Merge branch 'gue-next'

Tom Herbert says:

====================
gue: Remote checksum offload

This patch set implements remote checksum offload for
GUE, which is a mechanism that provides checksum offload of
encapsulated packets using rudimentary offload capabilities found in
most Network Interface Card (NIC) devices. The outer header checksum
for UDP is enabled in packets and, with some additional meta
information in the GUE header, a receiver is able to deduce the
checksum to be set for an inner encapsulated packet. Effectively this
offloads the computation of the inner checksum. Enabling the outer
checksum in encapsulation has the additional advantage that it covers
more of the packet than the inner checksum including the encapsulation
headers.

Remote checksum offload is described in:
http://tools.ietf.org/html/draft-herbert-remotecsumoffload-01

The GUE transmit and receive paths are modified to support the
remote checksum offload option. The option contains a checksum
offset and checksum start which are directly derived from values
set in stack when doing CHECKSUM_PARTIAL. On receipt of the option, the
operation is to calculate the packet checksum from "start" to end of
the packet (normally derived for checksum complete), and then set
the resultant value at checksum "offset" (the checksum field has
already been primed with the pseudo header). This emulates a NIC
that implements NETIF_F_HW_CSUM.

The primary purpose of this feature is to eliminate cost of performing
checksum calculation over a packet when encpasulating.

In this patch set:
  - Move fou_build_header into fou.c and split it into a couple of
    functions
  - Enable offloading of outer UDP checksum in encapsulation
  - Change udp_offload to support remote checksum offload, includes
    new GSO type and ensuring encapsulated layers (TCP) doesn't try to
    set a checksum covered by RCO
  - TX support for RCO with GUE. This is configured through ip_tunnel
    and set the option on transmit when packet being encapsulated is
    CHECKSUM_PARTIAL
  - RX support for RCO with GUE for normal and GRO paths. Includes
    resolving the offloaded checksum

v2:
  Address comments from davem: Move accounting for private option
  field in gue_encap_hlen to patch in which we add the remote checksum
  offload option.

Testing:

I ran performance numbers using netperf TCP_STREAM and TCP_RR with 200
streams, comparing GUE with and without remote checksum offload (doing
checksum-unnecessary to complete conversion in both cases). These
were run on mlnx4 and bnx2x. Some mlnx4 results are below.

GRE/GUE
    TCP_STREAM
      IPv4, with remote checksum offload
        9.71% TX CPU utilization
        7.42% RX CPU utilization
        36380 Mbps
      IPv4, without remote checksum offload
        12.40% TX CPU utilization
        7.36% RX CPU utilization
        36591 Mbps
    TCP_RR
      IPv4, with remote checksum offload
        77.79% CPU utilization
	91/144/216 90/95/99% latencies
        1.95127e+06 tps
      IPv4, without remote checksum offload
        78.70% CPU utilization
        89/152/297 90/95/99% latencies
        1.95458e+06 tps

IPIP/GUE
    TCP_STREAM
      With remote checksum offload
        10.30% TX CPU utilization
        7.43% RX CPU utilization
        36486 Mbps
      Without remote checksum offload
        12.47% TX CPU utilization
        7.49% RX CPU utilization
        36694 Mbps
    TCP_RR
      With remote checksum offload
        77.80% CPU utilization
        87/153/270 90/95/99% latencies
        1.98735e+06 tps
      Without remote checksum offload
        77.98% CPU utilization
        87/150/287 90/95/99% latencies
        1.98737e+06 tps

SIT/GUE
    TCP_STREAM
      With remote checksum offload
        9.68% TX CPU utilization
        7.36% RX CPU utilization
        35971 Mbps
      Without remote checksum offload
        12.95% TX CPU utilization
        8.04% RX CPU utilization
        36177 Mbps
    TCP_RR
      With remote checksum offload
        79.32% CPU utilization
        94/158/295 90/95/99% latencies
        1.88842e+06 tps
      Without remote checksum offload
        80.23% CPU utilization
        94/149/226 90/95/99% latencies
        1.90338e+06 tps

VXLAN
    TCP_STREAM
        35.03% TX CPU utilization
        20.85% RX CPU utilization
        36230 Mbps
    TCP_RR
        77.36% CPU utilization
        84/146/270 90/95/99% latencies
        2.08063e+06 tps

We can also look at CPU time in csum_partial using perf (with bnx2x
setup). For GRE with TCP_STREAM I see:

    With remote checksum offload
        0.33% TX
        1.81% RX
    Without remote checksum offload
        6.00% TX
        0.51% RX

