[PATCH v3 net-next 0/6] tcp: implement SACK compression

[PATCH v3 net-next 0/6] tcp: implement SACK compression

[Eric Dumazet]

When TCP receives an out-of-order packet, it immediately sends
a SACK packet, generating network load but also forcing the
receiver to send 1-MSS pathological packets, increasing its
RTX queue length/depth, and thus processing time.

Wifi networks suffer from this aggressive behavior, but generally
speaking, all these SACK packets add fuel to the fire when networks
are under congestion.

This patch series adds SACK compression, but the infrastructure
could be leveraged to also compress ACK in the future.

v2: Addressed Neal feedback.
    Added two sysctls to allow fine tuning, or even disabling the feature.

v3: take rtt = min(srtt, rcv_rtt) as Yuchung suggested, because rcv_rtt
    can be over estimated for RPC (or sender limited)

Eric Dumazet (6):
  tcp: use __sock_put() instead of sock_put() in tcp_clear_xmit_timers()
  tcp: do not force quickack when receiving out-of-order packets
  tcp: add SACK compression
  tcp: add TCPAckCompressed SNMP counter
  tcp: add tcp_comp_sack_delay_ns sysctl
  tcp: add tcp_comp_sack_nr sysctl

 Documentation/networking/ip-sysctl.txt | 13 +++++++++
 include/linux/tcp.h                    |  2 ++
 include/net/netns/ipv4.h               |  2 ++
 include/net/tcp.h                      |  5 +++-
 include/uapi/linux/snmp.h              |  1 +
 net/ipv4/proc.c                        |  1 +
 net/ipv4/sysctl_net_ipv4.c             | 17 ++++++++++++
 net/ipv4/tcp.c                         |  1 +
 net/ipv4/tcp_input.c                   | 38 ++++++++++++++++++++------
 net/ipv4/tcp_ipv4.c                    |  2 ++
 net/ipv4/tcp_output.c                  |  9 ++++++
 net/ipv4/tcp_timer.c                   | 25 +++++++++++++++++
 12 files changed, 107 insertions(+), 9 deletions(-)

-- 
2.17.0.441.gb46fe60e1d-goog

[PATCH v3 net-next 1/6] tcp: use __sock_put() instead of sock_put() in tcp_clear_xmit_timers()

[Eric Dumazet]

Socket can not disappear under us.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
---
 include/net/tcp.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/include/net/tcp.h b/include/net/tcp.h
index 6deb540297ccaa1f05ce633efe313d1ca2c15dd9..511bd0fde1dc1dd842598d083905b0425bcb05f8 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -559,7 +559,7 @@ void tcp_init_xmit_timers(struct sock *);
 static inline void tcp_clear_xmit_timers(struct sock *sk)
 {
 	if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
-		sock_put(sk);
+		__sock_put(sk);
 
 	inet_csk_clear_xmit_timers(sk);
 }
-- 
2.17.0.441.gb46fe60e1d-goog

[PATCH v3 net-next 2/6] tcp: do not force quickack when receiving out-of-order packets

[Eric Dumazet]

As explained in commit 9f9843a751d0 ("tcp: properly handle stretch
acks in slow start"), TCP stacks have to consider how many packets
are acknowledged in one single ACK, because of GRO, but also
because of ACK compression or losses.

We plan to add SACK compression in the following patch, we
must therefore not call tcp_enter_quickack_mode()

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
---
 net/ipv4/tcp_input.c | 2 --
 1 file changed, 2 deletions(-)

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 0bf032839548f8dccb7f24a6fb5a7d47ea29208b..f5622b250665178e44460fa2cd4a11af23dfb23d 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -4715,8 +4715,6 @@ static void tcp_data_queue(struct sock *sk, struct sk_buff *skb)
 	if (!before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt + tcp_receive_window(tp)))
 		goto out_of_window;
 
-	tcp_enter_quickack_mode(sk);
-
 	if (before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) {
 		/* Partial packet, seq < rcv_next < end_seq */
 		SOCK_DEBUG(sk, "partial packet: rcv_next %X seq %X - %X\n",
-- 
2.17.0.441.gb46fe60e1d-goog

[PATCH v3 net-next 3/6] tcp: add SACK compression

[Eric Dumazet]

When TCP receives an out-of-order packet, it immediately sends
a SACK packet, generating network load but also forcing the
receiver to send 1-MSS pathological packets, increasing its
RTX queue length/depth, and thus processing time.

Wifi networks suffer from this aggressive behavior, but generally
speaking, all these SACK packets add fuel to the fire when networks
are under congestion.

This patch adds a high resolution timer and tp->compressed_ack counter.

Instead of sending a SACK, we program this timer with a small delay,
based on RTT and capped to 1 ms :

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

If subsequent SACKs need to be sent while the timer has not yet
expired, we simply increment tp->compressed_ack.

When timer expires, a SACK is sent with the latest information.
Whenever an ACK is sent (if data is sent, or if in-order
data is received) timer is canceled.

Note that tcp_sack_new_ofo_skb() is able to force a SACK to be sent
if the sack blocks need to be shuffled, even if the timer has not
expired.

A new SNMP counter is added in the following patch.

Two other patches add sysctls to allow changing the 1,000,000 and 44
values that this commit hard-coded.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/linux/tcp.h   |  2 ++
 include/net/tcp.h     |  3 +++
 net/ipv4/tcp.c        |  1 +
 net/ipv4/tcp_input.c  | 35 +++++++++++++++++++++++++++++------
 net/ipv4/tcp_output.c |  7 +++++++
 net/ipv4/tcp_timer.c  | 25 +++++++++++++++++++++++++
 6 files changed, 67 insertions(+), 6 deletions(-)

diff --git a/include/linux/tcp.h b/include/linux/tcp.h
index 807776928cb8610fe97121fbc3c600b08d5d2991..72705eaf4b84060a45bf04d5170f389a18010eac 100644
--- a/include/linux/tcp.h
+++ b/include/linux/tcp.h
@@ -218,6 +218,7 @@ struct tcp_sock {
 		   reord:1;	 /* reordering detected */
 	} rack;
 	u16	advmss;		/* Advertised MSS			*/
+	u8	compressed_ack;
 	u32	chrono_start;	/* Start time in jiffies of a TCP chrono */
 	u32	chrono_stat[3];	/* Time in jiffies for chrono_stat stats */
 	u8	chrono_type:2,	/* current chronograph type */
@@ -297,6 +298,7 @@ struct tcp_sock {
 	u32	sacked_out;	/* SACK'd packets			*/
 
 	struct hrtimer	pacing_timer;
+	struct hrtimer	compressed_ack_timer;
 
 	/* from STCP, retrans queue hinting */
 	struct sk_buff* lost_skb_hint;
diff --git a/include/net/tcp.h b/include/net/tcp.h
index 511bd0fde1dc1dd842598d083905b0425bcb05f8..952d842a604a3ed79e1bf87a712db20a461c35a9 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -561,6 +561,9 @@ static inline void tcp_clear_xmit_timers(struct sock *sk)
 	if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
 		__sock_put(sk);
 
+	if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
+		__sock_put(sk);
+
 	inet_csk_clear_xmit_timers(sk);
 }
 
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 62b776f9003798eaf06992a4eb0914d17646aa61..0a2ea0bbf867271db05aedd7d48b193677664321 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -2595,6 +2595,7 @@ int tcp_disconnect(struct sock *sk, int flags)
 	dst_release(sk->sk_rx_dst);
 	sk->sk_rx_dst = NULL;
 	tcp_saved_syn_free(tp);
+	tp->compressed_ack = 0;
 
 	/* Clean up fastopen related fields */
 	tcp_free_fastopen_req(tp);
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index f5622b250665178e44460fa2cd4a11af23dfb23d..cc2ac5346b92b968593f919192d543384865bcb8 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -4249,6 +4249,8 @@ static void tcp_sack_new_ofo_skb(struct sock *sk, u32 seq, u32 end_seq)
 	 * If the sack array is full, forget about the last one.
 	 */
 	if (this_sack >= TCP_NUM_SACKS) {
+		if (tp->compressed_ack)
+			tcp_send_ack(sk);
 		this_sack--;
 		tp->rx_opt.num_sacks--;
 		sp--;
@@ -5081,6 +5083,7 @@ 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;
 
 	    /* More than one full frame received... */
 	if (((tp->rcv_nxt - tp->rcv_wup) > inet_csk(sk)->icsk_ack.rcv_mss &&
@@ -5092,15 +5095,35 @@ static void __tcp_ack_snd_check(struct sock *sk, int ofo_possible)
 	    (tp->rcv_nxt - tp->copied_seq < sk->sk_rcvlowat ||
 	     __tcp_select_window(sk) >= tp->rcv_wnd)) ||
 	    /* We ACK each frame or... */
-	    tcp_in_quickack_mode(sk) ||
-	    /* We have out of order data. */
-	    (ofo_possible && !RB_EMPTY_ROOT(&tp->out_of_order_queue))) {
-		/* Then ack it now */
+	    tcp_in_quickack_mode(sk)) {
+send_now:
 		tcp_send_ack(sk);
-	} else {
-		/* Else, send delayed ack. */
+		return;
+	}
+
+	if (!ofo_possible || RB_EMPTY_ROOT(&tp->out_of_order_queue)) {
 		tcp_send_delayed_ack(sk);
+		return;
 	}
+
+	if (!tcp_is_sack(tp) || tp->compressed_ack >= 44)
+		goto send_now;
+	tp->compressed_ack++;
+
+	if (hrtimer_is_queued(&tp->compressed_ack_timer))
+		return;
+
+	/* compress ack timer : 5 % of rtt, but no more than 1 ms */
+
+	rtt = tp->rcv_rtt_est.rtt_us;
+	if (tp->srtt_us && tp->srtt_us < rtt)
+		rtt = tp->srtt_us;
+
+	delay = min_t(unsigned long, NSEC_PER_MSEC,
+		      rtt * (NSEC_PER_USEC >> 3)/20);
+	sock_hold(sk);
+	hrtimer_start(&tp->compressed_ack_timer, ns_to_ktime(delay),
+		      HRTIMER_MODE_REL_PINNED_SOFT);
 }
 
 static inline void tcp_ack_snd_check(struct sock *sk)
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 0d8f950a9006598c70dbf51e281a3fe32dfaa234..7ee98aad82b758674ca7f3e90bd3fc165e8fcd45 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -162,6 +162,13 @@ static void tcp_event_data_sent(struct tcp_sock *tp,
 /* Account for an ACK we sent. */
 static inline void tcp_event_ack_sent(struct sock *sk, unsigned int pkts)
 {
+	struct tcp_sock *tp = tcp_sk(sk);
+
+	if (unlikely(tp->compressed_ack)) {
+		tp->compressed_ack = 0;
+		if (hrtimer_try_to_cancel(&tp->compressed_ack_timer) == 1)
+			__sock_put(sk);
+	}
 	tcp_dec_quickack_mode(sk, pkts);
 	inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
 }
diff --git a/net/ipv4/tcp_timer.c b/net/ipv4/tcp_timer.c
index 92bdf64fffae3a5be291ca419eb21276b4c8cbae..3b3611729928f77934e0298bb248e55c7a7c5def 100644
--- a/net/ipv4/tcp_timer.c
+++ b/net/ipv4/tcp_timer.c
@@ -708,6 +708,27 @@ static void tcp_keepalive_timer (struct timer_list *t)
 	sock_put(sk);
 }
 
+static enum hrtimer_restart tcp_compressed_ack_kick(struct hrtimer *timer)
+{
+	struct tcp_sock *tp = container_of(timer, struct tcp_sock, compressed_ack_timer);
+	struct sock *sk = (struct sock *)tp;
+
+	bh_lock_sock(sk);
+	if (!sock_owned_by_user(sk)) {
+		if (tp->compressed_ack)
+			tcp_send_ack(sk);
+	} else {
+		if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED,
+				      &sk->sk_tsq_flags))
+			sock_hold(sk);
+	}
+	bh_unlock_sock(sk);
+
+	sock_put(sk);
+
+	return HRTIMER_NORESTART;
+}
+
 void tcp_init_xmit_timers(struct sock *sk)
 {
 	inet_csk_init_xmit_timers(sk, &tcp_write_timer, &tcp_delack_timer,
@@ -715,4 +736,8 @@ void tcp_init_xmit_timers(struct sock *sk)
 	hrtimer_init(&tcp_sk(sk)->pacing_timer, CLOCK_MONOTONIC,
 		     HRTIMER_MODE_ABS_PINNED_SOFT);
 	tcp_sk(sk)->pacing_timer.function = tcp_pace_kick;
+
+	hrtimer_init(&tcp_sk(sk)->compressed_ack_timer, CLOCK_MONOTONIC,
+		     HRTIMER_MODE_REL_PINNED_SOFT);
+	tcp_sk(sk)->compressed_ack_timer.function = tcp_compressed_ack_kick;
 }
-- 
2.17.0.441.gb46fe60e1d-goog

[PATCH v3 net-next 4/6] tcp: add TCPAckCompressed SNMP counter

[Eric Dumazet]

This counter tracks number of ACK packets that the host has not sent,
thanks to ACK compression.

