Merge branch 'tcp_sack_rttm'

Kenneth Klette Jonassen says:

====================
tcp: SACK RTTM changes for congestion control

This patch series improves SACK RTT measurements for congestion control:
  o Picks the latest sequence SACKed for RTT, i.e. most accurate delay
    signal.
  o Calls the congestion control's pkts_acked hook with SACK RTTMs
    even when not sequentially ACKing new data.

V2: amend misleading comment
====================

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

net/ipv4/tcp_input.c

@@ -1130,7 +1130,12 @@ static bool tcp_check_dsack(struct sock *sk, const struct sk_buff *ack_skb,
 struct tcp_sacktag_state {
 	int	reord;
 	int	fack_count;
-	long	rtt_us; /* RTT measured by SACKing never-retransmitted data */
+	/* Timestamps for earliest and latest never-retransmitted segment
+	 * that was SACKed. RTO needs the earliest RTT to stay conservative,
+	 * but congestion control should still get an accurate delay signal.
+	 */
+	struct skb_mstamp first_sackt;
+	struct skb_mstamp last_sackt;
 	int	flag;
 };
 
@@ -1233,14 +1238,9 @@ static u8 tcp_sacktag_one(struct sock *sk,
 							   state->reord);
 				if (!after(end_seq, tp->high_seq))
 					state->flag |= FLAG_ORIG_SACK_ACKED;
-				/* Pick the earliest sequence sacked for RTT */
-				if (state->rtt_us < 0) {
-					struct skb_mstamp now;
-
-					skb_mstamp_get(&now);
-					state->rtt_us = skb_mstamp_us_delta(&now,
-								xmit_time);
-				}
+				if (state->first_sackt.v64 == 0)
+					state->first_sackt = *xmit_time;
+				state->last_sackt = *xmit_time;
 			}
 
 			if (sacked & TCPCB_LOST) {
@@ -1634,7 +1634,7 @@ static int tcp_sack_cache_ok(const struct tcp_sock *tp, const struct tcp_sack_bl
 
 static int
 tcp_sacktag_write_queue(struct sock *sk, const struct sk_buff *ack_skb,
-			u32 prior_snd_una, long *sack_rtt_us)
+			u32 prior_snd_una, struct tcp_sacktag_state *state)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
 	const unsigned char *ptr = (skb_transport_header(ack_skb) +
@@ -1642,7 +1642,6 @@ tcp_sacktag_write_queue(struct sock *sk, const struct sk_buff *ack_skb,
 	struct tcp_sack_block_wire *sp_wire = (struct tcp_sack_block_wire *)(ptr+2);
 	struct tcp_sack_block sp[TCP_NUM_SACKS];
 	struct tcp_sack_block *cache;
-	struct tcp_sacktag_state state;
 	struct sk_buff *skb;
 	int num_sacks = min(TCP_NUM_SACKS, (ptr[1] - TCPOLEN_SACK_BASE) >> 3);
 	int used_sacks;
@@ -1650,9 +1649,8 @@ tcp_sacktag_write_queue(struct sock *sk, const struct sk_buff *ack_skb,
 	int i, j;
 	int first_sack_index;
 
-	state.flag = 0;
-	state.reord = tp->packets_out;
-	state.rtt_us = -1L;
+	state->flag = 0;
+	state->reord = tp->packets_out;
 
 	if (!tp->sacked_out) {
 		if (WARN_ON(tp->fackets_out))
@@ -1663,7 +1661,7 @@ tcp_sacktag_write_queue(struct sock *sk, const struct sk_buff *ack_skb,
 	found_dup_sack = tcp_check_dsack(sk, ack_skb, sp_wire,
 					 num_sacks, prior_snd_una);
 	if (found_dup_sack)
-		state.flag |= FLAG_DSACKING_ACK;
+		state->flag |= FLAG_DSACKING_ACK;
 
 	/* Eliminate too old ACKs, but take into
 	 * account more or less fresh ones, they can
@@ -1728,7 +1726,7 @@ tcp_sacktag_write_queue(struct sock *sk, const struct sk_buff *ack_skb,
 	}
 
 	skb = tcp_write_queue_head(sk);
-	state.fack_count = 0;
+	state->fack_count = 0;
 	i = 0;
 
 	if (!tp->sacked_out) {
@@ -1762,10 +1760,10 @@ tcp_sacktag_write_queue(struct sock *sk, const struct sk_buff *ack_skb,
 
