On Mon, Mar 5, 2018 at 1:27 AM, Masahiro Yamada
<yamada.masah...@socionext.com> wrote:
> Sorry for chiming in late.
>
> I noticed this thread today,
> honestly, the commit made me upset.
>
>
> Can I suggest another way to make it less fragile?
> __attribute((...)) can be placed after 'struct'.
>
>
> So, we can write:
>
>
> struct __randomize_layout path {
>         struct vfsmount *mnt;
>         struct dentry *dentry;
> };
>
>
>   instead of
>
>
> struct path {
>         struct vfsmount *mnt;
>         struct dentry *dentry;
> } __randomize_layout;

Ugh. I had tried this after the struct _name_, not after "struct"
itself. This does fix it, though it remains fragile, as you mention.

> If we force the former notation,
> the undefined __randomize_layout results in a build error
> instead of silent broken code generation.
>
>
> It is true somebody can still place
> __randomize_layout after the closing brace,
> but can we check this by coccicheck or checkpatch.pl?
> (we can describe it in coding style documentation, of course)
>
>
> IMHO, we should not (ab)use include/linux/kconfig.h
> to bring in misc things.

I'm happy to send a patch that reverts the other changes and relocates
all the markings...

Linus, how would you like this to go?

-Kees

-- 
Kees Cook
Pixel Security

Reply via email to