Merge branch 'tcp-second-round-for-EDT-conversion'

Eric Dumazet says:

====================
tcp: second round for EDT conversion

First round of EDT patches left TCP stack in a non optimal state.

- High speed flows suffered from loss of performance, addressed
  by the first patch of this series.

- Second patch brings pacing to the current state of networking,
  since we now reach ~100 Gbit on a single TCP flow.

- Third patch implements a mitigation for scheduling delays,
  like the one we did in sch_fq in the past.

- Fourth patch removes one special case in sch_fq for ACK packets.

- Fifth patch removes a serious perfomance cost for TCP internal
  pacing. We should setup the high resolution timer only if
  really needed.

- Sixth patch fixes a typo in BBR.

- Last patch is one minor change in cdg congestion control.

Neal Cardwell also has a patch series fixing BBR after
EDT adoption.
====================

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

include/linux/tcp.h

@@ -249,6 +249,7 @@ struct tcp_sock {
 	u32	tlp_high_seq;	/* snd_nxt at the time of TLP retransmit. */
 
 	u64	tcp_wstamp_ns;	/* departure time for next sent data packet */
+	u64	tcp_clock_cache; /* cache last tcp_clock_ns() (see tcp_mstamp_refresh()) */
 
 /* RTT measurement */
 	u64	tcp_mstamp;	/* most recent packet received/sent */

include/net/sock.h

@@ -422,8 +422,8 @@ struct sock {
 	struct timer_list	sk_timer;
 	__u32			sk_priority;
 	__u32			sk_mark;
-	u32			sk_pacing_rate; /* bytes per second */
+	unsigned long		sk_pacing_rate; /* bytes per second */
-	u32			sk_max_pacing_rate;
+	unsigned long		sk_max_pacing_rate;
 	struct page_frag	sk_frag;
 	netdev_features_t	sk_route_caps;
 	netdev_features_t	sk_route_nocaps;

net/core/filter.c

@@ -3927,8 +3927,8 @@ BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
 			sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 			sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
 			break;
-		case SO_MAX_PACING_RATE:
+		case SO_MAX_PACING_RATE: /* 32bit version */
-			sk->sk_max_pacing_rate = val;
+			sk->sk_max_pacing_rate = (val == ~0U) ? ~0UL : val;
 			sk->sk_pacing_rate = min(sk->sk_pacing_rate,
 						 sk->sk_max_pacing_rate);
 			break;

net/core/sock.c

@@ -998,7 +998,7 @@ set_rcvbuf:
 			cmpxchg(&sk->sk_pacing_status,
 				SK_PACING_NONE,
 				SK_PACING_NEEDED);
-		sk->sk_max_pacing_rate = val;
+		sk->sk_max_pacing_rate = (val == ~0U) ? ~0UL : val;
 		sk->sk_pacing_rate = min(sk->sk_pacing_rate,
 					 sk->sk_max_pacing_rate);
 		break;
@@ -1336,7 +1336,8 @@ int sock_getsockopt(struct socket *sock, int level, int optname,
 #endif
 
 	case SO_MAX_PACING_RATE:
-		v.val = sk->sk_max_pacing_rate;
+		/* 32bit version */
+		v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
 		break;
 
 	case SO_INCOMING_CPU:
@@ -2810,8 +2811,8 @@ void sock_init_data(struct socket *sock, struct sock *sk)
 	sk->sk_ll_usec		=	sysctl_net_busy_read;
 #endif
 
-	sk->sk_max_pacing_rate = ~0U;
+	sk->sk_max_pacing_rate = ~0UL;
-	sk->sk_pacing_rate = ~0U;
+	sk->sk_pacing_rate = ~0UL;
 	sk->sk_pacing_shift = 10;
 	sk->sk_incoming_cpu = -1;
 

net/ipv4/tcp.c

@@ -3111,10 +3111,10 @@ void tcp_get_info(struct sock *sk, struct tcp_info *info)
 {
 	const struct tcp_sock *tp = tcp_sk(sk); /* iff sk_type == SOCK_STREAM */
 	const struct inet_connection_sock *icsk = inet_csk(sk);
+	unsigned long rate;
 	u32 now;
 	u64 rate64;
 	bool slow;
-	u32 rate;
 
 	memset(info, 0, sizeof(*info));
 	if (sk->sk_type != SOCK_STREAM)
@@ -3124,11 +3124,11 @@ void tcp_get_info(struct sock *sk, struct tcp_info *info)
 
