2006-05-16 02:12:57

by john stultz

Subject: [-rt potential bug?] Bad error path in futex_wake

Hey Ingo,
I've been trying to help Dinakar chase down a futex bug where
lookup_pi_state oopses due to this->pi_state being NULL
(kernel/futex.c:480).

No clue on why yet, but looking over the code, I noticed the following
odd error path:

static int futex_wake(u32 __user *uaddr, int nr_wake)
{
[snip]
down_read(&current->mm->mmap_sem);

ret = get_futex_key(uaddr, &key);
if (unlikely(ret != 0))
goto out;

hb = hash_futex(&key);
spin_lock(&hb->lock);
head = &hb->chain;

list_for_each_entry_safe(this, next, head, list) {
if (match_futex (&this->key, &key)) {
if (this->pi_state)
!!!!------> return -EINVAL;
wake_futex(this);
if (++ret >= nr_wake)
break;
}
}

spin_unlock(&hb->lock);
out:
up_read(&current->mm->mmap_sem);
return ret;
}

I'm not very familiar w/ the futex code, so this might be the right
thing, but it sure looks wrong.

Thoughts?

thanks
-john