Re: [PATCH] treewide: Replace zero-length arrays with flexible-array member

From: Kees Cook
Date: Tue Feb 11 2020 - 15:12:55 EST


On Tue, Feb 11, 2020 at 01:54:22PM -0600, Gustavo A. R. Silva wrote:
>
>
> On 2/11/20 13:38, Greg KH wrote:
> > On Tue, Feb 11, 2020 at 11:32:04AM -0800, Kees Cook wrote:
> >> On Tue, Feb 11, 2020 at 01:20:36PM -0600, Gustavo A. R. Silva wrote:
> >>>
> >>>
> >>> On 2/11/20 12:32, Greg KH wrote:
> >>>> On Tue, Feb 11, 2020 at 11:41:26AM -0600, Gustavo A. R. Silva wrote:
> >>>>> The current codebase makes use of the zero-length array language
> >>>>> extension to the C90 standard, but the preferred mechanism to declare
> >>>>> variable-length types such as these ones is a flexible array member[1][2],
> >>>>> introduced in C99:
> >>>>>
> >>>>> struct foo {
> >>>>> int stuff;
> >>>>> struct boo array[];
> >>>>> };
> >>>>>
> >>>>> By making use of the mechanism above, we will get a compiler warning
> >>>>> in case the flexible array does not occur last in the structure, which
> >>>>> will help us prevent some kind of undefined behavior bugs from being
> >>>>> unadvertenly introduced[3] to the codebase from now on.
> >>>>>
> >>>>> All these instances of code were found with the help of the following
> >>>>> Coccinelle script:
> >>>>>
> >>>>> @@
> >>>>> identifier S, member, array;
> >>>>> type T1, T2;
> >>>>> @@
> >>>>>
> >>>>> struct S {
> >>>>> ...
> >>>>> T1 member;
> >>>>> T2 array[
> >>>>> - 0
> >>>>> ];
> >>>>> };
> >>>>>
> >>>>> [1] https://gcc.gnu.org/onlinedocs/gcc/Zero-Length.html
> >>>>> [2] https://github.com/KSPP/linux/issues/21
> >>>>> [3] commit 76497732932f ("cxgb3/l2t: Fix undefined behaviour")
> >>>>>
> >>>>> NOTE: I'll carry this in my -next tree for the v5.6 merge window.
> >>>>
> >>>> Why not carve this up into per-subsystem patches so that we can apply
> >>>> them to our 5.7-rc1 trees and then you submit the "remaining" that don't
> >>>> somehow get merged at that timeframe for 5.7-rc2?
> >>>>
> >>>
> >>> Yep, sounds good. I'll do that.
> >>
> >> FWIW, I'd just like to point out that since this is a mechanical change
> >> with no code generation differences (unlike the pre-C90 1-byte array
> >> conversions), it's a way better use of everyone's time to just splat
> >> this in all at once.
> >>
> >> That said, it looks like Gustavo is up for it, but I'd like us to
> >> generally consider these kinds of mechanical changes as being easier to
> >> manage in a single patch. (Though getting Acks tends to be a bit
> >> harder...)
> >
> > Hey, if this is such a mechanical patch, let's get it to Linus now,
> > what's preventing that from being merged now?

Now would be a good time, yes. (Linus has wanted Acks for such things
sometimes, but those were more "risky" changes...)

> Well, the only thing is that this has never been in linux-next.

Hmm. Was it in one of your 0day-tested trees?

--
Kees Cook