GSO: Support partial segmentation offload

This patch adds support for something I am referring to as GSO partial.
The basic idea is that we can support a broader range of devices for
segmentation if we use fixed outer headers and have the hardware only
really deal with segmenting the inner header.  The idea behind the naming
is due to the fact that everything before csum_start will be fixed headers,
and everything after will be the region that is handled by hardware.

With the current implementation it allows us to add support for the
following GSO types with an inner TSO_MANGLEID or TSO6 offload:
NETIF_F_GSO_GRE
NETIF_F_GSO_GRE_CSUM
NETIF_F_GSO_IPIP
NETIF_F_GSO_SIT
NETIF_F_UDP_TUNNEL
NETIF_F_UDP_TUNNEL_CSUM

In the case of hardware that already supports tunneling we may be able to
extend this further to support TSO_TCPV4 without TSO_MANGLEID if the
hardware can support updating inner IPv4 headers.
Signed-off-by: NAlexander Duyck <aduyck@mirantis.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

include/linux/netdev_features.h

@@ -48,6 +48,10 @@ enum {
 	NETIF_F_GSO_SIT_BIT,		/* ... SIT tunnel with TSO */
 	NETIF_F_GSO_UDP_TUNNEL_BIT,	/* ... UDP TUNNEL with TSO */
 	NETIF_F_GSO_UDP_TUNNEL_CSUM_BIT,/* ... UDP TUNNEL with TSO & CSUM */
+	NETIF_F_GSO_PARTIAL_BIT,	/* ... Only segment inner-most L4
+					 *     in hardware and all other
+					 *     headers in software.
+					 */
 	NETIF_F_GSO_TUNNEL_REMCSUM_BIT, /* ... TUNNEL with TSO & REMCSUM */
 	/**/NETIF_F_GSO_LAST =		/* last bit, see GSO_MASK */
 		NETIF_F_GSO_TUNNEL_REMCSUM_BIT,
@@ -122,6 +126,7 @@ enum {
 #define NETIF_F_GSO_UDP_TUNNEL	__NETIF_F(GSO_UDP_TUNNEL)
 #define NETIF_F_GSO_UDP_TUNNEL_CSUM __NETIF_F(GSO_UDP_TUNNEL_CSUM)
 #define NETIF_F_TSO_MANGLEID	__NETIF_F(TSO_MANGLEID)
+#define NETIF_F_GSO_PARTIAL	 __NETIF_F(GSO_PARTIAL)
 #define NETIF_F_GSO_TUNNEL_REMCSUM __NETIF_F(GSO_TUNNEL_REMCSUM)
 #define NETIF_F_HW_VLAN_STAG_FILTER __NETIF_F(HW_VLAN_STAG_FILTER)
 #define NETIF_F_HW_VLAN_STAG_RX	__NETIF_F(HW_VLAN_STAG_RX)

include/linux/netdevice.h

@@ -1654,6 +1654,7 @@ struct net_device {
 	netdev_features_t	vlan_features;
 	netdev_features_t	hw_enc_features;
 	netdev_features_t	mpls_features;
+	netdev_features_t	gso_partial_features;
 
 	int			ifindex;
 	int			group;
@@ -4004,6 +4005,7 @@ static inline bool net_gso_ok(netdev_features_t features, int gso_type)
 	BUILD_BUG_ON(SKB_GSO_SIT     != (NETIF_F_GSO_SIT >> NETIF_F_GSO_SHIFT));
 	BUILD_BUG_ON(SKB_GSO_UDP_TUNNEL != (NETIF_F_GSO_UDP_TUNNEL >> NETIF_F_GSO_SHIFT));
 	BUILD_BUG_ON(SKB_GSO_UDP_TUNNEL_CSUM != (NETIF_F_GSO_UDP_TUNNEL_CSUM >> NETIF_F_GSO_SHIFT));
+	BUILD_BUG_ON(SKB_GSO_PARTIAL != (NETIF_F_GSO_PARTIAL >> NETIF_F_GSO_SHIFT));
 	BUILD_BUG_ON(SKB_GSO_TUNNEL_REMCSUM != (NETIF_F_GSO_TUNNEL_REMCSUM >> NETIF_F_GSO_SHIFT));
 
 	return (features & feature) == feature;

include/linux/skbuff.h

@@ -483,7 +483,9 @@ enum {
 
 	SKB_GSO_UDP_TUNNEL_CSUM = 1 << 12,
 
-	SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
+	SKB_GSO_PARTIAL = 1 << 13,
+
+	SKB_GSO_TUNNEL_REMCSUM = 1 << 14,
 };
 
