Bernd Ernesti wrote:

> On Fri Apr 28 19:21:03 2000, Manuel Bouyer wrote:
> > 
> > On Fri, Apr 28, 2000 at 11:46:45AM +0200, Soren S. Jorvang wrote:
> > > > We should *really* find the real bug for this !
> > > > This doesn't help with my alpha where the busy vnodes are on the root
> > > > filesystem.
> > > 
> > > The hangs at shutdown I am currently seeing (I think it used to
> > > be something different) are caused by cache_purgevfs() spinning
> > > over and over again.
> > 
> > Do you have softdep ?
> 
> Where I don't use softdep. The problem appeared before softdep went into
> the tree.
> 
> > The problem I have (and I think Bernd's one is the same) is that the
> > kernel fails to unmount the filesystems, but no hangs.
> 
> Correct. It says something about syncing the filesystem and prints
> 10-20 numbers and then gives up.

I've seen this off-and-on since 1.4C.  It went away somewhere around
1.4H and reappeared somewhere around 1.4S.  With the debugging enabled
in vfs_subr.c to print out busy vnodes, it's always a vnode on a MFS
that's hung.  It seems to make sense that if the user-land mount_mfs has
already been killed then an MFS couldn't flush any buffers, but I don't
know why the problem comes and goes.  A couple of weeks ago I tried a
simple ignoring of vnodes that were of type VT_MFS but that didn't work
(I've forgotten exactly what happened).

Simon.