 			/* Head todo? */
 			if (before(start_seq, cache->start_seq)) {
-				skb = tcp_sacktag_skip(skb, sk, &state,
+				skb = tcp_sacktag_skip(skb, sk, state,
 						       start_seq);
 				skb = tcp_sacktag_walk(skb, sk, next_dup,
-						       &state,
+						       state,
 						       start_seq,
 						       cache->start_seq,
 						       dup_sack);
@@ -1776,7 +1774,7 @@ tcp_sacktag_write_queue(struct sock *sk, const struct sk_buff *ack_skb,
 				goto advance_sp;
 
 			skb = tcp_maybe_skipping_dsack(skb, sk, next_dup,
-						       &state,
+						       state,
 						       cache->end_seq);
 
 			/* ...tail remains todo... */
@@ -1785,12 +1783,12 @@ tcp_sacktag_write_queue(struct sock *sk, const struct sk_buff *ack_skb,
 				skb = tcp_highest_sack(sk);
 				if (!skb)
 					break;
-				state.fack_count = tp->fackets_out;
+				state->fack_count = tp->fackets_out;
 				cache++;
 				goto walk;
 			}
 
-			skb = tcp_sacktag_skip(skb, sk, &state, cache->end_seq);
+			skb = tcp_sacktag_skip(skb, sk, state, cache->end_seq);
 			/* Check overlap against next cached too (past this one already) */
 			cache++;
 			continue;
@@ -1800,12 +1798,12 @@ tcp_sacktag_write_queue(struct sock *sk, const struct sk_buff *ack_skb,
 			skb = tcp_highest_sack(sk);
 			if (!skb)
 				break;
-			state.fack_count = tp->fackets_out;
+			state->fack_count = tp->fackets_out;
 		}
-		skb = tcp_sacktag_skip(skb, sk, &state, start_seq);
+		skb = tcp_sacktag_skip(skb, sk, state, start_seq);
 
 walk:
-		skb = tcp_sacktag_walk(skb, sk, next_dup, &state,
+		skb = tcp_sacktag_walk(skb, sk, next_dup, state,
 				       start_seq, end_seq, dup_sack);
 
 advance_sp:
@@ -1820,9 +1818,9 @@ advance_sp:
 	for (j = 0; j < used_sacks; j++)
 		tp->recv_sack_cache[i++] = sp[j];
 
-	if ((state.reord < tp->fackets_out) &&
+	if ((state->reord < tp->fackets_out) &&
 	    ((inet_csk(sk)->icsk_ca_state != TCP_CA_Loss) || tp->undo_marker))
-		tcp_update_reordering(sk, tp->fackets_out - state.reord, 0);
+		tcp_update_reordering(sk, tp->fackets_out - state->reord, 0);
 
 	tcp_mark_lost_retrans(sk);
 	tcp_verify_left_out(tp);
@@ -1834,8 +1832,7 @@ out:
 	WARN_ON((int)tp->retrans_out < 0);
 	WARN_ON((int)tcp_packets_in_flight(tp) < 0);
 #endif
-	*sack_rtt_us = state.rtt_us;
-	return state.flag;
+	return state->flag;
 }
 
 /* Limits sacked_out so that sum with lost_out isn't ever larger than
@@ -3052,7 +3049,8 @@ static void tcp_ack_tstamp(struct sock *sk, struct sk_buff *skb,
  * arrived at the other end.
  */
 static int tcp_clean_rtx_queue(struct sock *sk, int prior_fackets,
-			       u32 prior_snd_una, long sack_rtt_us)
+			       u32 prior_snd_una,
+			       struct tcp_sacktag_state *sack)
 {
 	const struct inet_connection_sock *icsk = inet_csk(sk);
 	struct skb_mstamp first_ackt, last_ackt, now;
@@ -3060,8 +3058,9 @@ static int tcp_clean_rtx_queue(struct sock *sk, int prior_fackets,
 	u32 prior_sacked = tp->sacked_out;
 	u32 reord = tp->packets_out;
 	bool fully_acked = true;
-	long ca_seq_rtt_us = -1L;
+	long sack_rtt_us = -1L;
 	long seq_rtt_us = -1L;
+	long ca_rtt_us = -1L;
 	struct sk_buff *skb;
 	u32 pkts_acked = 0;
 	bool rtt_update;
@@ -3150,15 +3149,16 @@ static int tcp_clean_rtx_queue(struct sock *sk, int prior_fackets,
 	skb_mstamp_get(&now);
 	if (likely(first_ackt.v64)) {
 		seq_rtt_us = skb_mstamp_us_delta(&now, &first_ackt);
-		ca_seq_rtt_us = skb_mstamp_us_delta(&now, &last_ackt);
+		ca_rtt_us = skb_mstamp_us_delta(&now, &last_ackt);
+	}
+	if (sack->first_sackt.v64) {
+		sack_rtt_us = skb_mstamp_us_delta(&now, &sack->first_sackt);
+		ca_rtt_us = skb_mstamp_us_delta(&now, &sack->last_sackt);
 	}
 