 	/* Report meaningful fields for all TCP states, including listeners */
 	rate = READ_ONCE(sk->sk_pacing_rate);
-	rate64 = rate != ~0U ? rate : ~0ULL;
+	rate64 = (rate != ~0UL) ? rate : ~0ULL;
 	info->tcpi_pacing_rate = rate64;
 
 	rate = READ_ONCE(sk->sk_max_pacing_rate);
-	rate64 = rate != ~0U ? rate : ~0ULL;
+	rate64 = (rate != ~0UL) ? rate : ~0ULL;
 	info->tcpi_max_pacing_rate = rate64;
 
 	info->tcpi_reordering = tp->reordering;
@@ -3254,8 +3254,8 @@ struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk)
 	const struct tcp_sock *tp = tcp_sk(sk);
 	struct sk_buff *stats;
 	struct tcp_info info;
+	unsigned long rate;
 	u64 rate64;
-	u32 rate;
 
 	stats = alloc_skb(tcp_opt_stats_get_size(), GFP_ATOMIC);
 	if (!stats)
@@ -3274,7 +3274,7 @@ struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk)
 			  tp->total_retrans, TCP_NLA_PAD);
 
 	rate = READ_ONCE(sk->sk_pacing_rate);
-	rate64 = rate != ~0U ? rate : ~0ULL;
+	rate64 = (rate != ~0UL) ? rate : ~0ULL;
 	nla_put_u64_64bit(stats, TCP_NLA_PACING_RATE, rate64, TCP_NLA_PAD);
 
 	rate64 = tcp_compute_delivery_rate(tp);

net/ipv4/tcp_bbr.c

@@ -129,7 +129,7 @@ static const u32 bbr_probe_rtt_mode_ms = 200;
 static const int bbr_min_tso_rate = 1200000;
 
 /* Pace at ~1% below estimated bw, on average, to reduce queue at bottleneck. */
-static const int bbr_pacing_marging_percent = 1;
+static const int bbr_pacing_margin_percent = 1;
 
 /* We use a high_gain value of 2/ln(2) because it's the smallest pacing gain
  * that will allow a smoothly increasing pacing rate that will double each RTT
@@ -214,12 +214,12 @@ static u64 bbr_rate_bytes_per_sec(struct sock *sk, u64 rate, int gain)
 	rate *= mss;
 	rate *= gain;
 	rate >>= BBR_SCALE;
-	rate *= USEC_PER_SEC / 100 * (100 - bbr_pacing_marging_percent);
+	rate *= USEC_PER_SEC / 100 * (100 - bbr_pacing_margin_percent);
 	return rate >> BW_SCALE;
 }
 
 /* Convert a BBR bw and gain factor to a pacing rate in bytes per second. */
-static u32 bbr_bw_to_pacing_rate(struct sock *sk, u32 bw, int gain)
+static unsigned long bbr_bw_to_pacing_rate(struct sock *sk, u32 bw, int gain)
 {
 	u64 rate = bw;
 
@@ -258,7 +258,7 @@ static void bbr_set_pacing_rate(struct sock *sk, u32 bw, int gain)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
 	struct bbr *bbr = inet_csk_ca(sk);
-	u32 rate = bbr_bw_to_pacing_rate(sk, bw, gain);
+	unsigned long rate = bbr_bw_to_pacing_rate(sk, bw, gain);
 
 	if (unlikely(!bbr->has_seen_rtt && tp->srtt_us))
 		bbr_init_pacing_rate_from_rtt(sk);
@@ -280,7 +280,7 @@ static u32 bbr_tso_segs_goal(struct sock *sk)
 	/* Sort of tcp_tso_autosize() but ignoring
 	 * driver provided sk_gso_max_size.
 	 */
-	bytes = min_t(u32, sk->sk_pacing_rate >> sk->sk_pacing_shift,
+	bytes = min_t(unsigned long, sk->sk_pacing_rate >> sk->sk_pacing_shift,
 		      GSO_MAX_SIZE - 1 - MAX_TCP_HEADER);
 	segs = max_t(u32, bytes / tp->mss_cache, bbr_min_tso_segs(sk));
 

