Hi,
If possible can you do the following:
o build the kernel with kdb support turned on
o use kgcc to build it, I have seen bad stacks with
some other compilers.
o turn back on oplocks and recreate the hang
o work out which processes are hung
o on the console use break (or Ctrl-A on a serial console) to
get into kdb.
o type bt <pid> on the processes in D state and send us the
output.
A serial console makes this so much easier, but you may not have the
facilities.. If you do then the output of the bta command would be
even better.
You can continue the system with go, or reset it with reboot at this point.
This problem may just be an oplock bug, but it may be an interaction with
xfs.
Steve
> Hi,
>
> today something strange happened on a system running 2.4.6-pre1 from the
> latest CVS tree together with kernel NFS V3 support and samba-2.2 with
> windows 2000 clients. After I tried to move a directory under windows
> 2000 from one location to another within the same XFS partition the
> windows 2000 client locked up. First I thought there might be a windows
> problem - but it turned out that the corresponding smbd process on the
> linux box got into the D state and this happened to all other processes
> accessing the target directory (of the move).
> After a while I had a couple of processes (like ls, bash, find ...) all
> in the D state from the moment on when they tried to stat this
> particular directory. There were no system messages describing a
> possible reason for the lockup. After a reboot everything was fine
> again.
> In the meantime I configured samba not to use kernel OP locks and the
> problem did not occur again. This could be a clue but I was not able to
> do further testing since the system is used for production.
>
> Kind regards
>
> Deti
|