Re: xfs oops (CVS-2004-05-15_05:00_UTC)

[Christoph Hellwig <hch@xxxxxxxxxxxxx>]

On Sun, Jun 27, 2004 at 04:41:09AM +0200, Krzysztof Rusocki wrote:
> > > Hmm, it looks like the pagebuf already got freed by pagebuf_iostrategy
> > > when we hit the next line.
> 
> Indeed, that seems to be it - in 2 hours I've got 26 cases where
> pb->pb_target was altered by pagebuf_iostrategy():
> 
> pagebuf_daemon: pb->pb_target changed! 0xc7ec7e34 -> 0x6b6b6b6b
> (yes that's my trivial check)
> 
> What _possibly_ may be found interesting - original pb_target value is
> constant through all those messages, I'm not familiar with kernel
> or xfs internals enough to say whether it's something usual or not...
> (are pagebufs allocated via slab allocator?)

pagebufs have their own slabm yes, and 0x6b6b6b6b is slab poisoning, aka
the pagebuf has been freed.  As for the previous value beeing the same
I guess you only have very few (maybe just one?) XFS filesystem, and 
pb_targets points to a per-blockdevice structure.

I looked through the code and the target going away (aka the filesystem
beeing unmounted) is also possible in theory.