net/ipv4/tcp_cdg.c

@@ -146,7 +146,7 @@ static void tcp_cdg_hystart_update(struct sock *sk)
 		return;
 
 	if (hystart_detect & HYSTART_ACK_TRAIN) {
-		u32 now_us = div_u64(local_clock(), NSEC_PER_USEC);
+		u32 now_us = tp->tcp_mstamp;
 
 		if (ca->last_ack == 0 || !tcp_is_cwnd_limited(sk)) {
 			ca->last_ack = now_us;

net/ipv4/tcp_output.c

@@ -52,9 +52,8 @@ void tcp_mstamp_refresh(struct tcp_sock *tp)
 {
 	u64 val = tcp_clock_ns();
 
-	/* departure time for next data packet */
+	if (val > tp->tcp_clock_cache)
-	if (val > tp->tcp_wstamp_ns)
+		tp->tcp_clock_cache = val;
-		tp->tcp_wstamp_ns = val;
 
 	val = div_u64(val, NSEC_PER_USEC);
 	if (val > tp->tcp_mstamp)
@@ -976,32 +975,26 @@ enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer)
 	return HRTIMER_NORESTART;
 }
 
-static void tcp_internal_pacing(struct sock *sk)
+static void tcp_update_skb_after_send(struct sock *sk, struct sk_buff *skb,
-{
+				      u64 prior_wstamp)
-	if (!tcp_needs_internal_pacing(sk))
-		return;
-	hrtimer_start(&tcp_sk(sk)->pacing_timer,
-		      ns_to_ktime(tcp_sk(sk)->tcp_wstamp_ns),
-		      HRTIMER_MODE_ABS_PINNED_SOFT);
-	sock_hold(sk);
-}
-
-static void tcp_update_skb_after_send(struct sock *sk, struct sk_buff *skb)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
 
 	skb->skb_mstamp_ns = tp->tcp_wstamp_ns;
 	if (sk->sk_pacing_status != SK_PACING_NONE) {
-		u32 rate = sk->sk_pacing_rate;
+		unsigned long rate = sk->sk_pacing_rate;
 
 		/* Original sch_fq does not pace first 10 MSS
 		 * Note that tp->data_segs_out overflows after 2^32 packets,
 		 * this is a minor annoyance.
 		 */
-		if (rate != ~0U && rate && tp->data_segs_out >= 10) {
+		if (rate != ~0UL && rate && tp->data_segs_out >= 10) {
-			tp->tcp_wstamp_ns += div_u64((u64)skb->len * NSEC_PER_SEC, rate);
+			u64 len_ns = div64_ul((u64)skb->len * NSEC_PER_SEC, rate);
+			u64 credit = tp->tcp_wstamp_ns - prior_wstamp;
 
-			tcp_internal_pacing(sk);
+			/* take into account OS jitter */
+			len_ns -= min_t(u64, len_ns / 2, credit);
+			tp->tcp_wstamp_ns += len_ns;
 		}
 	}
 	list_move_tail(&skb->tcp_tsorted_anchor, &tp->tsorted_sent_queue);
@@ -1030,6 +1023,7 @@ static int __tcp_transmit_skb(struct sock *sk, struct sk_buff *skb,
 	struct sk_buff *oskb = NULL;
 	struct tcp_md5sig_key *md5;
 	struct tcphdr *th;
+	u64 prior_wstamp;
 	int err;
 
 	BUG_ON(!skb || !tcp_skb_pcount(skb));
@@ -1050,6 +1044,10 @@ static int __tcp_transmit_skb(struct sock *sk, struct sk_buff *skb,
 		if (unlikely(!skb))
 			return -ENOBUFS;
 	}
+
+	prior_wstamp = tp->tcp_wstamp_ns;
+	tp->tcp_wstamp_ns = max(tp->tcp_wstamp_ns, tp->tcp_clock_cache);
+
 	skb->skb_mstamp_ns = tp->tcp_wstamp_ns;
 
 	inet = inet_sk(sk);
@@ -1166,7 +1164,7 @@ static int __tcp_transmit_skb(struct sock *sk, struct sk_buff *skb,
 		err = net_xmit_eval(err);
 	}
 	if (!err && oskb) {
-		tcp_update_skb_after_send(sk, oskb);
+		tcp_update_skb_after_send(sk, oskb, prior_wstamp);
 		tcp_rate_skb_sent(sk, oskb);
 	}
 	return err;
@@ -1701,7 +1699,8 @@ static u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
 {
 	u32 bytes, segs;
 
