Comment 22 for bug 1403152

I am able to reliably recreate this on both Ubuntu 14.04 with kernel 3.13.0-37-generic and Ubuntu 14.10 with kernel 3.16.0-23-generic. I have test code that can hit the bug, but it's not cleaned up enough to share. The steps to repro are:

Start a container (COW clone using overlayfs from a base/parent Ubuntu 14.04 container). Start clone container, mount an NFS share (remote from machine hosting container) to /mnt in container, create a file on the NFS share in container (dd if=/dev/zero of=/mnt/foo bs=1024 count=100000), stop container, destroy container.

I have a script that performs the above sequence over and over again, running 5 such clone containers in parallel at any given time. It only takes about 30s for the kernel to get wedged with:

unregister_netdevice: waiting for lo to become free. Usage count = 1

This same script is fine if you do scp instead of writing over NFS. It seems like NFS inside the container is somehow involved here.