I suspect the fact that time in csum_partial noticably increases
with remote checksum offload for RX is due to taking the cache miss on
the encapsulated header in that function. By similar reasoning, if on
the TX side the packet were not in cache (say we did a splice from a
file whose data was never touched by the CPU) the CPU savings for TX
would probably be more pronounced.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

9 files changed, 421 insertions, 111 deletions

diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index e48e5c02e877..700189604f3d 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -3013,7 +3013,7 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,
 		if (nskb->len == len + doffset)
 			goto perform_csum_check;
 
-		if (!sg) {
+		if (!sg && !nskb->remcsum_offload) {
 			nskb->ip_summed = CHECKSUM_NONE;
 			nskb->csum = skb_copy_and_csum_bits(head_skb, offset,
 							    skb_put(nskb, len),
@@ -3085,7 +3085,7 @@ skip_fraglist:
 		nskb->truesize += nskb->data_len;
 
 perform_csum_check:
-		if (!csum) {
+		if (!csum && !nskb->remcsum_offload) {
 			nskb->csum = skb_checksum(nskb, doffset,
 						  nskb->len - doffset, 0);
 			nskb->ip_summed = CHECKSUM_NONE;
diff --git a/net/ipv4/Kconfig b/net/ipv4/Kconfig
index e682b48e0709..bd2901604842 100644
--- a/net/ipv4/Kconfig
+++ b/net/ipv4/Kconfig
@@ -322,6 +322,15 @@ config NET_FOU
 	  network mechanisms and optimizations for UDP (such as ECMP
 	  and RSS) can be leveraged to provide better service.
 
+config NET_FOU_IP_TUNNELS
+	bool "IP: FOU encapsulation of IP tunnels"
+	depends on NET_IPIP || NET_IPGRE || IPV6_SIT
+	select NET_FOU
+	---help---
+	  Allow configuration of FOU or GUE encapsulation for IP tunnels.
+	  When this option is enabled IP tunnels can be configured to use
+	  FOU or GUE encapsulation.
+
 config GENEVE
 	tristate "Generic Network Virtualization Encapsulation (Geneve)"
 	depends on INET
diff --git a/net/ipv4/af_inet.c b/net/ipv4/af_inet.c
index 8b7fe5b03906..ed2c672c5b01 100644
--- a/net/ipv4/af_inet.c
+++ b/net/ipv4/af_inet.c
@@ -1222,6 +1222,7 @@ static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
 		       SKB_GSO_TCPV6 |
 		       SKB_GSO_UDP_TUNNEL |
 		       SKB_GSO_UDP_TUNNEL_CSUM |
+		       SKB_GSO_TUNNEL_REMCSUM |
 		       SKB_GSO_MPLS |
 		       0)))
 		goto out;
diff --git a/net/ipv4/fou.c b/net/ipv4/fou.c
index 32e78924e246..740ae099a0d9 100644
--- a/net/ipv4/fou.c
+++ b/net/ipv4/fou.c
@@ -38,21 +38,17 @@ static inline struct fou *fou_from_sock(struct sock *sk)
 	return sk->sk_user_data;
 }
 
-static int fou_udp_encap_recv_deliver(struct sk_buff *skb,
-				      u8 protocol, size_t len)
+static void fou_recv_pull(struct sk_buff *skb, size_t len)
 {
 	struct iphdr *iph = ip_hdr(skb);
 
 	/* Remove 'len' bytes from the packet (UDP header and
-	 * FOU header if present), modify the protocol to the one
-	 * we found, and then call rcv_encap.
+	 * FOU header if present).
 	 */
 	iph->tot_len = htons(ntohs(iph->tot_len) - len);
 	__skb_pull(skb, len);
 	skb_postpull_rcsum(skb, udp_hdr(skb), len);
 	skb_reset_transport_header(skb);
-
-	return -protocol;
 }
 
 static int fou_udp_recv(struct sock *sk, struct sk_buff *skb)
@@ -62,16 +58,78 @@ static int fou_udp_recv(struct sock *sk, struct sk_buff *skb)
 	if (!fou)
 		return 1;
 
