From: Helg <helg-openbsd@gmx.de>
Subject: Re: fuse: change termination behaviour
To: tech@openbsd.org
Date: Wed, 17 Sep 2025 08:18:55 +0200

On Mon, Sep 08, 2025 at 05:40:17PM +0200, Helg wrote:
> On Fri, Sep 12, 2025 at 02:04:58PM +0200, Claudio Jeker wrote:
> > On Wed, Sep 10, 2025 at 02:18:03AM +0200, Helg wrote:
> > > This patch addresses incompatibilities in the way that FUSE handles
> > > terminating a FUSE session.
> > > 
> >  
> > Diff looks generally OK. If this is expected behaviour we should probably
> > follow it.
> > 
> > My question is what happens when you kill the fuse userland process but
> > end up with the FS still mounted. Is there a chance that this will lockup
> > the machine (similar to unreachable NFS servers)?
> 
> Once the fuse device is closed, any operation on the old mount point
> will result in ENXIO - Device not configured.
> 
> > 
> > I assume that the idea is that you can restart the FUSE file system daemon
> > without remounting the file system (and keeping dirty buffers accross such
> > a restart). Is this correct?
> 
> I don't think that's the intended purpose and I don't know of any fuse
> file systems that attempts this. One advantage of this approach is that
> if the file system daemon crashes, any process that is attempting to
> read or write on that mount will fail with ENXIO. With automatic unmounting,
> programs don't see that the file system has been unmounted.
> 

Ping... not sure if that was an OK

Thanks,
Helg