On Friday 20 September 2002 12:55, Tadas ?elionis wrote:
> i have just experienced kernel crash/panic when tried to 'sync' disks when
> md reconstruction was running. (no logs :/)
>
> system was debian woody 2.4.20-pre7, md raid1 autodetect, 2 scsi disks on
> aic7xxx, SMP, filesystem btw was ext3 - not xfs
>
> there goes the strange part... when booting next time md started
> reconstructing arrays as it should, but filesystems seemed clean...
>
> then there was a lot of input/output errors etc, i fsck'ed manually, but
> system turned out unusable, ALL fs's were corrupted. had to reinstall.
>
> i don't know why all the fs's were corrupted - now testing that system
> with 2.4.19.
>
>
> wonder if those 'Warning - running *really* short on DMA buffers' have
> anything todo with this.
Hi,
I don't know what happened with your system. The only time when I had data/fs
corruption with md was on old 2.4.x kernels. The kernel delivered with RedHat
7.1, including the XFS enabled one from the RH 7.1 SGI XFS installer, were
giving me horrible corruptions on several different IDE hardware. That said,
I'm not too much surprised one can face this although I have no idea what
went wrong in your case. Having data transfer errors on IDE DMA trasnsfers
can likely result in a heavy damaged md and therefore the filesystem has no
chance to survive. All this toghether with enabled write caching on the disk
can really mess things up.
Sorry, that's not much help but it can give you an idea what others faced
before :)
Simon
>
> > > I must add that the system is Slackware 8.0 based. The MD support is
> > > compiled in kernel (no modules support). The partitions are of type
> >
> > RAID
> >
> > > autodetect so linux can detect them at boot time. Also the / and /home
> > > partitions are NOT on the md0 divice. Only /var is.
> >
> > I do not really have any suggestions for you as to why the log appeared
> > clean, maybe you really were not mounted when the power was pulled.
> > As for the filesystem, all you can do is run xfs_check when you get a
> > chance.
> >
> > Steve
|