-	return fou_udp_encap_recv_deliver(skb, fou->protocol,
-					  sizeof(struct udphdr));
+	fou_recv_pull(skb, sizeof(struct udphdr));
+
+	return -fou->protocol;
+}
+
+static struct guehdr *gue_remcsum(struct sk_buff *skb, struct guehdr *guehdr,
+				  void *data, int hdrlen, u8 ipproto)
+{
+	__be16 *pd = data;
+	u16 start = ntohs(pd[0]);
+	u16 offset = ntohs(pd[1]);
+	u16 poffset = 0;
+	u16 plen;
+	__wsum csum, delta;
+	__sum16 *psum;
+
+	if (skb->remcsum_offload) {
+		/* Already processed in GRO path */
+		skb->remcsum_offload = 0;
+		return guehdr;
+	}
+
+	if (start > skb->len - hdrlen ||
+	    offset > skb->len - hdrlen - sizeof(u16))
+		return NULL;
+
+	if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE))
+		__skb_checksum_complete(skb);
+
+	plen = hdrlen + offset + sizeof(u16);
+	if (!pskb_may_pull(skb, plen))
+		return NULL;
+	guehdr = (struct guehdr *)&udp_hdr(skb)[1];
+
+	if (ipproto == IPPROTO_IP && sizeof(struct iphdr) < plen) {
+		struct iphdr *ip = (struct iphdr *)(skb->data + hdrlen);
+
+		/* If next header happens to be IP we can skip that for the
+		 * checksum calculation since the IP header checksum is zero
+		 * if correct.
+		 */
+		poffset = ip->ihl * 4;
+	}
+
+	csum = csum_sub(skb->csum, skb_checksum(skb, poffset + hdrlen,
+						start - poffset - hdrlen, 0));
+
+	/* Set derived checksum in packet */
+	psum = (__sum16 *)(skb->data + hdrlen + offset);
+	delta = csum_sub(csum_fold(csum), *psum);
+	*psum = csum_fold(csum);
+
+	/* Adjust skb->csum since we changed the packet */
+	skb->csum = csum_add(skb->csum, delta);
+
+	return guehdr;
+}
+
+static int gue_control_message(struct sk_buff *skb, struct guehdr *guehdr)
+{
+	/* No support yet */
+	kfree_skb(skb);
+	return 0;
 }
 
 static int gue_udp_recv(struct sock *sk, struct sk_buff *skb)
 {
 	struct fou *fou = fou_from_sock(sk);
-	size_t len;
+	size_t len, optlen, hdrlen;
 	struct guehdr *guehdr;
-	struct udphdr *uh;
+	void *data;
+	u16 doffset = 0;
 
 	if (!fou)
 		return 1;
@@ -80,25 +138,61 @@ static int gue_udp_recv(struct sock *sk, struct sk_buff *skb)
 	if (!pskb_may_pull(skb, len))
 		goto drop;
 
-	uh = udp_hdr(skb);
-	guehdr = (struct guehdr *)&uh[1];
+	guehdr = (struct guehdr *)&udp_hdr(skb)[1];
+
+	optlen = guehdr->hlen << 2;
+	len += optlen;
 
-	len += guehdr->hlen << 2;
 	if (!pskb_may_pull(skb, len))
 		goto drop;
 
-	uh = udp_hdr(skb);
-	guehdr = (struct guehdr *)&uh[1];
+	/* guehdr may change after pull */
+	guehdr = (struct guehdr *)&udp_hdr(skb)[1];
 
-	if (guehdr->version != 0)
-		goto drop;
+	hdrlen = sizeof(struct guehdr) + optlen;
 
-	if (guehdr->flags) {
-		/* No support yet */
+	if (guehdr->version != 0 || validate_gue_flags(guehdr, optlen))
 		goto drop;
+
+	hdrlen = sizeof(struct guehdr) + optlen;
+
+	ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(skb)->tot_len) - len);
+
+	/* Pull UDP header now, skb->data points to guehdr */
+	__skb_pull(skb, sizeof(struct udphdr));
+
+	/* Pull csum through the guehdr now . This can be used if
+	 * there is a remote checksum offload.
+	 */
+	skb_postpull_rcsum(skb, udp_hdr(skb), len);
+
+	data = &guehdr[1];
+
+	if (guehdr->flags & GUE_FLAG_PRIV) {
+		__be32 flags = *(__be32 *)(data + doffset);
+
+		doffset += GUE_LEN_PRIV;
+
+		if (flags & GUE_PFLAG_REMCSUM) {
+			guehdr = gue_remcsum(skb, guehdr, data + doffset,
+					     hdrlen, guehdr->proto_ctype);
+			if (!guehdr)
+				goto drop;
+
+			data = &guehdr[1];
+
+			doffset += GUE_PLEN_REMCSUM;
+		}
 	}
 
