Merge branch 'tcp-bbr'

Neal Cardwell says:

====================
tcp: BBR congestion control algorithm

This patch series implements a new TCP congestion control algorithm:
BBR (Bottleneck Bandwidth and RTT). A paper with a detailed
description of BBR will be published in ACM Queue, September-October
2016, as "BBR: Congestion-Based Congestion Control". BBR is widely
deployed in production at Google.

The patch series starts with a set of supporting infrastructure
changes, including a few that extend the congestion control
framework. The last patch adds BBR as a TCP congestion control
module. Please see individual patches for the details.

- v3 -> v4:
 - Updated tcp_bbr.c in "tcp_bbr: add BBR congestion control"
   to use const to qualify all the constant parameters.
   Thanks to Stephen Hemminger.
 - In "tcp_bbr: add BBR congestion control", remove the bbr_rate_kbps()
   function, which had a 64-bit divide that would be problematic on some
   architectures, and just use bbr_rate_bytes_per_sec() directly.
   Thanks to Kenneth Klette Jonassen for suggesting this.
 - In "tcp: switch back to proper tcp_skb_cb size check in tcp_init()",
   switched from sizeof(skb->cb) to FIELD_SIZEOF.
   Thanks to Lance Richardson for suggesting this.
 - Updated "tcp_bbr: add BBR congestion control" commit message with
   performance data, more details about deployment at Google, and
   another reminder to use fq with BBR.
 - Updated tcp_bbr.c in "tcp_bbr: add BBR congestion control"
   to use MODULE_LICENSE("Dual BSD/GPL").

- v2 -> v3: fix another issue caught by build bots:
 - adjust rate_sample struct initialization syntax to allow gcc-4.4 to compile
   the "tcp: track data delivery rate for a TCP connection" patch; also
   adjusted some similar syntax in "tcp_bbr: add BBR congestion control"

- v1 -> v2: fix issues caught by build bots:
 - fix "tcp: export data delivery rate" to use rate64 instead of rate,
   so there is a 64-bit numerator for the do_div call
 - fix conflicting definitions for minmax caused by
   "tcp: use windowed min filter library for TCP min_rtt estimation"
   with a new commit:
   tcp: cdg: rename struct minmax in tcp_cdg.c to avoid a naming conflict
 - fix warning about the use of __packed in
   "tcp: track data delivery rate for a TCP connection",
   which involves the addition of a new commit:
   tcp: switch back to proper tcp_skb_cb size check in tcp_init()
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

include/linux/tcp.h

@@ -19,6 +19,7 @@
 
 
 #include <linux/skbuff.h>
+#include <linux/win_minmax.h>
 #include <net/sock.h>
 #include <net/inet_connection_sock.h>
 #include <net/inet_timewait_sock.h>
@@ -212,7 +213,8 @@ struct tcp_sock {
 		u8 reord;    /* reordering detected */
 	} rack;
 	u16	advmss;		/* Advertised MSS			*/
-	u8	unused;
+	u8	rate_app_limited:1,  /* rate_{delivered,interval_us} limited? */
+		unused:7;
 	u8	nonagle     : 4,/* Disable Nagle algorithm?             */
 		thin_lto    : 1,/* Use linear timeouts for thin streams */
 		thin_dupack : 1,/* Fast retransmit on first dupack      */
@@ -234,9 +236,7 @@ struct tcp_sock {
 	u32	mdev_max_us;	/* maximal mdev for the last rtt period	*/
 	u32	rttvar_us;	/* smoothed mdev_max			*/
 	u32	rtt_seq;	/* sequence number to update rttvar	*/
-	struct rtt_meas {
-		u32 rtt, ts;	/* RTT in usec and sampling time in jiffies. */
-	} rtt_min[3];
+	struct  minmax rtt_min;
 
 	u32	packets_out;	/* Packets which are "in flight"	*/
 	u32	retrans_out;	/* Retransmitted packets out		*/
@@ -268,6 +268,12 @@ struct tcp_sock {
 				 * receiver in Recovery. */
 	u32	prr_out;	/* Total number of pkts sent during Recovery. */
 	u32	delivered;	/* Total data packets delivered incl. rexmits */
+	u32	lost;		/* Total data packets lost incl. rexmits */
+	u32	app_limited;	/* limited until "delivered" reaches this val */
+	struct skb_mstamp first_tx_mstamp;  /* start of window send phase */
+	struct skb_mstamp delivered_mstamp; /* time we reached "delivered" */
+	u32	rate_delivered;    /* saved rate sample: packets delivered */
+	u32	rate_interval_us;  /* saved rate sample: time elapsed */
 
  	u32	rcv_wnd;	/* Current receiver window		*/
 	u32	write_seq;	/* Tail(+1) of data held in tcp send buffer */

include/linux/win_minmax.h

+/**
+ * lib/minmax.c: windowed min/max tracker by Kathleen Nichols.
+ *
+ */
+#ifndef MINMAX_H
+#define MINMAX_H
+
+#include <linux/types.h>
+
+/* A single data point for our parameterized min-max tracker */
+struct minmax_sample {
+	u32	t;	/* time measurement was taken */
+	u32	v;	/* value measured */
+};
+
+/* State for the parameterized min-max tracker */
+struct minmax {
+	struct minmax_sample s[3];
+};
+
+static inline u32 minmax_get(const struct minmax *m)
+{
+	return m->s[0].v;
+}
+
+static inline u32 minmax_reset(struct minmax *m, u32 t, u32 meas)
+{
+	struct minmax_sample val = { .t = t, .v = meas };
+
+	m->s[2] = m->s[1] = m->s[0] = val;
+	return m->s[0].v;
+}
+
+u32 minmax_running_max(struct minmax *m, u32 win, u32 t, u32 meas);
+u32 minmax_running_min(struct minmax *m, u32 win, u32 t, u32 meas);
+
+#endif

include/net/inet_connection_sock.h

@@ -134,8 +134,8 @@ struct inet_connection_sock {
 	} icsk_mtup;
 	u32			  icsk_user_timeout;
 
-	u64			  icsk_ca_priv[64 / sizeof(u64)];
-#define ICSK_CA_PRIV_SIZE      (8 * sizeof(u64))
+	u64			  icsk_ca_priv[88 / sizeof(u64)];
+#define ICSK_CA_PRIV_SIZE      (11 * sizeof(u64))
 };
 
 #define ICSK_TIME_RETRANS	1	/* Retransmit timer */

include/net/tcp.h

@@ -533,6 +533,8 @@ __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
 #endif
 /* tcp_output.c */
 
+u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
+		     int min_tso_segs);
 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
 			       int nonagle);
 bool tcp_may_send_now(struct sock *sk);
@@ -671,7 +673,7 @@ static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
 /* Minimum RTT in usec. ~0 means not available. */
 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
 {
-	return tp->rtt_min[0].rtt;
+	return minmax_get(&tp->rtt_min);
 }
 
 /* Compute the actual receive window we are currently advertising.
@@ -763,8 +765,16 @@ struct tcp_skb_cb {
 	__u32		ack_seq;	/* Sequence number ACK'd	*/
 	union {
 		struct {
-			/* There is space for up to 20 bytes */
-			__u32 in_flight;/* Bytes in flight when packet sent */
+			/* There is space for up to 24 bytes */
+			__u32 in_flight:30,/* Bytes in flight at transmit */
+			      is_app_limited:1, /* cwnd not fully used? */
+			      unused:1;
+			/* pkts S/ACKed so far upon tx of skb, incl retrans: */
+			__u32 delivered;
+			/* start of send pipeline phase */
+			struct skb_mstamp first_tx_mstamp;
+			/* when we reached the "delivered" count */
+			struct skb_mstamp delivered_mstamp;
 		} tx;   /* only used for outgoing skbs */
 		union {
 			struct inet_skb_parm	h4;
@@ -860,6 +870,27 @@ struct ack_sample {
 	u32 in_flight;
 };
 
+/* A rate sample measures the number of (original/retransmitted) data
+ * packets delivered "delivered" over an interval of time "interval_us".
+ * The tcp_rate.c code fills in the rate sample, and congestion
+ * control modules that define a cong_control function to run at the end
+ * of ACK processing can optionally chose to consult this sample when
+ * setting cwnd and pacing rate.
+ * A sample is invalid if "delivered" or "interval_us" is negative.
+ */
+struct rate_sample {
+	struct	skb_mstamp prior_mstamp; /* starting timestamp for interval */
+	u32  prior_delivered;	/* tp->delivered at "prior_mstamp" */
+	s32  delivered;		/* number of packets delivered over interval */
+	long interval_us;	/* time for tp->delivered to incr "delivered" */
+	long rtt_us;		/* RTT of last (S)ACKed packet (or -1) */
+	int  losses;		/* number of packets marked lost upon ACK */
+	u32  acked_sacked;	/* number of packets newly (S)ACKed upon ACK */
+	u32  prior_in_flight;	/* in flight before this ACK */
+	bool is_app_limited;	/* is sample from packet with bubble in pipe? */
+	bool is_retrans;	/* is sample from retransmission? */
+};
+
 struct tcp_congestion_ops {
 	struct list_head	list;
 	u32 key;
@@ -884,6 +915,14 @@ struct tcp_congestion_ops {
 	u32  (*undo_cwnd)(struct sock *sk);
 	/* hook for packet ack accounting (optional) */
 	void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
+	/* suggest number of segments for each skb to transmit (optional) */
+	u32 (*tso_segs_goal)(struct sock *sk);
+	/* returns the multiplier used in tcp_sndbuf_expand (optional) */
+	u32 (*sndbuf_expand)(struct sock *sk);
+	/* call when packets are delivered to update cwnd and pacing rate,
+	 * after all the ca_state processing. (optional)
+	 */
+	void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
 	/* get info for inet_diag (optional) */
 	size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
 			   union tcp_cc_info *info);
@@ -946,6 +985,14 @@ static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
 		icsk->icsk_ca_ops->cwnd_event(sk, event);
 }
 
+/* From tcp_rate.c */
+void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
+void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
+			    struct rate_sample *rs);
+void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
+		  struct skb_mstamp *now, struct rate_sample *rs);
+void tcp_rate_check_app_limited(struct sock *sk);
+
 /* These functions determine how the current flow behaves in respect of SACK
  * handling. SACK is negotiated with the peer, and therefore it can vary
  * between different flows.

include/uapi/linux/inet_diag.h

@@ -124,6 +124,7 @@ enum {
 	INET_DIAG_PEERS,
 	INET_DIAG_PAD,
 	INET_DIAG_MARK,
+	INET_DIAG_BBRINFO,
 	__INET_DIAG_MAX,
 };
 
@@ -157,8 +158,20 @@ struct tcp_dctcp_info {
 	__u32	dctcp_ab_tot;
 };
 
+/* INET_DIAG_BBRINFO */
+
+struct tcp_bbr_info {
+	/* u64 bw: max-filtered BW (app throughput) estimate in Byte per sec: */
+	__u32	bbr_bw_lo;		/* lower 32 bits of bw */
+	__u32	bbr_bw_hi;		/* upper 32 bits of bw */
+	__u32	bbr_min_rtt;		/* min-filtered RTT in uSec */
+	__u32	bbr_pacing_gain;	/* pacing gain shifted left 8 bits */
+	__u32	bbr_cwnd_gain;		/* cwnd gain shifted left 8 bits */
+};
+
 union tcp_cc_info {
 	struct tcpvegas_info	vegas;
 	struct tcp_dctcp_info	dctcp;
+	struct tcp_bbr_info	bbr;
 };
 #endif /* _UAPI_INET_DIAG_H_ */

include/uapi/linux/pkt_sched.h

@@ -792,6 +792,8 @@ enum {
 
 	TCA_FQ_ORPHAN_MASK,	/* mask applied to orphaned skb hashes */
 
+	TCA_FQ_LOW_RATE_THRESHOLD, /* per packet delay under this rate */
+
 	__TCA_FQ_MAX
 };
 

include/uapi/linux/tcp.h

@@ -167,6 +167,7 @@ struct tcp_info {
 	__u8	tcpi_backoff;
 	__u8	tcpi_options;
 	__u8	tcpi_snd_wscale : 4, tcpi_rcv_wscale : 4;
+	__u8	tcpi_delivery_rate_app_limited:1;
 
