Re: df stuck in tstile

[steve%prd.co.uk@localhost (Steve Blinkhorn)]

I think that is the clue I needed.   I suspected, on the basis of your
advice, that it's an nfs-mounted file system that's the problem, so
I've unmounted them one by one (there are four), and can't unmount the
fourth..   Does that ring true?

--
Steve Blinkhorn <steve%prd.co.uk@localhost>

You wrote:
> 
> On Wed, Jan 30, 2019 at 04:14:49PM +0000, Steve Blinkhorn wrote:
> > I think I'm probably out of my depth on this one.  I used to get this
> > kind of problem on the machines these replace, and became convinced
> > that that was a bad block issue (which got aired on one lof the other
> > lists - they were x86 macnines).  Once there was one process stuck in
> > tstile, the problem began to accumulate, and the only solution was a
> > periodic reboot.  IIRC the problem started with a df.
> 
> This usually means that some driver is still trying to write a file system
> buffer back to disk (or NFS server or whatever), or there is a bug in a
> drivers error path and e.g. you hit some error condition and the driver
> did not handle that correctly, so now some vnode or file system lock is
> still blocked and everything that waits for it to become free gets blocked
> in tstile.
> 
> The backtraces of various blocking kernel threads would give us a hint
> which driver to look at. Can you drop into the debugger and force a
> crash dump + reboot? Then we could start looking at the kernel crash
> and extract better info from it.
> 
> Martin
>