Re: [PATCH net-next] ip_gre: remove CRC flag from dev features in gre_gso_segment

[Xin Long <lucien.xin@gmail.com>]

On Wed, Nov 18, 2020 at 8:29 AM Jakub Kicinski <kuba@kernel.org> wrote:
>
> On Mon, 16 Nov 2020 17:15:47 +0800 Xin Long wrote:
> > This patch is to let it always do CRC checksum in sctp_gso_segment()
> > by removing CRC flag from the dev features in gre_gso_segment() for
> > SCTP over GRE, just as it does in Commit 527beb8ef9c0 ("udp: support
> > sctp over udp in skb_udp_tunnel_segment") for SCTP over UDP.
> >
> > It could set csum/csum_start in GSO CB properly in sctp_gso_segment()
> > after that commit, so it would do checksum with gso_make_checksum()
> > in gre_gso_segment(), and Commit 622e32b7d4a6 ("net: gre: recompute
> > gre csum for sctp over gre tunnels") can be reverted now.
> >
> > Signed-off-by: Xin Long <lucien.xin@gmail.com>
>
> Makes sense, but does GRE tunnels don't always have a csum.
Do you mean the GRE csum can be offloaded? If so, it seems for GRE tunnel
we need the similar one as:

commit 4bcb877d257c87298aedead1ffeaba0d5df1991d
Author: Tom Herbert <therbert@google.com>
Date:   Tue Nov 4 09:06:52 2014 -0800

    udp: Offload outer UDP tunnel csum if available

I will confirm and implement it in another patch.

>
> Is the current hardware not capable of generating CRC csums over
> encapsulated patches at all?
There is, but very rare. The thing is after doing CRC csum, the outer
GRE/UDP checksum will have to be recomputed, as it's NOT zero after
all fields for CRC scum are summed, which is different from the
common checksum. So if it's a GRE/UDP tunnel, the inner CRC csum
has to be done there even if the HW supports its offload.

>
> I guess UDP tunnels can be configured without the csums as well
> so the situation isn't much different.
>
> > diff --git a/net/ipv4/gre_offload.c b/net/ipv4/gre_offload.c
> > index e0a2465..a5935d4 100644
> > --- a/net/ipv4/gre_offload.c
> > +++ b/net/ipv4/gre_offload.c
> > @@ -15,12 +15,12 @@ static struct sk_buff *gre_gso_segment(struct sk_buff *skb,
> >                                      netdev_features_t features)
> >  {
> >       int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb);
> > -     bool need_csum, need_recompute_csum, gso_partial;
> >       struct sk_buff *segs = ERR_PTR(-EINVAL);
> >       u16 mac_offset = skb->mac_header;
> >       __be16 protocol = skb->protocol;
> >       u16 mac_len = skb->mac_len;
> >       int gre_offset, outer_hlen;
> > +     bool need_csum, gso_partial;
>
> Nit, rev xmas tree looks broken now.
Will fix it in v2, :D

Thanks.
>
> >       if (!skb->encapsulation)
> >               goto out;
> > @@ -41,10 +41,10 @@ static struct sk_buff *gre_gso_segment(struct sk_buff *skb,
> >       skb->protocol = skb->inner_protocol;
> >
> >       need_csum = !!(skb_shinfo(skb)->gso_type & SKB_GSO_GRE_CSUM);
> > -     need_recompute_csum = skb->csum_not_inet;
> >       skb->encap_hdr_csum = need_csum;
> >
> >       features &= skb->dev->hw_enc_features;
> > +     features &= ~NETIF_F_SCTP_CRC;
> >
> >       /* segment inner packet. */
> >       segs = skb_mac_gso_segment(skb, features);
> > @@ -99,15 +99,7 @@ static struct sk_buff *gre_gso_segment(struct sk_buff *skb,
> >               }
> >
> >               *(pcsum + 1) = 0;
> > -             if (need_recompute_csum && !skb_is_gso(skb)) {
> > -                     __wsum csum;
> > -
> > -                     csum = skb_checksum(skb, gre_offset,
> > -                                         skb->len - gre_offset, 0);
> > -                     *pcsum = csum_fold(csum);
> > -             } else {
> > -                     *pcsum = gso_make_checksum(skb, 0);
> > -             }
> > +             *pcsum = gso_make_checksum(skb, 0);
> >       } while ((skb = skb->next));
> >  out:
> >       return segs;
>