[OpenAFS] strange error in directory / volume

[Derrick J Brashear]

On Mon, 2 Dec 2002, Dave McMurtrie wrote:

> FWIW, this exact same thing happened to me.  I can't provide any of the 
> above info because I didn't spend much time on this since I'm (so far) the 
> only person at our entire university to ever complain about this.  I think 
> this happens when you run 2 gconf aware apps simultaneously from 2 
> different machines.

And most of your users probably only log into one machine at a time.

> An obvious problem here is that gconfd relies on 2 things that AFS doesn't 
> fully support:  file locking and hardlinks.
> 
> If you throw together a quick C or perl program that calls opendir() &
> readdir() you'll see that the directory entry for "ior" exists but if you
> try to stat() it, you'll get ENOENT.  I have no clue what could cause this 
> to happen.

the linux dentry caching bug would cause this to happen, but i wouldn't
expect to see it with modern clients (1.2.6 and 1.2.7 at least count as
"modern"). basically the kernel thinks a file is still there that afs
knows is not. but...

> I fixed my broken directory by running bos salvage on my volume and I
> stopped running gnome on 2 different machines simultaneously when they
> both look at the same homedir.

this is the other instance it can happen in. someone else posted salvage
logs, i assume same deal.