Re: [REGRESSION] ptrace broken from "cgroup: cgroup v2 freezer" (76f969e)

[Roman Gushchin <guro@fb.com>]

On Mon, May 13, 2019 at 06:38:14PM +0200, Oleg Nesterov wrote:
> On 05/13, Oleg Nesterov wrote:
> >
> > Probably we add leave_frozen(true) after freezable_schedule() for now, then
> > think try to make something better...
> 
> And again, this is what I thought ptrace_stop() does, somehow I didn't notice
> that the last version doesn't have leave_frozen() in ptrace_stop().
> 
> Perhaps we can do a bit better, change only tracehook_report_syscall_entry() and
> PTRACE_EVENT_EXIT/SECCOMP paths to do leave_frozen() ?
> 
> At first glance other callers look fine in that they can do nothing "interesting"
> befor get_signal(), but we need to re-check...

Hi Oleg!

Thank you for looking into it!

I've just check the following patch (see below). It solves the regression
and overall seems correct. But I need some more time to check.

Thanks!

--

diff --git a/kernel/signal.c b/kernel/signal.c
index 8607b11ff936..088b377ad439 100644
--- a/kernel/signal.c
+++ b/kernel/signal.c
@@ -2112,6 +2112,8 @@ static void ptrace_stop(int exit_code, int why, int clear_code, kernel_siginfo_t
                preempt_enable_no_resched();
                cgroup_enter_frozen();
                freezable_schedule();
+               if (info)
+                       cgroup_leave_frozen(true);
        } else {
                /*
                 * By the time we got the lock, our tracer went away.