cgroup v1 take into account mount root

[glipner]

In a machine with Cgroups V1, by default Docker doesn't enter a cgroup namespace (--cgroupns=host).
Inside the container, as subtree of the host cgroup hierarchy is mounted:

# cat /proc/1/mountinfo
712 709 0:34 /docker/<guid> /sys/fs/cgroup/cpu,cpuacct ro,nosuid,nodev,noexec,relatime master:16 - cgroup cgroup rw,cpu,cpuacct

Meaning that accessing /sys/fs/cgroup/<controller>/something inside the container is the same as accessing /sys/fs/cgroup/<controller>/docker/<guid>/something from the host.

Reading from /proc/pid/cgroup yields a path in the host mount:
3:cpu,cpuacct:/docker/<guid>.
To access this cgroup from within the container, we need to strtip the /docker/<guid> part.