On Tue, 10 Oct 2017 10:52:01 +0800 Wei Yang <richard.weiy...@gmail.com> wrote:
> When parent is NULL, get_slot_offset() returns almost the address of slot.
> This is an invalid value for offset.
>
> One possible scenario happens on deleting #0 index, when it is the only one
> in tree.
>
> Current behavior doesn't harm the system, because the offset will not be
> used when parent is NULL in the following procedure or parent is checked
> before get_slot_offset() called. While it is still not safe to return an
> invalid offset.
>
> This patch returns 0 when parent is NULL in get_slot_offset().
>
I'm confused. If parent=NULL, get_slot_offset() will crash the kernel.
So why "Current behavior doesn't harm the system"?
> --- a/lib/radix-tree.c
> +++ b/lib/radix-tree.c
> @@ -119,7 +119,7 @@ bool is_sibling_entry(const struct radix_tree_node
> *parent, void *node)
> static inline unsigned long
> get_slot_offset(const struct radix_tree_node *parent, void __rcu **slot)
> {
> - return slot - parent->slots;
> + return parent ? (slot - parent->slots):0;
> }
>
> static unsigned int radix_tree_descend(const struct radix_tree_node *parent,
