[PATCH v2 1/3] ethdev: add ICMPv6 ID and sequence

[Leo Xu (Networking SW)]

> 02/02/2023 19:33, Leo Xu (Networking SW):
> > > 31/01/2023 07:53, Leo Xu (Networking SW):
> > > > From: Thomas Monjalon <thomas at monjalon.net>
> > > > > 20/12/2022 08:44, Leo Xu:
> > > > > > +/**
> > > > > > + * ICMP6 header
> > > > > > + */
> > > > > > +struct rte_icmp6_hdr {
> > > > > > +     uint8_t type;
> > > > > > +     uint8_t code;
> > > > > > +     rte_be16_t checksum;
> > > > > > +} __rte_packed;
> > > > > > +
> > > > > > +/**
> > > > > > + * ICMP6 echo
> > > > > > + */
> > > > > > +struct rte_icmp6_echo {
> > > > > > +     struct rte_icmp6_hdr hdr;
> > > > > > +     rte_be16_t identifier;
> > > > > > +     rte_be16_t sequence;
> > > > > > +} __rte_packed;
> > > > >
> > > > > It is exactly the same as struct rte_icmp_hdr.
> > > > > Why not reuse it?
> > > > > Maybe introduce struct rte_icmp_base_hdr and define
> > > > > rte_icmp_echo_hdr as rte_icmp_hdr?
> > > >
> > > > Hi Thomas,
> > > > Looks like, using rte_icmp_hdr as base header for both icmp and
> > > > icmpv6 is
> > > not that good.
> > > > since, rte_icmp_hdr default their headers always having id and
> > > > sequence
> > > fields, which is not applicable for most other icmp6/icmp types packets.
> > > >
> > > > I may suggest to keep icmp and icmp6 structures independent
> > > > against each
> > > other, because, looks like these two protocols definitions do not
> > > share common base.
> > >
> > > What about introducing rte_icmp_base_hdr?
> > > We should try to avoid duplicating things.
> > >
> >
> > You mean introduce rte_icmp_base_hdr like following?
> > struct rte_icmp_base_hdr {
> >       uint8_t  icmp_type;
> >       uint8_t  icmp_code;
> >       rte_be16_t icmp_cksum;
> > } __rte_packed;
> >
> > And change the existing rte_icmp_hdr to be:
> > struct rte_icmp_hdr {
> >       rte_icmp_base_hdr bash_hdr;
> >       rte_be16_t icmp_ident;
> >       rte_be16_t icmp_seq_nb;
> > } __rte_packed;
> > #define rte_icmp6_echo struct rte_icmp_hdr;
> >
> > If it is, there will be some compatibilities issues, since we changed existing
> structure.
> >
> > Or, maybe I'm missing something.
> > Would you help to give more details about above comment?
> 
> Currently we have this:
> struct rte_icmp_hdr {
>     uint8_t  icmp_type;     /* ICMP packet type. */
>     uint8_t  icmp_code;     /* ICMP packet code. */
>     rte_be16_t icmp_cksum;  /* ICMP packet checksum. */
>     rte_be16_t icmp_ident;  /* ICMP packet identifier. */
>     rte_be16_t icmp_seq_nb; /* ICMP packet sequence number. */ }
> __rte_packed;
> 
> I agree we can move some fields in a base struct, it would change the API.
> We could manage with a union, but we would lose the benefit.
> It looks like we need to keep rte_icmp_hdr as is.
> So we need to duplicate and define new structs.
> 
> What about removing the "6" from the new structs, so it would apply both to
> IPv4 and IPv6?
> 
> struct rte_icmp_base_hdr {
>         uint8_t type;
>         uint8_t code;
>         rte_be16_t checksum;
> } __rte_packed;
> 
> struct rte_icmp_echo_hdr {
>         struct rte_icmp_base_hdr base;
>         rte_be16_t identifier;
>         rte_be16_t sequence;
> } __rte_packed;
> 
> 

I agree with that proposal.
Then, we can deem existing struct rte_icmp_hdr as old one, which should not be used in new app.
And looks like, these new defined structures can cover all ICMP4/6 formats.
Good idea!
I will update accordingly, in next patch.