At 09:15 5-4-2001 -0500, Steve Lord wrote:
> >
> >I did see your message, so something in the mail system is on a go slow
> >probably.
> >
> >I think the recent spate of oopses is all coming out of one code change,
> >I have been running tests, but so far do not have a fix which works over
> >NFS and locally....
>
> The weird part of my oops is that the only thing I do is "cd" in to the
xfs
> fs and out of it and then unmount it. It should never oops if there has
> been NO activity on this fs.
>
> Maybe that is the problem.
> It works ok when:
> mounting it without options
> using it
> unmounting.
>
> It barfs when:
> mounting it with logbufs=8,logbsize=32768 (came out of the faq, shouldn't
> it be logsize ???)
> doing nothing or touching a file
> unmounting it oopses (umount gives a segmentation fault)
>
> As far I can see it has something to do with the mount options.
> Let me try it with other options.
>
OK, drop the logbufs to 4 and I bet the problem goes away - have you done a
complete rebuild recently? Also how much memory do you have?
I have not a complete rebuild. This tree is from march 27.
I normally do not use the extra mount options on this testmachine.
I dragged in a spare hard disk because I have a server moving landing on me
in a week which will also include a move to xfs. So I was practicing a dump
and restore including performance tests with respect to mount options.
The machine is a PIII 450 with 128MB ram and a 15GB IDE disk.
You can drop the logbsize option, 32K should be the default, it is the size
of the in memory log buffers, nothing to do with on disk log size.
Ah okay.
[root@lsautom /mnt]# mount -t xfs /dev/hdc1 hdc
[root@lsautom /mnt]# touch hdc/foo
[root@lsautom /mnt]# umount hdc
[root@lsautom /mnt]#
[root@lsautom /mnt]#
[root@lsautom /mnt]# mount -t xfs -o logbufs=4 /dev/hdc1 hdc
[root@lsautom /mnt]# touch hdc/foo2
[root@lsautom /mnt]# umount hdc
[root@lsautom /mnt]# mount -t xfs -o logbufs=8 /dev/hdc1 hdc
[root@lsautom /mnt]# touch hdc/foo3
Segmentation fault
[root@lsautom /mnt]# umount hdc
umount: /mnt/hdc: device is busy
[root@lsautom /mnt]#
Check. Is something running out of ram with logbufs=8 then?
Wild guess, I know. I have been wrong more then once.
Cheers
--
Seth
Has anybody seen my lightbulb?
I _really_ need some light here.
|