In cgroup v1, the parent of a cgroup mount is on tmpfs
(or maybe any other fs but definitely not cgroupfs).
Use this to traverse up the tree until we reach non-cgroupfs.
This should work in any setups (nested container, non-standard
cgroup mounts) so issue [1] won't happen.
Theoretically, the only problematic case is when someone mounts
cpuset cgroupfs into a directory on another cgroupfs. Let's
assume people don't do that -- if they do, they will get other
error (e.g. inability to read cpuset.cpus file).
[1] https://github.com/opencontainers/runc/issues/1367
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
Kir Kolyshkin
c85cd2b325