Re: Cgroups: Cannot move certain kernel threads from root cgroup

[Serge Hallyn <serge.hallyn-Z7WLFzj8eWMS+FvcfC7Uqw@public.gmane.org>]

Quoting Avinash Chiganmi (avinash.ca-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org):
> Hi,
> 
> I created a "cpu" cgroup:
>      mkdir /dev/cgroup
>      mkdir /dev/cgroup/cpu
>      mount -t cgroup -o cpu cpu /dev/cgroup/cpu
>      cd /dev/cgroup/cpu/
>      mkdir low
> 
> Now, I try to move tasks to the sub-cgroup "low":
>      for task in `cat /dev/cgroup/cpu/tasks`; do   /bin/echo $task >
> /dev/cgroup/cpu/low/tasks || echo "Failed to move PID $task"; done
> 
> /bin/echo: write error: Invalid argument
> Failed to move PID 3
> /bin/echo: write error: Invalid argument
> Failed to move PID 4
> /bin/echo: write error: Invalid argument
> Failed to move PID 5
> /bin/echo: write error: Invalid argument
> Failed to move PID 6
> /bin/echo: write error: Invalid argument
> Failed to move PID 7
> /bin/echo: write error: Invalid argument
> Failed to move PID 8
> /bin/echo: write error: Invalid argument
> .
> .
> .
> /bin/echo: write error: Invalid argument
> Failed to move PID 58
> 
> I was able to move the "init" process itself. However, I failed to move
> specific kernel threads:
> 
> root         3     2  0 18:16 ?        00:00:00 [migration/0]
> root         4     2  0 18:16 ?        00:00:00 [sirq-high/0]
> root         5     2  0 18:16 ?        00:00:04 [sirq-timer/0]
> root         6     2  0 18:16 ?        00:00:00 [sirq-net-tx/0]
> root         7     2  0 18:16 ?        00:00:00 [sirq-net-rx/0]
> root         8     2  0 18:16 ?        00:00:00 [sirq-block/0]
> root         9     2  0 18:16 ?        00:00:00 [sirq-block-iopo]
> root        10     2  0 18:16 ?        00:00:00 [sirq-tasklet/0]
> root        11     2  0 18:16 ?        00:00:00 [sirq-sched/0]
> root        12     2  0 18:16 ?        00:00:00 [sirq-hrtimer/0]
> root        13     2  0 18:16 ?        00:00:04 [sirq-rcu/0]
> root        14     2  0 18:16 ?        00:00:00 [watchdog/0]
> root        15     2  0 18:16 ?        00:00:00 [desched/0]
> root        16     2  0 18:16 ?        00:00:00 [migration/1]
> root        17     2  0 18:16 ?        00:00:00 [sirq-high/1]
> root        18     2  0 18:16 ?        00:00:03 [sirq-timer/1]
> root        19     2  0 18:16 ?        00:00:00 [sirq-net-tx/1]
> root        20     2  0 18:16 ?        00:00:00 [sirq-net-rx/1]
> root        21     2  0 18:16 ?        00:00:00 [sirq-block/1]
> 
> 
> Is there something special about these threads that prevents them from
> being moved to a different cgroup? Is there something I am missing?
> 
> Thanks!
> Avi.

I'm guessing:

kernel/cgroup.c:attach_task_by_pid():

        /*
         * Workqueue threads may acquire PF_THREAD_BOUND and become
         * trapped in a cpuset, or RT worker may be born in a cgroup
         * with no rt_runtime allocated.  Just say no.
         */
        if (tsk == kthreadd_task || (tsk->flags & PF_THREAD_BOUND)) {
                ret = -EINVAL;
                rcu_read_unlock();
                goto out_unlock_cgroup;
        }