
etc., and have the full setup.
Maybe UML internals don't have to change?
Would it be possible to do something like specify "initrd=pxelinux.0"
on the command line, or something like that?
There would still need to be an option for pxelinux.0 to find
its config file.
Or, things could be cut a little later in the process, and instead
maybe make UML support an nfsroot command-line option?
My understanding of both PXE and UML is superficial, so sorry if
my suggestions are a bit incoherent, but hopefully my goal is clear.
From: Patrick \Petschge\ Kilian <petschge@we...> - 2003-06-30 20:34:01
Mark Anderson wrote:
> Having to deal with these large file system files on the host
> side can be a little frustrating as far as maintainance from
> the host side is concerned.
If you are using a 2.4.* kernel and have hostfs support compiled into the
kernel, there might be an easy way around the large image files.
First you unpack a existing image to a directory.
host# mkdir <new_uml_root>
host# mount <filesystem_image> /mnt/temp -o loop
host# cp -a /mnt/temp <new_uml_root>
host# umount /mnt/temp
When you are doing a
host# ls -R <new_uml_root>
you should see content of the image file.
Then you have to modify the <new_uml_root>/etc/fstab
so that / got a description like
none / hostfs defaults 1 1
Now you should be able to start the uml like this:
linux root=/dev/root rootflags=/path/to/<new_uml_root> rootfstype=hostfs
For further information have a look at
Patrick "Petschge" Kilian
PS.: I haven't tried this yet, because I'm using a 2.5.* kernel.
cat /dev/audio > install.rpm and whistle in the micro.