Re: [PATCH v5 2/4] fuse: Support fuse filesystems outside of init_user_ns

[Seth Forshee]

On Wed, Nov 12, 2014 at 10:22:54AM -0600, Seth Forshee wrote:
> On Wed, Nov 12, 2014 at 02:09:15PM +0100, Miklos Szeredi wrote:
> > On Tue, Nov 11, 2014 at 09:37:10AM -0600, Eric W. Biederman wrote:
> > 
> > > > Maybe I'm being dense, but can someone give a concrete example of such an
> > > > attack?
> > > 
> > > There are two variants of things at play here.
> > > 
> > > There is the classic if you don't freeze your context at open time when
> > > you pass that file descriptor to another process unexpected things can
> > > happen.  
> > > 
> > > An essentially harmless but extremely confusing example is what happens
> > > to a partial read when it stops halfway through a uid value and the next
> > > read on the same file descriptor is from a process in a different user
> > > namespace.  Which uid value should be returned to userspace.
> > 
> > Fuse device doesn't currently do partial reads, so that's a non-issue.
> > 
> > > Now if I am in a nefarious mood I can create a unprivileged user
> > > namespace, open /dev/fuse and mount a fuse filesystem.  Pass the file
> > > descriptor to /dev/fuse to a processes that is in the default user
> > > namespace (and thus can use any uid/gid).   With that file desctipor
> > > report that there is a setuid 0 exectuable on that file system.
> > 
> > Yes, and this would also be prevented by MNT_NOSUID, which would be a good idea
> > anyway.  I just don't see the reason we'd want to allow clearing MNT_NOSUID in a
> > private namespace.
> > 
> > So we don't currently see a use case for relaxing either the MNT_NOSUID
> > restriction or for relaxing the requirement on the user namespace the fuse
> > server is in.  Is that correct?
> > 
> > If so, we should leave both restrictions in place since that allows the greatest
> > flexibility in the future, is either of those needs to be relaxed.
> 
> I'm not aware of specific use cases for either at this point. However,
> Andy's patch [1] will limit suid to the set of namespaces where the user
> who mounted the filesystem already has privileges. Enforcing MNT_NOSUID
> will require enforcement in the vfs, and in that case we definitely need
> to decide whether the policy is to implicitly add the flag or fail the
> mount attempt if the flag is not present [2].

I asked around a bit, and it turns out there are use cases for nested
containers (i.e. a container within a container) where the rootfs for
the outer container mounts a filesystem containing the rootfs for the
inner container. If that mount is nosuid then suid utilities like ping
aren't going to work in the inner container.

So since there's a use case for suid in a userns mount and we have what
we belive are sufficient protections against using this as a vector to
get privileges outside the container, I'm planning to move ahead without
the MNT_NOSUID restriction. Any objections?

Thanks,
Seth

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/