Do not create cgroup dir name from combining subsystems

On some systems, when we mount some cgroup subsystems into
a same mountpoint, the name sequence of mount options and
cgroup directory name can not be the same.

For example, the mount option is cpuacct,cpu, but
mountpoint name is /sys/fs/cgroup/cpu,cpuacct. In current
runc, we set mount destination name from combining
subsystems, which comes from mount option from
/proc/self/mountinfo, so in my case the name would be
/sys/fs/cgroup/cpuacct,cpu, which is differernt from
host, and will break some applications.

Fix it by using directory name from host mountpoint.

Signed-off-by: Qiang Huang <[email protected]>

Author: hqhq

libcontainer/rootfs_linux.go

@@ -346,7 +346,7 @@ func getCgroupMounts(m *configs.Mount) ([]*configs.Mount, error) {
 		binds = append(binds, &configs.Mount{
 			Device:           "bind",
 			Source:           filepath.Join(mm.Mountpoint, relDir),
-			Destination:      filepath.Join(m.Destination, strings.Join(mm.Subsystems, ",")),
+			Destination:      filepath.Join(m.Destination, filepath.Base(mm.Mountpoint)),
 			Flags:            syscall.MS_BIND | syscall.MS_REC | m.Flags,
 			PropagationFlags: m.PropagationFlags,
 		})