tcp: add net.ipv4.tcp_comp_sack_rtt_percent

TCP SACK compression has been added in 2018 in commit
5d9f4262b7ea ("tcp: add SACK compression").

It is working great for WAN flows (with large RTT).
Wifi in particular gets a significant boost _when_ ACK are suppressed.

Add a new sysctl so that we can tune the very conservative 5 % value
that has been used so far in this formula, so that small RTT flows
can benefit from this feature.

delay = min ( 5 % of RTT, 1 ms)

This patch adds new tcp_comp_sack_rtt_percent sysctl
to ease experiments and tuning.

Given that we cap the delay to 1ms (tcp_comp_sack_delay_ns sysctl),
set the default value to 33 %.

Quoting Neal Cardwell ( https://lore.kernel.org/netdev/CADVnQymZ1tFnEA1Q=vtECs0=Db7zHQ8=+WCQtnhHFVbEOzjVnQ@mail.gmail.com/ )

The rationale for 33% is basically to try to facilitate pipelining,
where there are always at least 3 ACKs and 3 GSO/TSO skbs per SRTT, so
that the path can maintain a budget for 3 full-sized GSO/TSO skbs "in
flight" at all times:

+ 1 skb in the qdisc waiting to be sent by the NIC next
+ 1 skb being sent by the NIC (being serialized by the NIC out onto the wire)
+ 1 skb being received and aggregated by the receiver machine's
aggregation mechanism (some combination of LRO, GRO, and sack
compression)

Note that this is basically the same magic number (3) and the same
rationales as:

(a) tcp_tso_should_defer() ensuring that we defer sending data for no
longer than cwnd/tcp_tso_win_divisor (where tcp_tso_win_divisor = 3),
and
(b) bbr_quantization_budget() ensuring that cwnd is at least 3 GSO/TSO
skbs to maintain pipelining and full throughput at low RTTs

Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Link: https://patch.msgid.link/20251106115236.3450026-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

5 files changed, 40 insertions, 10 deletions

diff --git a/Documentation/networking/ip-sysctl.rst b/Documentation/networking/ip-sysctl.rst
index 7cd35bfd39e6..2bae61be1859 100644
--- a/Documentation/networking/ip-sysctl.rst
+++ b/Documentation/networking/ip-sysctl.rst
@@ -854,9 +854,18 @@ tcp_sack - BOOLEAN
 
 	Default: 1 (enabled)
 
+tcp_comp_sack_rtt_percent - INTEGER
+	Percentage of SRTT used for the compressed SACK feature.
+	See tcp_comp_sack_nr, tcp_comp_sack_delay_ns, tcp_comp_sack_slack_ns.
+
+	Possible values : 1 - 1000
+
+	Default : 33 %
+
 tcp_comp_sack_delay_ns - LONG INTEGER
-	TCP tries to reduce number of SACK sent, using a timer
-	based on 5% of SRTT, capped by this sysctl, in nano seconds.
+	TCP tries to reduce number of SACK sent, using a timer based
+	on tcp_comp_sack_rtt_percent of SRTT, capped by this sysctl
+	in nano seconds.
 	The default is 1ms, based on TSO autosizing period.
 
 	Default : 1,000,000 ns (1 ms)
diff --git a/include/net/netns/ipv4.h b/include/net/netns/ipv4.h
index 0e96c90e56c6..de9d36acc8e2 100644
--- a/include/net/netns/ipv4.h
+++ b/include/net/netns/ipv4.h
@@ -221,6 +221,7 @@ struct netns_ipv4 {
 	int sysctl_tcp_pacing_ss_ratio;
 	int sysctl_tcp_pacing_ca_ratio;
 	unsigned int sysctl_tcp_child_ehash_entries;
+	int sysctl_tcp_comp_sack_rtt_percent;
 	unsigned long sysctl_tcp_comp_sack_delay_ns;
 	unsigned long sysctl_tcp_comp_sack_slack_ns;
 	int sysctl_max_syn_backlog;
diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c
index 0c7c8f9041cb..35367f8e2da3 100644
--- a/net/ipv4/sysctl_net_ipv4.c
+++ b/net/ipv4/sysctl_net_ipv4.c
@@ -1452,6 +1452,15 @@ static struct ctl_table ipv4_net_table[] = {
 		.proc_handler	= proc_doulongvec_minmax,
 	},
 	{
+		.procname	= "tcp_comp_sack_rtt_percent",
+		.data		= &init_net.ipv4.sysctl_tcp_comp_sack_rtt_percent,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= SYSCTL_ONE,
+		.extra2		= SYSCTL_ONE_THOUSAND,
+	},
+	{
 		.procname	= "tcp_comp_sack_slack_ns",
 		.data		= &init_net.ipv4.sysctl_tcp_comp_sack_slack_ns,
 		.maxlen		= sizeof(unsigned long),
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 804ec56bdd24..9df5d7515605 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -5893,7 +5893,9 @@ static inline void tcp_data_snd_check(struct sock *sk)
 static void __tcp_ack_snd_check(struct sock *sk, int ofo_possible)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
-	unsigned long rtt, delay;
+	struct net *net = sock_net(sk);
+	unsigned long rtt;
+	u64 delay;
 