 	__u32	tcpi_rto;
 	__u32	tcpi_ato;
@@ -211,6 +212,8 @@ struct tcp_info {
 	__u32	tcpi_min_rtt;
 	__u32	tcpi_data_segs_in;	/* RFC4898 tcpEStatsDataSegsIn */
 	__u32	tcpi_data_segs_out;	/* RFC4898 tcpEStatsDataSegsOut */
+
+	__u64   tcpi_delivery_rate;
 };
 
 /* for TCP_MD5SIG socket option */

lib/Makefile

@@ -22,7 +22,7 @@ lib-y := ctype.o string.o vsprintf.o cmdline.o \
 	 sha1.o chacha20.o md5.o irq_regs.o argv_split.o \
 	 flex_proportions.o ratelimit.o show_mem.o \
 	 is_single_threaded.o plist.o decompress.o kobject_uevent.o \
-	 earlycpio.o seq_buf.o nmi_backtrace.o nodemask.o
+	 earlycpio.o seq_buf.o nmi_backtrace.o nodemask.o win_minmax.o
 
 lib-$(CONFIG_MMU) += ioremap.o
 lib-$(CONFIG_SMP) += cpumask.o

lib/win_minmax.c

+/**
+ * lib/minmax.c: windowed min/max tracker
+ *
+ * Kathleen Nichols' algorithm for tracking the minimum (or maximum)
+ * value of a data stream over some fixed time interval.  (E.g.,
+ * the minimum RTT over the past five minutes.) It uses constant
+ * space and constant time per update yet almost always delivers
+ * the same minimum as an implementation that has to keep all the
+ * data in the window.
+ *
+ * The algorithm keeps track of the best, 2nd best & 3rd best min
+ * values, maintaining an invariant that the measurement time of
+ * the n'th best >= n-1'th best. It also makes sure that the three
+ * values are widely separated in the time window since that bounds
+ * the worse case error when that data is monotonically increasing
+ * over the window.
+ *
+ * Upon getting a new min, we can forget everything earlier because
+ * it has no value - the new min is <= everything else in the window
+ * by definition and it's the most recent. So we restart fresh on
+ * every new min and overwrites 2nd & 3rd choices. The same property
+ * holds for 2nd & 3rd best.
+ */
+#include <linux/module.h>
+#include <linux/win_minmax.h>
+
+/* As time advances, update the 1st, 2nd, and 3rd choices. */
+static u32 minmax_subwin_update(struct minmax *m, u32 win,
+				const struct minmax_sample *val)
+{
+	u32 dt = val->t - m->s[0].t;
+
+	if (unlikely(dt > win)) {
+		/*
+		 * Passed entire window without a new val so make 2nd
+		 * choice the new val & 3rd choice the new 2nd choice.
+		 * we may have to iterate this since our 2nd choice
+		 * may also be outside the window (we checked on entry
+		 * that the third choice was in the window).
+		 */
+		m->s[0] = m->s[1];
+		m->s[1] = m->s[2];
+		m->s[2] = *val;
+		if (unlikely(val->t - m->s[0].t > win)) {
+			m->s[0] = m->s[1];
+			m->s[1] = m->s[2];
+			m->s[2] = *val;
+		}
+	} else if (unlikely(m->s[1].t == m->s[0].t) && dt > win/4) {
+		/*
+		 * We've passed a quarter of the window without a new val
+		 * so take a 2nd choice from the 2nd quarter of the window.
+		 */
+		m->s[2] = m->s[1] = *val;
+	} else if (unlikely(m->s[2].t == m->s[1].t) && dt > win/2) {
+		/*
+		 * We've passed half the window without finding a new val
+		 * so take a 3rd choice from the last half of the window
+		 */
+		m->s[2] = *val;
+	}
+	return m->s[0].v;
+}
+
+/* Check if new measurement updates the 1st, 2nd or 3rd choice max. */
+u32 minmax_running_max(struct minmax *m, u32 win, u32 t, u32 meas)
+{
+	struct minmax_sample val = { .t = t, .v = meas };
+
+	if (unlikely(val.v >= m->s[0].v) ||	  /* found new max? */
+	    unlikely(val.t - m->s[2].t > win))	  /* nothing left in window? */
+		return minmax_reset(m, t, meas);  /* forget earlier samples */
+
+	if (unlikely(val.v >= m->s[1].v))
+		m->s[2] = m->s[1] = val;
+	else if (unlikely(val.v >= m->s[2].v))
+		m->s[2] = val;
+
+	return minmax_subwin_update(m, win, &val);
+}
+EXPORT_SYMBOL(minmax_running_max);
+
+/* Check if new measurement updates the 1st, 2nd or 3rd choice min. */
+u32 minmax_running_min(struct minmax *m, u32 win, u32 t, u32 meas)
+{
+	struct minmax_sample val = { .t = t, .v = meas };
+
+	if (unlikely(val.v <= m->s[0].v) ||	  /* found new min? */
+	    unlikely(val.t - m->s[2].t > win))	  /* nothing left in window? */
+		return minmax_reset(m, t, meas);  /* forget earlier samples */
+
+	if (unlikely(val.v <= m->s[1].v))
+		m->s[2] = m->s[1] = val;
+	else if (unlikely(val.v <= m->s[2].v))
+		m->s[2] = val;
+
+	return minmax_subwin_update(m, win, &val);
+}

net/ipv4/Kconfig

@@ -640,6 +640,21 @@ config TCP_CONG_CDG
 	  D.A. Hayes and G. Armitage. "Revisiting TCP congestion control using
 	  delay gradients." In Networking 2011. Preprint: http://goo.gl/No3vdg
 
+config TCP_CONG_BBR
+	tristate "BBR TCP"
+	default n
+	---help---
+
+	BBR (Bottleneck Bandwidth and RTT) TCP congestion control aims to
+	maximize network utilization and minimize queues. It builds an explicit
+	model of the the bottleneck delivery rate and path round-trip
+	propagation delay. It tolerates packet loss and delay unrelated to
+	congestion. It can operate over LAN, WAN, cellular, wifi, or cable
+	modem links. It can coexist with flows that use loss-based congestion
+	control, and can operate with shallow buffers, deep buffers,
+	bufferbloat, policers, or AQM schemes that do not provide a delay
+	signal. It requires the fq ("Fair Queue") pacing packet scheduler.
+
 choice
 	prompt "Default TCP congestion control"
 	default DEFAULT_CUBIC
@@ -674,6 +689,9 @@ choice
 	config DEFAULT_CDG
 		bool "CDG" if TCP_CONG_CDG=y
 
+	config DEFAULT_BBR
+		bool "BBR" if TCP_CONG_BBR=y
+
 	config DEFAULT_RENO
 		bool "Reno"
 endchoice

net/ipv4/Makefile

@@ -8,7 +8,7 @@ obj-y     := route.o inetpeer.o protocol.o \
 	     inet_timewait_sock.o inet_connection_sock.o \
 	     tcp.o tcp_input.o tcp_output.o tcp_timer.o tcp_ipv4.o \
 	     tcp_minisocks.o tcp_cong.o tcp_metrics.o tcp_fastopen.o \
-	     tcp_recovery.o \
+	     tcp_rate.o tcp_recovery.o \
 	     tcp_offload.o datagram.o raw.o udp.o udplite.o \
 	     udp_offload.o arp.o icmp.o devinet.o af_inet.o igmp.o \
 	     fib_frontend.o fib_semantics.o fib_trie.o \
@@ -41,6 +41,7 @@ obj-$(CONFIG_INET_DIAG) += inet_diag.o
 obj-$(CONFIG_INET_TCP_DIAG) += tcp_diag.o
 obj-$(CONFIG_INET_UDP_DIAG) += udp_diag.o
 obj-$(CONFIG_NET_TCPPROBE) += tcp_probe.o
+obj-$(CONFIG_TCP_CONG_BBR) += tcp_bbr.o
 obj-$(CONFIG_TCP_CONG_BIC) += tcp_bic.o
 obj-$(CONFIG_TCP_CONG_CDG) += tcp_cdg.o
 obj-$(CONFIG_TCP_CONG_CUBIC) += tcp_cubic.o

net/ipv4/tcp.c

@@ -387,7 +387,7 @@ void tcp_init_sock(struct sock *sk)
 
 	icsk->icsk_rto = TCP_TIMEOUT_INIT;
 	tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
-	tp->rtt_min[0].rtt = ~0U;
+	minmax_reset(&tp->rtt_min, tcp_time_stamp, ~0U);
 
 	/* So many TCP implementations out there (incorrectly) count the
 	 * initial SYN frame in their delayed-ACK and congestion control
@@ -396,6 +396,9 @@ void tcp_init_sock(struct sock *sk)
 	 */
 	tp->snd_cwnd = TCP_INIT_CWND;
 
+	/* There's a bubble in the pipe until at least the first ACK. */
+	tp->app_limited = ~0U;
+
 	/* See draft-stevens-tcpca-spec-01 for discussion of the
 	 * initialization of these values.
 	 */
@@ -1014,6 +1017,9 @@ int tcp_sendpage(struct sock *sk, struct page *page, int offset,
 					flags);
 
 	lock_sock(sk);
+
+	tcp_rate_check_app_limited(sk);  /* is sending application-limited? */
+
 	res = do_tcp_sendpages(sk, page, offset, size, flags);
 	release_sock(sk);
 	return res;
@@ -1115,6 +1121,8 @@ int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
 
 	timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
 
+	tcp_rate_check_app_limited(sk);  /* is sending application-limited? */
+
 	/* Wait for a connection to finish. One exception is TCP Fast Open
 	 * (passive side) where data is allowed to be sent before a connection
 	 * is fully established.
@@ -2704,7 +2712,7 @@ 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);
-	u32 now = tcp_time_stamp;
+	u32 now = tcp_time_stamp, intv;
 	unsigned int start;
 	int notsent_bytes;
 	u64 rate64;
@@ -2794,6 +2802,15 @@ void tcp_get_info(struct sock *sk, struct tcp_info *info)
 	info->tcpi_min_rtt = tcp_min_rtt(tp);
 	info->tcpi_data_segs_in = tp->data_segs_in;
 	info->tcpi_data_segs_out = tp->data_segs_out;
+
+	info->tcpi_delivery_rate_app_limited = tp->rate_app_limited ? 1 : 0;
+	rate = READ_ONCE(tp->rate_delivered);
+	intv = READ_ONCE(tp->rate_interval_us);
+	if (rate && intv) {
+		rate64 = (u64)rate * tp->mss_cache * USEC_PER_SEC;
+		do_div(rate64, intv);
+		put_unaligned(rate64, &info->tcpi_delivery_rate);
+	}
 }
 EXPORT_SYMBOL_GPL(tcp_get_info);
 
@@ -3261,11 +3278,12 @@ static void __init tcp_init_mem(void)
 
 void __init tcp_init(void)
 {
-	unsigned long limit;
 	int max_rshare, max_wshare, cnt;
+	unsigned long limit;
 	unsigned int i;
 