 #if BITS_PER_LONG > 32
@@ -3591,7 +3593,10 @@ static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
  * Keeps track of level of encapsulation of network headers.
  */
 struct skb_gso_cb {
+	union {
 		int	mac_offset;
+		int	data_offset;
+	};
 	int	encap_level;
 	__wsum	csum;
 	__u16	csum_start;

net/core/dev.c

@@ -2711,6 +2711,19 @@ struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
 			return ERR_PTR(err);
 	}
 
+	/* Only report GSO partial support if it will enable us to
+	 * support segmentation on this frame without needing additional
+	 * work.
+	 */
+	if (features & NETIF_F_GSO_PARTIAL) {
+		netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
+		struct net_device *dev = skb->dev;
+
+		partial_features |= dev->features & dev->gso_partial_features;
+		if (!skb_gso_ok(skb, features | partial_features))
+			features &= ~NETIF_F_GSO_PARTIAL;
+	}
+
 	BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
 		     sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
 
@@ -2834,8 +2847,17 @@ static netdev_features_t gso_features_check(const struct sk_buff *skb,
 	if (gso_segs > dev->gso_max_segs)
 		return features & ~NETIF_F_GSO_MASK;
 
-	/* Make sure to clear the IPv4 ID mangling feature if
-	 * the IPv4 header has the potential to be fragmented.
+	/* Support for GSO partial features requires software
+	 * intervention before we can actually process the packets
+	 * so we need to strip support for any partial features now
+	 * and we can pull them back in after we have partially
+	 * segmented the frame.
+	 */
+	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
+		features &= ~dev->gso_partial_features;
+
+	/* Make sure to clear the IPv4 ID mangling feature if the
+	 * IPv4 header has the potential to be fragmented.
 	 */
 	if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
 		struct iphdr *iph = skb->encapsulation ?
@@ -6729,6 +6751,14 @@ static netdev_features_t netdev_fix_features(struct net_device *dev,
 		}
 	}
 
+	/* GSO partial features require GSO partial be set */
+	if ((features & dev->gso_partial_features) &&
+	    !(features & NETIF_F_GSO_PARTIAL)) {
+		netdev_dbg(dev,
+			   "Dropping partially supported GSO features since no GSO partial.\n");
+		features &= ~dev->gso_partial_features;
+	}
+
 #ifdef CONFIG_NET_RX_BUSY_POLL
 	if (dev->netdev_ops->ndo_busy_poll)
 		features |= NETIF_F_BUSY_POLL;
@@ -7011,7 +7041,7 @@ int register_netdevice(struct net_device *dev)
 
 	/* Make NETIF_F_SG inheritable to tunnel devices.
 	 */
-	dev->hw_enc_features |= NETIF_F_SG;
+	dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
 
 	/* Make NETIF_F_SG inheritable to MPLS.
 	 */

net/core/ethtool.c

@@ -88,6 +88,7 @@ static const char netdev_features_strings[NETDEV_FEATURE_COUNT][ETH_GSTRING_LEN]
 	[NETIF_F_GSO_SIT_BIT] =		 "tx-sit-segmentation",
 	[NETIF_F_GSO_UDP_TUNNEL_BIT] =	 "tx-udp_tnl-segmentation",
 	[NETIF_F_GSO_UDP_TUNNEL_CSUM_BIT] = "tx-udp_tnl-csum-segmentation",
+	[NETIF_F_GSO_PARTIAL_BIT] =	 "tx-gso-partial",
 
 	[NETIF_F_FCOE_CRC_BIT] =         "tx-checksum-fcoe-crc",
 	[NETIF_F_SCTP_CRC_BIT] =        "tx-checksum-sctp",

net/core/skbuff.c

@@ -3076,8 +3076,9 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,
 	struct sk_buff *frag_skb = head_skb;
 	unsigned int offset = doffset;
 	unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
+	unsigned int partial_segs = 0;
 	unsigned int headroom;
-	unsigned int len;
+	unsigned int len = head_skb->len;
 	__be16 proto;
 	bool csum;
 	int sg = !!(features & NETIF_F_SG);
@@ -3094,6 +3095,15 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,
 
 	csum = !!can_checksum_protocol(features, proto);
 
+	/* GSO partial only requires that we trim off any excess that
+	 * doesn't fit into an MSS sized block, so take care of that
+	 * now.
+	 */
+	if (features & NETIF_F_GSO_PARTIAL) {
+		partial_segs = len / mss;
+		mss *= partial_segs;
+	}
+
 	headroom = skb_headroom(head_skb);
 	pos = skb_headlen(head_skb);
 
@@ -3281,6 +3291,23 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,
 	 */
 	segs->prev = tail;
 
