Hello, Oleg.

On Tue, Nov 13, 2018 at 04:37:01PM +0100, Oleg Nesterov wrote:

On 11/12, Roman Gushchin wrote:

quoted

This patch implements freezer for cgroup v2. However the functionality
is similar, the interface is different to cgroup v1: it follows
cgroup v2 interface principles.

Oh, it seems that I actually need to apply this patch to (try to) understand
the details ;) Will try tomorrow.

Yeah, it's a bit of a head spin like everything in signal delivery /
ptrace paths.

...

OK, please forget for now, but perhaps it would be more clean to add
JOBCTL_TRAP_FREEZE to the JOBCTL_PENDING_MASK check in recalc_sigpending()
and change get_signal to check JOBCTL_TRAP_MASK | JOBCTL_TRAP_FREEZE; and
I am not even sure cgroup_freezer_enter() should live in do_jobctl_trap().

I'm sure you're aware of the context but just to refresh - one thing
which was really broken about cgroup1 freezer was that it piggybacked
on hibernation freezer and put frozen tasks in a state which is
undefined when seen from userspace - they're just stuck in D sleep
somewhere in the kernel.  That's fine when the whole system is not
gonna be running, but not when only a subportion is being frozen.

So, the primary goal of cgroup2 freezer is putting the tasks in an
equivalent state as jobctl stop.  It's a jobctl stop but controlled by
cgroup frozen state, meaning that they can be killed, PTRACE_SEIZE'd
and INTERRUPT'ed (PTRACE_ATTACH doesn't work as signal delivery should
be blocked but that's fine) and so on.

Thanks.

-- 
tejun