> On Nov 12, 2024, at 8:56 AM, Gleb Smirnoff <gleb...@freebsd.org> wrote:
>
> On Mon, Nov 04, 2024 at 02:56:33PM +0000, Mark Johnston wrote:
> M> diff --git a/sys/sys/socketvar.h b/sys/sys/socketvar.h
> M> index 40464f555816..fda8d23f5649 100644
> M> --- a/sys/sys/socketvar.h
> M> +++ b/sys/sys/socketvar.h
> M> @@ -615,7 +615,8 @@ struct xsocket {
> M> uint32_t so_qlimit;
> M> pid_t so_pgid;
> M> uid_t so_uid;
> M> - int32_t so_spare32[8];
> M> + int32_t so_fibnum;
> M> + int32_t so_spare32[7];
> M> int16_t so_type;
> M> int16_t so_options;
> M> int16_t so_linger;
>
> I would suggest not to spend the spares in the main branch.
> So this commit should add the field, keeping 8 spares. But
> the merge to stable/14 should use a spare.
>
> Otherwise we eventually will run out of spares in main
> and also nearly run out of spares in some stable/22 branch.
>
> I'd suggest to follow the policy of not spending spares
> in the main branch for all structures, not just xsocket.
IIUC, those spares are reserved for **future** new features while keep API /
ABI compatibility.
If we lack spares, or in a foresee future we will lack then we should reserve
large quantity of spares.
If we happen to run out of them, then no choices but we have to break the API /
ABI compatibility, then this should
be done in CURRENT.
Right now there are still 7 spares not used, and I guess it is not urgent to
add extra spares.
Please correct me if wrong.
Best regards,
Zhenlei
>
> --
> Gleb Smirnoff
