[OpenAFS] Re: Openafs kernel problems

[Andrew Deason]

On Tue, 4 Feb 2014 09:38:27 -0500
Dave Botsch <botsch@cnf.cornell.edu> wrote:

> I'm seeing a fair amount of these on a daily basis. At this point, I
> believe I'm seeing them mostly from my RHEL6/64 machines running the
> latest RHEL6.5 kernel and oafs 1.6.5.2 :

The way I'm reading this, it sounds like you're seeing multiple
instances of this on the same machine while the machine is booted. Is
that correct?

I believe this is a WARN_ON_ONCE warning, which should only happen once
while the machine is booted. But maybe I am misreading something or
looking at the wrong thing.

> backtrace:
> :WARNING: at fs/dcache.c:1249 d_splice_alias+0x16e/0x180() (Tainted: P           ---------------   )

I'm still waiting for redpatch.git to download, but from looking around
the kernel git, I would guess this is a variant of
7732a557b1342c6e6966efb5f07effcf99f56167, with an extra WARN_ON_ONCE to
detect when the relevant code path is hit. That commit was reverted in
Linux master because there apparently was a code path that could hit it
(without openafs); I assume Red Hat put it back in because they thought
it was impractical to hit, or they wanted to find out what it was.

But yeah, that looks like a problem. Our days of trying to dance around
dcache issues may be at an end, and so we may need to rework how we
interact with dcaches...

Oh, but if you want a workaround or something, you may be able to avoid
this if you can avoid accessing the same volume from different
mountpoints.

-- 
Andrew Deason
adeason@sinenomine.net