[OpenAFS-devel] Re: Crash on AIX 6.1 TL6 SP3

Andrew Deason adeason@sinenomine.net
Thu, 24 Feb 2011 10:57:51 -0600


On Thu, 24 Feb 2011 16:12:37 +0100 (MET)
Niklas Edmundsson <Niklas.Edmundsson@hpc2n.umu.se> wrote:

> My conclusion is that IBM has failed to keep binary module compat
> intact in AIX 6.1.

Well, you could try comparing the structure sizes and various field
offsets (notably m_next), or even flag values or mbuf contents after
initialization... just ideas to see what's going on.

> What would the reasonable approach be to do this? Assume that people
> are using the rc.afs as shipped with openafs and script it there? Or
> be more hard core and code the check into cfgafs?

_Ideally_ there would be some kernel service to provide this
information, so we can just fail the load on in-kernel checks. But I, of
course, don't know if anything like that exists.

afsd doing the check may be sufficient, but if we wanted to do better,
we could have afsd tell the kernel module what version we are on
startup, and the kernel module can compare the version info.

> In any case we need to stamp the build with info on what it was built
> on, and I think that the oslevel -s output is more generic and useful
> even though it's slooow.

That seems fine for the build, but I'm less sure about that for every
time you start the client...

-- 
Andrew Deason
adeason@sinenomine.net