On Fri, Jan 13, 2017 at 9:46 PM, James Bottomley <James.Bottomley@xxxxxxxxxxxxxxxxxxxxx> wrote: > I'd like to attend LSF/MM mostly to participate in the storage topics > discussion, but also because containers is starting to need some > features from the VFS. I am also interested in intersection of containers with file systems, but I haven't seen a topic proposal for it, so adding TOPIC tag to Jame's email. Specifically, many containers use overlayfs which makes the role of the traditional file system an infrastructure layer, much like block and mm. How the backing file system interacts with overlayfs may have significant effect on the end users. The most prominent example is how clone support in backing fs removes a big pain for large files copy up. Another example of rather simple backing file system support that can improve overlayfs containers user performance: - API to set d_type for of dentries to optimize overlay dir iteration [1] I guess James had more use cases in mind, so maybe others do as well. [1] http://www.spinics.net/lists/linux-unionfs/msg01401.html > Hopefully the container uid shifting is mostly > sorted out by the superblock user namespace, and I should have patches > to demonstrate this in shiftfs this month. However, trusted containers > are also becoming an issue and, since containers are file based, not > image based, we really need to sort out the IMA interfaces to be > correctly integrated into the VFS, so I'd like to be present in those > discussions as well. > > James >