Sample output :

$ nstat -n;sleep 1;nstat|egrep "IpInReceives|IpOutRequests|TcpInSegs|TcpOutSegs|TcpExtTCPAckCompressed"
IpInReceives                    123250             0.0
IpOutRequests                   3684               0.0
TcpInSegs                       123251             0.0
TcpOutSegs                      3684               0.0
TcpExtTCPAckCompressed          119252             0.0

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
---
 include/uapi/linux/snmp.h | 1 +
 net/ipv4/proc.c           | 1 +
 net/ipv4/tcp_output.c     | 2 ++
 3 files changed, 4 insertions(+)

diff --git a/include/uapi/linux/snmp.h b/include/uapi/linux/snmp.h
index d02e859301ff499dd72a1c0e1b56bed10a9397a6..750d89120335eb489f698191edb6c5110969fa8c 100644
--- a/include/uapi/linux/snmp.h
+++ b/include/uapi/linux/snmp.h
@@ -278,6 +278,7 @@ enum
 	LINUX_MIB_TCPMTUPSUCCESS,		/* TCPMTUPSuccess */
 	LINUX_MIB_TCPDELIVERED,			/* TCPDelivered */
 	LINUX_MIB_TCPDELIVEREDCE,		/* TCPDeliveredCE */
+	LINUX_MIB_TCPACKCOMPRESSED,		/* TCPAckCompressed */
 	__LINUX_MIB_MAX
 };
 
diff --git a/net/ipv4/proc.c b/net/ipv4/proc.c
index 261b71d0ccc5c17c6032bf67eb8f842006766e64..6c1ff89a60fa0a3485dcc71fafc799e798d5dc11 100644
--- a/net/ipv4/proc.c
+++ b/net/ipv4/proc.c
@@ -298,6 +298,7 @@ static const struct snmp_mib snmp4_net_list[] = {
 	SNMP_MIB_ITEM("TCPMTUPSuccess", LINUX_MIB_TCPMTUPSUCCESS),
 	SNMP_MIB_ITEM("TCPDelivered", LINUX_MIB_TCPDELIVERED),
 	SNMP_MIB_ITEM("TCPDeliveredCE", LINUX_MIB_TCPDELIVEREDCE),
+	SNMP_MIB_ITEM("TCPAckCompressed", LINUX_MIB_TCPACKCOMPRESSED),
 	SNMP_MIB_SENTINEL
 };
 
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 7ee98aad82b758674ca7f3e90bd3fc165e8fcd45..437bb7ceba7fd388abac1c12f2920b02be77bad9 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -165,6 +165,8 @@ static inline void tcp_event_ack_sent(struct sock *sk, unsigned int pkts)
 	struct tcp_sock *tp = tcp_sk(sk);
 
 	if (unlikely(tp->compressed_ack)) {
+		NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPACKCOMPRESSED,
+			      tp->compressed_ack);
 		tp->compressed_ack = 0;
 		if (hrtimer_try_to_cancel(&tp->compressed_ack_timer) == 1)
 			__sock_put(sk);
-- 
2.17.0.441.gb46fe60e1d-goog

[PATCH v3 net-next 5/6] tcp: add tcp_comp_sack_delay_ns sysctl

[Eric Dumazet]

This per netns sysctl allows for TCP SACK compression fine-tuning.

Its default value is 1,000,000, or 1 ms to meet TSO autosizing period.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 Documentation/networking/ip-sysctl.txt | 7 +++++++
 include/net/netns/ipv4.h               | 1 +
 net/ipv4/sysctl_net_ipv4.c             | 7 +++++++
 net/ipv4/tcp_input.c                   | 4 ++--
 net/ipv4/tcp_ipv4.c                    | 1 +
 5 files changed, 18 insertions(+), 2 deletions(-)

diff --git a/Documentation/networking/ip-sysctl.txt b/Documentation/networking/ip-sysctl.txt
index ea304a23c8d72c92a925d0c107bfd2bcfbbb92ec..7ba952959bca0eee4ecf81fb5837e17790db0fde 100644
--- a/Documentation/networking/ip-sysctl.txt
+++ b/Documentation/networking/ip-sysctl.txt
@@ -525,6 +525,13 @@ tcp_rmem - vector of 3 INTEGERs: min, default, max
 tcp_sack - BOOLEAN
 	Enable select acknowledgments (SACKS).
 