-	sock_skb_cb_check_size(sizeof(struct tcp_skb_cb));
+	BUILD_BUG_ON(sizeof(struct tcp_skb_cb) >
+		     FIELD_SIZEOF(struct sk_buff, cb));
 
 	percpu_counter_init(&tcp_sockets_allocated, 0, GFP_KERNEL);
 	percpu_counter_init(&tcp_orphan_count, 0, GFP_KERNEL);

net/ipv4/tcp_bbr.c

+/* Bottleneck Bandwidth and RTT (BBR) congestion control
+ *
+ * BBR congestion control computes the sending rate based on the delivery
+ * rate (throughput) estimated from ACKs. In a nutshell:
+ *
+ *   On each ACK, update our model of the network path:
+ *      bottleneck_bandwidth = windowed_max(delivered / elapsed, 10 round trips)
+ *      min_rtt = windowed_min(rtt, 10 seconds)
+ *   pacing_rate = pacing_gain * bottleneck_bandwidth
+ *   cwnd = max(cwnd_gain * bottleneck_bandwidth * min_rtt, 4)
+ *
+ * The core algorithm does not react directly to packet losses or delays,
+ * although BBR may adjust the size of next send per ACK when loss is
+ * observed, or adjust the sending rate if it estimates there is a
+ * traffic policer, in order to keep the drop rate reasonable.
+ *
+ * BBR is described in detail in:
+ *   "BBR: Congestion-Based Congestion Control",
+ *   Neal Cardwell, Yuchung Cheng, C. Stephen Gunn, Soheil Hassas Yeganeh,
+ *   Van Jacobson. ACM Queue, Vol. 14 No. 5, September-October 2016.
+ *
+ * There is a public e-mail list for discussing BBR development and testing:
+ *   https://groups.google.com/forum/#!forum/bbr-dev
+ *
+ * NOTE: BBR *must* be used with the fq qdisc ("man tc-fq") with pacing enabled,
+ * since pacing is integral to the BBR design and implementation.
+ * BBR without pacing would not function properly, and may incur unnecessary
+ * high packet loss rates.
+ */
+#include <linux/module.h>
+#include <net/tcp.h>
+#include <linux/inet_diag.h>
+#include <linux/inet.h>
+#include <linux/random.h>
+#include <linux/win_minmax.h>
+
+/* Scale factor for rate in pkt/uSec unit to avoid truncation in bandwidth
+ * estimation. The rate unit ~= (1500 bytes / 1 usec / 2^24) ~= 715 bps.
+ * This handles bandwidths from 0.06pps (715bps) to 256Mpps (3Tbps) in a u32.
+ * Since the minimum window is >=4 packets, the lower bound isn't
+ * an issue. The upper bound isn't an issue with existing technologies.
+ */
+#define BW_SCALE 24
+#define BW_UNIT (1 << BW_SCALE)
+
+#define BBR_SCALE 8	/* scaling factor for fractions in BBR (e.g. gains) */
+#define BBR_UNIT (1 << BBR_SCALE)
+
+/* BBR has the following modes for deciding how fast to send: */
+enum bbr_mode {
+	BBR_STARTUP,	/* ramp up sending rate rapidly to fill pipe */
+	BBR_DRAIN,	/* drain any queue created during startup */
+	BBR_PROBE_BW,	/* discover, share bw: pace around estimated bw */
+	BBR_PROBE_RTT,	/* cut cwnd to min to probe min_rtt */
+};
+
+/* BBR congestion control block */
+struct bbr {
+	u32	min_rtt_us;	        /* min RTT in min_rtt_win_sec window */
+	u32	min_rtt_stamp;	        /* timestamp of min_rtt_us */
+	u32	probe_rtt_done_stamp;   /* end time for BBR_PROBE_RTT mode */
+	struct minmax bw;	/* Max recent delivery rate in pkts/uS << 24 */
+	u32	rtt_cnt;	    /* count of packet-timed rounds elapsed */
+	u32     next_rtt_delivered; /* scb->tx.delivered at end of round */
+	struct skb_mstamp cycle_mstamp;  /* time of this cycle phase start */
+	u32     mode:3,		     /* current bbr_mode in state machine */
+		prev_ca_state:3,     /* CA state on previous ACK */
+		packet_conservation:1,  /* use packet conservation? */
+		restore_cwnd:1,	     /* decided to revert cwnd to old value */
+		round_start:1,	     /* start of packet-timed tx->ack round? */
+		tso_segs_goal:7,     /* segments we want in each skb we send */
+		idle_restart:1,	     /* restarting after idle? */
+		probe_rtt_round_done:1,  /* a BBR_PROBE_RTT round at 4 pkts? */
+		unused:5,
+		lt_is_sampling:1,    /* taking long-term ("LT") samples now? */
+		lt_rtt_cnt:7,	     /* round trips in long-term interval */
+		lt_use_bw:1;	     /* use lt_bw as our bw estimate? */
+	u32	lt_bw;		     /* LT est delivery rate in pkts/uS << 24 */
+	u32	lt_last_delivered;   /* LT intvl start: tp->delivered */
+	u32	lt_last_stamp;	     /* LT intvl start: tp->delivered_mstamp */
+	u32	lt_last_lost;	     /* LT intvl start: tp->lost */
+	u32	pacing_gain:10,	/* current gain for setting pacing rate */
+		cwnd_gain:10,	/* current gain for setting cwnd */
+		full_bw_cnt:3,	/* number of rounds without large bw gains */
+		cycle_idx:3,	/* current index in pacing_gain cycle array */
+		unused_b:6;
+	u32	prior_cwnd;	/* prior cwnd upon entering loss recovery */
+	u32	full_bw;	/* recent bw, to estimate if pipe is full */
+};
+
+#define CYCLE_LEN	8	/* number of phases in a pacing gain cycle */
+
+/* Window length of bw filter (in rounds): */
+static const int bbr_bw_rtts = CYCLE_LEN + 2;
+/* Window length of min_rtt filter (in sec): */
+static const u32 bbr_min_rtt_win_sec = 10;
+/* Minimum time (in ms) spent at bbr_cwnd_min_target in BBR_PROBE_RTT mode: */
+static const u32 bbr_probe_rtt_mode_ms = 200;
+/* Skip TSO below the following bandwidth (bits/sec): */
+static const int bbr_min_tso_rate = 1200000;
+
+/* 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
+ * and send the same number of packets per RTT that an un-paced, slow-starting
+ * Reno or CUBIC flow would:
+ */
+static const int bbr_high_gain  = BBR_UNIT * 2885 / 1000 + 1;
+/* The pacing gain of 1/high_gain in BBR_DRAIN is calculated to typically drain
+ * the queue created in BBR_STARTUP in a single round:
+ */
+static const int bbr_drain_gain = BBR_UNIT * 1000 / 2885;
+/* The gain for deriving steady-state cwnd tolerates delayed/stretched ACKs: */
+static const int bbr_cwnd_gain  = BBR_UNIT * 2;
+/* The pacing_gain values for the PROBE_BW gain cycle, to discover/share bw: */
+static const int bbr_pacing_gain[] = {
+	BBR_UNIT * 5 / 4,	/* probe for more available bw */
+	BBR_UNIT * 3 / 4,	/* drain queue and/or yield bw to other flows */
+	BBR_UNIT, BBR_UNIT, BBR_UNIT,	/* cruise at 1.0*bw to utilize pipe, */
+	BBR_UNIT, BBR_UNIT, BBR_UNIT	/* without creating excess queue... */
+};
+/* Randomize the starting gain cycling phase over N phases: */
+static const u32 bbr_cycle_rand = 7;
+
+/* Try to keep at least this many packets in flight, if things go smoothly. For
+ * smooth functioning, a sliding window protocol ACKing every other packet
+ * needs at least 4 packets in flight:
+ */
+static const u32 bbr_cwnd_min_target = 4;
+
+/* To estimate if BBR_STARTUP mode (i.e. high_gain) has filled pipe... */
+/* If bw has increased significantly (1.25x), there may be more bw available: */
+static const u32 bbr_full_bw_thresh = BBR_UNIT * 5 / 4;
+/* But after 3 rounds w/o significant bw growth, estimate pipe is full: */
+static const u32 bbr_full_bw_cnt = 3;
+
+/* "long-term" ("LT") bandwidth estimator parameters... */
+/* The minimum number of rounds in an LT bw sampling interval: */
+static const u32 bbr_lt_intvl_min_rtts = 4;
+/* If lost/delivered ratio > 20%, interval is "lossy" and we may be policed: */
+static const u32 bbr_lt_loss_thresh = 50;
+/* If 2 intervals have a bw ratio <= 1/8, their bw is "consistent": */
+static const u32 bbr_lt_bw_ratio = BBR_UNIT / 8;
+/* If 2 intervals have a bw diff <= 4 Kbit/sec their bw is "consistent": */
+static const u32 bbr_lt_bw_diff = 4000 / 8;
+/* If we estimate we're policed, use lt_bw for this many round trips: */
+static const u32 bbr_lt_bw_max_rtts = 48;
+
+/* Do we estimate that STARTUP filled the pipe? */
+static bool bbr_full_bw_reached(const struct sock *sk)
+{
+	const struct bbr *bbr = inet_csk_ca(sk);
+
+	return bbr->full_bw_cnt >= bbr_full_bw_cnt;
+}
+
+/* Return the windowed max recent bandwidth sample, in pkts/uS << BW_SCALE. */
+static u32 bbr_max_bw(const struct sock *sk)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	return minmax_get(&bbr->bw);
+}
+
+/* Return the estimated bandwidth of the path, in pkts/uS << BW_SCALE. */
+static u32 bbr_bw(const struct sock *sk)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	return bbr->lt_use_bw ? bbr->lt_bw : bbr_max_bw(sk);
+}
+
+/* Return rate in bytes per second, optionally with a gain.
+ * The order here is chosen carefully to avoid overflow of u64. This should
+ * work for input rates of up to 2.9Tbit/sec and gain of 2.89x.
+ */
+static u64 bbr_rate_bytes_per_sec(struct sock *sk, u64 rate, int gain)
+{
+	rate *= tcp_mss_to_mtu(sk, tcp_sk(sk)->mss_cache);
+	rate *= gain;
+	rate >>= BBR_SCALE;
+	rate *= USEC_PER_SEC;
+	return rate >> BW_SCALE;
+}
+
+/* Pace using current bw estimate and a gain factor. In order to help drive the
+ * network toward lower queues while maintaining high utilization and low
+ * latency, the average pacing rate aims to be slightly (~1%) lower than the
+ * estimated bandwidth. This is an important aspect of the design. In this
+ * implementation this slightly lower pacing rate is achieved implicitly by not
+ * including link-layer headers in the packet size used for the pacing rate.
+ */
+static void bbr_set_pacing_rate(struct sock *sk, u32 bw, int gain)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+	u64 rate = bw;
+
+	rate = bbr_rate_bytes_per_sec(sk, rate, gain);
+	rate = min_t(u64, rate, sk->sk_max_pacing_rate);
+	if (bbr->mode != BBR_STARTUP || rate > sk->sk_pacing_rate)
+		sk->sk_pacing_rate = rate;
+}
+
+/* Return count of segments we want in the skbs we send, or 0 for default. */
+static u32 bbr_tso_segs_goal(struct sock *sk)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	return bbr->tso_segs_goal;
+}
+
+static void bbr_set_tso_segs_goal(struct sock *sk)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+	u32 min_segs;
+
+	min_segs = sk->sk_pacing_rate < (bbr_min_tso_rate >> 3) ? 1 : 2;
+	bbr->tso_segs_goal = min(tcp_tso_autosize(sk, tp->mss_cache, min_segs),
+				 0x7FU);
+}
+
+/* Save "last known good" cwnd so we can restore it after losses or PROBE_RTT */
+static void bbr_save_cwnd(struct sock *sk)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	if (bbr->prev_ca_state < TCP_CA_Recovery && bbr->mode != BBR_PROBE_RTT)
+		bbr->prior_cwnd = tp->snd_cwnd;  /* this cwnd is good enough */
+	else  /* loss recovery or BBR_PROBE_RTT have temporarily cut cwnd */
+		bbr->prior_cwnd = max(bbr->prior_cwnd, tp->snd_cwnd);
+}
+
+static void bbr_cwnd_event(struct sock *sk, enum tcp_ca_event event)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	if (event == CA_EVENT_TX_START && tp->app_limited) {
+		bbr->idle_restart = 1;
+		/* Avoid pointless buffer overflows: pace at est. bw if we don't
+		 * need more speed (we're restarting from idle and app-limited).
+		 */
+		if (bbr->mode == BBR_PROBE_BW)
+			bbr_set_pacing_rate(sk, bbr_bw(sk), BBR_UNIT);
+	}
+}
+
+/* Find target cwnd. Right-size the cwnd based on min RTT and the
+ * estimated bottleneck bandwidth:
+ *
+ * cwnd = bw * min_rtt * gain = BDP * gain
+ *
+ * The key factor, gain, controls the amount of queue. While a small gain
+ * builds a smaller queue, it becomes more vulnerable to noise in RTT
+ * measurements (e.g., delayed ACKs or other ACK compression effects). This
+ * noise may cause BBR to under-estimate the rate.
+ *
+ * To achieve full performance in high-speed paths, we budget enough cwnd to