-	return fou_udp_encap_recv_deliver(skb, guehdr->next_hdr, len);
+	if (unlikely(guehdr->control))
+		return gue_control_message(skb, guehdr);
+
+	__skb_pull(skb, hdrlen);
+	skb_reset_transport_header(skb);
+
+	return -guehdr->proto_ctype;
+
 drop:
 	kfree_skb(skb);
 	return 0;
@@ -147,6 +241,66 @@ out_unlock:
 	return err;
 }
 
+static struct guehdr *gue_gro_remcsum(struct sk_buff *skb, unsigned int off,
+				      struct guehdr *guehdr, void *data,
+				      size_t hdrlen, u8 ipproto)
+{
+	__be16 *pd = data;
+	u16 start = ntohs(pd[0]);
+	u16 offset = ntohs(pd[1]);
+	u16 poffset = 0;
+	u16 plen;
+	void *ptr;
+	__wsum csum, delta;
+	__sum16 *psum;
+
+	if (skb->remcsum_offload)
+		return guehdr;
+
+	if (start > skb_gro_len(skb) - hdrlen ||
+	    offset > skb_gro_len(skb) - hdrlen - sizeof(u16) ||
+	    !NAPI_GRO_CB(skb)->csum_valid || skb->remcsum_offload)
+		return NULL;
+
+	plen = hdrlen + offset + sizeof(u16);
+
+	/* Pull checksum that will be written */
+	if (skb_gro_header_hard(skb, off + plen)) {
+		guehdr = skb_gro_header_slow(skb, off + plen, off);
+		if (!guehdr)
+			return NULL;
+	}
+
+	ptr = (void *)guehdr + hdrlen;
+
+	if (ipproto == IPPROTO_IP &&
+	    (hdrlen + sizeof(struct iphdr) < plen)) {
+		struct iphdr *ip = (struct iphdr *)(ptr + hdrlen);
+
+		/* If next header happens to be IP we can skip
+		 * that for the checksum calculation since the
+		 * IP header checksum is zero if correct.
+		 */
+		poffset = ip->ihl * 4;
+	}
+
+	csum = csum_sub(NAPI_GRO_CB(skb)->csum,
+			csum_partial(ptr + poffset, start - poffset, 0));
+
+	/* Set derived checksum in packet */
+	psum = (__sum16 *)(ptr + offset);
+	delta = csum_sub(csum_fold(csum), *psum);
+	*psum = csum_fold(csum);
+
+	/* Adjust skb->csum since we changed the packet */
+	skb->csum = csum_add(skb->csum, delta);
+	NAPI_GRO_CB(skb)->csum = csum_add(NAPI_GRO_CB(skb)->csum, delta);
+
+	skb->remcsum_offload = 1;
+
+	return guehdr;
+}
+
 static struct sk_buff **gue_gro_receive(struct sk_buff **head,
 					struct sk_buff *skb)
 {
@@ -154,38 +308,64 @@ static struct sk_buff **gue_gro_receive(struct sk_buff **head,
 	const struct net_offload *ops;
 	struct sk_buff **pp = NULL;
 	struct sk_buff *p;
-	u8 proto;
 	struct guehdr *guehdr;
-	unsigned int hlen, guehlen;
-	unsigned int off;
+	size_t len, optlen, hdrlen, off;
+	void *data;
+	u16 doffset = 0;
 	int flush = 1;
 
 	off = skb_gro_offset(skb);
-	hlen = off + sizeof(*guehdr);
+	len = off + sizeof(*guehdr);
+
 	guehdr = skb_gro_header_fast(skb, off);
-	if (skb_gro_header_hard(skb, hlen)) {
-		guehdr = skb_gro_header_slow(skb, hlen, off);
+	if (skb_gro_header_hard(skb, len)) {
+		guehdr = skb_gro_header_slow(skb, len, off);
 		if (unlikely(!guehdr))
 			goto out;
 	}
 
-	proto = guehdr->next_hdr;
+	optlen = guehdr->hlen << 2;
+	len += optlen;
 
-	rcu_read_lock();
-	offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
-	ops = rcu_dereference(offloads[proto]);
-	if (WARN_ON(!ops || !ops->callbacks.gro_receive))
-		goto out_unlock;
+	if (skb_gro_header_hard(skb, len)) {
+		guehdr = skb_gro_header_slow(skb, len, off);
+		if (unlikely(!guehdr))
+			goto out;
+	}
 
-	guehlen = sizeof(*guehdr) + (guehdr->hlen << 2);
+	if (unlikely(guehdr->control) || guehdr->version != 0 ||
+	    validate_gue_flags(guehdr, optlen))
+		goto out;
 