+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.
+	The default is 1ms, based on TSO autosizing period.
+
+	Default : 1,000,000 ns (1 ms)
+
 tcp_slow_start_after_idle - BOOLEAN
 	If set, provide RFC2861 behavior and time out the congestion
 	window after an idle period.  An idle period is defined at
diff --git a/include/net/netns/ipv4.h b/include/net/netns/ipv4.h
index 8491bc9c86b1553ab603e4363e8e38ca7ff547e0..927318243cfaa2ddd8eb423c6ba6e66253f771d3 100644
--- a/include/net/netns/ipv4.h
+++ b/include/net/netns/ipv4.h
@@ -160,6 +160,7 @@ struct netns_ipv4 {
 	int sysctl_tcp_pacing_ca_ratio;
 	int sysctl_tcp_wmem[3];
 	int sysctl_tcp_rmem[3];
+	unsigned long sysctl_tcp_comp_sack_delay_ns;
 	struct inet_timewait_death_row tcp_death_row;
 	int sysctl_max_syn_backlog;
 	int sysctl_tcp_fastopen;
diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c
index 4b195bac8ac0eefe0a224528ad854338c4f8e6e3..11fbfdc1566eca95f91360522178295318277588 100644
--- a/net/ipv4/sysctl_net_ipv4.c
+++ b/net/ipv4/sysctl_net_ipv4.c
@@ -1151,6 +1151,13 @@ static struct ctl_table ipv4_net_table[] = {
 		.proc_handler	= proc_dointvec_minmax,
 		.extra1		= &one,
 	},
+	{
+		.procname	= "tcp_comp_sack_delay_ns",
+		.data		= &init_net.ipv4.sysctl_tcp_comp_sack_delay_ns,
+		.maxlen		= sizeof(unsigned long),
+		.mode		= 0644,
+		.proc_handler	= proc_doulongvec_minmax,
+	},
 	{
 		.procname	= "udp_rmem_min",
 		.data		= &init_net.ipv4.sysctl_udp_rmem_min,
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index cc2ac5346b92b968593f919192d543384865bcb8..6a1dae38c9558c7bc9dd31e9f16c4e8ea8c78149 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -5113,13 +5113,13 @@ static void __tcp_ack_snd_check(struct sock *sk, int ofo_possible)
 	if (hrtimer_is_queued(&tp->compressed_ack_timer))
 		return;
 
-	/* compress ack timer : 5 % of rtt, but no more than 1 ms */
+	/* compress ack timer : 5 % 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, NSEC_PER_MSEC,
+	delay = min_t(unsigned long, sock_net(sk)->ipv4.sysctl_tcp_comp_sack_delay_ns,
 		      rtt * (NSEC_PER_USEC >> 3)/20);
 	sock_hold(sk);
 	hrtimer_start(&tp->compressed_ack_timer, ns_to_ktime(delay),
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index caf23de88f8a369c2038cecd34ce42c522487e90..a3f4647341db2eb5a63c3e9f1e8b93099aedadab 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -2572,6 +2572,7 @@ static int __net_init tcp_sk_init(struct net *net)
 		       init_net.ipv4.sysctl_tcp_wmem,
 		       sizeof(init_net.ipv4.sysctl_tcp_wmem));
 	}
+	net->ipv4.sysctl_tcp_comp_sack_delay_ns = NSEC_PER_MSEC;
 	net->ipv4.sysctl_tcp_fastopen = TFO_CLIENT_ENABLE;
 	spin_lock_init(&net->ipv4.tcp_fastopen_ctx_lock);
 	net->ipv4.sysctl_tcp_fastopen_blackhole_timeout = 60 * 60;
-- 
2.17.0.441.gb46fe60e1d-goog

[PATCH v3 net-next 6/6] tcp: add tcp_comp_sack_nr sysctl

[Eric Dumazet]

This per netns sysctl allows for TCP SACK compression fine-tuning.

