[PATCH nf 1/2] netfilter: flowtable: fix inline vlan encapsulation in xmit path

[PATCH nf 1/2] netfilter: flowtable: fix inline vlan encapsulation in xmit path

[Pablo Neira Ayuso]

Several issues in the inline vlan support:

- The layer 2 encapsulation representation in the tuple takes encap[0] as
  the outer header and encap[1] as the inner header as seen from the ingress
  path. Reverse the encap loop to push first the inner then the outer vlan
  header.

- Postpone pushing the layer 2 header once destination device is known.
  This allows to calculate the needed hearoom via LL_RESERVED_SPACE to
  accommodate the layer 2 headers.

- Add and use nf_flow_vlan_push() as suggested by Eric Woudstra, this
  is a simplified version of skb_vlan_push() for egress path only.

Fixes: c653d5a78f34 ("netfilter: flowtable: inline vlan encapsulation in xmit path")
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
---
Tested with ip netns enviroment both with single vlan and vlan-in-vlan, it
needs a bit of tidy up before submission for selftests.

I originally tested this submission with a script that I discover it was wrong:
netns configuration with vlan-in-vlan devices but that only used a single vlan
device.

 net/netfilter/nf_flow_table_ip.c | 91 ++++++++++++++++++++++----------
 1 file changed, 63 insertions(+), 28 deletions(-)

diff --git a/net/netfilter/nf_flow_table_ip.c b/net/netfilter/nf_flow_table_ip.c
index dbd7644fdbeb..1573189a23a1 100644
--- a/net/netfilter/nf_flow_table_ip.c
+++ b/net/netfilter/nf_flow_table_ip.c
@@ -462,23 +462,6 @@ static void nf_flow_encap_pop(struct nf_flowtable_ctx *ctx,
 		nf_flow_ip_tunnel_pop(ctx, skb);
 }
 
-struct nf_flow_xmit {
-	const void		*dest;
-	const void		*source;
-	struct net_device	*outdev;
-};
-
-static unsigned int nf_flow_queue_xmit(struct net *net, struct sk_buff *skb,
-				       struct nf_flow_xmit *xmit)
-{
-	skb->dev = xmit->outdev;
-	dev_hard_header(skb, skb->dev, ntohs(skb->protocol),
-			xmit->dest, xmit->source, skb->len);
-	dev_queue_xmit(skb);
-
-	return NF_STOLEN;
-}
-
 static struct flow_offload_tuple_rhash *
 nf_flow_offload_lookup(struct nf_flowtable_ctx *ctx,
 		       struct nf_flowtable *flow_table, struct sk_buff *skb)
@@ -544,6 +527,31 @@ static int nf_flow_offload_forward(struct nf_flowtable_ctx *ctx,
 	return 1;
 }
 
+/* Similar to skb_vlan_push. */
+static int nf_flow_vlan_push(struct sk_buff *skb, __be16 proto, u16 id,
+			     u32 needed_headroom)
+{
+	if (skb_vlan_tag_present(skb)) {
+		struct vlan_hdr *vhdr;
+
+		if (skb_cow_head(skb, needed_headroom + VLAN_HLEN))
+			return -1;
+
+		skb_push(skb, VLAN_HLEN);
+		if (skb_mac_header_was_set(skb))
+			skb->mac_header -= VLAN_HLEN;
+
+		vhdr = (struct vlan_hdr *)skb->data;
+		skb->network_header -= VLAN_HLEN;
+		vhdr->h_vlan_TCI = htons(skb_vlan_tag_get(skb));
+		vhdr->h_vlan_encapsulated_proto = skb->protocol;
+		skb->protocol = skb->vlan_proto;
+	}
+	__vlan_hwaccel_put_tag(skb, proto, id);
+
+	return 0;
+}
+
 static int nf_flow_pppoe_push(struct sk_buff *skb, u16 id)
 {
 	int data_len = skb->len + sizeof(__be16);
@@ -730,17 +738,19 @@ static int nf_flow_tunnel_v6_push(struct net *net, struct sk_buff *skb,
 }
 
 static int nf_flow_encap_push(struct sk_buff *skb,
-			      struct flow_offload_tuple *tuple)
+			      struct flow_offload_tuple *tuple,
+			      struct net_device *outdev)
 {
+	u32 needed_headroom = LL_RESERVED_SPACE(outdev);
 	int i;
 