+ * fit full-sized skbs in-flight on both end hosts to fully utilize the path:
+ *   - one skb in sending host Qdisc,
+ *   - one skb in sending host TSO/GSO engine
+ *   - one skb being received by receiver host LRO/GRO/delayed-ACK engine
+ * Don't worry, at low rates (bbr_min_tso_rate) this won't bloat cwnd because
+ * in such cases tso_segs_goal is 1. The minimum cwnd is 4 packets,
+ * which allows 2 outstanding 2-packet sequences, to try to keep pipe
+ * full even with ACK-every-other-packet delayed ACKs.
+ */
+static u32 bbr_target_cwnd(struct sock *sk, u32 bw, int gain)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+	u32 cwnd;
+	u64 w;
+
+	/* If we've never had a valid RTT sample, cap cwnd at the initial
+	 * default. This should only happen when the connection is not using TCP
+	 * timestamps and has retransmitted all of the SYN/SYNACK/data packets
+	 * ACKed so far. In this case, an RTO can cut cwnd to 1, in which
+	 * case we need to slow-start up toward something safe: TCP_INIT_CWND.
+	 */
+	if (unlikely(bbr->min_rtt_us == ~0U))	 /* no valid RTT samples yet? */
+		return TCP_INIT_CWND;  /* be safe: cap at default initial cwnd*/
+
+	w = (u64)bw * bbr->min_rtt_us;
+
+	/* Apply a gain to the given value, then remove the BW_SCALE shift. */
+	cwnd = (((w * gain) >> BBR_SCALE) + BW_UNIT - 1) / BW_UNIT;
+
+	/* Allow enough full-sized skbs in flight to utilize end systems. */
+	cwnd += 3 * bbr->tso_segs_goal;
+
+	/* Reduce delayed ACKs by rounding up cwnd to the next even number. */
+	cwnd = (cwnd + 1) & ~1U;
+
+	return cwnd;
+}
+
+/* An optimization in BBR to reduce losses: On the first round of recovery, we
+ * follow the packet conservation principle: send P packets per P packets acked.
+ * After that, we slow-start and send at most 2*P packets per P packets acked.
+ * After recovery finishes, or upon undo, we restore the cwnd we had when
+ * recovery started (capped by the target cwnd based on estimated BDP).
+ *
+ * TODO(ycheng/ncardwell): implement a rate-based approach.
+ */
+static bool bbr_set_cwnd_to_recover_or_restore(
+	struct sock *sk, const struct rate_sample *rs, u32 acked, u32 *new_cwnd)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+	u8 prev_state = bbr->prev_ca_state, state = inet_csk(sk)->icsk_ca_state;
+	u32 cwnd = tp->snd_cwnd;
+
+	/* An ACK for P pkts should release at most 2*P packets. We do this
+	 * in two steps. First, here we deduct the number of lost packets.
+	 * Then, in bbr_set_cwnd() we slow start up toward the target cwnd.
+	 */
+	if (rs->losses > 0)
+		cwnd = max_t(s32, cwnd - rs->losses, 1);
+
+	if (state == TCP_CA_Recovery && prev_state != TCP_CA_Recovery) {
+		/* Starting 1st round of Recovery, so do packet conservation. */
+		bbr->packet_conservation = 1;
+		bbr->next_rtt_delivered = tp->delivered;  /* start round now */
+		/* Cut unused cwnd from app behavior, TSQ, or TSO deferral: */
+		cwnd = tcp_packets_in_flight(tp) + acked;
+	} else if (prev_state >= TCP_CA_Recovery && state < TCP_CA_Recovery) {
+		/* Exiting loss recovery; restore cwnd saved before recovery. */
+		bbr->restore_cwnd = 1;
+		bbr->packet_conservation = 0;
+	}
+	bbr->prev_ca_state = state;
+
+	if (bbr->restore_cwnd) {
+		/* Restore cwnd after exiting loss recovery or PROBE_RTT. */
+		cwnd = max(cwnd, bbr->prior_cwnd);
+		bbr->restore_cwnd = 0;
+	}
+
+	if (bbr->packet_conservation) {
+		*new_cwnd = max(cwnd, tcp_packets_in_flight(tp) + acked);
+		return true;	/* yes, using packet conservation */
+	}
+	*new_cwnd = cwnd;
+	return false;
+}
+
+/* Slow-start up toward target cwnd (if bw estimate is growing, or packet loss
+ * has drawn us down below target), or snap down to target if we're above it.
+ */
+static void bbr_set_cwnd(struct sock *sk, const struct rate_sample *rs,
+			 u32 acked, u32 bw, int gain)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+	u32 cwnd = 0, target_cwnd = 0;
+
+	if (!acked)
+		return;
+
+	if (bbr_set_cwnd_to_recover_or_restore(sk, rs, acked, &cwnd))
+		goto done;
+
+	/* If we're below target cwnd, slow start cwnd toward target cwnd. */
+	target_cwnd = bbr_target_cwnd(sk, bw, gain);
+	if (bbr_full_bw_reached(sk))  /* only cut cwnd if we filled the pipe */
+		cwnd = min(cwnd + acked, target_cwnd);
+	else if (cwnd < target_cwnd || tp->delivered < TCP_INIT_CWND)
+		cwnd = cwnd + acked;
+	cwnd = max(cwnd, bbr_cwnd_min_target);
+
+done:
+	tp->snd_cwnd = min(cwnd, tp->snd_cwnd_clamp);	/* apply global cap */
+	if (bbr->mode == BBR_PROBE_RTT)  /* drain queue, refresh min_rtt */
+		tp->snd_cwnd = min(tp->snd_cwnd, bbr_cwnd_min_target);
+}
+
+/* End cycle phase if it's time and/or we hit the phase's in-flight target. */
+static bool bbr_is_next_cycle_phase(struct sock *sk,
+				    const struct rate_sample *rs)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+	bool is_full_length =
+		skb_mstamp_us_delta(&tp->delivered_mstamp, &bbr->cycle_mstamp) >
+		bbr->min_rtt_us;
+	u32 inflight, bw;
+
+	/* The pacing_gain of 1.0 paces at the estimated bw to try to fully
+	 * use the pipe without increasing the queue.
+	 */
+	if (bbr->pacing_gain == BBR_UNIT)
+		return is_full_length;		/* just use wall clock time */
+
+	inflight = rs->prior_in_flight;  /* what was in-flight before ACK? */
+	bw = bbr_max_bw(sk);
+
+	/* A pacing_gain > 1.0 probes for bw by trying to raise inflight to at
+	 * least pacing_gain*BDP; this may take more than min_rtt if min_rtt is
+	 * small (e.g. on a LAN). We do not persist if packets are lost, since
+	 * a path with small buffers may not hold that much.
+	 */
+	if (bbr->pacing_gain > BBR_UNIT)
+		return is_full_length &&
+			(rs->losses ||  /* perhaps pacing_gain*BDP won't fit */
+			 inflight >= bbr_target_cwnd(sk, bw, bbr->pacing_gain));
+
+	/* A pacing_gain < 1.0 tries to drain extra queue we added if bw
+	 * probing didn't find more bw. If inflight falls to match BDP then we
+	 * estimate queue is drained; persisting would underutilize the pipe.
+	 */
+	return is_full_length ||
+		inflight <= bbr_target_cwnd(sk, bw, BBR_UNIT);
+}
+
+static void bbr_advance_cycle_phase(struct sock *sk)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	bbr->cycle_idx = (bbr->cycle_idx + 1) & (CYCLE_LEN - 1);
+	bbr->cycle_mstamp = tp->delivered_mstamp;
+	bbr->pacing_gain = bbr_pacing_gain[bbr->cycle_idx];
+}
+
+/* Gain cycling: cycle pacing gain to converge to fair share of available bw. */
+static void bbr_update_cycle_phase(struct sock *sk,
+				   const struct rate_sample *rs)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	if ((bbr->mode == BBR_PROBE_BW) && !bbr->lt_use_bw &&
+	    bbr_is_next_cycle_phase(sk, rs))
+		bbr_advance_cycle_phase(sk);
+}
+
+static void bbr_reset_startup_mode(struct sock *sk)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	bbr->mode = BBR_STARTUP;
+	bbr->pacing_gain = bbr_high_gain;
+	bbr->cwnd_gain	 = bbr_high_gain;
+}
+
+static void bbr_reset_probe_bw_mode(struct sock *sk)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	bbr->mode = BBR_PROBE_BW;
+	bbr->pacing_gain = BBR_UNIT;
+	bbr->cwnd_gain = bbr_cwnd_gain;
+	bbr->cycle_idx = CYCLE_LEN - 1 - prandom_u32_max(bbr_cycle_rand);
+	bbr_advance_cycle_phase(sk);	/* flip to next phase of gain cycle */
+}
+
+static void bbr_reset_mode(struct sock *sk)
+{
+	if (!bbr_full_bw_reached(sk))
+		bbr_reset_startup_mode(sk);
+	else
+		bbr_reset_probe_bw_mode(sk);
+}
+
+/* Start a new long-term sampling interval. */
+static void bbr_reset_lt_bw_sampling_interval(struct sock *sk)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	bbr->lt_last_stamp = tp->delivered_mstamp.stamp_jiffies;
+	bbr->lt_last_delivered = tp->delivered;
+	bbr->lt_last_lost = tp->lost;
+	bbr->lt_rtt_cnt = 0;
+}
+
+/* Completely reset long-term bandwidth sampling. */
+static void bbr_reset_lt_bw_sampling(struct sock *sk)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	bbr->lt_bw = 0;
+	bbr->lt_use_bw = 0;
+	bbr->lt_is_sampling = false;
+	bbr_reset_lt_bw_sampling_interval(sk);
+}
+
+/* Long-term bw sampling interval is done. Estimate whether we're policed. */
+static void bbr_lt_bw_interval_done(struct sock *sk, u32 bw)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+	u32 diff;
+
+	if (bbr->lt_bw) {  /* do we have bw from a previous interval? */
+		/* Is new bw close to the lt_bw from the previous interval? */
+		diff = abs(bw - bbr->lt_bw);
+		if ((diff * BBR_UNIT <= bbr_lt_bw_ratio * bbr->lt_bw) ||
+		    (bbr_rate_bytes_per_sec(sk, diff, BBR_UNIT) <=
+		     bbr_lt_bw_diff)) {
+			/* All criteria are met; estimate we're policed. */
+			bbr->lt_bw = (bw + bbr->lt_bw) >> 1;  /* avg 2 intvls */
+			bbr->lt_use_bw = 1;
+			bbr->pacing_gain = BBR_UNIT;  /* try to avoid drops */
+			bbr->lt_rtt_cnt = 0;
+			return;
+		}
+	}
+	bbr->lt_bw = bw;
+	bbr_reset_lt_bw_sampling_interval(sk);
+}
+
+/* Token-bucket traffic policers are common (see "An Internet-Wide Analysis of
+ * Traffic Policing", SIGCOMM 2016). BBR detects token-bucket policers and
+ * explicitly models their policed rate, to reduce unnecessary losses. We
+ * estimate that we're policed if we see 2 consecutive sampling intervals with
+ * consistent throughput and high packet loss. If we think we're being policed,
+ * set lt_bw to the "long-term" average delivery rate from those 2 intervals.
+ */
+static void bbr_lt_bw_sampling(struct sock *sk, const struct rate_sample *rs)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+	u32 lost, delivered;
+	u64 bw;
+	s32 t;
+
+	if (bbr->lt_use_bw) {	/* already using long-term rate, lt_bw? */
+		if (bbr->mode == BBR_PROBE_BW && bbr->round_start &&
+		    ++bbr->lt_rtt_cnt >= bbr_lt_bw_max_rtts) {
+			bbr_reset_lt_bw_sampling(sk);    /* stop using lt_bw */
+			bbr_reset_probe_bw_mode(sk);  /* restart gain cycling */
+		}
+		return;
+	}
+
+	/* Wait for the first loss before sampling, to let the policer exhaust
+	 * its tokens and estimate the steady-state rate allowed by the policer.
+	 * Starting samples earlier includes bursts that over-estimate the bw.
+	 */
+	if (!bbr->lt_is_sampling) {
+		if (!rs->losses)
+			return;
+		bbr_reset_lt_bw_sampling_interval(sk);
+		bbr->lt_is_sampling = true;
+	}
+
+	/* To avoid underestimates, reset sampling if we run out of data. */
+	if (rs->is_app_limited) {
+		bbr_reset_lt_bw_sampling(sk);
+		return;
+	}
+
+	if (bbr->round_start)
+		bbr->lt_rtt_cnt++;	/* count round trips in this interval */
+	if (bbr->lt_rtt_cnt < bbr_lt_intvl_min_rtts)
+		return;		/* sampling interval needs to be longer */
+	if (bbr->lt_rtt_cnt > 4 * bbr_lt_intvl_min_rtts) {
+		bbr_reset_lt_bw_sampling(sk);  /* interval is too long */
+		return;
+	}
+
+	/* End sampling interval when a packet is lost, so we estimate the
+	 * policer tokens were exhausted. Stopping the sampling before the
+	 * tokens are exhausted under-estimates the policed rate.
+	 */
+	if (!rs->losses)
+		return;
+
+	/* Calculate packets lost and delivered in sampling interval. */
+	lost = tp->lost - bbr->lt_last_lost;
+	delivered = tp->delivered - bbr->lt_last_delivered;
+	/* Is loss rate (lost/delivered) >= lt_loss_thresh? If not, wait. */
+	if (!delivered || (lost << BBR_SCALE) < bbr_lt_loss_thresh * delivered)
