[OpenAFS] Fedora kernel builds

Simon Wilkinson sxw@inf.ed.ac.uk
Sat, 15 Mar 2008 22:43:25 +0000

>> Neither do I.  Do you have any suggestions as to
>> how one could either change the headlong rush to
>> GPL_ONLY in Linux (seems unlikely to me, given
>> the agendas),

I'm not convinced that things are (currently) that bad. We have an  
issue at present that a couple of debugging options make some core  
kernel features unavailable to us. One of those (RCU) was always  
going to disappear due, ironically enough, to IBM's patent license on  
the RCU technology. With regards to the other problem, I think  
there's a pretty strong argument that mutex_lock_nested() being GPL  
only is a bug, as it's such a core operating system feature. Whether  
we can successfully make that argument or not, is another matter.

One of the advantages of all of the work that's happened in the Linux  
cache manager is that it's moved AFS closer to being a 'standard'  
Linux filesystem - using standard kernel vnodes, rather than growing  
our own, moving away from using iget() to access local cache files,  
and using keyrings rather than patching the syscall table reduces our  
dependency on volatile non-public interfaces, and should help with  
portability in the long term.

If the worst comes to the worst, and core interfaces change in a way  
that means we can no longer use them, there are still other options  
open to us.

> I would rather spend our limited resources getting
> kernel modules developed for the BSDs than invest
> further in Linux which is run by folks who care
> more about agendas then technical issues.

This is a dangerous approach. Linux is by far the most prevalent of  
the free-Unixen. If OpenAFS was to stop supporting Linux, sites  
wouldn't use that as a reason to migrate away from Linux, they'd use  
it as a reason to pick a different file system.