This limits number of SACK that can be compressed.
Using 0 disables SACK compression.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 Documentation/networking/ip-sysctl.txt |  6 ++++++
 include/net/netns/ipv4.h               |  1 +
 net/ipv4/sysctl_net_ipv4.c             | 10 ++++++++++
 net/ipv4/tcp_input.c                   |  3 ++-
 net/ipv4/tcp_ipv4.c                    |  1 +
 5 files changed, 20 insertions(+), 1 deletion(-)

diff --git a/Documentation/networking/ip-sysctl.txt b/Documentation/networking/ip-sysctl.txt
index 7ba952959bca0eee4ecf81fb5837e17790db0fde..924bd51327b7a8dff3503d7afccdd54e1eb5c29b 100644
--- a/Documentation/networking/ip-sysctl.txt
+++ b/Documentation/networking/ip-sysctl.txt
@@ -532,6 +532,12 @@ tcp_comp_sack_delay_ns - LONG INTEGER
 
 	Default : 1,000,000 ns (1 ms)
 
+tcp_comp_sack_nr - INTEGER
+	Max numer of SACK that can be compressed.
+	Using 0 disables SACK compression.
+
+	Detault : 44
+
 tcp_slow_start_after_idle - BOOLEAN
 	If set, provide RFC2861 behavior and time out the congestion
 	window after an idle period.  An idle period is defined at
diff --git a/include/net/netns/ipv4.h b/include/net/netns/ipv4.h
index 927318243cfaa2ddd8eb423c6ba6e66253f771d3..661348f23ea5a3a9320b2cafcd17e23960214771 100644
--- a/include/net/netns/ipv4.h
+++ b/include/net/netns/ipv4.h
@@ -160,6 +160,7 @@ struct netns_ipv4 {
 	int sysctl_tcp_pacing_ca_ratio;
 	int sysctl_tcp_wmem[3];
 	int sysctl_tcp_rmem[3];
+	int sysctl_tcp_comp_sack_nr;
 	unsigned long sysctl_tcp_comp_sack_delay_ns;
 	struct inet_timewait_death_row tcp_death_row;
 	int sysctl_max_syn_backlog;
diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c
index 11fbfdc1566eca95f91360522178295318277588..d2eed3ddcb0a1ad9778d96d46c685f6c60b93d8d 100644
--- a/net/ipv4/sysctl_net_ipv4.c
+++ b/net/ipv4/sysctl_net_ipv4.c
@@ -46,6 +46,7 @@ static int tcp_syn_retries_min = 1;
 static int tcp_syn_retries_max = MAX_TCP_SYNCNT;
 static int ip_ping_group_range_min[] = { 0, 0 };
 static int ip_ping_group_range_max[] = { GID_T_MAX, GID_T_MAX };
+static int comp_sack_nr_max = 255;
 
 /* obsolete */
 static int sysctl_tcp_low_latency __read_mostly;
@@ -1158,6 +1159,15 @@ static struct ctl_table ipv4_net_table[] = {
 		.mode		= 0644,
 		.proc_handler	= proc_doulongvec_minmax,
 	},
+	{
+		.procname	= "tcp_comp_sack_nr",
+		.data		= &init_net.ipv4.sysctl_tcp_comp_sack_nr,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &zero,
+		.extra2		= &comp_sack_nr_max,
+	},
 	{
 		.procname	= "udp_rmem_min",
 		.data		= &init_net.ipv4.sysctl_udp_rmem_min,
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 6a1dae38c9558c7bc9dd31e9f16c4e8ea8c78149..aebb29ab2fdf2ceaa182cd11928f145a886149ff 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -5106,7 +5106,8 @@ static void __tcp_ack_snd_check(struct sock *sk, int ofo_possible)
 		return;
 	}
 
-	if (!tcp_is_sack(tp) || tp->compressed_ack >= 44)
+	if (!tcp_is_sack(tp) ||
+	    tp->compressed_ack >= sock_net(sk)->ipv4.sysctl_tcp_comp_sack_nr)
 		goto send_now;
 	tp->compressed_ack++;
 
