From 93e16ea025d234d0ed01d9dc9c819257a2159bb6 Mon Sep 17 00:00:00 2001 From: Eric Dumazet Date: Fri, 1 Mar 2024 19:37:37 +0000 Subject: [PATCH 1/4] net: gro: rename skb_gro_header_hard() skb_gro_header_hard() is renamed to skb_gro_may_pull() to match the convention used by common helpers like pskb_may_pull(). This means the condition is inverted: if (skb_gro_header_hard(skb, hlen)) slow_path(); becomes: if (!skb_gro_may_pull(skb, hlen)) slow_path(); Signed-off-by: Eric Dumazet Acked-by: Paolo Abeni Signed-off-by: Paolo Abeni --- drivers/net/geneve.c | 2 +- include/net/gro.h | 7 ++++--- net/core/gro.c | 2 +- net/ipv4/fou_core.c | 2 +- net/ipv4/gre_offload.c | 2 +- net/ipv4/tcp_offload.c | 2 +- 6 files changed, 9 insertions(+), 8 deletions(-) diff --git a/drivers/net/geneve.c b/drivers/net/geneve.c index 6f3f9b446b1d..e25e0a31126c 100644 --- a/drivers/net/geneve.c +++ b/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; diff --git a/include/net/gro.h b/include/net/gro.h index b435f0ddbf64..ffc2c96d263b 100644 --- a/include/net/gro.h +++ b/include/net/gro.h @@ -145,9 +145,10 @@ static inline void *skb_gro_header_fast(struct sk_buff *skb, 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; + return hlen <= NAPI_GRO_CB(skb)->frag0_len; } static inline void skb_gro_frag0_invalidate(struct sk_buff *skb) @@ -172,7 +173,7 @@ static inline void *skb_gro_header(struct sk_buff *skb, 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; } diff --git a/net/core/gro.c b/net/core/gro.c index 0759277dc14e..927ccf681490 100644 --- a/net/core/gro.c +++ b/net/core/gro.c @@ -700,7 +700,7 @@ 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", diff --git a/net/ipv4/fou_core.c b/net/ipv4/fou_core.c index 0c41076e31ed..a8494f796dca 100644 --- a/net/ipv4/fou_core.c +++ b/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; diff --git a/net/ipv4/gre_offload.c b/net/ipv4/gre_offload.c index 311e70bfce40..5028c72d494a 100644 --- a/net/ipv4/gre_offload.c +++ b/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; diff --git a/net/ipv4/tcp_offload.c b/net/ipv4/tcp_offload.c index 8311c38267b5..875456efc92d 100644 --- a/net/ipv4/tcp_offload.c +++ b/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; From bd56a29c7a4ebcd3ca69505a2e676449e60965f3 Mon Sep 17 00:00:00 2001 From: Eric Dumazet Date: Fri, 1 Mar 2024 19:37:38 +0000 Subject: [PATCH 2/4] net: gro: change skb_gro_network_header() Change skb_gro_network_header() to accept a const sk_buff and to no longer check if frag0 is NULL or not. This allows to remove skb_gro_frag0_invalidate() which is seen in profiles when header-split is enabled. sk_buff parameter is constified for skb_gro_header_fast(), inet_gro_compute_pseudo() and ip6_gro_compute_pseudo(). Signed-off-by: Eric Dumazet Acked-by: Paolo Abeni Signed-off-by: Paolo Abeni --- include/net/gro.h | 27 ++++++++++++--------------- 1 file changed, 12 insertions(+), 15 deletions(-) diff --git a/include/net/gro.h b/include/net/gro.h index ffc2c96d263b..3c3666e46b30 100644 --- a/include/net/gro.h +++ b/include/net/gro.h @@ -139,7 +139,7 @@ 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; @@ -151,24 +151,17 @@ static inline bool skb_gro_may_pull(const struct sk_buff *skb, return hlen <= NAPI_GRO_CB(skb)->frag0_len; } -static inline void skb_gro_frag0_invalidate(struct sk_buff *skb) -{ - NAPI_GRO_CB(skb)->frag0 = NULL; - NAPI_GRO_CB(skb)->frag0_len = 0; -} - static inline void *skb_gro_header_slow(struct sk_buff *skb, unsigned int hlen, unsigned int offset) { 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; @@ -178,13 +171,16 @@ static inline void *skb_gro_header(struct sk_buff *skb, 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); @@ -422,7 +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); From c7583e9f768eeb82f2531c8372584ba89cfade8b Mon Sep 17 00:00:00 2001 From: Eric Dumazet Date: Fri, 1 Mar 2024 19:37:39 +0000 Subject: [PATCH 3/4] net: gro: enable fast path for more cases Currently the so-called GRO fast path is only enabled for napi_frags_skb() callers. After the prior patch, we no longer have to clear frag0 whenever we pulled bytes to skb->head. We therefore can initialize frag0 to skb->data so that GRO fast path can be used in the following additional cases: - Drivers using header split (populating skb->data with headers, and having payload in one or more page fragments). - Drivers not using any page frag (entire packet is in skb->data) Add a likely() in skb_gro_may_pull() to help the compiler to generate better code if possible. Signed-off-by: Eric Dumazet Acked-by: Paolo Abeni Signed-off-by: Paolo Abeni --- include/net/gro.h | 2 +- net/core/gro.c | 23 ++++++++++++++++------- 2 files changed, 17 insertions(+), 8 deletions(-) diff --git a/include/net/gro.h b/include/net/gro.h index 3c3666e46b30..2b58671a6549 100644 --- a/include/net/gro.h +++ b/include/net/gro.h @@ -148,7 +148,7 @@ static inline void *skb_gro_header_fast(const struct sk_buff *skb, static inline bool skb_gro_may_pull(const struct sk_buff *skb, unsigned int hlen) { - return hlen <= NAPI_GRO_CB(skb)->frag0_len; + return likely(hlen <= NAPI_GRO_CB(skb)->frag0_len); } static inline void *skb_gro_header_slow(struct sk_buff *skb, unsigned int hlen, diff --git a/net/core/gro.c b/net/core/gro.c index 927ccf681490..6a0edbd826a1 100644 --- a/net/core/gro.c +++ b/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; - if (!skb_headlen(skb) && pinfo->nr_frags && - !PageHighMem(skb_frag_page(frag0)) && + pinfo = skb_shinfo(skb); + frag0 = &pinfo->frags[0]; + + 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, @@ -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; } From 8f78010b701d8fdc290063f29fefc09aaaca4085 Mon Sep 17 00:00:00 2001 From: Eric Dumazet Date: Fri, 1 Mar 2024 19:37:40 +0000 Subject: [PATCH 4/4] tcp: gro: micro optimizations in tcp[4]_gro_complete() In tcp_gro_complete() : Moving the skb->inner_transport_header setting allows the compiler to reuse the previously loaded value of skb->transport_header. Caching skb_shinfo() avoids duplications as well. In tcp4_gro_complete(), doing a single change on skb_shinfo(skb)->gso_type also generates better code. Signed-off-by: Eric Dumazet Acked-by: Paolo Abeni Signed-off-by: Paolo Abeni --- net/ipv4/tcp_offload.c | 17 +++++++++-------- 1 file changed, 9 insertions(+), 8 deletions(-) diff --git a/net/ipv4/tcp_offload.c b/net/ipv4/tcp_offload.c index 875456efc92d..b955ab3b236d 100644 --- a/net/ipv4/tcp_offload.c +++ b/net/ipv4/tcp_offload.c @@ -299,18 +299,20 @@ out: 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;