[OpenAFS] 2.6?

[Jeffrey Hutzelman]

On Tuesday, January 27, 2004 18:53:29 +0000 Paul Blackburn 
<mpb@est.ibm.com> wrote:

> Please could you explain what are the problems
> with porting OpenAFS to the 2.6 kernel?

There are three major problems.  The first is that there is apparently a 
much higher degree of concurrency in the 2.6 kernel, requiring locking we 
don't currently have.

The remaining two problems can be summarized as "they hate us".

Specifically, they hate us, so we can't hook our syscall any more, which 
means all of our existing cache manager setup/control, pioctl, and KTC 
management interfaces need to be replaced.

And, they hate us, so we can't have PAG's (we can't hook setgroups anymore, 
either, so we can't do it ourselves, but Linux doesn't provide such a 
mechanism either).

Eventually, these problems will be resolved.


We have a student who is working on dealing with the user/kernel interfaces 
this semester.  He's only just begun to look at it in detail, so it's a 
little early to tell what the replacement interfaces will look like or when 
the work will be available.

I believe some work has been done wrt PAG's, and I know the NFSv4 community 
is interested in this issue as well.  Last I knew, Linus had indicated that 
only a reasonably general (not AFS-specific) solution would be appropriate 
for inclusion in the kernel; at that point, pre-2.6 releases were in 
feature freeze, and such a new mechanism would not have made it in.  I 
don't know if anyone has written code since then, or if such patches would 
be accepted into the 2.6 tree now.

-- Jeff