In case you're curious I encountered this problem most frequently (and
reproduced it most easily) by using xfs_db interactively whilst performing a
lot of I/O (mostly writes) on the same partition.
It was much harder to reproduce the problem when running a one-off command like
'xfs_db -r /dev/sda1 -c frag' or similar. However I did find that I could
cause it eventually if I ran that type of command repeatedly in a loop.
I found xfs_db to be a useful tool so hopefully this will get fixed soon! :-)
From: Jan Derfinak [mailto:ja@xxxxxxxxxxxx]
Sent: Tue 2/10/2004 5:58 PM
To: Christoph Hellwig
Cc: Smith, Andy P.; linux-xfs@xxxxxxxxxxx
Subject: Re: XFS_DB Problem (Corruption of in-memory data detected.)
On Tue, 10 Feb 2004, Christoph Hellwig wrote:
> > never had problem. I use 2.6 kernel now. What is wrong on read-only
> > access?
> The block device and XFS's metadata use the same pages but they don't
> talk to each other enough, so you can get into cases where a page used
> by XFS is flushed and the blockdevice code scribbles over the page which
> will get you inmemory corruption.
I suggest to notice this in xfs_db man page. From description of '-r' option
isn't clear that it is dangerous.