> > > I'm not saying fuse is worthless. It is a nice toy for single-user
> > > systems. But I do not think we should be merging "allow ordinary users
> > > to mount their own fuse's" before issues above are fixed.
> > 
> > I think multi user systems are not all that interesting.  And I
> > suspect very few of them want reliably working suspend/hibernate
> > (which they wouldn't get due to other issues anyway), or have weird
> > shutdown scripts which stop when they are unable to umount
> > filesystems.
> 
> Weird shutdown scripts? I believe all shutdown scripts have this issue
> -- if you want to [cleanly] unmount your / filesystem, you need all
> the opens for write closed, right...? Which self-deadlocked fused
> holding files open will prevent.
> 
> > For paranoid sysadmins, I suggest not enabling fuse for unprivileged
> > users, which is pretty easy to do: just don't set /dev/fuse to be
> > world read-writable (which is the default BTW).
> > 
> > So your reasons just don't warrant a big effort involving VFS hacking,
> > etc.  Patches are of course welcome.
> 
> Well, I believe code with obscure, but almost impossible to fix
> problems should not be merged...

That code _has_ been merged, something like 3 years ago, and is doing
fine, thank you.

The unprivileged mounts code, which we should be discussing, doesn't
change anything about that, except to not require another suid-root
utility.  Many distributions enabling unprivileged mounting by default
_now_, so it's not as if there's some great danger in doing this
slightly differently.

> Anyway, I believe it would be fair to mention kill -9 no longer
> working and shutdown/hibernation/multiuser problems it implies in the
> changelogs and probably sysctl documentation or how is this enabled.

Sure.

Miklos
_______________________________________________
Containers mailing list
Containers@lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers