Re: [PATCH] mm: oom: Fix race condition between oom_badness and do_exit of task

[Tetsuo Handa <penguin-kernel@xxxxxxxxxxxxxxxxxxx>]

On 2018/03/08 13:51, Kohli, Gaurav wrote:
> On 3/8/2018 2:26 AM, David Rientjes wrote:
> 
>> On Wed, 7 Mar 2018, Gaurav Kohli wrote:
>>
>>> diff --git a/mm/oom_kill.c b/mm/oom_kill.c
>>> index 6fd9773..5f4cc4b 100644
>>> --- a/mm/oom_kill.c
>>> +++ b/mm/oom_kill.c
>>> @@ -114,9 +114,11 @@ struct task_struct *find_lock_task_mm(struct task_struct *p)
>>>         for_each_thread(p, t) {
>>>           task_lock(t);
>>> +        get_task_struct(t);
>>>           if (likely(t->mm))
>>>               goto found;
>>>           task_unlock(t);
>>> +        put_task_struct(t);
>>>       }
>>>       t = NULL;
>>>   found:
>> We hold rcu_read_lock() here, so perhaps only do get_task_struct() before
>> doing rcu_read_unlock() and we have a non-NULL t?
> 
> Here rcu_read_lock will not help, as our task may change due to below algo:
> 
> for_each_thread(p, t) {
>          task_lock(t);
> +        get_task_struct(t);
>          if (likely(t->mm))
>              goto found;
>          task_unlock(t);
> +        put_task_struct(t)
> 
> 
> So only we can increase usage counter here only at the current task.

static int proc_single_show(struct seq_file *m, void *v)
{
	struct inode *inode = m->private;
	struct pid_namespace *ns;
	struct pid *pid;
	struct task_struct *task;
	int ret;

	ns = inode->i_sb->s_fs_info;
	pid = proc_pid(inode);
	task = get_pid_task(pid, PIDTYPE_PID); /* get_task_struct() is called upon success. */
	if (!task)
		return -ESRCH;

	ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);

	put_task_struct(task);
	return ret;
}

static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
			  struct pid *pid, struct task_struct *task)
{
	unsigned long totalpages = totalram_pages + total_swap_pages;
	unsigned long points = 0;

	points = oom_badness(task, NULL, NULL, totalpages) *
			     1000 / totalpages; /* task->usage > 0 due to proc_single_show() */
	seq_printf(m, "%lu\n", points);

	return 0;
}

struct task_struct *find_lock_task_mm(struct task_struct *p) /* p->usage > 0 */
{
	struct task_struct *t;

	rcu_read_lock();

	for_each_thread(p, t) {
		task_lock(t);
		if (likely(t->mm))
			goto found;
		task_unlock(t);
	}
	t = NULL;
found:
	rcu_read_unlock();

	return t; /* t->usage > 0 even if t != p because t->mm != NULL */
}

t->alloc_lock is still held when leaving find_lock_task_mm(), which means
that t->mm != NULL. But nothing prevents t from setting t->mm = NULL at
exit_mm() from do_exit() and calling exit_creds() from __put_task_struct(t)
after task_unlock(t) is called. Seems difficult to trigger race window. Maybe
something has preempted because oom_badness() becomes outside of RCU grace
period upon leaving find_lock_task_mm() when called from proc_oom_score().