Merge branch 'net-zerocopy-refine'

Willem de Bruijn says:

====================
zerocopy refinements

1/4 is a small optimization follow-up to the earlier fix to skb_segment:
    check skb state once per skb, instead of once per frag.
2/4 makes behavior more consistent between standard and zerocopy send:
    set the PSH bit when hitting MAX_SKB_FRAGS. This helps GRO.
3/4 resolves a surprising inconsistency in notification:
    because small packets were not stored in frags, they would not set
    the copied error code over loopback. This change also optimizes
    the path by removing copying and making tso_fragment cheaper.
4/4 follows-up to 3/4 by no longer allocated now unused memory.
    this was actually already in RFC patches, but dropped as I pared
    down the patch set during revisions.
====================

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

2 files changed, 24 insertions, 14 deletions

diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index a3cb0be4c6f3..00b0757830e2 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -3656,6 +3656,10 @@ normal:
 		skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags &
 					      SKBTX_SHARED_FRAG;
 
+		if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
+		    skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
+			goto err;
+
 		while (pos < offset + len) {
 			if (i >= nfrags) {
 				BUG_ON(skb_headlen(list_skb));
@@ -3667,6 +3671,11 @@ normal:
 
 				BUG_ON(!nfrags);
 
+				if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
+				    skb_zerocopy_clone(nskb, frag_skb,
+						       GFP_ATOMIC))
+					goto err;
+
 				list_skb = list_skb->next;
 			}
 
@@ -3678,11 +3687,6 @@ normal:
 				goto err;
 			}
 
-			if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC)))
-				goto err;
-			if (skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
-				goto err;
-
 			*nskb_frag = *frag;
 			__skb_frag_ref(nskb_frag);
 			size = skb_frag_size(nskb_frag);
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 67d39b79c801..7ac583a2b9fe 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -1104,12 +1104,15 @@ static int linear_payload_sz(bool first_skb)
 	return 0;
 }
 
-static int select_size(const struct sock *sk, bool sg, bool first_skb)
+static int select_size(const struct sock *sk, bool sg, bool first_skb, bool zc)
 {
 	const struct tcp_sock *tp = tcp_sk(sk);
 	int tmp = tp->mss_cache;
 
 	if (sg) {
+		if (zc)
+			return 0;
+
 		if (sk_can_gso(sk)) {
 			tmp = linear_payload_sz(first_skb);
 		} else {
@@ -1186,7 +1189,7 @@ int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)
 	int flags, err, copied = 0;
 	int mss_now = 0, size_goal, copied_syn = 0;
 	bool process_backlog = false;
-	bool sg;
+	bool sg, zc = false;
 	long timeo;
 
 	flags = msg->msg_flags;
@@ -1204,7 +1207,8 @@ int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)
 			goto out_err;
 		}
 
-		if (!(sk_check_csum_caps(sk) && sk->sk_route_caps & NETIF_F_SG))
+		zc = sk_check_csum_caps(sk) && sk->sk_route_caps & NETIF_F_SG;
+		if (!zc)
 			uarg->zerocopy = 0;
 	}
 
@@ -1281,6 +1285,7 @@ restart:
 
 		if (copy <= 0 || !tcp_skb_can_collapse_to(skb)) {
 			bool first_skb;
+			int linear;
 
 new_segment:
 			/* Allocate new segment. If the interface is SG,
@@ -1294,9 +1299,8 @@ new_segment:
 				goto restart;
 			}
 			first_skb = tcp_rtx_and_write_queues_empty(sk);
-			skb = sk_stream_alloc_skb(sk,
-						  select_size(sk, sg, first_skb),
-						  sk->sk_allocation,
+			linear = select_size(sk, sg, first_skb, zc);
+			skb = sk_stream_alloc_skb(sk, linear, sk->sk_allocation,
 						  first_skb);
 			if (!skb)
 				goto wait_for_memory;
@@ -1325,13 +1329,13 @@ new_segment:
 			copy = msg_data_left(msg);
 
 		/* Where to copy to? */
-		if (skb_availroom(skb) > 0) {
+		if (skb_availroom(skb) > 0 && !zc) {
 			/* We have some space in skb head. Superb! */
 			copy = min_t(int, copy, skb_availroom(skb));
 			err = skb_add_data_nocache(sk, skb, &msg->msg_iter, copy);
 			if (err)
 				goto do_fault;
-		} else if (!uarg || !uarg->zerocopy) {
+		} else if (!zc) {
 			bool merge = true;
 			int i = skb_shinfo(skb)->nr_frags;
 			struct page_frag *pfrag = sk_page_frag(sk);
@@ -1371,8 +1375,10 @@ new_segment:
 			pfrag->offset += copy;
 		} else {
 			err = skb_zerocopy_iter_stream(sk, skb, msg, copy, uarg);
-			if (err == -EMSGSIZE || err == -EEXIST)
+			if (err == -EMSGSIZE || err == -EEXIST) {
+				tcp_mark_push(tp, skb);
 				goto new_segment;
+			}
 			if (err < 0)
 				goto do_error;
 			copy = err;