3 files changed, 88 insertions, 4 deletions

diff --git a/net/ipv4/tcp_bbr.c b/net/ipv4/tcp_bbr.c
index b89bce4c721e..92b045c72163 100644
--- a/net/ipv4/tcp_bbr.c
+++ b/net/ipv4/tcp_bbr.c
@@ -52,10 +52,9 @@
  * There is a public e-mail list for discussing BBR development and testing:
  *   https://groups.google.com/forum/#!forum/bbr-dev
  *
- * NOTE: BBR *must* be used with the fq qdisc ("man tc-fq") with pacing enabled,
- * since pacing is integral to the BBR design and implementation.
- * BBR without pacing would not function properly, and may incur unnecessary
- * high packet loss rates.
+ * NOTE: BBR might be used with the fq qdisc ("man tc-fq") with pacing enabled,
+ * otherwise TCP stack falls back to an internal pacing using one high
+ * resolution timer per TCP socket and may use more resources.
  */
 #include <linux/module.h>
 #include <net/tcp.h>
@@ -830,6 +829,8 @@ static void bbr_init(struct sock *sk)
 	bbr->cycle_idx = 0;
 	bbr_reset_lt_bw_sampling(sk);
 	bbr_reset_startup_mode(sk);
+
+	cmpxchg(&sk->sk_pacing_status, SK_PACING_NONE, SK_PACING_NEEDED);
 }
 
 static u32 bbr_sndbuf_expand(struct sock *sk)
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 4858e190f6ac..a32172d69a03 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -904,6 +904,72 @@ out:
 	sk_free(sk);
 }
 
+/* Note: Called under hard irq.
+ * We can not call TCP stack right away.
+ */
+enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer)
+{
+	struct tcp_sock *tp = container_of(timer, struct tcp_sock, pacing_timer);
+	struct sock *sk = (struct sock *)tp;
+	unsigned long nval, oval;
+
+	for (oval = READ_ONCE(sk->sk_tsq_flags);; oval = nval) {
+		struct tsq_tasklet *tsq;
+		bool empty;
+
+		if (oval & TSQF_QUEUED)
+			break;
+
+		nval = (oval & ~TSQF_THROTTLED) | TSQF_QUEUED | TCPF_TSQ_DEFERRED;
+		nval = cmpxchg(&sk->sk_tsq_flags, oval, nval);
+		if (nval != oval)
+			continue;
+
+		if (!atomic_inc_not_zero(&sk->sk_wmem_alloc))
+			break;
+		/* queue this socket to tasklet queue */
+		tsq = this_cpu_ptr(&tsq_tasklet);
+		empty = list_empty(&tsq->head);
+		list_add(&tp->tsq_node, &tsq->head);
+		if (empty)
+			tasklet_schedule(&tsq->tasklet);
+		break;
+	}
+	return HRTIMER_NORESTART;
+}
+
+/* BBR congestion control needs pacing.
+ * Same remark for SO_MAX_PACING_RATE.
+ * sch_fq packet scheduler is efficiently handling pacing,
+ * but is not always installed/used.
+ * Return true if TCP stack should pace packets itself.
+ */
+static bool tcp_needs_internal_pacing(const struct sock *sk)
+{
+	return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
+}
+
+static void tcp_internal_pacing(struct sock *sk, const struct sk_buff *skb)
+{
+	u64 len_ns;
+	u32 rate;
+
+	if (!tcp_needs_internal_pacing(sk))
+		return;
+	rate = sk->sk_pacing_rate;
+	if (!rate || rate == ~0U)
+		return;
+
+	/* Should account for header sizes as sch_fq does,
+	 * but lets make things simple.
+	 */
+	len_ns = (u64)skb->len * NSEC_PER_SEC;
+	do_div(len_ns, rate);
+	hrtimer_start(&tcp_sk(sk)->pacing_timer,
+		      ktime_add_ns(ktime_get(), len_ns),
+		      HRTIMER_MODE_ABS_PINNED);
+}
+
 /* This routine actually transmits TCP packets queued in by
  * tcp_do_sendmsg().  This is used by both the initial
  * transmission and possible later retransmissions.
@@ -1034,6 +1100,7 @@ static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it,
 	if (skb->len != tcp_header_size) {
 		tcp_event_data_sent(tp, sk);
 		tp->data_segs_out += tcp_skb_pcount(skb);
+		tcp_internal_pacing(sk, skb);
 	}
 
 	if (after(tcb->end_seq, tp->snd_nxt) || tcb->seq == tcb->end_seq)
@@ -2086,6 +2153,12 @@ static int tcp_mtu_probe(struct sock *sk)
 	return -1;
 }
 
+static bool tcp_pacing_check(const struct sock *sk)
+{
+	return tcp_needs_internal_pacing(sk) &&
+	       hrtimer_active(&tcp_sk(sk)->pacing_timer);
+}
+
 /* TCP Small Queues :
  * Control number of packets in qdisc/devices to two packets / or ~1 ms.
  * (These limits are doubled for retransmits)
@@ -2210,6 +2283,9 @@ static bool tcp_write_xmit(struct sock *sk, unsigned int mss_now, int nonagle,
 	while ((skb = tcp_send_head(sk))) {
 		unsigned int limit;
 
+		if (tcp_pacing_check(sk))
+			break;
+
 		tso_segs = tcp_init_tso_segs(skb, mss_now);
 		BUG_ON(!tso_segs);
 
@@ -2878,6 +2954,10 @@ void tcp_xmit_retransmit_queue(struct sock *sk)
 
 		if (skb == tcp_send_head(sk))
 			break;
+
+		if (tcp_pacing_check(sk))
+			break;
+
 		/* we could do better than to assign each time */
 		if (!hole)
 			tp->retransmit_skb_hint = skb;
diff --git a/net/ipv4/tcp_timer.c b/net/ipv4/tcp_timer.c
index 14672543cf0b..86934bcf685a 100644
--- a/net/ipv4/tcp_timer.c
+++ b/net/ipv4/tcp_timer.c
@@ -710,4 +710,7 @@ void tcp_init_xmit_timers(struct sock *sk)
 {
 	inet_csk_init_xmit_timers(sk, &tcp_write_timer, &tcp_delack_timer,
 				  &tcp_keepalive_timer);
+	hrtimer_init(&tcp_sk(sk)->pacing_timer, CLOCK_MONOTONIC,
+		     HRTIMER_MODE_ABS_PINNED);
+	tcp_sk(sk)->pacing_timer.function = tcp_pace_kick;
 }