net: implement lockless SO_MAX_PACING_RATE

SO_MAX_PACING_RATE setsockopt() does not need to hold
the socket lock, because sk->sk_pacing_rate readers
can run fine if the value is changed by other threads,
after adding READ_ONCE() accessors.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

include/trace/events/mptcp.h

@@ -44,7 +44,7 @@ TRACE_EVENT(mptcp_subflow_get_send,
 		ssk = mptcp_subflow_tcp_sock(subflow);
 		if (ssk && sk_fullsock(ssk)) {
 			__entry->snd_wnd = tcp_sk(ssk)->snd_wnd;
-			__entry->pace = ssk->sk_pacing_rate;
+			__entry->pace = READ_ONCE(ssk->sk_pacing_rate);
 		} else {
 			__entry->snd_wnd = 0;
 			__entry->pace = 0;

net/core/sock.c

@@ -1160,6 +1160,27 @@ int sk_setsockopt(struct sock *sk, int level, int optname,
 		WRITE_ONCE(sk->sk_busy_poll_budget, val);
 		return 0;
 #endif
+	case SO_MAX_PACING_RATE:
+		{
+		unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
+		unsigned long pacing_rate;
+
+		if (sizeof(ulval) != sizeof(val) &&
+		    optlen >= sizeof(ulval) &&
+		    copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
+			return -EFAULT;
+		}
+		if (ulval != ~0UL)
+			cmpxchg(&sk->sk_pacing_status,
+				SK_PACING_NONE,
+				SK_PACING_NEEDED);
+		/* Pairs with READ_ONCE() from sk_getsockopt() */
+		WRITE_ONCE(sk->sk_max_pacing_rate, ulval);
+		pacing_rate = READ_ONCE(sk->sk_pacing_rate);
+		if (ulval < pacing_rate)
+			WRITE_ONCE(sk->sk_pacing_rate, ulval);
+		return 0;
+		}
 	}
 
 	sockopt_lock_sock(sk);
@@ -1423,25 +1444,6 @@ int sk_setsockopt(struct sock *sk, int level, int optname,
 		break;
 
 
-	case SO_MAX_PACING_RATE:
-		{
-		unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
-
-		if (sizeof(ulval) != sizeof(val) &&
-		    optlen >= sizeof(ulval) &&
-		    copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
-			ret = -EFAULT;
-			break;
-		}
-		if (ulval != ~0UL)
-			cmpxchg(&sk->sk_pacing_status,
-				SK_PACING_NONE,
-				SK_PACING_NEEDED);
-		/* Pairs with READ_ONCE() from sk_getsockopt() */
-		WRITE_ONCE(sk->sk_max_pacing_rate, ulval);
-		sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
-		break;
-		}
 	case SO_INCOMING_CPU:
 		reuseport_update_incoming_cpu(sk, val);
 		break;

net/ipv4/tcp_bbr.c

@@ -258,7 +258,7 @@ static unsigned long bbr_bw_to_pacing_rate(struct sock *sk, u32 bw, int gain)
 	u64 rate = bw;
 
 	rate = bbr_rate_bytes_per_sec(sk, rate, gain);
-	rate = min_t(u64, rate, sk->sk_max_pacing_rate);
+	rate = min_t(u64, rate, READ_ONCE(sk->sk_max_pacing_rate));
 	return rate;
 }
 
@@ -278,7 +278,8 @@ static void bbr_init_pacing_rate_from_rtt(struct sock *sk)
 	}
 	bw = (u64)tcp_snd_cwnd(tp) * BW_UNIT;
 	do_div(bw, rtt_us);
-	sk->sk_pacing_rate = bbr_bw_to_pacing_rate(sk, bw, bbr_high_gain);
+	WRITE_ONCE(sk->sk_pacing_rate,
+		   bbr_bw_to_pacing_rate(sk, bw, bbr_high_gain));
 }
 
 /* Pace using current bw estimate and a gain factor. */
@@ -290,14 +291,14 @@ static void bbr_set_pacing_rate(struct sock *sk, u32 bw, int gain)
 
 	if (unlikely(!bbr->has_seen_rtt && tp->srtt_us))
 		bbr_init_pacing_rate_from_rtt(sk);
-	if (bbr_full_bw_reached(sk) || rate > sk->sk_pacing_rate)
-		sk->sk_pacing_rate = rate;
+	if (bbr_full_bw_reached(sk) || rate > READ_ONCE(sk->sk_pacing_rate))
+		WRITE_ONCE(sk->sk_pacing_rate, rate);
 }
 
