Merge branch 'inet_csums'

Tom Herbert says:

====================
net: Checksum offload changes - Part IV

I am working on overhauling RX checksum offload. Goals of this effort
are:

- Specify what exactly it means when driver returns CHECKSUM_UNNECESSARY
- Preserve CHECKSUM_COMPLETE through encapsulation layers
- Don't do skb_checksum more than once per packet
- Unify GRO and non-GRO csum verification as much as possible
- Unify the checksum functions (checksum_init)
- Simply code

What is in this fourth patch set:

- Preserve CHECKSUM_COMPLETE instead of changing it to
  CHECKSUM_UNNECESSARY. This allows correct reuse in validating multiple
  csums in a packet.
- When SW needs to compute the packet checksum, save it as
  CHECKSUM_COMPLETE. Also mark that checksum was compute by SW.
- Add skb_gro_postpull_rcsum to udp and vxlan to make GRO work with
  CHECKSUM_COMPLETE.

v2: Removed patch setting skb_encapsulation when validating checksum
    in tcp_gro_receive

Please review carefully and test if possible, mucking with basic
checksum functions is always a little precarious :-)
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/vxlan.c

@@ -565,6 +565,7 @@ static struct sk_buff **vxlan_gro_receive(struct sk_buff **head, struct sk_buff
 			goto out;
 	}
 	skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
+	skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 
 	off_eth = skb_gro_offset(skb);
 	hlen = off_eth + sizeof(*eh);
@@ -599,6 +600,7 @@ static struct sk_buff **vxlan_gro_receive(struct sk_buff **head, struct sk_buff
 	}
 
 	skb_gro_pull(skb, sizeof(*eh)); /* pull inner eth header */
+	skb_gro_postpull_rcsum(skb, eh, sizeof(*eh));
 	pp = ptype->callbacks.gro_receive(head, skb);
 
 out_unlock:

include/linux/skbuff.h

@@ -572,7 +572,9 @@ struct sk_buff {
 	 */
 	__u8			encapsulation:1;
 	__u8			encap_hdr_csum:1;
-	/* 5/7 bit hole (depending on ndisc_nodetype presence) */
+	__u8			csum_valid:1;
+	__u8			csum_complete_sw:1;
+	/* 3/5 bit hole (depending on ndisc_nodetype presence) */
 	kmemcheck_bitfield_end(flags2);
 
 #if defined CONFIG_NET_DMA || defined CONFIG_NET_RX_BUSY_POLL
@@ -2735,7 +2737,7 @@ __sum16 __skb_checksum_complete(struct sk_buff *skb);
 
 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
 {
-	return skb->ip_summed & CHECKSUM_UNNECESSARY;
+	return ((skb->ip_summed & CHECKSUM_UNNECESSARY) || skb->csum_valid);
 }
 
 /**
@@ -2769,10 +2771,8 @@ static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
 						  bool zero_okay,
 						  __sum16 check)
 {
-	if (skb_csum_unnecessary(skb)) {
-		return false;
-	} else if (zero_okay && !check) {
-		skb->ip_summed = CHECKSUM_UNNECESSARY;
+	if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
+		skb->csum_valid = 1;
 		return false;
 	}
 
@@ -2799,15 +2799,20 @@ static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
 {
 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
 		if (!csum_fold(csum_add(psum, skb->csum))) {
-			skb->ip_summed = CHECKSUM_UNNECESSARY;
+			skb->csum_valid = 1;
 			return 0;
 		}
 	}
 
 	skb->csum = psum;
 
-	if (complete || skb->len <= CHECKSUM_BREAK)
-		return __skb_checksum_complete(skb);
+	if (complete || skb->len <= CHECKSUM_BREAK) {
+		__sum16 csum;
+
+		csum = __skb_checksum_complete(skb);
+		skb->csum_valid = !csum;
+		return csum;
+	}
 
 	return 0;
 }
@@ -2831,6 +2836,7 @@ static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
 				zero_okay, check, compute_pseudo)	\
 ({									\
 	__sum16 __ret = 0;						\
+	skb->csum_valid = 0;						\
 	if (__skb_checksum_validate_needed(skb, zero_okay, check))	\
 		__ret = __skb_checksum_validate_complete(skb,		\
 				complete, compute_pseudo(skb, proto));	\

net/core/datagram.c

@@ -739,11 +739,15 @@ __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
 	__sum16 sum;
 
 	sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
-	if (likely(!sum)) {
-		if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
+	if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && !sum &&
+	    !skb->csum_complete_sw)
 		netdev_rx_csum_fault(skb->dev);
-		skb->ip_summed = CHECKSUM_UNNECESSARY;
-	}
+
+	/* Save checksum complete for later use */
+	skb->csum = sum;
+	skb->ip_summed = CHECKSUM_COMPLETE;
+	skb->csum_complete_sw = 1;
+
 	return sum;
 }
 EXPORT_SYMBOL(__skb_checksum_complete_head);

net/ipv4/gre_offload.c

@@ -131,10 +131,12 @@ static __sum16 gro_skb_checksum(struct sk_buff *skb)
 		csum_partial(skb->data, skb_gro_offset(skb), 0));
 	sum = csum_fold(NAPI_GRO_CB(skb)->csum);
 	if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE)) {
-		if (unlikely(!sum))
+		if (unlikely(!sum) && !skb->csum_complete_sw)
 			netdev_rx_csum_fault(skb->dev);
-	} else
+	} else {
 		skb->ip_summed = CHECKSUM_COMPLETE;
+		skb->csum_complete_sw = 1;
+	}
 
 	return sum;
 }

net/ipv4/udp_offload.c

@@ -200,6 +200,7 @@ static struct sk_buff **udp_gro_receive(struct sk_buff **head, struct sk_buff *s
 	}
 
 	skb_gro_pull(skb, sizeof(struct udphdr)); /* pull encapsulating udp header */
+	skb_gro_postpull_rcsum(skb, uh, sizeof(struct udphdr));
 	pp = uo_priv->offload->callbacks.gro_receive(head, skb);
 
 out_unlock:

net/sunrpc/socklib.c

@@ -173,7 +173,8 @@ int csum_partial_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 		return -1;
 	if (csum_fold(desc.csum))
 		return -1;
-	if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
+	if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
+	    !skb->csum_complete_sw)
 		netdev_rx_csum_fault(skb->dev);
 	return 0;
 no_checksum: