[OpenAFS-devel] afs_dentry_iput patch (linux)

Neulinger, Nathan R. nneul@umr.edu
Wed, 24 Jan 2001 15:50:33 -0600


In my case, refCount is usually around 65535 or so. That could possibly
indicate a underflow/overflow of 16 bit integer condition?

-- Nathan

> -----Original Message-----
> From: Chas Williams [mailto:chas@cmf.nrl.navy.mil]
> Sent: Wednesday, January 24, 2001 2:07 PM
> To: Neulinger, Nathan R.
> Cc: openafs-devel@openafs.org
> Subject: Re: [OpenAFS-devel] afs_dentry_iput patch (linux) 
> 
> 
> >Ok. While I admit I like the fact that you are deleting code 
> that generates
> >an error message on some of my clients (web and other 
> servers) - I wonder if
> >that is a good approach or if it will just hide my problem. :)
> 
> well... it will just move the error actually.  i am looking 
> at this same
> bug.  i only see it on SMP systems.  i believe xxx_iput() needs to 
> atleast a spin_lock.  looking at osi_iput (a 2.4 specific version)
> it seems to me that osi_iput could be called again on the 
> same inode between
> the entry and afs_delete_inode.
> 
> i basically see 'Bad refCount 0...' which makes me thing that
> afs_delete_inode() doesnt get finished before iput() is called again.
> at one point, i put lock_kernel()/unlock_kernel() afs_delete_inode
> but i suspect it needs to be a higher up.
> 
> not gospel, just some thoughts.
>