Re: [PATCH] selftest: net: fix variable sized type not at the end of struct warnings

[Simon Horman <horms@kernel.org>]

On Sat, Nov 08, 2025 at 10:59:59PM +0530, Ankit Khushwaha wrote:
> On Tue, Oct 28, 2025 at 05:23:12PM +0000, Simon Horman wrote:
> > On Mon, Oct 27, 2025 at 10:38:56AM +0530, Ankit Khushwaha wrote:
> > > Some network selftests defined variable-sized types defined at the end of
> > > struct causing -Wgnu-variable-sized-type-not-at-end warning.
> > > 
> > > warning:
> > > timestamping.c:285:18: warning: field 'cm' with variable sized type 
> > > 'struct cmsghdr' not at the end of a struct or class is a GNU 
> > > extension [-Wgnu-variable-sized-type-not-at-end]
> > >   285 |                 struct cmsghdr cm;
> > >       |                                ^
> > > 
> > > ipsec.c:835:5: warning: field 'u' with variable sized type 'union 
> > > (unnamed union at ipsec.c:831:3)' not at the end of a struct or class 
> > > is a GNU extension [-Wgnu-variable-sized-type-not-at-end]
> > >   835 |                 } u;
> > >       |                   ^
> > > 
> > > This patch move these field at the end of struct to fix these warnings.
> > > 
> > > Signed-off-by: Ankit Khushwaha <ankitkhushwaha.linux@gmail.com>
> > 
> > Hi Ankit,
> > 
> > I don't believe this change is correct.
> > 
> > I think that the intention of the code is the char arrays (buf and control)
> > provide the buffer space for the variable-length trailing field
> > of the preceding structure. Where we basically have a header followed
> > by data. But your patch would place the before the header.
> >
> Hi Simon,
> So if buf and control providing the buffer space, then i think it is
> better to suppress `-Wgnu-variable-sized-type-not-at-end` warning 
> within this block of code.
> 
> 	#pragma GCC diagnostic push
> 	#pragma GCC diagnostic ignored "-Wgnu-variable-sized-type-not-at-end"

I'm unsure of the attitude towards using #pragma like this in kernel code,
but certainly it would be a new one for me.

> 
> 	struct {
> 		union {
> 			struct xfrm_algo        alg;
> 			struct xfrm_algo_aead   aead;
> 			struct xfrm_algo_auth   auth;
> 		} u;
> 		char buf[XFRM_ALGO_KEY_BUF_SIZE];
> 	} alg = {};
> 
> 	#pragma GCC diagnostic pop
> 
> I think this would be fine.

In my view, the most promising approach I am aware of is using
TRAILING_OVERLAP(), as illustrated in [1]. However, that
approach was recently rejected, so I guess that means
it doesn't have much promise after all.

[1] https://lore.kernel.org/netdev/aPdx4iPK4-KIhjFq@kspp/