-	hlen = off + guehlen;
-	if (skb_gro_header_hard(skb, hlen)) {
-		guehdr = skb_gro_header_slow(skb, hlen, off);
-		if (unlikely(!guehdr))
-			goto out_unlock;
+	hdrlen = sizeof(*guehdr) + optlen;
+
+	/* Adjust NAPI_GRO_CB(skb)->csum to account for guehdr,
+	 * this is needed if there is a remote checkcsum offload.
+	 */
+	skb_gro_postpull_rcsum(skb, guehdr, hdrlen);
+
+	data = &guehdr[1];
+
+	if (guehdr->flags & GUE_FLAG_PRIV) {
+		__be32 flags = *(__be32 *)(data + doffset);
+
+		doffset += GUE_LEN_PRIV;
+
+		if (flags & GUE_PFLAG_REMCSUM) {
+			guehdr = gue_gro_remcsum(skb, off, guehdr,
+						 data + doffset, hdrlen,
+						 guehdr->proto_ctype);
+			if (!guehdr)
+				goto out;
+
+			data = &guehdr[1];
+
+			doffset += GUE_PLEN_REMCSUM;
+		}
 	}
 
+	skb_gro_pull(skb, hdrlen);
+
 	flush = 0;
 
 	for (p = *head; p; p = p->next) {
@@ -197,7 +377,7 @@ static struct sk_buff **gue_gro_receive(struct sk_buff **head,
 		guehdr2 = (struct guehdr *)(p->data + off);
 
 		/* Compare base GUE header to be equal (covers
-		 * hlen, version, next_hdr, and flags.
+		 * hlen, version, proto_ctype, and flags.
 		 */
 		if (guehdr->word != guehdr2->word) {
 			NAPI_GRO_CB(p)->same_flow = 0;
@@ -212,10 +392,11 @@ static struct sk_buff **gue_gro_receive(struct sk_buff **head,
 		}
 	}
 
-	skb_gro_pull(skb, guehlen);
-
-	/* Adjusted NAPI_GRO_CB(skb)->csum after skb_gro_pull()*/
-	skb_gro_postpull_rcsum(skb, guehdr, guehlen);
+	rcu_read_lock();
+	offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
+	ops = rcu_dereference(offloads[guehdr->proto_ctype]);
+	if (WARN_ON(!ops || !ops->callbacks.gro_receive))
+		goto out_unlock;
 
 	pp = ops->callbacks.gro_receive(head, skb);
 
@@ -236,7 +417,7 @@ static int gue_gro_complete(struct sk_buff *skb, int nhoff)
 	u8 proto;
 	int err = -ENOENT;
 
-	proto = guehdr->next_hdr;
+	proto = guehdr->proto_ctype;
 
 	guehlen = sizeof(*guehdr) + (guehdr->hlen << 2);
 
@@ -487,6 +668,125 @@ static const struct genl_ops fou_nl_ops[] = {
 	},
 };
 
