Merge branch 'sfc-conntrack-offload'

Edward Cree says:

====================
sfc: support conntrack NAT offload

The EF100 MAE supports performing NAT (and NPT) on packets which match in
 the conntrack table.  This series adds that capability to the driver.
====================

Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

4 files changed, 101 insertions, 3 deletions

diff --git a/drivers/net/ethernet/sfc/mae.c b/drivers/net/ethernet/sfc/mae.c
index 021980a958b7..10709d828a63 100644
--- a/drivers/net/ethernet/sfc/mae.c
+++ b/drivers/net/ethernet/sfc/mae.c
@@ -1291,10 +1291,11 @@ int efx_mae_alloc_action_set(struct efx_nic *efx, struct efx_tc_action_set *act)
 	size_t outlen;
 	int rc;
 
-	MCDI_POPULATE_DWORD_4(inbuf, MAE_ACTION_SET_ALLOC_IN_FLAGS,
+	MCDI_POPULATE_DWORD_5(inbuf, MAE_ACTION_SET_ALLOC_IN_FLAGS,
 			      MAE_ACTION_SET_ALLOC_IN_VLAN_PUSH, act->vlan_push,
 			      MAE_ACTION_SET_ALLOC_IN_VLAN_POP, act->vlan_pop,
 			      MAE_ACTION_SET_ALLOC_IN_DECAP, act->decap,
+			      MAE_ACTION_SET_ALLOC_IN_DO_NAT, act->do_nat,
 			      MAE_ACTION_SET_ALLOC_IN_DO_DECR_IP_TTL,
 			      act->do_ttl_dec);
 
diff --git a/drivers/net/ethernet/sfc/tc.c b/drivers/net/ethernet/sfc/tc.c
index 3d76b7598631..6db3d7ed3a86 100644
--- a/drivers/net/ethernet/sfc/tc.c
+++ b/drivers/net/ethernet/sfc/tc.c
@@ -2457,6 +2457,14 @@ static int efx_tc_flower_replace(struct efx_nic *efx,
 			NL_SET_ERR_MSG_MOD(extack, "Cannot offload tunnel decap action without tunnel device");
 			rc = -EOPNOTSUPP;
 			goto release;
+		case FLOW_ACTION_CT:
+			if (fa->ct.action != TCA_CT_ACT_NAT) {
+				rc = -EOPNOTSUPP;
+				NL_SET_ERR_MSG_FMT_MOD(extack, "Can only offload CT 'nat' action in RHS rules, not %d", fa->ct.action);
+				goto release;
+			}
+			act->do_nat = 1;
+			break;
 		default:
 			NL_SET_ERR_MSG_FMT_MOD(extack, "Unhandled action %u",
 					       fa->id);
diff --git a/drivers/net/ethernet/sfc/tc.h b/drivers/net/ethernet/sfc/tc.h
index 86e38ea7988c..7b5190078bee 100644
--- a/drivers/net/ethernet/sfc/tc.h
+++ b/drivers/net/ethernet/sfc/tc.h
@@ -48,6 +48,7 @@ struct efx_tc_encap_action; /* see tc_encap_actions.h */
  * @vlan_push: the number of vlan headers to push
  * @vlan_pop: the number of vlan headers to pop
  * @decap: used to indicate a tunnel header decapsulation should take place
+ * @do_nat: perform NAT/NPT with values returned by conntrack match
  * @do_ttl_dec: used to indicate IP TTL / Hop Limit should be decremented
  * @deliver: used to indicate a deliver action should take place
  * @vlan_tci: tci fields for vlan push actions
@@ -68,6 +69,7 @@ struct efx_tc_action_set {
 	u16 vlan_push:2;
 	u16 vlan_pop:2;
 	u16 decap:1;
+	u16 do_nat:1;
 	u16 do_ttl_dec:1;
 	u16 deliver:1;
 	__be16 vlan_tci[2];
diff --git a/drivers/net/ethernet/sfc/tc_conntrack.c b/drivers/net/ethernet/sfc/tc_conntrack.c
index 44bb57670340..d90206f27161 100644
--- a/drivers/net/ethernet/sfc/tc_conntrack.c
+++ b/drivers/net/ethernet/sfc/tc_conntrack.c
@@ -276,10 +276,84 @@ static int efx_tc_ct_parse_match(struct efx_nic *efx, struct flow_rule *fr,
 	return 0;
 }
 