 /* override sysctl_tcp_min_tso_segs */
 __bpf_kfunc static u32 bbr_min_tso_segs(struct sock *sk)
 {
-	return sk->sk_pacing_rate < (bbr_min_tso_rate >> 3) ? 1 : 2;
+	return READ_ONCE(sk->sk_pacing_rate) < (bbr_min_tso_rate >> 3) ? 1 : 2;
 }
 
 static u32 bbr_tso_segs_goal(struct sock *sk)
@@ -309,7 +310,7 @@ static u32 bbr_tso_segs_goal(struct sock *sk)
 	 * driver provided sk_gso_max_size.
 	 */
 	bytes = min_t(unsigned long,
-		      sk->sk_pacing_rate >> READ_ONCE(sk->sk_pacing_shift),
+		      READ_ONCE(sk->sk_pacing_rate) >> READ_ONCE(sk->sk_pacing_shift),
 		      GSO_LEGACY_MAX_SIZE - 1 - MAX_TCP_HEADER);
 	segs = max_t(u32, bytes / tp->mss_cache, bbr_min_tso_segs(sk));
 

net/ipv4/tcp_input.c

@@ -927,8 +927,8 @@ static void tcp_update_pacing_rate(struct sock *sk)
 	 * without any lock. We want to make sure compiler wont store
 	 * intermediate values in this location.
 	 */
-	WRITE_ONCE(sk->sk_pacing_rate, min_t(u64, rate,
-					     sk->sk_max_pacing_rate));
+	WRITE_ONCE(sk->sk_pacing_rate,
+		   min_t(u64, rate, READ_ONCE(sk->sk_max_pacing_rate)));
 }
 
 /* Calculate rto without backoff.  This is the second half of Van Jacobson's

net/ipv4/tcp_output.c

@@ -1201,7 +1201,7 @@ static void tcp_update_skb_after_send(struct sock *sk, struct sk_buff *skb,
 	struct tcp_sock *tp = tcp_sk(sk);
 
 	if (sk->sk_pacing_status != SK_PACING_NONE) {
-		unsigned long rate = sk->sk_pacing_rate;
+		unsigned long rate = READ_ONCE(sk->sk_pacing_rate);
 
 		/* Original sch_fq does not pace first 10 MSS
 		 * Note that tp->data_segs_out overflows after 2^32 packets,
@@ -1973,7 +1973,7 @@ static u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
 	unsigned long bytes;
 	u32 r;
 
-	bytes = sk->sk_pacing_rate >> READ_ONCE(sk->sk_pacing_shift);
+	bytes = READ_ONCE(sk->sk_pacing_rate) >> READ_ONCE(sk->sk_pacing_shift);
 
 	r = tcp_min_rtt(tcp_sk(sk)) >> READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_tso_rtt_log);
 	if (r < BITS_PER_TYPE(sk->sk_gso_max_size))
@@ -2553,7 +2553,7 @@ static bool tcp_small_queue_check(struct sock *sk, const struct sk_buff *skb,
 
 	limit = max_t(unsigned long,
 		      2 * skb->truesize,
-		      sk->sk_pacing_rate >> READ_ONCE(sk->sk_pacing_shift));
+		      READ_ONCE(sk->sk_pacing_rate) >> READ_ONCE(sk->sk_pacing_shift));
 	if (sk->sk_pacing_status == SK_PACING_NONE)
 		limit = min_t(unsigned long, limit,
 			      READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_limit_output_bytes));
@@ -2561,7 +2561,8 @@ static bool tcp_small_queue_check(struct sock *sk, const struct sk_buff *skb,
 
 	if (static_branch_unlikely(&tcp_tx_delay_enabled) &&
 	    tcp_sk(sk)->tcp_tx_delay) {
-		u64 extra_bytes = (u64)sk->sk_pacing_rate * tcp_sk(sk)->tcp_tx_delay;
+		u64 extra_bytes = (u64)READ_ONCE(sk->sk_pacing_rate) *
+				  tcp_sk(sk)->tcp_tx_delay;
 
 		/* TSQ is based on skb truesize sum (sk_wmem_alloc), so we
 		 * approximate our needs assuming an ~100% skb->truesize overhead.

net/sched/sch_fq.c

@@ -668,7 +668,7 @@ static struct sk_buff *fq_dequeue(struct Qdisc *sch)
 	 */
 	if (!skb->tstamp) {
 		if (skb->sk)
-			rate = min(skb->sk->sk_pacing_rate, rate);
+			rate = min(READ_ONCE(skb->sk->sk_pacing_rate), rate);
 
 		if (rate <= q->low_rate_threshold) {
 			f->credit = 0;