Hi Thomas, > Use strace to find the syscall which fails.
This led to the problem; thank you for the hint. Once you see it it's so obvious. It fails upon unpacking the devices.tar.gz because by default one is not allowed to create device nodes (mknod) inside a guest; which makes sense.
It would be nice if debootstrap would provide an error message and not just silently die. I might take the time to file an enhancement with the developers. Right now, it seems to die or skip some steps, then the make-fai-nfsroot script just continues and runs into that dpkg-divert problem, which is entirely unrelated; it's just the result of the half-way done and unnoticed debootstrap.
As long as one owns the vserver host (and not just the guest) the original problem can be solved by putting MKNOD into the /etc/vservers/<servername>/bcapabilities file. (Sorry to anyone who just rented a vserver as many people do these days.) That will make make-fai-nfsroot succeed.
Note: It might be necessary to allow the fai vserver guest some more freedom besides MKNOD in order to be able to really use it as a FAI install server. I personally only needed to have the (amd64) nfsroot created to port it onto an existing (i386) install server, so I did not care about the rest. The CAP_SYS_ADMIN capability is usually needed to allow the guest to do any mounting, which is also part of some of the scripts involved in FAI.
Thank you for your help on this. Regards, Torsten
