[Ocfs2-users] o2fsck_mark_cluster_allocated: Internal logicfaliure!! duplicate cluster

[Mark Fasheh]

On Fri, Aug 17, 2007 at 02:52:10PM +0100, paul fretter (TOC) wrote:
> I think things could be very broken on our shared disk.

Can you describe your hardware setup? What you're using for shared storage,
etc.


> I finally managed to get fsck to run without –F, but I am still seeing loads of
> the aforementioned error messages during pass2, and each time I run fsck the
> same errors for the same clusters/files are repeated.

Can you post a full log of your fsck output? If it's too big for e-mail, you
can attach it to a bugzilla bug. The same goes for any of the other info I
ask you for later in this e-mail.


> Some stuff has been recovered to lost+found, but there are a load of broken
> file/dir entries that I cannot do anything with, as the error messages from
> fsck.
> 
> If I ‘ls –al’  one of the broken files it seems all the meta-information has
> gone.
> 
> e.g.
> 
> ls –al
> 
> ?---------               ?              ?              ?              ?
> hmmer.hmm

Yeah - as you note, it seems that the directory entries are pretty fried.

Does debugfs.ocfs2 give you anything? Could you run it against the shared
disk and use the 'ls', 'cd' and 'stat' commands against some of the bad
directories / inodes? If we get some choice bits from there, it might tell
us more about what's going on.

Also, for completeness sake would you mind posting the contents of all your
nodes cluster.conf files?


> To be a bit more specific, when I try to run fcsk.ocfs2 like this I get an
> error
> 
> #fsck.ocfs2 -yf /dev/sde
> 
> fsck.ocfs2: Trylock failed while locking down the cluster
> 
>  
> 
> So in desperation I am running it with –F as well.
> 
>  
> 
> There is no-one else using the cluster.

Did you ever have the software confirm that. You can check if other nodes
are mounted on a file system by running:

	mounted.ocfs2 -f /dev/sde


To be honest - it sounds like that's exactly what might've happened. Of
course, now that fsck.ocfs2 is running without -F, we might never know for
sure :/
	--Mark

--
Mark Fasheh
Senior Software Developer, Oracle
mark.fasheh at oracle.com