+		return;
+
+	/* Find average delivery rate in this sampling interval. */
+	t = (s32)(tp->delivered_mstamp.stamp_jiffies - bbr->lt_last_stamp);
+	if (t < 1)
+		return;		/* interval is less than one jiffy, so wait */
+	t = jiffies_to_usecs(t);
+	/* Interval long enough for jiffies_to_usecs() to return a bogus 0? */
+	if (t < 1) {
+		bbr_reset_lt_bw_sampling(sk);  /* interval too long; reset */
+		return;
+	}
+	bw = (u64)delivered * BW_UNIT;
+	do_div(bw, t);
+	bbr_lt_bw_interval_done(sk, bw);
+}
+
+/* Estimate the bandwidth based on how fast packets are delivered */
+static void bbr_update_bw(struct sock *sk, const struct rate_sample *rs)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+	u64 bw;
+
+	bbr->round_start = 0;
+	if (rs->delivered < 0 || rs->interval_us <= 0)
+		return; /* Not a valid observation */
+
+	/* See if we've reached the next RTT */
+	if (!before(rs->prior_delivered, bbr->next_rtt_delivered)) {
+		bbr->next_rtt_delivered = tp->delivered;
+		bbr->rtt_cnt++;
+		bbr->round_start = 1;
+		bbr->packet_conservation = 0;
+	}
+
+	bbr_lt_bw_sampling(sk, rs);
+
+	/* Divide delivered by the interval to find a (lower bound) bottleneck
+	 * bandwidth sample. Delivered is in packets and interval_us in uS and
+	 * ratio will be <<1 for most connections. So delivered is first scaled.
+	 */
+	bw = (u64)rs->delivered * BW_UNIT;
+	do_div(bw, rs->interval_us);
+
+	/* If this sample is application-limited, it is likely to have a very
+	 * low delivered count that represents application behavior rather than
+	 * the available network rate. Such a sample could drag down estimated
+	 * bw, causing needless slow-down. Thus, to continue to send at the
+	 * last measured network rate, we filter out app-limited samples unless
+	 * they describe the path bw at least as well as our bw model.
+	 *
+	 * So the goal during app-limited phase is to proceed with the best
+	 * network rate no matter how long. We automatically leave this
+	 * phase when app writes faster than the network can deliver :)
+	 */
+	if (!rs->is_app_limited || bw >= bbr_max_bw(sk)) {
+		/* Incorporate new sample into our max bw filter. */
+		minmax_running_max(&bbr->bw, bbr_bw_rtts, bbr->rtt_cnt, bw);
+	}
+}
+
+/* Estimate when the pipe is full, using the change in delivery rate: BBR
+ * estimates that STARTUP filled the pipe if the estimated bw hasn't changed by
+ * at least bbr_full_bw_thresh (25%) after bbr_full_bw_cnt (3) non-app-limited
+ * rounds. Why 3 rounds: 1: rwin autotuning grows the rwin, 2: we fill the
+ * higher rwin, 3: we get higher delivery rate samples. Or transient
+ * cross-traffic or radio noise can go away. CUBIC Hystart shares a similar
+ * design goal, but uses delay and inter-ACK spacing instead of bandwidth.
+ */
+static void bbr_check_full_bw_reached(struct sock *sk,
+				      const struct rate_sample *rs)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+	u32 bw_thresh;
+
+	if (bbr_full_bw_reached(sk) || !bbr->round_start || rs->is_app_limited)
+		return;
+
+	bw_thresh = (u64)bbr->full_bw * bbr_full_bw_thresh >> BBR_SCALE;
+	if (bbr_max_bw(sk) >= bw_thresh) {
+		bbr->full_bw = bbr_max_bw(sk);
+		bbr->full_bw_cnt = 0;
+		return;
+	}
+	++bbr->full_bw_cnt;
+}
+
+/* If pipe is probably full, drain the queue and then enter steady-state. */
+static void bbr_check_drain(struct sock *sk, const struct rate_sample *rs)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	if (bbr->mode == BBR_STARTUP && bbr_full_bw_reached(sk)) {
+		bbr->mode = BBR_DRAIN;	/* drain queue we created */
+		bbr->pacing_gain = bbr_drain_gain;	/* pace slow to drain */
+		bbr->cwnd_gain = bbr_high_gain;	/* maintain cwnd */
+	}	/* fall through to check if in-flight is already small: */
+	if (bbr->mode == BBR_DRAIN &&
+	    tcp_packets_in_flight(tcp_sk(sk)) <=
+	    bbr_target_cwnd(sk, bbr_max_bw(sk), BBR_UNIT))
+		bbr_reset_probe_bw_mode(sk);  /* we estimate queue is drained */
+}
+
+/* The goal of PROBE_RTT mode is to have BBR flows cooperatively and
+ * periodically drain the bottleneck queue, to converge to measure the true
+ * min_rtt (unloaded propagation delay). This allows the flows to keep queues
+ * small (reducing queuing delay and packet loss) and achieve fairness among
+ * BBR flows.
+ *
+ * The min_rtt filter window is 10 seconds. When the min_rtt estimate expires,
+ * we enter PROBE_RTT mode and cap the cwnd at bbr_cwnd_min_target=4 packets.
+ * After at least bbr_probe_rtt_mode_ms=200ms and at least one packet-timed
+ * round trip elapsed with that flight size <= 4, we leave PROBE_RTT mode and
+ * re-enter the previous mode. BBR uses 200ms to approximately bound the
+ * performance penalty of PROBE_RTT's cwnd capping to roughly 2% (200ms/10s).
+ *
+ * Note that flows need only pay 2% if they are busy sending over the last 10
+ * seconds. Interactive applications (e.g., Web, RPCs, video chunks) often have
+ * natural silences or low-rate periods within 10 seconds where the rate is low
+ * enough for long enough to drain its queue in the bottleneck. We pick up
+ * these min RTT measurements opportunistically with our min_rtt filter. :-)
+ */
+static void bbr_update_min_rtt(struct sock *sk, const struct rate_sample *rs)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+	bool filter_expired;
+
+	/* Track min RTT seen in the min_rtt_win_sec filter window: */
+	filter_expired = after(tcp_time_stamp,
+			       bbr->min_rtt_stamp + bbr_min_rtt_win_sec * HZ);
+	if (rs->rtt_us >= 0 &&
+	    (rs->rtt_us <= bbr->min_rtt_us || filter_expired)) {
+		bbr->min_rtt_us = rs->rtt_us;
+		bbr->min_rtt_stamp = tcp_time_stamp;
+	}
+
+	if (bbr_probe_rtt_mode_ms > 0 && filter_expired &&
+	    !bbr->idle_restart && bbr->mode != BBR_PROBE_RTT) {
+		bbr->mode = BBR_PROBE_RTT;  /* dip, drain queue */
+		bbr->pacing_gain = BBR_UNIT;
+		bbr->cwnd_gain = BBR_UNIT;
+		bbr_save_cwnd(sk);  /* note cwnd so we can restore it */
+		bbr->probe_rtt_done_stamp = 0;
+	}
+
+	if (bbr->mode == BBR_PROBE_RTT) {
+		/* Ignore low rate samples during this mode. */
+		tp->app_limited =
+			(tp->delivered + tcp_packets_in_flight(tp)) ? : 1;
+		/* Maintain min packets in flight for max(200 ms, 1 round). */
+		if (!bbr->probe_rtt_done_stamp &&
+		    tcp_packets_in_flight(tp) <= bbr_cwnd_min_target) {
+			bbr->probe_rtt_done_stamp = tcp_time_stamp +
+				msecs_to_jiffies(bbr_probe_rtt_mode_ms);
+			bbr->probe_rtt_round_done = 0;
+			bbr->next_rtt_delivered = tp->delivered;
+		} else if (bbr->probe_rtt_done_stamp) {
+			if (bbr->round_start)
+				bbr->probe_rtt_round_done = 1;
+			if (bbr->probe_rtt_round_done &&
+			    after(tcp_time_stamp, bbr->probe_rtt_done_stamp)) {
+				bbr->min_rtt_stamp = tcp_time_stamp;
+				bbr->restore_cwnd = 1;  /* snap to prior_cwnd */
+				bbr_reset_mode(sk);
+			}
+		}
+	}
+	bbr->idle_restart = 0;
+}
+
+static void bbr_update_model(struct sock *sk, const struct rate_sample *rs)
+{
+	bbr_update_bw(sk, rs);
+	bbr_update_cycle_phase(sk, rs);
+	bbr_check_full_bw_reached(sk, rs);
+	bbr_check_drain(sk, rs);
+	bbr_update_min_rtt(sk, rs);
+}
+
+static void bbr_main(struct sock *sk, const struct rate_sample *rs)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+	u32 bw;
+
+	bbr_update_model(sk, rs);
+
+	bw = bbr_bw(sk);
+	bbr_set_pacing_rate(sk, bw, bbr->pacing_gain);
+	bbr_set_tso_segs_goal(sk);
+	bbr_set_cwnd(sk, rs, rs->acked_sacked, bw, bbr->cwnd_gain);
+}
+
+static void bbr_init(struct sock *sk)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct bbr *bbr = inet_csk_ca(sk);
+	u64 bw;
+
+	bbr->prior_cwnd = 0;
+	bbr->tso_segs_goal = 0;	 /* default segs per skb until first ACK */
+	bbr->rtt_cnt = 0;
+	bbr->next_rtt_delivered = 0;
+	bbr->prev_ca_state = TCP_CA_Open;
+	bbr->packet_conservation = 0;
+
+	bbr->probe_rtt_done_stamp = 0;
+	bbr->probe_rtt_round_done = 0;
+	bbr->min_rtt_us = tcp_min_rtt(tp);
+	bbr->min_rtt_stamp = tcp_time_stamp;
+
+	minmax_reset(&bbr->bw, bbr->rtt_cnt, 0);  /* init max bw to 0 */
+
+	/* Initialize pacing rate to: high_gain * init_cwnd / RTT. */
+	bw = (u64)tp->snd_cwnd * BW_UNIT;
+	do_div(bw, (tp->srtt_us >> 3) ? : USEC_PER_MSEC);
+	sk->sk_pacing_rate = 0;		/* force an update of sk_pacing_rate */
+	bbr_set_pacing_rate(sk, bw, bbr_high_gain);
+
+	bbr->restore_cwnd = 0;
+	bbr->round_start = 0;
+	bbr->idle_restart = 0;
+	bbr->full_bw = 0;
+	bbr->full_bw_cnt = 0;
+	bbr->cycle_mstamp.v64 = 0;
+	bbr->cycle_idx = 0;
+	bbr_reset_lt_bw_sampling(sk);
+	bbr_reset_startup_mode(sk);
+}
+
+static u32 bbr_sndbuf_expand(struct sock *sk)
+{
+	/* Provision 3 * cwnd since BBR may slow-start even during recovery. */
+	return 3;
+}
+
+/* In theory BBR does not need to undo the cwnd since it does not
+ * always reduce cwnd on losses (see bbr_main()). Keep it for now.
+ */
+static u32 bbr_undo_cwnd(struct sock *sk)
+{
+	return tcp_sk(sk)->snd_cwnd;
+}
+
+/* Entering loss recovery, so save cwnd for when we exit or undo recovery. */
+static u32 bbr_ssthresh(struct sock *sk)
+{
+	bbr_save_cwnd(sk);
+	return TCP_INFINITE_SSTHRESH;	 /* BBR does not use ssthresh */
+}
+
+static size_t bbr_get_info(struct sock *sk, u32 ext, int *attr,
+			   union tcp_cc_info *info)
+{
+	if (ext & (1 << (INET_DIAG_BBRINFO - 1)) ||
+	    ext & (1 << (INET_DIAG_VEGASINFO - 1))) {
+		struct tcp_sock *tp = tcp_sk(sk);
+		struct bbr *bbr = inet_csk_ca(sk);
+		u64 bw = bbr_bw(sk);
+
+		bw = bw * tp->mss_cache * USEC_PER_SEC >> BW_SCALE;
+		memset(&info->bbr, 0, sizeof(info->bbr));
+		info->bbr.bbr_bw_lo		= (u32)bw;
+		info->bbr.bbr_bw_hi		= (u32)(bw >> 32);
+		info->bbr.bbr_min_rtt		= bbr->min_rtt_us;
+		info->bbr.bbr_pacing_gain	= bbr->pacing_gain;
+		info->bbr.bbr_cwnd_gain		= bbr->cwnd_gain;
+		*attr = INET_DIAG_BBRINFO;
+		return sizeof(info->bbr);
+	}
+	return 0;
+}
+
+static void bbr_set_state(struct sock *sk, u8 new_state)
+{
+	struct bbr *bbr = inet_csk_ca(sk);
+
+	if (new_state == TCP_CA_Loss) {
+		struct rate_sample rs = { .losses = 1 };
+
+		bbr->prev_ca_state = TCP_CA_Loss;
+		bbr->full_bw = 0;
+		bbr->round_start = 1;	/* treat RTO like end of a round */
+		bbr_lt_bw_sampling(sk, &rs);
+	}
+}
+
+static struct tcp_congestion_ops tcp_bbr_cong_ops __read_mostly = {
+	.flags		= TCP_CONG_NON_RESTRICTED,
+	.name		= "bbr",
+	.owner		= THIS_MODULE,
+	.init		= bbr_init,
+	.cong_control	= bbr_main,
+	.sndbuf_expand	= bbr_sndbuf_expand,
+	.undo_cwnd	= bbr_undo_cwnd,
+	.cwnd_event	= bbr_cwnd_event,
+	.ssthresh	= bbr_ssthresh,
+	.tso_segs_goal	= bbr_tso_segs_goal,
+	.get_info	= bbr_get_info,
+	.set_state	= bbr_set_state,
+};
+
+static int __init bbr_register(void)
+{
+	BUILD_BUG_ON(sizeof(struct bbr) > ICSK_CA_PRIV_SIZE);
+	return tcp_register_congestion_control(&tcp_bbr_cong_ops);
+}
+
+static void __exit bbr_unregister(void)
+{
+	tcp_unregister_congestion_control(&tcp_bbr_cong_ops);
+}
+
+module_init(bbr_register);
+module_exit(bbr_unregister);
+
+MODULE_AUTHOR("Van Jacobson <vanj@google.com>");
+MODULE_AUTHOR("Neal Cardwell <ncardwell@google.com>");
+MODULE_AUTHOR("Yuchung Cheng <ycheng@google.com>");
+MODULE_AUTHOR("Soheil Hassas Yeganeh <soheil@google.com>");
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_DESCRIPTION("TCP BBR (Bottleneck Bandwidth and RTT)");