-	for (i = 0; i < tuple->encap_num; i++) {
+	for (i = tuple->encap_num - 1; i >= 0; i--) {
 		switch (tuple->encap[i].proto) {
 		case htons(ETH_P_8021Q):
 		case htons(ETH_P_8021AD):
-			skb_reset_mac_header(skb);
-			if (skb_vlan_push(skb, tuple->encap[i].proto,
-					  tuple->encap[i].id) < 0)
+			if (nf_flow_vlan_push(skb, tuple->encap[i].proto,
+					      tuple->encap[i].id,
+					      needed_headroom) < 0)
 				return -1;
 			break;
 		case htons(ETH_P_PPP_SES):
@@ -753,6 +763,35 @@ static int nf_flow_encap_push(struct sk_buff *skb,
 	return 0;
 }
 
+struct nf_flow_xmit {
+	const void		*dest;
+	const void		*source;
+	struct net_device	*outdev;
+	struct flow_offload_tuple *tuple;
+};
+
+static void __nf_flow_queue_xmit(struct net *net, struct sk_buff *skb,
+				    struct nf_flow_xmit *xmit)
+{
+	skb->dev = xmit->outdev;
+	dev_hard_header(skb, skb->dev, ntohs(skb->protocol),
+			xmit->dest, xmit->source, skb->len);
+	dev_queue_xmit(skb);
+}
+
+static unsigned int nf_flow_queue_xmit(struct net *net, struct sk_buff *skb,
+				       struct nf_flow_xmit *xmit)
+{
+	if (xmit->tuple->encap_num) {
+		if (nf_flow_encap_push(skb, xmit->tuple, xmit->outdev) < 0)
+			return NF_DROP;
+	}
+
+	__nf_flow_queue_xmit(net, skb, xmit);
+
+	return NF_STOLEN;
+}
+
 unsigned int
 nf_flow_offload_ip_hook(void *priv, struct sk_buff *skb,
 			const struct nf_hook_state *state)
@@ -797,9 +836,6 @@ nf_flow_offload_ip_hook(void *priv, struct sk_buff *skb,
 	if (nf_flow_tunnel_v4_push(state->net, skb, other_tuple, &ip_daddr) < 0)
 		return NF_DROP;
 
-	if (nf_flow_encap_push(skb, other_tuple) < 0)
-		return NF_DROP;
-
 	switch (tuplehash->tuple.xmit_type) {
 	case FLOW_OFFLOAD_XMIT_NEIGH:
 		rt = dst_rtable(tuplehash->tuple.dst_cache);
@@ -829,6 +865,7 @@ nf_flow_offload_ip_hook(void *priv, struct sk_buff *skb,
 		WARN_ON_ONCE(1);
 		return NF_DROP;
 	}
+	xmit.tuple = other_tuple;
 
 	return nf_flow_queue_xmit(state->net, skb, &xmit);
 }
@@ -1119,9 +1156,6 @@ nf_flow_offload_ipv6_hook(void *priv, struct sk_buff *skb,
 				   &ip6_daddr, encap_limit) < 0)
 		return NF_DROP;
 
-	if (nf_flow_encap_push(skb, other_tuple) < 0)
-		return NF_DROP;
-
 	switch (tuplehash->tuple.xmit_type) {
 	case FLOW_OFFLOAD_XMIT_NEIGH:
 		rt = dst_rt6_info(tuplehash->tuple.dst_cache);
@@ -1151,6 +1185,7 @@ nf_flow_offload_ipv6_hook(void *priv, struct sk_buff *skb,
 		WARN_ON_ONCE(1);
 		return NF_DROP;
 	}
+	xmit.tuple = other_tuple;
 
 	return nf_flow_queue_xmit(state->net, skb, &xmit);
 }
-- 
2.47.3

[PATCH nf 2/2] netfilter: flowtable: fix inline pppoe encapsulation in xmit path

[Pablo Neira Ayuso]

Address two issues in the inline pppoe encapsulation:

- Add needs_gso_segment flag to segment PPPoE packets in software
  given that there is no GSO support for this.

