On Tue, Feb 08, 2022 at 11:30:03PM +0900, Toshiaki Makita wrote:
> On 2022/02/08 2:56, Pablo Neira Ayuso wrote:
> > On Thu, Feb 03, 2022 at 08:59:39PM +0900, Toshiaki Makita wrote:
[...]
> > > diff --git a/net/netfilter/nf_flow_table_ip.c b/net/netfilter/nf_flow_table_ip.c
> > > index 889cf88..48e2f58 100644
> > > --- a/net/netfilter/nf_flow_table_ip.c
> > > +++ b/net/netfilter/nf_flow_table_ip.c
[...]
> > > @@ -202,15 +209,25 @@ static int nf_flow_tuple_ip(struct sk_buff *skb, const struct net_device *dev,
> > > if (!pskb_may_pull(skb, thoff + *hdrsize))
> > > return -1;
> > > + if (ipproto == IPPROTO_GRE) {
> >
> > No ifdef here? Maybe remove these ifdef everywhere?
>
> I wanted to avoid adding many ifdefs and I expect this to be compiled out
> when CONFIG_NF_CT_PROTO_GRE=n as this block is unreachable anyway. It rather
> may have been unintuitive though.
>
> Removing all of these ifdefs will cause inconsistent behavior between
> CONFIG_NF_CT_PROTO_GRE=n/y.
> When CONFIG_NF_CT_PROTO_GRE=n, conntrack cannot determine GRE version, thus
> it will track GREv1 without key infomation, and the flow will be offloaded.
> When CONFIG_NF_CT_PROTO_GRE=y, GREv1 will have key information and will not
> be offloaded.
> I wanted to just refuse offloading of GRE to avoid this inconsistency.
> Anyway this kind of inconsistency seems to happen in software conntrack, so
> if you'd like to remove ifdefs, I will do.
Good point, thanks for explaining. LGTM.
[...]
> > > diff --git a/net/netfilter/nft_flow_offload.c b/net/netfilter/nft_flow_offload.c
> > > index 0af34ad..731b5d8 100644
> > > --- a/net/netfilter/nft_flow_offload.c
> > > +++ b/net/netfilter/nft_flow_offload.c
> > > @@ -298,6 +298,19 @@ static void nft_flow_offload_eval(const struct nft_expr *expr,
> > > break;
> > > case IPPROTO_UDP:
> > > break;
> > > +#ifdef CONFIG_NF_CT_PROTO_GRE
> > > + case IPPROTO_GRE: {
> > > + struct nf_conntrack_tuple *tuple;
> > > +
> > > + if (ct->status & IPS_NAT_MASK)
> > > + goto out;
> >
> > Why this NAT check?
>
> NAT requires more work. I'd like to start with a minimal GRE support.
> Maybe we can add NAT support later.
Oh well, yes, no problem.