net/ipv4/tcp_cdg.c

@@ -56,7 +56,7 @@ MODULE_PARM_DESC(use_shadow, "use shadow window heuristic");
 module_param(use_tolerance, bool, 0644);
 MODULE_PARM_DESC(use_tolerance, "use loss tolerance heuristic");
 
-struct minmax {
+struct cdg_minmax {
 	union {
 		struct {
 			s32 min;
@@ -74,10 +74,10 @@ enum cdg_state {
 };
 
 struct cdg {
-	struct minmax rtt;
-	struct minmax rtt_prev;
-	struct minmax *gradients;
-	struct minmax gsum;
+	struct cdg_minmax rtt;
+	struct cdg_minmax rtt_prev;
+	struct cdg_minmax *gradients;
+	struct cdg_minmax gsum;
 	bool gfilled;
 	u8  tail;
 	u8  state;
@@ -353,7 +353,7 @@ static void tcp_cdg_cwnd_event(struct sock *sk, const enum tcp_ca_event ev)
 {
 	struct cdg *ca = inet_csk_ca(sk);
 	struct tcp_sock *tp = tcp_sk(sk);
-	struct minmax *gradients;
+	struct cdg_minmax *gradients;
 
 	switch (ev) {
 	case CA_EVENT_CWND_RESTART:

net/ipv4/tcp_cong.c

@@ -69,7 +69,7 @@ int tcp_register_congestion_control(struct tcp_congestion_ops *ca)
 	int ret = 0;
 
 	/* all algorithms must implement ssthresh and cong_avoid ops */
-	if (!ca->ssthresh || !ca->cong_avoid) {
+	if (!ca->ssthresh || !(ca->cong_avoid || ca->cong_control)) {
 		pr_err("%s does not implement required ops\n", ca->name);
 		return -EINVAL;
 	}

net/ipv4/tcp_input.c

@@ -289,6 +289,7 @@ static bool tcp_ecn_rcv_ecn_echo(const struct tcp_sock *tp, const struct tcphdr
 static void tcp_sndbuf_expand(struct sock *sk)
 {
 	const struct tcp_sock *tp = tcp_sk(sk);
+	const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
 	int sndmem, per_mss;
 	u32 nr_segs;
 
@@ -309,7 +310,8 @@ static void tcp_sndbuf_expand(struct sock *sk)
 	 * Cubic needs 1.7 factor, rounded to 2 to include
 	 * extra cushion (application might react slowly to POLLOUT)
 	 */
-	sndmem = 2 * nr_segs * per_mss;
+	sndmem = ca_ops->sndbuf_expand ? ca_ops->sndbuf_expand(sk) : 2;
+	sndmem *= nr_segs * per_mss;
 
 	if (sk->sk_sndbuf < sndmem)
 		sk->sk_sndbuf = min(sndmem, sysctl_tcp_wmem[2]);
@@ -899,12 +901,29 @@ static void tcp_verify_retransmit_hint(struct tcp_sock *tp, struct sk_buff *skb)
 		tp->retransmit_high = TCP_SKB_CB(skb)->end_seq;
 }
 
+/* Sum the number of packets on the wire we have marked as lost.
+ * There are two cases we care about here:
+ * a) Packet hasn't been marked lost (nor retransmitted),
+ *    and this is the first loss.
+ * b) Packet has been marked both lost and retransmitted,
+ *    and this means we think it was lost again.
+ */
+static void tcp_sum_lost(struct tcp_sock *tp, struct sk_buff *skb)
+{
+	__u8 sacked = TCP_SKB_CB(skb)->sacked;
+
+	if (!(sacked & TCPCB_LOST) ||
+	    ((sacked & TCPCB_LOST) && (sacked & TCPCB_SACKED_RETRANS)))
+		tp->lost += tcp_skb_pcount(skb);
+}
+
 static void tcp_skb_mark_lost(struct tcp_sock *tp, struct sk_buff *skb)
 {
 	if (!(TCP_SKB_CB(skb)->sacked & (TCPCB_LOST|TCPCB_SACKED_ACKED))) {
 		tcp_verify_retransmit_hint(tp, skb);
 
 		tp->lost_out += tcp_skb_pcount(skb);
+		tcp_sum_lost(tp, skb);
 		TCP_SKB_CB(skb)->sacked |= TCPCB_LOST;
 	}
 }
@@ -913,6 +932,7 @@ void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb)
 {
 	tcp_verify_retransmit_hint(tp, skb);
 
+	tcp_sum_lost(tp, skb);
 	if (!(TCP_SKB_CB(skb)->sacked & (TCPCB_LOST|TCPCB_SACKED_ACKED))) {
 		tp->lost_out += tcp_skb_pcount(skb);
 		TCP_SKB_CB(skb)->sacked |= TCPCB_LOST;
@@ -1094,6 +1114,7 @@ struct tcp_sacktag_state {
 	 */
 	struct skb_mstamp first_sackt;
 	struct skb_mstamp last_sackt;
+	struct rate_sample *rate;
 	int	flag;
 };
 
@@ -1261,6 +1282,7 @@ static bool tcp_shifted_skb(struct sock *sk, struct sk_buff *skb,
 	tcp_sacktag_one(sk, state, TCP_SKB_CB(skb)->sacked,
 			start_seq, end_seq, dup_sack, pcount,
 			&skb->skb_mstamp);
+	tcp_rate_skb_delivered(sk, skb, state->rate);
 
 	if (skb == tp->lost_skb_hint)
 		tp->lost_cnt_hint += pcount;
@@ -1311,6 +1333,9 @@ static bool tcp_shifted_skb(struct sock *sk, struct sk_buff *skb,
 		tcp_advance_highest_sack(sk, skb);
 
 	tcp_skb_collapse_tstamp(prev, skb);
+	if (unlikely(TCP_SKB_CB(prev)->tx.delivered_mstamp.v64))
+		TCP_SKB_CB(prev)->tx.delivered_mstamp.v64 = 0;
+
 	tcp_unlink_write_queue(skb, sk);
 	sk_wmem_free_skb(sk, skb);
 
