On 03/21, Mandeep Singh Baines wrote:

Oleg Nesterov (oleg-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org) wrote:

quoted

+/*
+ * rcu-safe, but should start at ->group_leader.
+ * thread_group_leader(g) protects against the race with exec which
+ * removes the leader from list.
+ * smp_rmb() pairs with implicit mb() implied by unlock + lock in
+ * de_thread()->release_task() path.
+ */
+#define while_each_thread_rcu(g, t)				\
+	while ((t = next_thread(t)) != g &&			\
+		({ smp_rmb(); thread_group_leader(g); }))
+

Couldn't you miss the exec_thread if:

t = exec_thread && !thread_group_leader(g)

Yes, we already discussed this, iirc.

I was going to write that this is fine, but then I changed my mind.
Indeed, it is not good while_each_thread_rcu() can miss the new leader.

Could we change do_prlimit()? Especially since its slow path.

But do_prlimit() is correct. It sees the unhashed task under tasklist,
task->group_leader should be correct.

I really like you're earlier solution (ignoring barrier):

#define while_each_thread(g, t) \
	while (t->group_leader == g->group_leader && (t = next_thread(t)) != g)

The problem is "ignoring barrier and races".

OK, I'll try to think again.

Oleg.