- Use FLOW_OFFLOAD_XMIT_DIRECT since neighbour cache is not available
  in point-to-point device, use the hardware address that is obtained
  via flowtable path discovery (ie. fill_forward_path).

Fixes: 18d27bed0880 ("netfilter: flowtable: inline pppoe encapsulation in xmit path")
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
---
Tested locally with a pppoe server in my network and pppoe client in a netns inside VM,
it shows an improvement of 70% in bandwith with iperf3 when forwarding pppoe traffic
through the flowtable here compared to classic forwarding.

 include/net/netfilter/nf_flow_table.h |  4 +++-
 net/netfilter/nf_flow_table_core.c    |  1 +
 net/netfilter/nf_flow_table_ip.c      | 34 ++++++++++++++++++++++++---
 net/netfilter/nf_flow_table_path.c    |  7 +++++-
 4 files changed, 41 insertions(+), 5 deletions(-)

diff --git a/include/net/netfilter/nf_flow_table.h b/include/net/netfilter/nf_flow_table.h
index b09c11c048d5..7b23b245a5a8 100644
--- a/include/net/netfilter/nf_flow_table.h
+++ b/include/net/netfilter/nf_flow_table.h
@@ -148,9 +148,10 @@ struct flow_offload_tuple {
 	/* All members above are keys for lookups, see flow_offload_hash(). */
 	struct { }			__hash;
 
-	u8				dir:2,
+	u16				dir:2,
 					xmit_type:3,
 					encap_num:2,
+					needs_gso_segment:1,
 					tun_num:2,
 					in_vlan_ingress:2;
 	u16				mtu;
@@ -232,6 +233,7 @@ struct nf_flow_route {
 			u32			hw_ifindex;
 			u8			h_source[ETH_ALEN];
 			u8			h_dest[ETH_ALEN];
+			u8			needs_gso_segment:1;
 		} out;
 		enum flow_offload_xmit_type	xmit_type;
 	} tuple[FLOW_OFFLOAD_DIR_MAX];
diff --git a/net/netfilter/nf_flow_table_core.c b/net/netfilter/nf_flow_table_core.c
index 2c4140e6f53c..785d8c244a77 100644
--- a/net/netfilter/nf_flow_table_core.c
+++ b/net/netfilter/nf_flow_table_core.c
@@ -122,6 +122,7 @@ static int flow_offload_fill_route(struct flow_offload *flow,
 
 	flow_tuple->tun = route->tuple[dir].in.tun;
 	flow_tuple->encap_num = route->tuple[dir].in.num_encaps;
+	flow_tuple->needs_gso_segment = route->tuple[dir].out.needs_gso_segment;
 	flow_tuple->tun_num = route->tuple[dir].in.num_tuns;
 
 	switch (route->tuple[dir].xmit_type) {
diff --git a/net/netfilter/nf_flow_table_ip.c b/net/netfilter/nf_flow_table_ip.c
index 1573189a23a1..c3aa5df9b724 100644
--- a/net/netfilter/nf_flow_table_ip.c
+++ b/net/netfilter/nf_flow_table_ip.c
@@ -552,7 +552,8 @@ static int nf_flow_vlan_push(struct sk_buff *skb, __be16 proto, u16 id,
 	return 0;
 }
 
-static int nf_flow_pppoe_push(struct sk_buff *skb, u16 id)
+static int nf_flow_pppoe_push(struct sk_buff *skb, u16 id,
+			      u32 needed_headroom)
 {
 	int data_len = skb->len + sizeof(__be16);
 	struct ppp_hdr {
@@ -561,7 +562,7 @@ static int nf_flow_pppoe_push(struct sk_buff *skb, u16 id)
 	} *ph;
 	__be16 proto;
 
