Re: [Patch net-next] ipv6: fix a potential NULL deref

[Eric Dumazet <eric.dumazet@gmail.com>]

On Mon, 2012-10-29 at 15:25 +0800, Cong Wang wrote:

> Oh, yeah, gcc should be smart enough to do calculation without deref it
> given it has the offset and the address. And dst happens to be first
> field of rt, so offset is 0, &rt->dst should be NULL too if rt is NULL.
> 

There is no dereference, even if gcc was dumb, since dst is an embedded
struct, not a pointer to a struct.

This wont change in a near future.

> But this will be a problem if someone moved dst inside rt, as there is
> no comment saying dst has to be the first one?

I dont think this placement will change in a near future, it would break
lot of things.

I guess we can use a BUILD_BUG_ON() instead of a comment that
could be ignored.

Note we could move dst in rtable, if we change the NULL test in
dst_release with some if ((unsigned long)dst < 4000) condition

I'll send a patch against ip_rt_put()