Merge branch 'net-gro-cleanups-and-fast-path-refinement'

Eric Dumazet says:

====================
net: gro: cleanups and fast path refinement

Current GRO stack has a 'fast path' for a subset of drivers,
users of napi_frags_skb().

With TCP zerocopy/direct uses, header split at receive is becoming
more important, and GRO fast path is disabled.

This series makes GRO (a bit) more efficient for almost all use cases.
====================

Link: https://lore.kernel.org/r/20240301193740.3436871-1-edumazet@google.comSigned-off-by: Paolo Abeni <pabeni@redhat.com>

drivers/net/geneve.c

@@ -508,7 +508,7 @@ static struct sk_buff *geneve_gro_receive(struct sock *sk,
 	gh_len = geneve_hlen(gh);
 
 	hlen = off_gnv + gh_len;
-	if (skb_gro_header_hard(skb, hlen)) {
+	if (!skb_gro_may_pull(skb, hlen)) {
 		gh = skb_gro_header_slow(skb, hlen, off_gnv);
 		if (unlikely(!gh))
 			goto out;

include/net/gro.h

@@ -139,21 +139,16 @@ static inline void skb_gro_pull(struct sk_buff *skb, unsigned int len)
 	NAPI_GRO_CB(skb)->data_offset += len;
 }
 
-static inline void *skb_gro_header_fast(struct sk_buff *skb,
+static inline void *skb_gro_header_fast(const struct sk_buff *skb,
 					unsigned int offset)
 {
 	return NAPI_GRO_CB(skb)->frag0 + offset;
 }
 
-static inline int skb_gro_header_hard(struct sk_buff *skb, unsigned int hlen)
+static inline bool skb_gro_may_pull(const struct sk_buff *skb,
+				    unsigned int hlen)
 {
-	return NAPI_GRO_CB(skb)->frag0_len < hlen;
-}
-
-static inline void skb_gro_frag0_invalidate(struct sk_buff *skb)
-{
-	NAPI_GRO_CB(skb)->frag0 = NULL;
-	NAPI_GRO_CB(skb)->frag0_len = 0;
+	return likely(hlen <= NAPI_GRO_CB(skb)->frag0_len);
 }
 
 static inline void *skb_gro_header_slow(struct sk_buff *skb, unsigned int hlen,
@@ -162,28 +157,30 @@ static inline void *skb_gro_header_slow(struct sk_buff *skb, unsigned int hlen,
 	if (!pskb_may_pull(skb, hlen))
 		return NULL;
 
-	skb_gro_frag0_invalidate(skb);
 	return skb->data + offset;
 }
 
-static inline void *skb_gro_header(struct sk_buff *skb,
-					unsigned int hlen, unsigned int offset)
+static inline void *skb_gro_header(struct sk_buff *skb, unsigned int hlen,
+				   unsigned int offset)
 {
 	void *ptr;
 
 	ptr = skb_gro_header_fast(skb, offset);
-	if (skb_gro_header_hard(skb, hlen))
+	if (!skb_gro_may_pull(skb, hlen))
 		ptr = skb_gro_header_slow(skb, hlen, offset);
 	return ptr;
 }
 
-static inline void *skb_gro_network_header(struct sk_buff *skb)
+static inline void *skb_gro_network_header(const struct sk_buff *skb)
 {
-	return (NAPI_GRO_CB(skb)->frag0 ?: skb->data) +
-	       skb_network_offset(skb);
+	if (skb_gro_may_pull(skb, skb_gro_offset(skb)))
+		return skb_gro_header_fast(skb, skb_network_offset(skb));
+
+	return skb_network_header(skb);
 }
 
-static inline __wsum inet_gro_compute_pseudo(struct sk_buff *skb, int proto)
+static inline __wsum inet_gro_compute_pseudo(const struct sk_buff *skb,
+					     int proto)
 {
 	const struct iphdr *iph = skb_gro_network_header(skb);
 
@@ -421,7 +418,8 @@ static inline struct udphdr *udp_gro_udphdr(struct sk_buff *skb)
 	return uh;
 }
 
-static inline __wsum ip6_gro_compute_pseudo(struct sk_buff *skb, int proto)
+static inline __wsum ip6_gro_compute_pseudo(const struct sk_buff *skb,
+					    int proto)
 {
 	const struct ipv6hdr *iph = skb_gro_network_header(skb);
 

net/core/gro.c

@@ -369,15 +369,21 @@ static void gro_list_prepare(const struct list_head *head,
 
 static inline void skb_gro_reset_offset(struct sk_buff *skb, u32 nhoff)
 {
-	const struct skb_shared_info *pinfo = skb_shinfo(skb);
-	const skb_frag_t *frag0 = &pinfo->frags[0];
+	const struct skb_shared_info *pinfo;
+	const skb_frag_t *frag0;
+	unsigned int headlen;
 
