On Fri, 18 Feb 2022, Pablo Neira Ayuso wrote: > On Fri, Feb 18, 2022 at 12:55:07AM +0100, Pablo Neira Ayuso wrote: > > On Thu, Feb 17, 2022 at 08:27:08PM -0300, Marcelo Ricardo Leitner wrote: > > > On Thu, Feb 17, 2022 at 02:55:27PM +0100, Pablo Neira Ayuso wrote: > > > > On Thu, Feb 17, 2022 at 11:34:24AM +0200, Paul Blakey wrote: > > > > > After cited commit optimizted hw insertion, flow table entries are > > > > > populated with ifindex information which was intended to only be used > > > > > for HW offload. This tuple ifindex is hashed in the flow table key, so > > > > > it must be filled for lookup to be successful. But tuple ifindex is only > > > > > relevant for the netfilter flowtables (nft), so it's not filled in > > > > > act_ct flow table lookup, resulting in lookup failure, and no SW > > > > > offload and no offload teardown for TCP connection FIN/RST packets. > > > > > > > > > > To fix this, allow flow tables that don't hash the ifindex. > > > > > Netfilter flow tables will keep using ifindex for a more specific > > > > > offload, while act_ct will not. > > > > > > > > Using iif == zero should be enough to specify not set? > > > > > > You mean, when searching, if search input iif == zero, to simply not > > > check it? That seems dangerous somehow. > > > > dev_new_index() does not allocate ifindex as zero. > > > > Anyway, @Paul: could you add a tc_ifidx field instead in the union > > right after __hash instead to fix 9795ded7f924? > > I mean this incomplete patch below: > > diff --git a/include/net/netfilter/nf_flow_table.h b/include/net/netfilter/nf_flow_table.h > index a3647fadf1cc..d4fa4f716f68 100644 > --- a/include/net/netfilter/nf_flow_table.h > +++ b/include/net/netfilter/nf_flow_table.h > @@ -142,6 +142,7 @@ struct flow_offload_tuple { > u8 h_source[ETH_ALEN]; > u8 h_dest[ETH_ALEN]; > } out; > + u32 tc_ifidx; > }; > }; > > You will need to update nf_flow_rule_match() to set key->meta.ingress_ifindex to > use tc_ifidx if it is set to non-zero value. > I understand how it could fix the original issue, but I don't think this is better, because it makes tuple less generic. What you suggested with using 0 to avoid needing the new flag is good enough for me, and is cleaner in my opinion. I'll send the == 0 one as V2 for chance you agree, and if you want to change to this, I won't mind sending it as V3.