+static void fou_build_udp(struct sk_buff *skb, struct ip_tunnel_encap *e,
+			  struct flowi4 *fl4, u8 *protocol, __be16 sport)
+{
+	struct udphdr *uh;
+
+	skb_push(skb, sizeof(struct udphdr));
+	skb_reset_transport_header(skb);
+
+	uh = udp_hdr(skb);
+
+	uh->dest = e->dport;
+	uh->source = sport;
+	uh->len = htons(skb->len);
+	uh->check = 0;
+	udp_set_csum(!(e->flags & TUNNEL_ENCAP_FLAG_CSUM), skb,
+		     fl4->saddr, fl4->daddr, skb->len);
+
+	*protocol = IPPROTO_UDP;
+}
+
+int fou_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
+		     u8 *protocol, struct flowi4 *fl4)
+{
+	bool csum = !!(e->flags & TUNNEL_ENCAP_FLAG_CSUM);
+	int type = csum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
+	__be16 sport;
+
+	skb = iptunnel_handle_offloads(skb, csum, type);
+
+	if (IS_ERR(skb))
+		return PTR_ERR(skb);
+
+	sport = e->sport ? : udp_flow_src_port(dev_net(skb->dev),
+					       skb, 0, 0, false);
+	fou_build_udp(skb, e, fl4, protocol, sport);
+
+	return 0;
+}
+EXPORT_SYMBOL(fou_build_header);
+
+int gue_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
+		     u8 *protocol, struct flowi4 *fl4)
+{
+	bool csum = !!(e->flags & TUNNEL_ENCAP_FLAG_CSUM);
+	int type = csum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
+	struct guehdr *guehdr;
+	size_t hdrlen, optlen = 0;
+	__be16 sport;
+	void *data;
+	bool need_priv = false;
+
+	if ((e->flags & TUNNEL_ENCAP_FLAG_REMCSUM) &&
+	    skb->ip_summed == CHECKSUM_PARTIAL) {
+		csum = false;
+		optlen += GUE_PLEN_REMCSUM;
+		type |= SKB_GSO_TUNNEL_REMCSUM;
+		need_priv = true;
+	}
+
+	optlen += need_priv ? GUE_LEN_PRIV : 0;
+
+	skb = iptunnel_handle_offloads(skb, csum, type);
+
+	if (IS_ERR(skb))
+		return PTR_ERR(skb);
+
+	/* Get source port (based on flow hash) before skb_push */
+	sport = e->sport ? : udp_flow_src_port(dev_net(skb->dev),
+					       skb, 0, 0, false);
+
+	hdrlen = sizeof(struct guehdr) + optlen;
+
+	skb_push(skb, hdrlen);
+
+	guehdr = (struct guehdr *)skb->data;
+
+	guehdr->control = 0;
+	guehdr->version = 0;
+	guehdr->hlen = optlen >> 2;
+	guehdr->flags = 0;
+	guehdr->proto_ctype = *protocol;
+
+	data = &guehdr[1];
+
+	if (need_priv) {
+		__be32 *flags = data;
+
+		guehdr->flags |= GUE_FLAG_PRIV;
+		*flags = 0;
+		data += GUE_LEN_PRIV;
+
+		if (type & SKB_GSO_TUNNEL_REMCSUM) {
+			u16 csum_start = skb_checksum_start_offset(skb);
+			__be16 *pd = data;
+
+			if (csum_start < hdrlen)
+				return -EINVAL;
+
+			csum_start -= hdrlen;
+			pd[0] = htons(csum_start);
+			pd[1] = htons(csum_start + skb->csum_offset);
+
+			if (!skb_is_gso(skb)) {
+				skb->ip_summed = CHECKSUM_NONE;
+				skb->encapsulation = 0;
+			}
+
+			*flags |= GUE_PFLAG_REMCSUM;
+			data += GUE_PLEN_REMCSUM;
+		}
+
+	}
+
+	fou_build_udp(skb, e, fl4, protocol, sport);
+
+	return 0;
+}
+EXPORT_SYMBOL(gue_build_header);
+
 static int __init fou_init(void)
 {
 	int ret;
diff --git a/net/ipv4/ip_tunnel.c b/net/ipv4/ip_tunnel.c
index 0bb8e141eacc..c3587e1c8b82 100644
--- a/net/ipv4/ip_tunnel.c
+++ b/net/ipv4/ip_tunnel.c
@@ -56,7 +56,10 @@
 #include <net/netns/generic.h>
 #include <net/rtnetlink.h>
 #include <net/udp.h>
-#include <net/gue.h>
+
+#if IS_ENABLED(CONFIG_NET_FOU)
+#include <net/fou.h>
+#endif
 
 #if IS_ENABLED(CONFIG_IPV6)
 #include <net/ipv6.h>
@@ -494,10 +497,12 @@ static int ip_encap_hlen(struct ip_tunnel_encap *e)
 	switch (e->type) {
 	case TUNNEL_ENCAP_NONE:
 		return 0;
+#if IS_ENABLED(CONFIG_NET_FOU)
 	case TUNNEL_ENCAP_FOU:
-		return sizeof(struct udphdr);
+		return fou_encap_hlen(e);
 	case TUNNEL_ENCAP_GUE:
-		return sizeof(struct udphdr) + sizeof(struct guehdr);
+		return gue_encap_hlen(e);
+#endif
 	default:
 		return -EINVAL;
 	}
@@ -526,60 +531,18 @@ int ip_tunnel_encap_setup(struct ip_tunnel *t,
 }
 EXPORT_SYMBOL_GPL(ip_tunnel_encap_setup);
 