@@ -1540,6 +1565,7 @@ static struct sk_buff *tcp_sacktag_walk(struct sk_buff *skb, struct sock *sk,
 						dup_sack,
 						tcp_skb_pcount(skb),
 						&skb->skb_mstamp);
+			tcp_rate_skb_delivered(sk, skb, state->rate);
 
 			if (!before(TCP_SKB_CB(skb)->seq,
 				    tcp_highest_sack_seq(tp)))
@@ -1622,8 +1648,10 @@ 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)
+	if (found_dup_sack) {
 		state->flag |= FLAG_DSACKING_ACK;
+		tp->delivered++; /* A spurious retransmission is delivered */
+	}
 
 	/* Eliminate too old ACKs, but take into
 	 * account more or less fresh ones, they can
@@ -1890,6 +1918,7 @@ void tcp_enter_loss(struct sock *sk)
 	struct sk_buff *skb;
 	bool new_recovery = icsk->icsk_ca_state < TCP_CA_Recovery;
 	bool is_reneg;			/* is receiver reneging on SACKs? */
+	bool mark_lost;
 
 	/* Reduce ssthresh if it has not yet been made inside this window. */
 	if (icsk->icsk_ca_state <= TCP_CA_Disorder ||
@@ -1923,8 +1952,12 @@ void tcp_enter_loss(struct sock *sk)
 		if (skb == tcp_send_head(sk))
 			break;
 
+		mark_lost = (!(TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED) ||
+			     is_reneg);
+		if (mark_lost)
+			tcp_sum_lost(tp, skb);
 		TCP_SKB_CB(skb)->sacked &= (~TCPCB_TAGBITS)|TCPCB_SACKED_ACKED;
-		if (!(TCP_SKB_CB(skb)->sacked&TCPCB_SACKED_ACKED) || is_reneg) {
+		if (mark_lost) {
 			TCP_SKB_CB(skb)->sacked &= ~TCPCB_SACKED_ACKED;
 			TCP_SKB_CB(skb)->sacked |= TCPCB_LOST;
 			tp->lost_out += tcp_skb_pcount(skb);
@@ -2503,6 +2536,9 @@ static inline void tcp_end_cwnd_reduction(struct sock *sk)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
 
+	if (inet_csk(sk)->icsk_ca_ops->cong_control)
+		return;
+
 	/* Reset cwnd to ssthresh in CWR or Recovery (unless it's undone) */
 	if (inet_csk(sk)->icsk_ca_state == TCP_CA_CWR ||
 	    (tp->undo_marker && tp->snd_ssthresh < TCP_INFINITE_SSTHRESH)) {
@@ -2879,67 +2915,13 @@ static void tcp_fastretrans_alert(struct sock *sk, const int acked,
 	*rexmit = REXMIT_LOST;
 }
 
-/* Kathleen Nichols' algorithm for tracking the minimum value of
- * a data stream over some fixed time interval. (E.g., the minimum
- * RTT over the past five minutes.) It uses constant space and constant
- * time per update yet almost always delivers the same minimum as an
- * implementation that has to keep all the data in the window.
- *
- * The algorithm keeps track of the best, 2nd best & 3rd best min
- * values, maintaining an invariant that the measurement time of the
- * n'th best >= n-1'th best. It also makes sure that the three values
- * are widely separated in the time window since that bounds the worse
- * case error when that data is monotonically increasing over the window.
- *
- * Upon getting a new min, we can forget everything earlier because it
- * has no value - the new min is <= everything else in the window by
- * definition and it's the most recent. So we restart fresh on every new min
- * and overwrites 2nd & 3rd choices. The same property holds for 2nd & 3rd
- * best.
- */
 static void tcp_update_rtt_min(struct sock *sk, u32 rtt_us)
 {
-	const u32 now = tcp_time_stamp, wlen = sysctl_tcp_min_rtt_wlen * HZ;
-	struct rtt_meas *m = tcp_sk(sk)->rtt_min;
-	struct rtt_meas rttm = {
-		.rtt = likely(rtt_us) ? rtt_us : jiffies_to_usecs(1),
-		.ts = now,
-	};
-	u32 elapsed;
-
-	/* Check if the new measurement updates the 1st, 2nd, or 3rd choices */
-	if (unlikely(rttm.rtt <= m[0].rtt))
-		m[0] = m[1] = m[2] = rttm;
-	else if (rttm.rtt <= m[1].rtt)
-		m[1] = m[2] = rttm;
-	else if (rttm.rtt <= m[2].rtt)
-		m[2] = rttm;
-
-	elapsed = now - m[0].ts;
-	if (unlikely(elapsed > wlen)) {
-		/* Passed entire window without a new min so make 2nd choice
-		 * the new min & 3rd choice the new 2nd. So forth and so on.
-		 */
-		m[0] = m[1];
-		m[1] = m[2];
-		m[2] = rttm;
-		if (now - m[0].ts > wlen) {
-			m[0] = m[1];
-			m[1] = rttm;
-			if (now - m[0].ts > wlen)
-				m[0] = rttm;
-		}
-	} else if (m[1].ts == m[0].ts && elapsed > wlen / 4) {
-		/* Passed a quarter of the window without a new min so
-		 * take 2nd choice from the 2nd quarter of the window.
-		 */
-		m[2] = m[1] = rttm;
-	} else if (m[2].ts == m[1].ts && elapsed > wlen / 2) {
-		/* Passed half the window without a new min so take the 3rd
-		 * choice from the last half of the window.
-		 */
-		m[2] = rttm;
-	}
+	struct tcp_sock *tp = tcp_sk(sk);
+	u32 wlen = sysctl_tcp_min_rtt_wlen * HZ;
+
+	minmax_running_min(&tp->rtt_min, wlen, tcp_time_stamp,
+			   rtt_us ? : jiffies_to_usecs(1));
 }
 
 static inline bool tcp_ack_update_rtt(struct sock *sk, const int flag,
@@ -3102,10 +3084,11 @@ static void tcp_ack_tstamp(struct sock *sk, struct sk_buff *skb,
  */
 static int tcp_clean_rtx_queue(struct sock *sk, int prior_fackets,
 			       u32 prior_snd_una, int *acked,
-			       struct tcp_sacktag_state *sack)
+			       struct tcp_sacktag_state *sack,
+			       struct skb_mstamp *now)
 {
 	const struct inet_connection_sock *icsk = inet_csk(sk);
-	struct skb_mstamp first_ackt, last_ackt, now;
+	struct skb_mstamp first_ackt, last_ackt;
 	struct tcp_sock *tp = tcp_sk(sk);
 	u32 prior_sacked = tp->sacked_out;
 	u32 reord = tp->packets_out;
@@ -3137,7 +3120,6 @@ static int tcp_clean_rtx_queue(struct sock *sk, int prior_fackets,
 			acked_pcount = tcp_tso_acked(sk, skb);
 			if (!acked_pcount)
 				break;
-
 			fully_acked = false;
 		} else {
 			/* Speedup tcp_unlink_write_queue() and next loop */
@@ -3173,6 +3155,7 @@ static int tcp_clean_rtx_queue(struct sock *sk, int prior_fackets,
 
 		tp->packets_out -= acked_pcount;
 		pkts_acked += acked_pcount;
+		tcp_rate_skb_delivered(sk, skb, sack->rate);
 
 		/* Initial outgoing SYN's get put onto the write_queue
 		 * just like anything else we transmit.  It is not
@@ -3205,16 +3188,15 @@ static int tcp_clean_rtx_queue(struct sock *sk, int prior_fackets,
 	if (skb && (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED))
 		flag |= FLAG_SACK_RENEGING;
 
-	skb_mstamp_get(&now);
 	if (likely(first_ackt.v64) && !(flag & FLAG_RETRANS_DATA_ACKED)) {
-		seq_rtt_us = skb_mstamp_us_delta(&now, &first_ackt);
-		ca_rtt_us = skb_mstamp_us_delta(&now, &last_ackt);
+		seq_rtt_us = skb_mstamp_us_delta(now, &first_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);
+		sack_rtt_us = skb_mstamp_us_delta(now, &sack->first_sackt);
+		ca_rtt_us = skb_mstamp_us_delta(now, &sack->last_sackt);
 	}
-
+	sack->rate->rtt_us = ca_rtt_us; /* RTT of last (S)ACKed packet, or -1 */
 	rtt_update = tcp_ack_update_rtt(sk, flag, seq_rtt_us, sack_rtt_us,
 					ca_rtt_us);
 
@@ -3242,7 +3224,7 @@ static int tcp_clean_rtx_queue(struct sock *sk, int prior_fackets,
 		tp->fackets_out -= min(pkts_acked, tp->fackets_out);
 
 	} else if (skb && rtt_update && sack_rtt_us >= 0 &&
-		   sack_rtt_us > skb_mstamp_us_delta(&now, &skb->skb_mstamp)) {
+		   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
 		 * after when the head was last (re)transmitted. Otherwise the
 		 * timeout may continue to extend in loss recovery.
@@ -3333,8 +3315,15 @@ static inline bool tcp_may_raise_cwnd(const struct sock *sk, const int flag)
  * information. All transmission or retransmission are delayed afterwards.
  */
 static void tcp_cong_control(struct sock *sk, u32 ack, u32 acked_sacked,
-			     int flag)
+			     int flag, const struct rate_sample *rs)
 {
+	const struct inet_connection_sock *icsk = inet_csk(sk);
+
+	if (icsk->icsk_ca_ops->cong_control) {
+		icsk->icsk_ca_ops->cong_control(sk, rs);
+		return;
+	}
+
 	if (tcp_in_cwnd_reduction(sk)) {
 		/* Reduce cwnd if state mandates */
 		tcp_cwnd_reduction(sk, acked_sacked, flag);
@@ -3579,17 +3568,21 @@ 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;
+	struct rate_sample rs = { .prior_delivered = 0 };
 	u32 prior_snd_una = tp->snd_una;
 	u32 ack_seq = TCP_SKB_CB(skb)->seq;
 	u32 ack = TCP_SKB_CB(skb)->ack_seq;
 	bool is_dupack = false;
 	u32 prior_fackets;
 	int prior_packets = tp->packets_out;
-	u32 prior_delivered = tp->delivered;
+	u32 delivered = tp->delivered;
+	u32 lost = tp->lost;
 	int acked = 0; /* Number of packets newly acked */
 	int rexmit = REXMIT_NONE; /* Flag to (re)transmit to recover losses */
+	struct skb_mstamp now;
 
 	sack_state.first_sackt.v64 = 0;
+	sack_state.rate = &rs;
 
 	/* We very likely will need to access write queue head. */
 	prefetchw(sk->sk_write_queue.next);
@@ -3612,6 +3605,8 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag)
 	if (after(ack, tp->snd_nxt))
 		goto invalid_ack;
 
+	skb_mstamp_get(&now);
+
 	if (icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
 	    icsk->icsk_pending == ICSK_TIME_LOSS_PROBE)
 		tcp_rearm_rto(sk);
@@ -3622,6 +3617,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag)
 	}
 
 	prior_fackets = tp->fackets_out;
+	rs.prior_in_flight = tcp_packets_in_flight(tp);
 
