[OpenAFS] Openafs 1.2.5/Linux 2.4.18 problem (dd bs>1500kb)

[Steven N. Hirsch]

On Fri, 19 Jul 2002, Valentin v. Seggern wrote:

> I believe there is a strange bug in Openafs 1.2.5.
>  
> I do "dd if=/dev/zero  of=/afs/rz/somewhere bs=1500k count=10" it
> works well.
>  
> Everything below that works fine, too. I can increase count as well
> but as soon as I raise bs to bs=2048k dd will not work anymore
> (starting with count=3). I can not kill the process nor can I stop
> afs (dd is still using /afs). The only solution is rebooting.
>  
> I experience this on Linux 2.4.18 and Openafs 1.2.5.
>  
> I start afsd with -memcache -daemons 2 -volumes 50 -stat 300
> -logfile /var/log/afs. 
> 
> dd runs directly on the afs server. /var/log/afs is emtpy and afs is 
> not reporting errors in any of the logfiles.
>  
> On another unrelated computer (Linux 2.2.19, Openafs 1.2.1, -stat
> 300 -dcache 100 -daemons 2 -volumes 50 w/o diskcache) this works quite
> well.
>  
> So is this a diskcache/openafs or linux kernel problem?

Do you have the client cache in an ext3 filesystem?  I have noticed some 
very random and hard-to-reproduce strangeness with this combination - 
usually taking the form of processes stuck in 'D' or 'S' states.  Once I 
removed the has_journal flag and remounted as ext2, all the problems 
vanished.

Steve