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index a3f4647341db2eb5a63c3e9f1e8b93099aedadab..adbdb503db0c983ef4185f83b138aa51bafd15bf 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -2573,6 +2573,7 @@ static int __net_init tcp_sk_init(struct net *net)
 		       sizeof(init_net.ipv4.sysctl_tcp_wmem));
 	}
 	net->ipv4.sysctl_tcp_comp_sack_delay_ns = NSEC_PER_MSEC;
+	net->ipv4.sysctl_tcp_comp_sack_nr = 44;
 	net->ipv4.sysctl_tcp_fastopen = TFO_CLIENT_ENABLE;
 	spin_lock_init(&net->ipv4.tcp_fastopen_ctx_lock);
 	net->ipv4.sysctl_tcp_fastopen_blackhole_timeout = 60 * 60;
-- 
2.17.0.441.gb46fe60e1d-goog

Re: [PATCH v3 net-next 3/6] tcp: add SACK compression

[Neal Cardwell]

On Thu, May 17, 2018 at 5:47 PM Eric Dumazet <edumazet@google.com> wrote:

> When TCP receives an out-of-order packet, it immediately sends
> a SACK packet, generating network load but also forcing the
> receiver to send 1-MSS pathological packets, increasing its
> RTX queue length/depth, and thus processing time.

> Wifi networks suffer from this aggressive behavior, but generally
> speaking, all these SACK packets add fuel to the fire when networks
> are under congestion.

> This patch adds a high resolution timer and tp->compressed_ack counter.

> Instead of sending a SACK, we program this timer with a small delay,
> based on RTT and capped to 1 ms :

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

> If subsequent SACKs need to be sent while the timer has not yet
> expired, we simply increment tp->compressed_ack.

> When timer expires, a SACK is sent with the latest information.
> Whenever an ACK is sent (if data is sent, or if in-order
> data is received) timer is canceled.

> Note that tcp_sack_new_ofo_skb() is able to force a SACK to be sent
> if the sack blocks need to be shuffled, even if the timer has not
> expired.

> A new SNMP counter is added in the following patch.

> Two other patches add sysctls to allow changing the 1,000,000 and 44
> values that this commit hard-coded.

> Signed-off-by: Eric Dumazet <edumazet@google.com>
> ---

Very nice. I like the constants and the min(rcv_rtt, srtt).

Acked-by: Neal Cardwell <ncardwell@google.com>

Thanks!

neal

Re: [PATCH v3 net-next 5/6] tcp: add tcp_comp_sack_delay_ns sysctl

[Neal Cardwell]

On Thu, May 17, 2018 at 5:47 PM Eric Dumazet <edumazet@google.com> wrote:

> This per netns sysctl allows for TCP SACK compression fine-tuning.

> Its default value is 1,000,000, or 1 ms to meet TSO autosizing period.

> Signed-off-by: Eric Dumazet <edumazet@google.com>
> ---

Acked-by: Neal Cardwell <ncardwell@google.com>

Thanks!

neal

Re: [PATCH v3 net-next 6/6] tcp: add tcp_comp_sack_nr sysctl

[Neal Cardwell]

On Thu, May 17, 2018 at 5:47 PM Eric Dumazet <edumazet@google.com> wrote:

> This per netns sysctl allows for TCP SACK compression fine-tuning.

> This limits number of SACK that can be compressed.
> Using 0 disables SACK compression.

> Signed-off-by: Eric Dumazet <edumazet@google.com>
> ---

Acked-by: Neal Cardwell <ncardwell@google.com>

Thanks!

neal

Re: [PATCH v3 net-next 3/6] tcp: add SACK compression

[Yuchung Cheng]

On Thu, May 17, 2018 at 2:57 PM, Neal Cardwell <ncardwell@google.com> wrote:
> On Thu, May 17, 2018 at 5:47 PM Eric Dumazet <edumazet@google.com> wrote:
>
>> When TCP receives an out-of-order packet, it immediately sends
>> a SACK packet, generating network load but also forcing the
>> receiver to send 1-MSS pathological packets, increasing its
>> RTX queue length/depth, and thus processing time.
>
>> Wifi networks suffer from this aggressive behavior, but generally
>> speaking, all these SACK packets add fuel to the fire when networks
>> are under congestion.
>
>> This patch adds a high resolution timer and tp->compressed_ack counter.
>
>> Instead of sending a SACK, we program this timer with a small delay,
>> based on RTT and capped to 1 ms :
>
>>          delay = min ( 5 % of RTT, 1 ms)
>
>> If subsequent SACKs need to be sent while the timer has not yet
>> expired, we simply increment tp->compressed_ack.
>
>> When timer expires, a SACK is sent with the latest information.
>> Whenever an ACK is sent (if data is sent, or if in-order
>> data is received) timer is canceled.
>
>> Note that tcp_sack_new_ofo_skb() is able to force a SACK to be sent
>> if the sack blocks need to be shuffled, even if the timer has not
>> expired.
>
>> A new SNMP counter is added in the following patch.
>
>> Two other patches add sysctls to allow changing the 1,000,000 and 44
>> values that this commit hard-coded.
>
>> Signed-off-by: Eric Dumazet <edumazet@google.com>
>> ---
>
> Very nice. I like the constants and the min(rcv_rtt, srtt).
>
> Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>

