tcp: do not pace pure ack packets

When we added pacing to TCP, we decided to let sch_fq take care
of actual pacing.

All TCP had to do was to compute sk->pacing_rate using simple formula:

sk->pacing_rate = 2 * cwnd * mss / rtt

It works well for senders (bulk flows), but not very well for receivers
or even RPC :

cwnd on the receiver can be less than 10, rtt can be around 100ms, so we
can end up pacing ACK packets, slowing down the sender.

Really, only the sender should pace, according to its own logic.

Instead of adding a new bit in skb, or call yet another flow
dissection, we tweak skb->truesize to a small value (2), and
we instruct sch_fq to use new helper and not pace pure ack.

Note this also helps TCP small queue, as ack packets present
in qdisc/NIC do not prevent sending a data packet (RPC workload)

This helps to reduce tx completion overhead, ack packets can use regular
sock_wfree() instead of tcp_wfree() which is a bit more expensive.

This has no impact in the case packets are sent to loopback interface,
as we do not coalesce ack packets (were we would detect skb->truesize
lie)

In case netem (with a delay) is used, skb_orphan_partial() also sets
skb->truesize to 1.

This patch is a combination of two patches we used for about one year at
Google.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

include/net/tcp.h

@@ -1713,4 +1713,19 @@ static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
 	return dopt;
 }
 
+/* locally generated TCP pure ACKs have skb->truesize == 2
+ * (check tcp_send_ack() in net/ipv4/tcp_output.c )
+ * This is much faster than dissecting the packet to find out.
+ * (Think of GRE encapsulations, IPv4, IPv6, ...)
+ */
+static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
+{
+	return skb->truesize == 2;
+}
+
+static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
+{
+	skb->truesize = 2;
+}
+
 #endif	/* _TCP_H */

net/ipv4/tcp_output.c

@@ -948,7 +948,7 @@ static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it,
 
 	skb_orphan(skb);
 	skb->sk = sk;
-	skb->destructor = tcp_wfree;
+	skb->destructor = skb_is_tcp_pure_ack(skb) ? sock_wfree : tcp_wfree;
 	skb_set_hash_from_sk(skb, sk);
 	atomic_add(skb->truesize, &sk->sk_wmem_alloc);
 
@@ -3265,6 +3265,14 @@ void tcp_send_ack(struct sock *sk)
 	skb_reserve(buff, MAX_TCP_HEADER);
 	tcp_init_nondata_skb(buff, tcp_acceptable_seq(sk), TCPHDR_ACK);
 
+	/* We do not want pure acks influencing TCP Small Queues or fq/pacing
+	 * too much.
+	 * SKB_TRUESIZE(max(1 .. 66, MAX_TCP_HEADER)) is unfortunately ~784
+	 * We also avoid tcp_wfree() overhead (cache line miss accessing
+	 * tp->tsq_flags) by using regular sock_wfree()
+	 */
+	skb_set_tcp_pure_ack(buff);
+
 	/* Send it off, this clears delayed acks for us. */
 	skb_mstamp_get(&buff->skb_mstamp);
 	tcp_transmit_skb(sk, buff, 0, sk_gfp_atomic(sk, GFP_ATOMIC));

net/sched/sch_fq.c

@@ -52,6 +52,7 @@
 #include <net/pkt_sched.h>
 #include <net/sock.h>
 #include <net/tcp_states.h>
+#include <net/tcp.h>
 
 /*
  * Per flow structure, dynamically allocated
@@ -445,7 +446,9 @@ static struct sk_buff *fq_dequeue(struct Qdisc *sch)
 		goto begin;
 	}
 
-	if (unlikely(f->head && now < f->time_next_packet)) {
+	skb = f->head;
+	if (unlikely(skb && now < f->time_next_packet &&
+		     !skb_is_tcp_pure_ack(skb))) {
 		head->first = f->next;
 		fq_flow_set_throttled(q, f);
 		goto begin;
@@ -464,12 +467,15 @@ static struct sk_buff *fq_dequeue(struct Qdisc *sch)
 		goto begin;
 	}
 	prefetch(&skb->end);
-	f->time_next_packet = now;
 	f->credit -= qdisc_pkt_len(skb);
 
 	if (f->credit > 0 || !q->rate_enable)
 		goto out;
 
+	/* Do not pace locally generated ack packets */
+	if (skb_is_tcp_pure_ack(skb))
+		goto out;
+
 	rate = q->flow_max_rate;
 	if (skb->sk)
 		rate = min(skb->sk->sk_pacing_rate, rate);