On Mon, Jan 6, 2020 at 7:35 PM Vivek Goyal <vgoyal@redhat.com> wrote:
On Mon, Jan 06, 2020 at 05:27:05PM +0000, Ernst, Eric wrote:
[CC linux-unionfs@vger.kernel.org and amir]
Hi Miklos,
One of the popular use cases for Kata Containers is running docker-in-docker. That is, a container image is run which in turn will make use of a container runtime to do a container build.
When combined with virtio-fs, we end up with a configuration like: xfs/ext4 -> overlayfs -> virtio-fs -> overlayfs
As discussed in [1], per overlayfs spec: "The upper filesystem will normally be writable and if it is it must support the creation of trusted.* extended attributes, and must provide valid d_type in readdir responses, so NFS is not suitable."
I don't know exaactly the reasons why NFS is not supported as upper. Are above only two reasons? These might work with virtio-fs depending on underlying filesystem. If yes, should we check for these properties at mount time (instead of relying on dentry flags only, ovl_dentry_remote()).
I feel there is more to it.
NFS also has these automount points, that we currently can't cope with in overlayfs. And there's revalidation, which we reject on upper simply because overlayfs currently doesn't call ->revalidate() on upper. Not that we would not be able to, it's just something that probably needs some thought. Virtio-fs does not yet have the magic automount thing (which would be useful to resolve inode number conflicts), but it does have revalidate. In the virtio-fs case, not calling ->revalidate() should not be a problem, so it's safe to try out this configuration by adding a hack to disable the remote check in case of a virtio-fs upper. Thanks, Miklos