-	bytes = min(sk->sk_pacing_rate >> sk->sk_pacing_shift,
+	bytes = min_t(unsigned long,
+		      sk->sk_pacing_rate >> sk->sk_pacing_shift,
 		      sk->sk_gso_max_size - 1 - MAX_TCP_HEADER);
 
 	/* Goal is to send at least one packet per ms,
@@ -2175,10 +2174,23 @@ static int tcp_mtu_probe(struct sock *sk)
 	return -1;
 }
 
-static bool tcp_pacing_check(const struct sock *sk)
+static bool tcp_pacing_check(struct sock *sk)
 {
-	return tcp_needs_internal_pacing(sk) &&
+	struct tcp_sock *tp = tcp_sk(sk);
-	       hrtimer_is_queued(&tcp_sk(sk)->pacing_timer);
+
+	if (!tcp_needs_internal_pacing(sk))
+		return false;
+
+	if (tp->tcp_wstamp_ns <= tp->tcp_clock_cache)
+		return false;
+
+	if (!hrtimer_is_queued(&tp->pacing_timer)) {
+		hrtimer_start(&tp->pacing_timer,
+			      ns_to_ktime(tp->tcp_wstamp_ns),
+			      HRTIMER_MODE_ABS_PINNED_SOFT);
+		sock_hold(sk);
+	}
+	return true;
 }
 
 /* TCP Small Queues :
@@ -2195,10 +2207,12 @@ static bool tcp_pacing_check(const struct sock *sk)
 static bool tcp_small_queue_check(struct sock *sk, const struct sk_buff *skb,
 				  unsigned int factor)
 {
-	unsigned int limit;
+	unsigned long limit;
 
-	limit = max(2 * skb->truesize, sk->sk_pacing_rate >> sk->sk_pacing_shift);
+	limit = max_t(unsigned long,
-	limit = min_t(u32, limit,
+		      2 * skb->truesize,
+		      sk->sk_pacing_rate >> sk->sk_pacing_shift);
+	limit = min_t(unsigned long, limit,
 		      sock_net(sk)->ipv4.sysctl_tcp_limit_output_bytes);
 	limit <<= factor;
 
@@ -2315,7 +2329,7 @@ static bool tcp_write_xmit(struct sock *sk, unsigned int mss_now, int nonagle,
 
 		if (unlikely(tp->repair) && tp->repair_queue == TCP_SEND_QUEUE) {
 			/* "skb_mstamp" is used as a start point for the retransmit timer */
-			tcp_update_skb_after_send(sk, skb);
+			tcp_update_skb_after_send(sk, skb, tp->tcp_wstamp_ns);
 			goto repair; /* Skip network transmission */
 		}
 
@@ -2890,7 +2904,7 @@ int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs)
 		} tcp_skb_tsorted_restore(skb);
 
 		if (!err) {
-			tcp_update_skb_after_send(sk, skb);
+			tcp_update_skb_after_send(sk, skb, tp->tcp_wstamp_ns);
 			tcp_rate_skb_sent(sk, skb);
 		}
 	} else {

net/ipv4/tcp_timer.c

@@ -360,7 +360,7 @@ static void tcp_probe_timer(struct sock *sk)
 	 */
 	start_ts = tcp_skb_timestamp(skb);
 	if (!start_ts)
-		skb->skb_mstamp_ns = tp->tcp_wstamp_ns;
+		skb->skb_mstamp_ns = tp->tcp_clock_cache;
 	else if (icsk->icsk_user_timeout &&
 		 (s32)(tcp_time_stamp(tp) - start_ts) > icsk->icsk_user_timeout)
 		goto abort;

