lkml.org 
[lkml]   [2014]   [Oct]   [23]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: introduce task_rcu_dereference?
On 10/23, Oleg Nesterov wrote:
>
> Damn.

Yes.

> On 10/22, Oleg Nesterov wrote:
> >
> > +struct task_struct *task_rcu_dereference(struct task_struct **ptask)
> > +{
> > + struct task_struct *task;
> > + struct sighand_struct *sighand;
> > +
> > + task = rcu_dereference(*ptask);
> > + if (!task)
> > + return NULL;
> > +
> > + /* If it fails the check below must fail too */
> > + probe_slab_address(&task->sighand, sighand);
> > + /*
> > + * Pairs with atomic_dec_and_test() in put_task_struct(task).
> > + * If we have read the freed/reused memory, we must see that
> > + * the pointer was updated. The caller might want to retry in
> > + * this case.
> > + */
> > + smp_rmb();
> > + if (unlikely(task != ACCESS_ONCE(*ptask)))
> > + return ERR_PTR(-EAGAIN);
>
> This is not exactly right. task == *ptask can be false positive.
>
> It can be freed, then resused (so that sighand != NULL can be false
> positive), then freed again, and then reused again as task_struct.
>
> This is not that bad, we still can safely use this task_struct, but
> the comment should be updated. Plus -EINVAL below can be wrong in
> this case although this minor.

Yes.

> Yeees, SLAB_DESTTROY_BY_RCU closes this race. Not sure why I'd like
> to avoid it, but I do ;)

Argh. I only meant that SLAB_DESTTROY_BY_RCU can make the comments
simpler. "closes this race" applies too "check below must fail too"
too. Sorry if I confused you.

"task == *ptask can be false positive" is true with or without
SLAB_DESTTROY_BY_RCU, and this needs a good comment. Yes, it can't
be reused twice, but still we can't 100% trust the "sighand != NULL"
check.

So let me repeat, SDBR can only turn probe_slab_address() into a plain
load.

But I can't think properly today, will try to recheck tomorrow and send
v2.

Oleg.



\
 
 \ /
  Last update: 2014-10-23 21:01    [W:0.073 / U:12.604 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site