Re: How do I kill a process that is locked to a resource

Nathan Bryant (nathan@burgessinc.com)
Fri, 20 Dec 1996 14:22:18 -0500 (EST)


On Fri, 20 Dec 1996, Tom Zerucha wrote:

> My SCSI disk is having problems. This usually means that whatever process
> has it open is busy, and cannot be killed in any manner, and so I cannot
> umount the disk (or more properly, shutdown will hang forever at one of
> these two points). So, I have to fsck every other open device after
> hitting reboot.
>
> Somewhere I saw that there would be a umount "force" option. I think
> there should be that, and also a kill -REALLY_DO_IT option, or maybe
> integrate this stuff into fuser.

Hacking the kernel to "really kill" something is not the solution. An
unkillable process indicates an error somewhere else.

I recently had a similar problem with an Advansys SCSI card. A SCSI bus
timeout would occur, the mid-layer SCSI code would try to reset the card,
and then the process would hang forever. It turned out that there was a
bug in the low-level Advansys driver's handling of synchronous SCSI bus
resets. Bob Frey <bobf@unix.advansys.com> sent me a patch which fixed the
Advansys driver and allowed the process to recover from the timeouts.

Are you using an Advansys card by any chance? If so, I think I've still
got the patch lying around.

If not, let's see the messages you're getting from the kernel. There may
well be a bug in the SCSI code that's preventing the process from
recovering.

+-----------------------+---------------------------------------+
| Nathan Bryant | Unsolicited commercial e-mail WILL be |
| nathan@burgessinc.com | charged an $80/hr proofreading fee. |
+-----------------------+---------------------------------------+