Panic

[Robert Watson]

On Tue, 1 Aug 2006, Robert Watson wrote:

> On Tue, 25 Jul 2006, Graham Menhennitt wrote:
>
>>>> Fatal trap 12: page fault while in kernel mode
>>>>> current process = 479 (mountd)
>>> 
>>> I have the same panic reproducibly. Shutting off nfs_server_enable (i.e. 
>>> mountd) in rc.conf prevents it. This is with 6-STABLE cvsupped yesterday. 
>>> I'll get some more info and follow up the PR.
>>> 
>> I rebuilt my kernel (to enable debugging) and now it doesn't panic. So it 
>> seems that an old kernel (from around the end of May) with a new mountd 
>> (from Sunday) will crash. But a new kernel with a new mountd won't.
>
> FYI, I've managed to reproduce this on a 7-CURRENT kernel, so will try to 
> take a look at this in detail in the next few days.  It looks like a race 
> during socket connect/accept for UNIX domain sockets, likely involving 
> simultaneous close, which may be a sign of a bug in mountd (or the like) 
> that triggers it. Of course, the kernel shouldn't panic under those 
> circumstances. :-)

On further reflection, this is simply a bug in the UNIX domain socket code, 
and has to do with a race between an attempt to connect to a socket and the 
socket being closed, such as may happen during a reboot.  I'll do some more 
digging and see about a possible fix for this.

Thanks,

Robert N M Watson
Computer Laboratory
University of Cambridge