+	/* Update GSO info on first skb in partial sequence. */
+	if (partial_segs) {
+		int type = skb_shinfo(head_skb)->gso_type;
+
+		/* Update type to add partial and then remove dodgy if set */
+		type |= SKB_GSO_PARTIAL;
+		type &= ~SKB_GSO_DODGY;
+
+		/* Update GSO info and prepare to start updating headers on
+		 * our way back down the stack of protocols.
+		 */
+		skb_shinfo(segs)->gso_size = skb_shinfo(head_skb)->gso_size;
+		skb_shinfo(segs)->gso_segs = partial_segs;
+		skb_shinfo(segs)->gso_type = type;
+		SKB_GSO_CB(segs)->data_offset = skb_headroom(segs) + doffset;
+	}
+
 	/* Following permits correct backpressure, for protocols
 	 * using skb_set_owner_w().
 	 * Idea is to tranfert ownership from head_skb to last segment.

net/ipv4/af_inet.c

@@ -1200,7 +1200,7 @@ static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
 	const struct net_offload *ops;
 	unsigned int offset = 0;
 	struct iphdr *iph;
-	int proto;
+	int proto, tot_len;
 	int nhoff;
 	int ihl;
 	int id;
@@ -1219,6 +1219,7 @@ static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
 		       SKB_GSO_UDP_TUNNEL_CSUM |
 		       SKB_GSO_TCP_FIXEDID |
 		       SKB_GSO_TUNNEL_REMCSUM |
+		       SKB_GSO_PARTIAL |
 		       0)))
 		goto out;
 
@@ -1273,10 +1274,21 @@ static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
 			if (skb->next)
 				iph->frag_off |= htons(IP_MF);
 			offset += skb->len - nhoff - ihl;
-		} else if (!fixedid) {
+			tot_len = skb->len - nhoff;
+		} else if (skb_is_gso(skb)) {
+			if (!fixedid) {
+				iph->id = htons(id);
+				id += skb_shinfo(skb)->gso_segs;
+			}
+			tot_len = skb_shinfo(skb)->gso_size +
+				  SKB_GSO_CB(skb)->data_offset +
+				  skb->head - (unsigned char *)iph;
+		} else {
+			if (!fixedid)
 				iph->id = htons(id++);
+			tot_len = skb->len - nhoff;
 		}
-		iph->tot_len = htons(skb->len - nhoff);
+		iph->tot_len = htons(tot_len);
 		ip_send_check(iph);
 		if (encap)
 			skb_reset_inner_headers(skb);

net/ipv4/gre_offload.c

@@ -36,7 +36,8 @@ static struct sk_buff *gre_gso_segment(struct sk_buff *skb,
 				  SKB_GSO_GRE |
 				  SKB_GSO_GRE_CSUM |
 				  SKB_GSO_IPIP |
-				  SKB_GSO_SIT)))
+				  SKB_GSO_SIT |
+				  SKB_GSO_PARTIAL)))
 		goto out;
 
 	if (!skb->encapsulation)
@@ -87,7 +88,7 @@ static struct sk_buff *gre_gso_segment(struct sk_buff *skb,
 	skb = segs;
 	do {
 		struct gre_base_hdr *greh;
-		__be32 *pcsum;
+		__sum16 *pcsum;
 
 		/* Set up inner headers if we are offloading inner checksum */
 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
@@ -107,10 +108,25 @@ static struct sk_buff *gre_gso_segment(struct sk_buff *skb,
 			continue;
 
 		greh = (struct gre_base_hdr *)skb_transport_header(skb);
-		pcsum = (__be32 *)(greh + 1);
+		pcsum = (__sum16 *)(greh + 1);
 
+		if (skb_is_gso(skb)) {
+			unsigned int partial_adj;
+
+			/* Adjust checksum to account for the fact that
+			 * the partial checksum is based on actual size
+			 * whereas headers should be based on MSS size.
+			 */
+			partial_adj = skb->len + skb_headroom(skb) -
+				      SKB_GSO_CB(skb)->data_offset -
+				      skb_shinfo(skb)->gso_size;
+			*pcsum = ~csum_fold((__force __wsum)htonl(partial_adj));
+		} else {
 			*pcsum = 0;
-		*(__sum16 *)pcsum = gso_make_checksum(skb, 0);
+		}
+
+		*(pcsum + 1) = 0;
+		*pcsum = gso_make_checksum(skb, 0);
 	} while ((skb = skb->next));
 out:
 	return segs;

net/ipv4/tcp_offload.c

@@ -109,6 +109,12 @@ struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
 		goto out;
 	}
 