+/**
+ * struct efx_tc_ct_mangler_state - tracks which fields have been pedited
+ *
+ * @ipv4: IP source or destination addr has been set
+ * @tcpudp: TCP/UDP source or destination port has been set
+ */
+struct efx_tc_ct_mangler_state {
+	u8 ipv4:1;
+	u8 tcpudp:1;
+};
+
+static int efx_tc_ct_mangle(struct efx_nic *efx, struct efx_tc_ct_entry *conn,
+			    const struct flow_action_entry *fa,
+			    struct efx_tc_ct_mangler_state *mung)
+{
+	/* Is this the first mangle we've processed for this rule? */
+	bool first = !(mung->ipv4 || mung->tcpudp);
+	bool dnat = false;
+
+	switch (fa->mangle.htype) {
+	case FLOW_ACT_MANGLE_HDR_TYPE_IP4:
+		switch (fa->mangle.offset) {
+		case offsetof(struct iphdr, daddr):
+			dnat = true;
+			fallthrough;
+		case offsetof(struct iphdr, saddr):
+			if (fa->mangle.mask)
+				return -EOPNOTSUPP;
+			conn->nat_ip = htonl(fa->mangle.val);
+			mung->ipv4 = 1;
+			break;
+		default:
+			return -EOPNOTSUPP;
+		}
+		break;
+	case FLOW_ACT_MANGLE_HDR_TYPE_TCP:
+	case FLOW_ACT_MANGLE_HDR_TYPE_UDP:
+		/* Both struct tcphdr and struct udphdr start with
+		 *	__be16 source;
+		 *	__be16 dest;
+		 * so we can use the same code for both.
+		 */
+		switch (fa->mangle.offset) {
+		case offsetof(struct tcphdr, dest):
+			BUILD_BUG_ON(offsetof(struct tcphdr, dest) !=
+				     offsetof(struct udphdr, dest));
+			dnat = true;
+			fallthrough;
+		case offsetof(struct tcphdr, source):
+			BUILD_BUG_ON(offsetof(struct tcphdr, source) !=
+				     offsetof(struct udphdr, source));
+			if (~fa->mangle.mask != 0xffff)
+				return -EOPNOTSUPP;
+			conn->l4_natport = htons(fa->mangle.val);
+			mung->tcpudp = 1;
+			break;
+		default:
+			return -EOPNOTSUPP;
+		}
+		break;
+	default:
+		return -EOPNOTSUPP;
+	}
+	/* first mangle tells us whether this is SNAT or DNAT;
+	 * subsequent mangles must match that
+	 */
+	if (first)
+		conn->dnat = dnat;
+	else if (conn->dnat != dnat)
+		return -EOPNOTSUPP;
+	return 0;
+}
+
 static int efx_tc_ct_replace(struct efx_tc_ct_zone *ct_zone,
 			     struct flow_cls_offload *tc)
 {
 	struct flow_rule *fr = flow_cls_offload_flow_rule(tc);
+	struct efx_tc_ct_mangler_state mung = {};
 	struct efx_tc_ct_entry *conn, *old;
 	struct efx_nic *efx = ct_zone->efx;
 	const struct flow_action_entry *fa;
@@ -326,6 +400,17 @@ static int efx_tc_ct_replace(struct efx_tc_ct_zone *ct_zone,
 				goto release;
 			}
 			break;
+		case FLOW_ACTION_MANGLE:
+			if (conn->eth_proto != htons(ETH_P_IP)) {
+				netif_dbg(efx, drv, efx->net_dev,
+					  "NAT only supported for IPv4\n");
+				rc = -EOPNOTSUPP;
+				goto release;
+			}
+			rc = efx_tc_ct_mangle(efx, conn, fa, &mung);
+			if (rc)
+				goto release;
+			break;
 		default:
 			netif_dbg(efx, drv, efx->net_dev,
 				  "Unhandled action %u for conntrack\n", fa->id);
@@ -335,8 +420,10 @@ static int efx_tc_ct_replace(struct efx_tc_ct_zone *ct_zone,
 	}
 
 	/* fill in defaults for unmangled values */
-	conn->nat_ip = conn->dnat ? conn->dst_ip : conn->src_ip;
-	conn->l4_natport = conn->dnat ? conn->l4_dport : conn->l4_sport;
+	if (!mung.ipv4)
+		conn->nat_ip = conn->dnat ? conn->dst_ip : conn->src_ip;
+	if (!mung.tcpudp)
+		conn->l4_natport = conn->dnat ? conn->l4_dport : conn->l4_sport;
 
 	cnt = efx_tc_flower_allocate_counter(efx, EFX_TC_COUNTER_TYPE_CT);
 	if (IS_ERR(cnt)) {