Re: [PATCH] sys_getppid oopses on debug kernel

[Dave Hansen <haveblue@us.ibm.com>]

On Tue, 2006-08-08 at 14:22 +0400, Kirill Korotaev wrote:
> sys_getppid() optimization can access a freed memory.
> On kernels with DEBUG_SLAB turned ON, this results in
> Oops.
...
> +#else
> +	/*
> +	 * ->real_parent could be released before dereference and
> +	 * we accessed freed kernel memory, which faults with debugging on.
> +	 * Keep it simple and stupid.
> +	 */
> +	read_lock(&tasklist_lock);
> +	pid = current->group_leader->real_parent->tgid;
> +	read_unlock(&tasklist_lock);
> +#endif
>  	return pid;
>  }

Accessing freed memory is a bug, always, not just *only* when slab
debugging is on, right?  Doesn't this mean we could get junk, or that
the reader could potentially run off a bad pointer?

It seems that this patch only papers over the problem in the case when
it is observed, but doesn't really even fix the normal case.

Could we use a seqlock to determine when real_parent is in flux, and
re-read real_parent until we get a consistent one?  We could use in in
lieu of the existing for() loop.

-- Dave