-static int fou_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
-			    size_t hdr_len, u8 *protocol, struct flowi4 *fl4)
-{
-	struct udphdr *uh;
-	__be16 sport;
-	bool csum = !!(e->flags & TUNNEL_ENCAP_FLAG_CSUM);
-	int type = csum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
-
-	skb = iptunnel_handle_offloads(skb, csum, type);
-
-	if (IS_ERR(skb))
-		return PTR_ERR(skb);
-
-	/* Get length and hash before making space in skb */
-
-	sport = e->sport ? : udp_flow_src_port(dev_net(skb->dev),
-					       skb, 0, 0, false);
-
-	skb_push(skb, hdr_len);
-
-	skb_reset_transport_header(skb);
-	uh = udp_hdr(skb);
-
-	if (e->type == TUNNEL_ENCAP_GUE) {
-		struct guehdr *guehdr = (struct guehdr *)&uh[1];
-
-		guehdr->version = 0;
-		guehdr->hlen = 0;
-		guehdr->flags = 0;
-		guehdr->next_hdr = *protocol;
-	}
-
-	uh->dest = e->dport;
-	uh->source = sport;
-	uh->len = htons(skb->len);
-	uh->check = 0;
-	udp_set_csum(!(e->flags & TUNNEL_ENCAP_FLAG_CSUM), skb,
-		     fl4->saddr, fl4->daddr, skb->len);
-
-	*protocol = IPPROTO_UDP;
-
-	return 0;
-}
-
 int ip_tunnel_encap(struct sk_buff *skb, struct ip_tunnel *t,
 		    u8 *protocol, struct flowi4 *fl4)
 {
 	switch (t->encap.type) {
 	case TUNNEL_ENCAP_NONE:
 		return 0;
+#if IS_ENABLED(CONFIG_NET_FOU)
 	case TUNNEL_ENCAP_FOU:
+		return fou_build_header(skb, &t->encap, protocol, fl4);
 	case TUNNEL_ENCAP_GUE:
-		return fou_build_header(skb, &t->encap, t->encap_hlen,
-					protocol, fl4);
+		return gue_build_header(skb, &t->encap, protocol, fl4);
+#endif
 	default:
 		return -EINVAL;
 	}
diff --git a/net/ipv4/tcp_offload.c b/net/ipv4/tcp_offload.c
index 5b90f2f447a5..a1b2a5624f91 100644
--- a/net/ipv4/tcp_offload.c
+++ b/net/ipv4/tcp_offload.c
@@ -97,6 +97,7 @@ struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
 			       SKB_GSO_MPLS |
 			       SKB_GSO_UDP_TUNNEL |
 			       SKB_GSO_UDP_TUNNEL_CSUM |
+			       SKB_GSO_TUNNEL_REMCSUM |
 			       0) ||
 			     !(type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))))
 			goto out;
diff --git a/net/ipv4/udp_offload.c b/net/ipv4/udp_offload.c
index 6480cea7aa53..0a5a70d0e84c 100644
--- a/net/ipv4/udp_offload.c
+++ b/net/ipv4/udp_offload.c
@@ -29,7 +29,7 @@ static struct sk_buff *__skb_udp_tunnel_segment(struct sk_buff *skb,
 	netdev_features_t features,
 	struct sk_buff *(*gso_inner_segment)(struct sk_buff *skb,
 					     netdev_features_t features),
-	__be16 new_protocol)
+	__be16 new_protocol, bool is_ipv6)
 {
 	struct sk_buff *segs = ERR_PTR(-EINVAL);
 	u16 mac_offset = skb->mac_header;
@@ -39,7 +39,10 @@ static struct sk_buff *__skb_udp_tunnel_segment(struct sk_buff *skb,
 	netdev_features_t enc_features;
 	int udp_offset, outer_hlen;
 	unsigned int oldlen;
-	bool need_csum;
+	bool need_csum = !!(skb_shinfo(skb)->gso_type &
+			    SKB_GSO_UDP_TUNNEL_CSUM);
+	bool remcsum = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TUNNEL_REMCSUM);
+	bool offload_csum = false, dont_encap = (need_csum || remcsum);
 
 	oldlen = (u16)~skb->len;
 
@@ -52,10 +55,13 @@ static struct sk_buff *__skb_udp_tunnel_segment(struct sk_buff *skb,
 	skb_set_network_header(skb, skb_inner_network_offset(skb));
 	skb->mac_len = skb_inner_network_offset(skb);
 	skb->protocol = new_protocol;
+	skb->encap_hdr_csum = need_csum;
+	skb->remcsum_offload = remcsum;
 
-	need_csum = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM);
-	if (need_csum)
-		skb->encap_hdr_csum = 1;
+	/* Try to offload checksum if possible */
+	offload_csum = !!(need_csum &&
+			  (skb->dev->features &
+			   (is_ipv6 ? NETIF_F_V6_CSUM : NETIF_F_V4_CSUM)));
 