-	if (skb_cow_head(skb, PPPOE_SES_HLEN))
+	if (skb_cow_head(skb, needed_headroom + PPPOE_SES_HLEN))
 		return -1;
 
 	switch (skb->protocol) {
@@ -754,7 +755,8 @@ static int nf_flow_encap_push(struct sk_buff *skb,
 				return -1;
 			break;
 		case htons(ETH_P_PPP_SES):
-			if (nf_flow_pppoe_push(skb, tuple->encap[i].id) < 0)
+			if (nf_flow_pppoe_push(skb, tuple->encap[i].id,
+					       needed_headroom) < 0)
 				return -1;
 			break;
 		}
@@ -779,10 +781,36 @@ static void __nf_flow_queue_xmit(struct net *net, struct sk_buff *skb,
 	dev_queue_xmit(skb);
 }
 
+static unsigned int nf_flow_encap_gso_xmit(struct net *net, struct sk_buff *skb,
+					   struct nf_flow_xmit *xmit)
+{
+	struct sk_buff *segs, *nskb;
+
+	segs = skb_gso_segment(skb, 0);
+	if (IS_ERR(segs) || !segs)
+		return NF_DROP;
+
+	skb_list_walk_safe(segs, segs, nskb) {
+		skb_mark_not_on_list(segs);
+
+		if (nf_flow_encap_push(segs, xmit->tuple, xmit->outdev) < 0) {
+			kfree_skb_list(nskb);
+			return NF_DROP;
+		}
+		__nf_flow_queue_xmit(net, segs, xmit);
+	}
+	consume_skb(skb);
+
+	return NF_STOLEN;
+}
+
 static unsigned int nf_flow_queue_xmit(struct net *net, struct sk_buff *skb,
 				       struct nf_flow_xmit *xmit)
 {
 	if (xmit->tuple->encap_num) {
+		if (skb_is_gso(skb) && xmit->tuple->needs_gso_segment)
+			return nf_flow_encap_gso_xmit(net, skb, xmit);
+
 		if (nf_flow_encap_push(skb, xmit->tuple, xmit->outdev) < 0)
 			return NF_DROP;
 	}
diff --git a/net/netfilter/nf_flow_table_path.c b/net/netfilter/nf_flow_table_path.c
index 6bb9579dcc2a..37f7b590c531 100644
--- a/net/netfilter/nf_flow_table_path.c
+++ b/net/netfilter/nf_flow_table_path.c
@@ -86,6 +86,7 @@ struct nft_forward_info {
 	u8 ingress_vlans;
 	u8 h_source[ETH_ALEN];
 	u8 h_dest[ETH_ALEN];
+	bool needs_gso_segment;
 	enum flow_offload_xmit_type xmit_type;
 };
 
@@ -138,8 +139,11 @@ static void nft_dev_path_info(const struct net_device_path_stack *stack,
 					path->encap.proto;
 				info->num_encaps++;
 			}
-			if (path->type == DEV_PATH_PPPOE)
+			if (path->type == DEV_PATH_PPPOE) {
 				memcpy(info->h_dest, path->encap.h_dest, ETH_ALEN);
+				info->xmit_type = FLOW_OFFLOAD_XMIT_DIRECT;
+				info->needs_gso_segment = 1;
+			}
 			break;
 		case DEV_PATH_BRIDGE:
 			if (is_zero_ether_addr(info->h_source))
@@ -279,6 +283,7 @@ static void nft_dev_forward_path(const struct nft_pktinfo *pkt,
 		memcpy(route->tuple[dir].out.h_dest, info.h_dest, ETH_ALEN);
 		route->tuple[dir].xmit_type = info.xmit_type;
 	}
+	route->tuple[!dir].out.needs_gso_segment = info.needs_gso_segment;
 }
 
 int nft_flow_route(const struct nft_pktinfo *pkt, const struct nf_conn *ct,
-- 
2.47.3

Re: [PATCH nf 2/2] netfilter: flowtable: fix inline pppoe encapsulation in xmit path

[Qingfang Deng]

On Fri, 24 Apr 2026 12:07:59 +0200, Pablo Neira Ayuso wrote:
> Address two issues in the inline pppoe encapsulation:
>
> - Add needs_gso_segment flag to segment PPPoE packets in software
>   given that there is no GSO support for this.

FYI, there is a pending PPPoE GSO patch:

https://lore.kernel.org/netdev/20260326081127.61229-1-dqfext@gmail.com/

Regards,
Qingfang