>  > with hoard a client machine will continue to work, if the server is down
>  > for a few minutes.
>  > 
>  > so, what's left ?
>  >  - i can't mount / as coda filesystem
> 
> Yup, good project.

i will look at the initrd stuff.
if coda is mounted as /coda, can i mount /coda/tmp from nfs ?
guess : mount operations are in the vfs layer, so it works ?

>  >  - i need a few config files in each pc's /etc to differ 
>  > 	(inittab, XF86Config and Xserver)
> 
> We can adapt Coda for that easily -- compare the @sys facility.  There
> is also general Linux VFS code to do expansion of special names: see
> linux/fs/nametrans.c.  That's a cool idea. 

whats the @sys facility ?

problem : it doesn't work. dentry stuff in 2.1.40 broke it, and nobody
fixed it. 

nametrans works like this:
/etc/inittab			(default)
/etc/inittab#host=fox#		(file for host fox)
/etc/inittab#type=server#	(for every machine that is compiled with
					type=server)

i would propose to do:
/etc/inittab -> "inittab#host#"		(a symlink)
/etc/inittab#host#		(may not exist, would everything to fail)
/etc/inittab#fox#		(if the symlink is broken, it should look
				 for #.*# in the symlink, and replace #host#
				 with #`hostname`#).

that would be faster (special lookups only on "broken" symlinks),
and will make these things obsolete : nametrans allows to restrict the
nametrans to some special uid or gid, and to disable it via an
environment variable. this is done for speed, and programs like nfs or
tar(backup) that need to work korrectly. with an "evaluated symlink"
aproach this wouldn't be necessary.

drawbacks : loss of some flexibility. maybe.

will look at the kernel next week...

andreas