+	/* GSO partial only requires splitting the frame into an MSS
+	 * multiple and possibly a remainder.  So update the mss now.
+	 */
+	if (features & NETIF_F_GSO_PARTIAL)
+		mss = skb->len - (skb->len % mss);
+
 	copy_destructor = gso_skb->destructor == tcp_wfree;
 	ooo_okay = gso_skb->ooo_okay;
 	/* All segments but the first should have ooo_okay cleared */
@@ -133,7 +139,7 @@ struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
 	newcheck = ~csum_fold((__force __wsum)((__force u32)th->check +
 					       (__force u32)delta));
 
-	do {
+	while (skb->next) {
 		th->fin = th->psh = 0;
 		th->check = newcheck;
 
@@ -153,7 +159,7 @@ struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
 
 		th->seq = htonl(seq);
 		th->cwr = 0;
-	} while (skb->next);
+	}
 
 	/* Following permits TCP Small Queues to work well with GSO :
 	 * The callback to TCP stack will be called at the time last frag

net/ipv4/udp_offload.c

@@ -39,8 +39,11 @@ static struct sk_buff *__skb_udp_tunnel_segment(struct sk_buff *skb,
 	 * 16 bit length field due to the header being added outside of an
 	 * IP or IPv6 frame that was already limited to 64K - 1.
 	 */
-	partial = csum_sub(csum_unfold(uh->check),
-			   (__force __wsum)htonl(skb->len));
+	if (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)
+		partial = (__force __wsum)uh->len;
+	else
+		partial = (__force __wsum)htonl(skb->len);
+	partial = csum_sub(csum_unfold(uh->check), partial);
 
 	/* setup inner skb. */
 	skb->encapsulation = 0;
@@ -89,7 +92,7 @@ static struct sk_buff *__skb_udp_tunnel_segment(struct sk_buff *skb,
 	udp_offset = outer_hlen - tnl_hlen;
 	skb = segs;
 	do {
-		__be16 len;
+		unsigned int len;
 
 		if (remcsum)
 			skb->ip_summed = CHECKSUM_NONE;
@@ -107,14 +110,26 @@ static struct sk_buff *__skb_udp_tunnel_segment(struct sk_buff *skb,
 		skb_reset_mac_header(skb);
 		skb_set_network_header(skb, mac_len);
 		skb_set_transport_header(skb, udp_offset);
-		len = htons(skb->len - udp_offset);
+		len = skb->len - udp_offset;
 		uh = udp_hdr(skb);
-		uh->len = len;
+
+		/* If we are only performing partial GSO the inner header
+		 * will be using a length value equal to only one MSS sized
+		 * segment instead of the entire frame.
+		 */
+		if (skb_is_gso(skb)) {
+			uh->len = htons(skb_shinfo(skb)->gso_size +
+					SKB_GSO_CB(skb)->data_offset +
+					skb->head - (unsigned char *)uh);
+		} else {
+			uh->len = htons(len);
+		}
 
 		if (!need_csum)
 			continue;
 
-		uh->check = ~csum_fold(csum_add(partial, (__force __wsum)len));
+		uh->check = ~csum_fold(csum_add(partial,
+				       (__force __wsum)htonl(len)));
 
 		if (skb->encapsulation || !offload_csum) {
 			uh->check = gso_make_checksum(skb, ~uh->check);

net/ipv6/ip6_offload.c

@@ -63,6 +63,7 @@ static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb,
 	int proto;
 	struct frag_hdr *fptr;
 	unsigned int unfrag_ip6hlen;
+	unsigned int payload_len;
 	u8 *prevhdr;
 	int offset = 0;
 	bool encap, udpfrag;
@@ -82,6 +83,7 @@ static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb,
 		       SKB_GSO_UDP_TUNNEL |
 		       SKB_GSO_UDP_TUNNEL_CSUM |
 		       SKB_GSO_TUNNEL_REMCSUM |
+		       SKB_GSO_PARTIAL |
 		       0)))
 		goto out;
 
@@ -118,7 +120,13 @@ static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb,
 
 	for (skb = segs; skb; skb = skb->next) {
 		ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff);
-		ipv6h->payload_len = htons(skb->len - nhoff - sizeof(*ipv6h));
+		if (skb_is_gso(skb))
+			payload_len = skb_shinfo(skb)->gso_size +
+				      SKB_GSO_CB(skb)->data_offset +
+				      skb->head - (unsigned char *)(ipv6h + 1);
+		else
+			payload_len = skb->len - nhoff - sizeof(*ipv6h);
+		ipv6h->payload_len = htons(payload_len);
 		skb->network_header = (u8 *)ipv6h - skb->head;
 
 		if (udpfrag) {