 	    /* More than one full frame received... */
 	if (((tp->rcv_nxt - tp->rcv_wup) > inet_csk(sk)->icsk_ack.rcv_mss &&
@@ -5912,7 +5914,7 @@ static void __tcp_ack_snd_check(struct sock *sk, int ofo_possible)
 		 * Defer the ack until tcp_release_cb().
 		 */
 		if (sock_owned_by_user_nocheck(sk) &&
-		    READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_backlog_ack_defer)) {
+		    READ_ONCE(net->ipv4.sysctl_tcp_backlog_ack_defer)) {
 			set_bit(TCP_ACK_DEFERRED, &sk->sk_tsq_flags);
 			return;
 		}
@@ -5927,7 +5929,7 @@ send_now:
 	}
 
 	if (!tcp_is_sack(tp) ||
-	    tp->compressed_ack >= READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_comp_sack_nr))
+	    tp->compressed_ack >= READ_ONCE(net->ipv4.sysctl_tcp_comp_sack_nr))
 		goto send_now;
 
 	if (tp->compressed_ack_rcv_nxt != tp->rcv_nxt) {
@@ -5942,18 +5944,26 @@ send_now:
 	if (hrtimer_is_queued(&tp->compressed_ack_timer))
 		return;
 
-	/* compress ack timer : 5 % of rtt, but no more than tcp_comp_sack_delay_ns */
+	/* compress ack timer : comp_sack_rtt_percent of rtt,
+	 * but no more than tcp_comp_sack_delay_ns.
+	 */
 
 	rtt = tp->rcv_rtt_est.rtt_us;
 	if (tp->srtt_us && tp->srtt_us < rtt)
 		rtt = tp->srtt_us;
 
-	delay = min_t(unsigned long,
-		      READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_comp_sack_delay_ns),
-		      rtt * (NSEC_PER_USEC >> 3)/20);
+	/* delay = (rtt >> 3) * NSEC_PER_USEC * comp_sack_rtt_percent / 100
+	 * ->
+	 * delay = rtt * 1.25 * comp_sack_rtt_percent
+	 */
+	delay = (u64)(rtt + (rtt >> 2)) *
+		READ_ONCE(net->ipv4.sysctl_tcp_comp_sack_rtt_percent);
+
+	delay = min(delay, READ_ONCE(net->ipv4.sysctl_tcp_comp_sack_delay_ns));
+
 	sock_hold(sk);
 	hrtimer_start_range_ns(&tp->compressed_ack_timer, ns_to_ktime(delay),
-			       READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_comp_sack_slack_ns),
+			       READ_ONCE(net->ipv4.sysctl_tcp_comp_sack_slack_ns),
 			       HRTIMER_MODE_REL_PINNED_SOFT);
 }
 
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index 0cfebac33a91..a7d9fec2950b 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -3595,6 +3595,7 @@ static int __net_init tcp_sk_init(struct net *net)
 	net->ipv4.sysctl_tcp_comp_sack_delay_ns = NSEC_PER_MSEC;
 	net->ipv4.sysctl_tcp_comp_sack_slack_ns = 100 * NSEC_PER_USEC;
 	net->ipv4.sysctl_tcp_comp_sack_nr = 44;
+	net->ipv4.sysctl_tcp_comp_sack_rtt_percent = 33;
 	net->ipv4.sysctl_tcp_backlog_ack_defer = 1;
 	net->ipv4.sysctl_tcp_fastopen = TFO_CLIENT_ENABLE;
 	net->ipv4.sysctl_tcp_fastopen_blackhole_timeout = 0;