Re: [PATCH 1/1] mm, oom_adj: don't loop through tasks in __set_oom_adj when not necessary

Oleg Nesterov [off-list ref] writes:

On 08/20, Oleg Nesterov wrote:

quoted

On 08/20, Eric W. Biederman wrote:

quoted

--- a/fs/exec.c
+++ b/fs/exec.c

@@ -1139,6 +1139,10 @@ static int exec_mmap(struct mm_struct *mm)
 	vmacache_flush(tsk);
 	task_unlock(tsk);
 	if (old_mm) {
+		mm->oom_score_adj = old_mm->oom_score_adj;
+		mm->oom_score_adj_min = old_mm->oom_score_adj_min;
+		if (tsk->vfork_done)
+			mm->oom_score_adj = tsk->vfork_oom_score_adj;

too late, ->vfork_done is NULL after mm_release().

And this can race with __set_oom_adj(). Yes, the current code is racy too,
but this change adds another race, __set_oom_adj() could already observe
->mm != NULL and update mm->oom_score_adj.

  ^^^^^^^^^^^^

I meant ->mm == new_mm.

And another problem. Suppose we have

	if (!vfork()) {
		change_oom_score();
		exec();
	}

the parent can be killed before the child execs, in this case vfork_oom_score_adj
will be lost.

Yes.

Looking at include/uapi/linux/oom.h it appears that there are a lot of
oom_score_adj values that are reserved.  So it should be completely
possible to initialize vfork_oom_score_adj to -32768 aka SHRT_MIN, and
use that as a flag to see if it is active or not.

Likewise for vfork_oom_score_adj_min if we need to duplicate that one as
well.


That deals with that entire class of race.  We still have races during
exec about vfork_done being cleared before the new ->mm == new_mm.
While that is worth fixing is an independent issue.  

Eric