- ZFS changes depend directly on Linux kernel supporting our new namespace
- Current namespace framework means namespace additions are invasive
clone(2)
flags parameter is nearly out of available bits for new namespaces.
- Design of namespace changes needs to be considered useful for non-ZFS consumers to be palatable to upstream
- Must have a good case for using one of the last available
clone(2)
flags
Instead of adding new namespace directly, introduce an extensible framework for namespaces. This would allow external modules to act as namespace providers.
The last clone(2)
flag could indicate an additional parameter to be read for namespace information. This solves an existing problem for upstream and is more likely to be accepted.
We would then only have to provide a separate external module for our own namespace, which is easier to maintain outside of the kernel.