 	NAPI_GRO_CB(skb)->data_offset = 0;
-	NAPI_GRO_CB(skb)->frag0 = NULL;
-	NAPI_GRO_CB(skb)->frag0_len = 0;
+	headlen = skb_headlen(skb);
+	NAPI_GRO_CB(skb)->frag0 = skb->data;
+	NAPI_GRO_CB(skb)->frag0_len = headlen;
+	if (headlen)
+		return;
+
+	pinfo = skb_shinfo(skb);
+	frag0 = &pinfo->frags[0];
 
-	if (!skb_headlen(skb) && pinfo->nr_frags &&
-	    !PageHighMem(skb_frag_page(frag0)) &&
+	if (pinfo->nr_frags && !PageHighMem(skb_frag_page(frag0)) &&
 	    (!NET_IP_ALIGN || !((skb_frag_off(frag0) + nhoff) & 3))) {
 		NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
 		NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
@@ -700,7 +706,7 @@ static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
 	skb_reset_mac_header(skb);
 	skb_gro_reset_offset(skb, hlen);
 
-	if (unlikely(skb_gro_header_hard(skb, hlen))) {
+	if (unlikely(!skb_gro_may_pull(skb, hlen))) {
 		eth = skb_gro_header_slow(skb, hlen, 0);
 		if (unlikely(!eth)) {
 			net_warn_ratelimited("%s: dropping impossible skb from %s\n",
@@ -710,7 +716,10 @@ static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
 		}
 	} else {
 		eth = (const struct ethhdr *)skb->data;
-		gro_pull_from_frag0(skb, hlen);
+
+		if (NAPI_GRO_CB(skb)->frag0 != skb->data)
+			gro_pull_from_frag0(skb, hlen);
+
 		NAPI_GRO_CB(skb)->frag0 += hlen;
 		NAPI_GRO_CB(skb)->frag0_len -= hlen;
 	}

net/ipv4/fou_core.c

@@ -351,7 +351,7 @@ static struct sk_buff *gue_gro_receive(struct sock *sk,
 	optlen = guehdr->hlen << 2;
 	len += optlen;
 
-	if (skb_gro_header_hard(skb, len)) {
+	if (!skb_gro_may_pull(skb, len)) {
 		guehdr = skb_gro_header_slow(skb, len, off);
 		if (unlikely(!guehdr))
 			goto out;

net/ipv4/gre_offload.c

@@ -174,7 +174,7 @@ static struct sk_buff *gre_gro_receive(struct list_head *head,
 		grehlen += GRE_HEADER_SECTION;
 
 	hlen = off + grehlen;
-	if (skb_gro_header_hard(skb, hlen)) {
+	if (!skb_gro_may_pull(skb, hlen)) {
 		greh = skb_gro_header_slow(skb, hlen, off);
 		if (unlikely(!greh))
 			goto out;

net/ipv4/tcp_offload.c

@@ -204,7 +204,7 @@ struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb)
 		goto out;
 
 	hlen = off + thlen;
-	if (skb_gro_header_hard(skb, hlen)) {
+	if (!skb_gro_may_pull(skb, hlen)) {
 		th = skb_gro_header_slow(skb, hlen, off);
 		if (unlikely(!th))
 			goto out;
@@ -299,18 +299,20 @@ struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb)
 void tcp_gro_complete(struct sk_buff *skb)
 {
 	struct tcphdr *th = tcp_hdr(skb);
+	struct skb_shared_info *shinfo;
+
+	if (skb->encapsulation)
+		skb->inner_transport_header = skb->transport_header;
 
 	skb->csum_start = (unsigned char *)th - skb->head;
 	skb->csum_offset = offsetof(struct tcphdr, check);
 	skb->ip_summed = CHECKSUM_PARTIAL;
 
-	skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
+	shinfo = skb_shinfo(skb);
+	shinfo->gso_segs = NAPI_GRO_CB(skb)->count;
 
 	if (th->cwr)
-		skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
-
-	if (skb->encapsulation)
-		skb->inner_transport_header = skb->transport_header;
+		shinfo->gso_type |= SKB_GSO_TCP_ECN;
 }
 EXPORT_SYMBOL(tcp_gro_complete);
 
@@ -335,10 +337,9 @@ INDIRECT_CALLABLE_SCOPE int tcp4_gro_complete(struct sk_buff *skb, int thoff)
 
 	th->check = ~tcp_v4_check(skb->len - thoff, iph->saddr,
 				  iph->daddr, 0);
-	skb_shinfo(skb)->gso_type |= SKB_GSO_TCPV4;
 
-	if (NAPI_GRO_CB(skb)->is_atomic)
-		skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_FIXEDID;
+	skb_shinfo(skb)->gso_type |= SKB_GSO_TCPV4 |
+			(NAPI_GRO_CB(skb)->is_atomic * SKB_GSO_TCP_FIXEDID);
 
 	tcp_gro_complete(skb);
 	return 0;