[Virtio-fs] [PATCH] virtiofsd: Prevent multiply running with same vhost_user_socket

[Masayoshi Mizuma]

Hi Stefan,

On Fri, Aug 02, 2019 at 04:39:30PM +0100, Stefan Hajnoczi wrote:
> On Thu, Aug 01, 2019 at 11:13:48AM -0400, Masayoshi Mizuma wrote:
> > From: Masayoshi Mizuma <m.mizuma at jp.fujitsu.com>
> > 
> > virtiofsd can run multiply even if the vhost_user_socket is
> > same path.
> > 
> >   ]# ./virtiofsd -o vhost_user_socket=/tmp/vhostqemu -o source=/tmp/share &
> >   [1] 244965
> >   virtio_session_mount: Waiting for vhost-user socket connection...
> >   ]# ./virtiofsd -o vhost_user_socket=/tmp/vhostqemu -o source=/tmp/share &
> >   [2] 244966
> >   virtio_session_mount: Waiting for vhost-user socket connection...
> >   ]#
> > 
> > The file access from the guest works because the second virtiofsd
> > handles that, however, the user will get confused the situation
> > and may be the cause of unexpected problem. So it's better to
> > prevent the multiple running.
> > 
> > Change unlink() of the socket to the exit of virtiofsd so that the
> > latter virtiofsd can fail on bind() as EADDRINUSE.
> > 
> > Signed-off-by: Masayoshi Mizuma <m.mizuma at jp.fujitsu.com>
> > ---
> >  contrib/virtiofsd/fuse_lowlevel.c | 1 +
> >  contrib/virtiofsd/fuse_virtio.c   | 1 -
> >  2 files changed, 1 insertion(+), 1 deletion(-)
> 
> This patch makes the UNIX domain socket act similar to a pidfile or
> lockfile.  Does the user need to clean up the UNIX domain socket
> manually before virtiofsd can restart again after a crash?
> 
> This can be a robustness/reliability problem because we must assume that
> crashes happen and recovery should not require human intervention.
> 
> On the other hand, there are plenty of daemons that have long-lived UNIX
> domain socket paths.  How do they handle crash recovery?

It's good point, thanks.
I think if the socket ramains due to the former crash, then the socket
should be unlinked by virtiofsd, not manualy.

virtiofsd should detect whether the socket is in use by the other
virtiofsd or just remains by crash.
I'll try to implement that...

Thanks!
Masa