net/sched/sch_fq.c

@@ -92,8 +92,8 @@ struct fq_sched_data {
 	u32		quantum;
 	u32		initial_quantum;
 	u32		flow_refill_delay;
-	u32		flow_max_rate;	/* optional max rate per flow */
 	u32		flow_plimit;	/* max packets per flow */
+	unsigned long	flow_max_rate;	/* optional max rate per flow */
 	u32		orphan_mask;	/* mask for orphaned skb */
 	u32		low_rate_threshold;
 	struct rb_root	*fq_root;
@@ -416,7 +416,8 @@ static struct sk_buff *fq_dequeue(struct Qdisc *sch)
 	struct fq_flow_head *head;
 	struct sk_buff *skb;
 	struct fq_flow *f;
-	u32 rate, plen;
+	unsigned long rate;
+	u32 plen;
 
 	skb = fq_dequeue_head(sch, &q->internal);
 	if (skb)
@@ -443,7 +444,7 @@ begin:
 	}
 
 	skb = f->head;
-	if (skb && !skb_is_tcp_pure_ack(skb)) {
+	if (skb) {
 		u64 time_next_packet = max_t(u64, ktime_to_ns(skb->tstamp),
 					     f->time_next_packet);
 
@@ -485,11 +486,11 @@ begin:
 		if (f->credit > 0)
 			goto out;
 	}
-	if (rate != ~0U) {
+	if (rate != ~0UL) {
 		u64 len = (u64)plen * NSEC_PER_SEC;
 
 		if (likely(rate))
-			do_div(len, rate);
+			len = div64_ul(len, rate);
 		/* Since socket rate can change later,
 		 * clamp the delay to 1 second.
 		 * Really, providers of too big packets should be fixed !
@@ -701,9 +702,11 @@ static int fq_change(struct Qdisc *sch, struct nlattr *opt,
 		pr_warn_ratelimited("sch_fq: defrate %u ignored.\n",
 				    nla_get_u32(tb[TCA_FQ_FLOW_DEFAULT_RATE]));
 
-	if (tb[TCA_FQ_FLOW_MAX_RATE])
+	if (tb[TCA_FQ_FLOW_MAX_RATE]) {
-		q->flow_max_rate = nla_get_u32(tb[TCA_FQ_FLOW_MAX_RATE]);
+		u32 rate = nla_get_u32(tb[TCA_FQ_FLOW_MAX_RATE]);
 
+		q->flow_max_rate = (rate == ~0U) ? ~0UL : rate;
+	}
 	if (tb[TCA_FQ_LOW_RATE_THRESHOLD])
 		q->low_rate_threshold =
 			nla_get_u32(tb[TCA_FQ_LOW_RATE_THRESHOLD]);
@@ -766,7 +769,7 @@ static int fq_init(struct Qdisc *sch, struct nlattr *opt,
 	q->quantum		= 2 * psched_mtu(qdisc_dev(sch));
 	q->initial_quantum	= 10 * psched_mtu(qdisc_dev(sch));
 	q->flow_refill_delay	= msecs_to_jiffies(40);
-	q->flow_max_rate	= ~0U;
+	q->flow_max_rate	= ~0UL;
 	q->time_next_delayed_flow = ~0ULL;
 	q->rate_enable		= 1;
 	q->new_flows.first	= NULL;
@@ -802,7 +805,8 @@ static int fq_dump(struct Qdisc *sch, struct sk_buff *skb)
 	    nla_put_u32(skb, TCA_FQ_QUANTUM, q->quantum) ||
 	    nla_put_u32(skb, TCA_FQ_INITIAL_QUANTUM, q->initial_quantum) ||
 	    nla_put_u32(skb, TCA_FQ_RATE_ENABLE, q->rate_enable) ||
-	    nla_put_u32(skb, TCA_FQ_FLOW_MAX_RATE, q->flow_max_rate) ||
+	    nla_put_u32(skb, TCA_FQ_FLOW_MAX_RATE,
+			min_t(unsigned long, q->flow_max_rate, ~0U)) ||
 	    nla_put_u32(skb, TCA_FQ_FLOW_REFILL_DELAY,
 			jiffies_to_usecs(q->flow_refill_delay)) ||
 	    nla_put_u32(skb, TCA_FQ_ORPHAN_MASK, q->orphan_mask) ||