 	/* segment inner packet. */
 	enc_features = skb->dev->hw_enc_features & features;
@@ -72,11 +78,21 @@ static struct sk_buff *__skb_udp_tunnel_segment(struct sk_buff *skb,
 	do {
 		struct udphdr *uh;
 		int len;
-
-		skb_reset_inner_headers(skb);
-		skb->encapsulation = 1;
+		__be32 delta;
+
+		if (dont_encap) {
+			skb->encapsulation = 0;
+			skb->ip_summed = CHECKSUM_NONE;
+		} else {
+			/* Only set up inner headers if we might be offloading
+			 * inner checksum.
+			 */
+			skb_reset_inner_headers(skb);
+			skb->encapsulation = 1;
+		}
 
 		skb->mac_len = mac_len;
+		skb->protocol = protocol;
 
 		skb_push(skb, outer_hlen);
 		skb_reset_mac_header(skb);
@@ -86,19 +102,36 @@ static struct sk_buff *__skb_udp_tunnel_segment(struct sk_buff *skb,
 		uh = udp_hdr(skb);
 		uh->len = htons(len);
 
-		if (need_csum) {
-			__be32 delta = htonl(oldlen + len);
+		if (!need_csum)
+			continue;
 
-			uh->check = ~csum_fold((__force __wsum)
-					       ((__force u32)uh->check +
-						(__force u32)delta));
+		delta = htonl(oldlen + len);
+
+		uh->check = ~csum_fold((__force __wsum)
+				       ((__force u32)uh->check +
+					(__force u32)delta));
+		if (offload_csum) {
+			skb->ip_summed = CHECKSUM_PARTIAL;
+			skb->csum_start = skb_transport_header(skb) - skb->head;
+			skb->csum_offset = offsetof(struct udphdr, check);
+		} else if (remcsum) {
+			/* Need to calculate checksum from scratch,
+			 * inner checksums are never when doing
+			 * remote_checksum_offload.
+			 */
+
+			skb->csum = skb_checksum(skb, udp_offset,
+						 skb->len - udp_offset,
+						 0);
+			uh->check = csum_fold(skb->csum);
+			if (uh->check == 0)
+				uh->check = CSUM_MANGLED_0;
+		} else {
 			uh->check = gso_make_checksum(skb, ~uh->check);
 
 			if (uh->check == 0)
 				uh->check = CSUM_MANGLED_0;
 		}
-
-		skb->protocol = protocol;
 	} while ((skb = skb->next));
 out:
 	return segs;
@@ -134,7 +167,7 @@ struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
 	}
 
 	segs = __skb_udp_tunnel_segment(skb, features, gso_inner_segment,
-					protocol);
+					protocol, is_ipv6);
 
 out_unlock:
 	rcu_read_unlock();
@@ -172,6 +205,7 @@ static struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
 		if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY |
 				      SKB_GSO_UDP_TUNNEL |
 				      SKB_GSO_UDP_TUNNEL_CSUM |
+				      SKB_GSO_TUNNEL_REMCSUM |
 				      SKB_GSO_IPIP |
 				      SKB_GSO_GRE | SKB_GSO_GRE_CSUM |
 				      SKB_GSO_MPLS) ||
diff --git a/net/ipv6/ip6_offload.c b/net/ipv6/ip6_offload.c
index a071563a7e6e..e9767079a360 100644
--- a/net/ipv6/ip6_offload.c
+++ b/net/ipv6/ip6_offload.c
@@ -78,6 +78,7 @@ static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb,
 		       SKB_GSO_SIT |
 		       SKB_GSO_UDP_TUNNEL |
 		       SKB_GSO_UDP_TUNNEL_CSUM |
+		       SKB_GSO_TUNNEL_REMCSUM |
 		       SKB_GSO_MPLS |
 		       SKB_GSO_TCPV6 |
 		       0)))
diff --git a/net/ipv6/udp_offload.c b/net/ipv6/udp_offload.c
index 6b8f543f6ac6..637ba2e438b7 100644
--- a/net/ipv6/udp_offload.c
+++ b/net/ipv6/udp_offload.c
@@ -42,6 +42,7 @@ static struct sk_buff *udp6_ufo_fragment(struct sk_buff *skb,
 				      SKB_GSO_DODGY |
 				      SKB_GSO_UDP_TUNNEL |
 				      SKB_GSO_UDP_TUNNEL_CSUM |
+				      SKB_GSO_TUNNEL_REMCSUM |
 				      SKB_GSO_GRE |
 				      SKB_GSO_GRE_CSUM |
 				      SKB_GSO_IPIP |