Re: bug in memcg oom-killer results in a hung syscall in another process in the same cgroup

[Konstantin Khlebnikov]

On 12.07.2016 18:35, Shayan Pooya wrote:
> > > With strace, when running 500 concurrent mem-hog tasks on the same
> > > kernel, 33 of them failed with:
> > >
> > > strace: ../sysdeps/nptl/fork.c:136: __libc_fork: Assertion
> > > `THREAD_GETMEM (self, tid) != ppid' failed.
> > >
> > > Which is: https://sourceware.org/bugzilla/show_bug.cgi?id=15392
> > > And discussed before at: https://lkml.org/lkml/2015/2/6/470 but that
> > > patch was not accepted.
> >
> > OK, so the problem is that the oom killed task doesn't report the futex
> > release properly? If yes then I fail to see how that is memcg specific.
> > Could you try to clarify what you consider a bug again, please? I am not
> > really sure I understand this report.
>
> It looks like it is just a very easy way to reproduce the problem that
> Konstantin described in that lkml thread. That patch was not accepted
> and I see no other fixes for that issue upstream. Here is a copy of
> his root-cause analysis from said thread:
>
> Whole sequence looks like: task calls fork, glibc calls syscall clone with
> CLONE_CHILD_SETTID and passes pointer to TLS THREAD_SELF->tid as argument.
> Child task gets read-only copy of VM including TLS. Child calls put_user()
> to handle CLONE_CHILD_SETTID from schedule_tail(). put_user() trigger page
> fault and it fails because do_wp_page()  hits memcg limit without invoking
> OOM-killer because this is page-fault from kernel-space.  Put_user returns
> -EFAULT, which is ignored.  Child returns into user-space and catches here
> assert (THREAD_GETMEM (self, tid) != ppid), glibc tries to print something
> but hangs on deadlock on internal locks. Halt and catch fire.

Yep. Bug still not fixed in upstream. In our kernel I've plugged it with this:

--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -2808,8 +2808,9 @@ asmlinkage __visible void schedule_tail(struct task_struct *prev)
        balance_callback(rq);
        preempt_enable();

-       if (current->set_child_tid)
-               put_user(task_pid_vnr(current), current->set_child_tid);
+       if (current->set_child_tid &&
+           put_user(task_pid_vnr(current), current->set_child_tid))
+               force_sig(SIGSEGV, current);
 }

Add Oleg into CC. IIRR he had some ideas how to fix this. =)

--
Konstantin