Re: BUG_ON(nd->inode != parent->d_inode);

[Linus Torvalds]

On Fri, Mar 8, 2013 at 3:07 PM, Dave Jones <davej@xxxxxxxxxx> wrote:
>
> Ok, got something more meaningful out of the lookup_slow trace.
>
> [   66.082984] parent->dname.name  (06b6b6b6b6b6b6b)
> [   66.083637] parent =
>
> At first I thought AH-HA! SLAB POISON!
> But look closer.. it's shifted by 8 bits.

Or just the high byte has been cleared.

But yeah, if the parent has been free'd then that certainly explains
why the "impossible" test of

  nd->inode != parent->d_inode

would trigger. And it would explain any odd crashes at lookup time
too. In particular, the NULL pointer one you reference seems to be
dir->i_op->lookup being NULL, so calling it (understandable) ends up
doing bad things.

I really don't understand how the parent could be free'd early.
Dentries are freed by RCU, and the dentry lookup code is some of the
most well-tested out there. I don't see how /proc could mess that up,
unless it just completely screws up some refcounting thing or other.

                 Linus
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/