 	rtt_update = tcp_ack_update_rtt(sk, flag, seq_rtt_us, sack_rtt_us);
 
 	if (flag & FLAG_ACKED) {
-		const struct tcp_congestion_ops *ca_ops
-			= inet_csk(sk)->icsk_ca_ops;
-
 		tcp_rearm_rto(sk);
 		if (unlikely(icsk->icsk_mtup.probe_size &&
 			     !after(tp->mtu_probe.probe_seq_end, tp->snd_una))) {
@@ -3181,11 +3181,6 @@ static int tcp_clean_rtx_queue(struct sock *sk, int prior_fackets,
 
 		tp->fackets_out -= min(pkts_acked, tp->fackets_out);
 
-		if (ca_ops->pkts_acked) {
-			long rtt_us = min_t(ulong, ca_seq_rtt_us, sack_rtt_us);
-			ca_ops->pkts_acked(sk, pkts_acked, rtt_us);
-		}
-
 	} else if (skb && rtt_update && sack_rtt_us >= 0 &&
 		   sack_rtt_us > skb_mstamp_us_delta(&now, &skb->skb_mstamp)) {
 		/* Do not re-arm RTO if the sack RTT is measured from data sent
@@ -3195,6 +3190,9 @@ static int tcp_clean_rtx_queue(struct sock *sk, int prior_fackets,
 		tcp_rearm_rto(sk);
 	}
 
+	if (icsk->icsk_ca_ops->pkts_acked)
+		icsk->icsk_ca_ops->pkts_acked(sk, pkts_acked, ca_rtt_us);
+
 #if FASTRETRANS_DEBUG > 0
 	WARN_ON((int)tp->sacked_out < 0);
 	WARN_ON((int)tp->lost_out < 0);
@@ -3459,6 +3457,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag)
 {
 	struct inet_connection_sock *icsk = inet_csk(sk);
 	struct tcp_sock *tp = tcp_sk(sk);
+	struct tcp_sacktag_state sack_state;
 	u32 prior_snd_una = tp->snd_una;
 	u32 ack_seq = TCP_SKB_CB(skb)->seq;
 	u32 ack = TCP_SKB_CB(skb)->ack_seq;
@@ -3467,7 +3466,8 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag)
 	int prior_packets = tp->packets_out;
 	const int prior_unsacked = tp->packets_out - tp->sacked_out;
 	int acked = 0; /* Number of packets newly acked */
-	long sack_rtt_us = -1L;
+
+	sack_state.first_sackt.v64 = 0;
 
 	/* We very likely will need to access write queue head. */
 	prefetchw(sk->sk_write_queue.next);
@@ -3531,7 +3531,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag)
 
 		if (TCP_SKB_CB(skb)->sacked)
 			flag |= tcp_sacktag_write_queue(sk, skb, prior_snd_una,
-							&sack_rtt_us);
+							&sack_state);
 
 		if (tcp_ecn_rcv_ecn_echo(tp, tcp_hdr(skb))) {
 			flag |= FLAG_ECE;
@@ -3556,7 +3556,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag)
 	/* See if we can take anything off of the retransmit queue. */
 	acked = tp->packets_out;
 	flag |= tcp_clean_rtx_queue(sk, prior_fackets, prior_snd_una,
-				    sack_rtt_us);
+				    &sack_state);
 	acked -= tp->packets_out;
 
 	/* Advance cwnd if state allows */
@@ -3608,7 +3608,7 @@ old_ack:
 	 */
 	if (TCP_SKB_CB(skb)->sacked) {
 		flag |= tcp_sacktag_write_queue(sk, skb, prior_snd_una,
-						&sack_rtt_us);
+						&sack_state);
 		tcp_fastretrans_alert(sk, acked, prior_unsacked,
 				      is_dupack, flag);
 	}