When rootless runc exits, failed to remove cgroupv2 with sub-cgroup

[chenk008]

In case of rootless, cgroup2 mount into container /sys/fs/cgroup. After I created sub-cgroup test, and then exited. A error message shown

ERRO[0029] remove /sys/fs/cgroup/user.slice/user-1000.slice/[email protected]/user.slice/runc-foo.scope: device or resource busy

Reproduce:

runc spec --rootless
// add cgroupfs mount into config.json
{
      "destination": "/sys/fs/cgroup",
      "type": "cgroup",
      "source": "cgroup",
      "options": [
        "nosuid",
        "noexec",
        "nodev",
        "relatime"
      ]
}
runc --systemd-cgroup run foo
// in the container
mkdir /sys/fs/cgroup/test
exit

systemd version: 239 (239-41.el8_3)
kernel version: 4.18.0-305.12.1.el8_4.x86_64
linux distro: centos 8

[chenk008]

I think it relates to

runc/libcontainer/cgroups/systemd/v2.go

Line 305 in e999e29

err := os.Remove(m.path)

[kolyshkin]

Ah, so you're using cgroup v2 on CentOS 8. We do not test this configuration, this is why we do not catch this bug.