Great work. Hopefully this would save middle-boxes' from handling
TCP-ACK themselves.

>
> Thanks!
>
> neal

Re: [PATCH v3 net-next 3/6] tcp: add SACK compression

[Toke Høiland-Jørgensen]

Eric Dumazet <edumazet@google.com> writes:

> When TCP receives an out-of-order packet, it immediately sends
> a SACK packet, generating network load but also forcing the
> receiver to send 1-MSS pathological packets, increasing its
> RTX queue length/depth, and thus processing time.
>
> Wifi networks suffer from this aggressive behavior, but generally
> speaking, all these SACK packets add fuel to the fire when networks
> are under congestion.
>
> This patch adds a high resolution timer and tp->compressed_ack counter.
>
> Instead of sending a SACK, we program this timer with a small delay,
> based on RTT and capped to 1 ms :
>
> 	delay = min ( 5 % of RTT, 1 ms)
>
> If subsequent SACKs need to be sent while the timer has not yet
> expired, we simply increment tp->compressed_ack.
>
> When timer expires, a SACK is sent with the latest information.
> Whenever an ACK is sent (if data is sent, or if in-order
> data is received) timer is canceled.
>
> Note that tcp_sack_new_ofo_skb() is able to force a SACK to be sent
> if the sack blocks need to be shuffled, even if the timer has not
> expired.
>
> A new SNMP counter is added in the following patch.
>
> Two other patches add sysctls to allow changing the 1,000,000 and 44
> values that this commit hard-coded.
>
> Signed-off-by: Eric Dumazet <edumazet@google.com>

Acked-by: Toke Høiland-Jørgensen <toke@toke.dk>

Re: [PATCH v3 net-next 0/6] tcp: implement SACK compression

[David Miller]

From: Eric Dumazet <edumazet@google.com>
Date: Thu, 17 May 2018 14:47:23 -0700

> When TCP receives an out-of-order packet, it immediately sends
> a SACK packet, generating network load but also forcing the
> receiver to send 1-MSS pathological packets, increasing its
> RTX queue length/depth, and thus processing time.
> 
> Wifi networks suffer from this aggressive behavior, but generally
> speaking, all these SACK packets add fuel to the fire when networks
> are under congestion.
> 
> This patch series adds SACK compression, but the infrastructure
> could be leveraged to also compress ACK in the future.
> 
> v2: Addressed Neal feedback.
>     Added two sysctls to allow fine tuning, or even disabling the feature.
> 
> v3: take rtt = min(srtt, rcv_rtt) as Yuchung suggested, because rcv_rtt
>     can be over estimated for RPC (or sender limited)

This looks great, series applied, thanks Eric.

So now we handle locally terminated traffic well, however we still need
something that handles traffic flowing through the system.  And for that
reason the CAKE ACK filter still has merit.

I completely agree with you that it has to be implemented properly,
parse all TCP options and the SACK blocks, and pass the ACKs through
when there are options it doesn't understand or the SACK/timestampe/etc.
update contains new information that must be passed along and not dropper.

Thanks again.

Re: [PATCH v3 net-next 0/6] tcp: implement SACK compression

[Eric Dumazet]

On 05/18/2018 08:42 AM, David Miller wrote:

> This looks great, series applied, thanks Eric.
> 
> So now we handle locally terminated traffic well, however we still need
> something that handles traffic flowing through the system.  And for that
> reason the CAKE ACK filter still has merit.
> 
> I completely agree with you that it has to be implemented properly,
> parse all TCP options and the SACK blocks, and pass the ACKs through
> when there are options it doesn't understand or the SACK/timestampe/etc.
> update contains new information that must be passed along and not dropper.
> 
> Thanks again.
> 

Be assured I will carefully review Cake ACK filter ;)

Thanks David !