 	/* ts_recent update must be made after we are sure that the packet
 	 * is in window.
@@ -3677,7 +3673,7 @@ 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. */
 	flag |= tcp_clean_rtx_queue(sk, prior_fackets, prior_snd_una, &acked,
-				    &sack_state);
+				    &sack_state, &now);
 
 	if (tcp_ack_is_dubious(sk, flag)) {
 		is_dupack = !(flag & (FLAG_SND_UNA_ADVANCED | FLAG_NOT_DUP));
@@ -3694,7 +3690,10 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag)
 
 	if (icsk->icsk_pending == ICSK_TIME_RETRANS)
 		tcp_schedule_loss_probe(sk);
-	tcp_cong_control(sk, ack, tp->delivered - prior_delivered, flag);
+	delivered = tp->delivered - delivered;	/* freshly ACKed or SACKed */
+	lost = tp->lost - lost;			/* freshly marked lost */
+	tcp_rate_gen(sk, delivered, lost, &now, &rs);
+	tcp_cong_control(sk, ack, delivered, flag, &rs);
 	tcp_xmit_recovery(sk, rexmit);
 	return 1;
 
@@ -5993,6 +5992,7 @@ int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb)
 		} else
 			tcp_init_metrics(sk);
 
+		if (!inet_csk(sk)->icsk_ca_ops->cong_control)
 			tcp_update_pacing_rate(sk);
 
 		/* Prevent spurious tcp_cwnd_restart() on first data packet */

net/ipv4/tcp_minisocks.c

@@ -464,7 +464,7 @@ struct sock *tcp_create_openreq_child(const struct sock *sk,
 
 		newtp->srtt_us = 0;
 		newtp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
-		newtp->rtt_min[0].rtt = ~0U;
+		minmax_reset(&newtp->rtt_min, tcp_time_stamp, ~0U);
 		newicsk->icsk_rto = TCP_TIMEOUT_INIT;
 
 		newtp->packets_out = 0;
@@ -487,6 +487,9 @@ struct sock *tcp_create_openreq_child(const struct sock *sk,
 		newtp->snd_cwnd = TCP_INIT_CWND;
 		newtp->snd_cwnd_cnt = 0;
 
+		/* There's a bubble in the pipe until at least the first ACK. */
+		newtp->app_limited = ~0U;
+
 		tcp_init_xmit_timers(newsk);
 		newtp->write_seq = newtp->pushed_seq = treq->snt_isn + 1;
 

net/ipv4/tcp_output.c

@@ -918,6 +918,7 @@ static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it,
 		skb_mstamp_get(&skb->skb_mstamp);
 		TCP_SKB_CB(skb)->tx.in_flight = TCP_SKB_CB(skb)->end_seq
 			- tp->snd_una;
+		tcp_rate_skb_sent(sk, skb);
 
 		if (unlikely(skb_cloned(skb)))
 			skb = pskb_copy(skb, gfp_mask);
@@ -1213,6 +1214,9 @@ int tcp_fragment(struct sock *sk, struct sk_buff *skb, u32 len,
 	tcp_set_skb_tso_segs(skb, mss_now);
 	tcp_set_skb_tso_segs(buff, mss_now);
 
+	/* Update delivered info for the new segment */
+	TCP_SKB_CB(buff)->tx = TCP_SKB_CB(skb)->tx;
+
 	/* If this packet has been sent out already, we must
 	 * adjust the various packet counters.
 	 */
@@ -1358,6 +1362,7 @@ int tcp_mss_to_mtu(struct sock *sk, int mss)
 	}
 	return mtu;
 }
+EXPORT_SYMBOL(tcp_mss_to_mtu);
 
 /* MTU probing init per socket */
 void tcp_mtup_init(struct sock *sk)
@@ -1545,7 +1550,8 @@ static bool tcp_nagle_check(bool partial, const struct tcp_sock *tp,
 /* Return how many segs we'd like on a TSO packet,
  * to send one TSO packet per ms
  */
-static u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now)
+u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
+		     int min_tso_segs)
 {
 	u32 bytes, segs;
 
@@ -1557,10 +1563,23 @@ static u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now)
 	 * This preserves ACK clocking and is consistent
 	 * with tcp_tso_should_defer() heuristic.
 	 */
-	segs = max_t(u32, bytes / mss_now, sysctl_tcp_min_tso_segs);
+	segs = max_t(u32, bytes / mss_now, min_tso_segs);
 
 	return min_t(u32, segs, sk->sk_gso_max_segs);
 }
+EXPORT_SYMBOL(tcp_tso_autosize);
+
+/* Return the number of segments we want in the skb we are transmitting.
+ * See if congestion control module wants to decide; otherwise, autosize.
+ */
+static u32 tcp_tso_segs(struct sock *sk, unsigned int mss_now)
+{
+	const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
+	u32 tso_segs = ca_ops->tso_segs_goal ? ca_ops->tso_segs_goal(sk) : 0;
+
+	return tso_segs ? :
+		tcp_tso_autosize(sk, mss_now, sysctl_tcp_min_tso_segs);
+}
 
 /* Returns the portion of skb which can be sent right away */
 static unsigned int tcp_mss_split_point(const struct sock *sk,
@@ -2057,7 +2076,7 @@ static bool tcp_write_xmit(struct sock *sk, unsigned int mss_now, int nonagle,
 		}
 	}
 
-	max_segs = tcp_tso_autosize(sk, mss_now);
+	max_segs = tcp_tso_segs(sk, mss_now);
 	while ((skb = tcp_send_head(sk))) {
 		unsigned int limit;
 
@@ -2774,7 +2793,7 @@ void tcp_xmit_retransmit_queue(struct sock *sk)
 		last_lost = tp->snd_una;
 	}
 
-	max_segs = tcp_tso_autosize(sk, tcp_current_mss(sk));
+	max_segs = tcp_tso_segs(sk, tcp_current_mss(sk));
 	tcp_for_write_queue_from(skb, sk) {
 		__u8 sacked;
 		int segs;

net/ipv4/tcp_rate.c

+#include <net/tcp.h>
+
+/* The bandwidth estimator estimates the rate at which the network
+ * can currently deliver outbound data packets for this flow. At a high
+ * level, it operates by taking a delivery rate sample for each ACK.
+ *
+ * A rate sample records the rate at which the network delivered packets
+ * for this flow, calculated over the time interval between the transmission
+ * of a data packet and the acknowledgment of that packet.
+ *
+ * Specifically, over the interval between each transmit and corresponding ACK,
+ * the estimator generates a delivery rate sample. Typically it uses the rate
+ * at which packets were acknowledged. However, the approach of using only the
+ * acknowledgment rate faces a challenge under the prevalent ACK decimation or
+ * compression: packets can temporarily appear to be delivered much quicker
+ * than the bottleneck rate. Since it is physically impossible to do that in a
+ * sustained fashion, when the estimator notices that the ACK rate is faster
+ * than the transmit rate, it uses the latter:
+ *
+ *    send_rate = #pkts_delivered/(last_snd_time - first_snd_time)
+ *    ack_rate  = #pkts_delivered/(last_ack_time - first_ack_time)
+ *    bw = min(send_rate, ack_rate)
+ *
+ * Notice the estimator essentially estimates the goodput, not always the
+ * network bottleneck link rate when the sending or receiving is limited by
+ * other factors like applications or receiver window limits.  The estimator
+ * deliberately avoids using the inter-packet spacing approach because that
+ * approach requires a large number of samples and sophisticated filtering.
+ *
+ * TCP flows can often be application-limited in request/response workloads.
+ * The estimator marks a bandwidth sample as application-limited if there
+ * was some moment during the sampled window of packets when there was no data
+ * ready to send in the write queue.
+ */
+
+/* Snapshot the current delivery information in the skb, to generate
+ * a rate sample later when the skb is (s)acked in tcp_rate_skb_delivered().
+ */
+void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+
+	 /* In general we need to start delivery rate samples from the
+	  * time we received the most recent ACK, to ensure we include
+	  * the full time the network needs to deliver all in-flight
+	  * packets. If there are no packets in flight yet, then we
+	  * know that any ACKs after now indicate that the network was
+	  * able to deliver those packets completely in the sampling
+	  * interval between now and the next ACK.
+	  *
+	  * Note that we use packets_out instead of tcp_packets_in_flight(tp)
+	  * because the latter is a guess based on RTO and loss-marking
+	  * heuristics. We don't want spurious RTOs or loss markings to cause
+	  * a spuriously small time interval, causing a spuriously high
+	  * bandwidth estimate.
+	  */
+	if (!tp->packets_out) {
+		tp->first_tx_mstamp  = skb->skb_mstamp;
+		tp->delivered_mstamp = skb->skb_mstamp;
+	}
+
+	TCP_SKB_CB(skb)->tx.first_tx_mstamp	= tp->first_tx_mstamp;
+	TCP_SKB_CB(skb)->tx.delivered_mstamp	= tp->delivered_mstamp;
+	TCP_SKB_CB(skb)->tx.delivered		= tp->delivered;
+	TCP_SKB_CB(skb)->tx.is_app_limited	= tp->app_limited ? 1 : 0;
+}
+
+/* When an skb is sacked or acked, we fill in the rate sample with the (prior)
+ * delivery information when the skb was last transmitted.
+ *
+ * If an ACK (s)acks multiple skbs (e.g., stretched-acks), this function is
+ * called multiple times. We favor the information from the most recently
+ * sent skb, i.e., the skb with the highest prior_delivered count.
+ */
+void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
+			    struct rate_sample *rs)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct tcp_skb_cb *scb = TCP_SKB_CB(skb);
+
+	if (!scb->tx.delivered_mstamp.v64)
+		return;
+
+	if (!rs->prior_delivered ||
+	    after(scb->tx.delivered, rs->prior_delivered)) {
+		rs->prior_delivered  = scb->tx.delivered;
+		rs->prior_mstamp     = scb->tx.delivered_mstamp;
+		rs->is_app_limited   = scb->tx.is_app_limited;
+		rs->is_retrans	     = scb->sacked & TCPCB_RETRANS;
+
+		/* Find the duration of the "send phase" of this window: */
+		rs->interval_us      = skb_mstamp_us_delta(
+						&skb->skb_mstamp,
+						&scb->tx.first_tx_mstamp);
+
+		/* Record send time of most recently ACKed packet: */
+		tp->first_tx_mstamp  = skb->skb_mstamp;
+	}
+	/* Mark off the skb delivered once it's sacked to avoid being
+	 * used again when it's cumulatively acked. For acked packets
+	 * we don't need to reset since it'll be freed soon.
+	 */
+	if (scb->sacked & TCPCB_SACKED_ACKED)
+		scb->tx.delivered_mstamp.v64 = 0;
+}
+
+/* Update the connection delivery information and generate a rate sample. */
+void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
+		  struct skb_mstamp *now, struct rate_sample *rs)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	u32 snd_us, ack_us;
+
+	/* Clear app limited if bubble is acked and gone. */
+	if (tp->app_limited && after(tp->delivered, tp->app_limited))
+		tp->app_limited = 0;
+
+	/* TODO: there are multiple places throughout tcp_ack() to get
+	 * current time. Refactor the code using a new "tcp_acktag_state"
+	 * to carry current time, flags, stats like "tcp_sacktag_state".
+	 */
+	if (delivered)
+		tp->delivered_mstamp = *now;
+
+	rs->acked_sacked = delivered;	/* freshly ACKed or SACKed */
+	rs->losses = lost;		/* freshly marked lost */
+	/* Return an invalid sample if no timing information is available. */
+	if (!rs->prior_mstamp.v64) {
+		rs->delivered = -1;
+		rs->interval_us = -1;
+		return;
+	}
+	rs->delivered   = tp->delivered - rs->prior_delivered;
+
+	/* Model sending data and receiving ACKs as separate pipeline phases
+	 * for a window. Usually the ACK phase is longer, but with ACK
+	 * compression the send phase can be longer. To be safe we use the
+	 * longer phase.
+	 */
+	snd_us = rs->interval_us;				/* send phase */
+	ack_us = skb_mstamp_us_delta(now, &rs->prior_mstamp);	/* ack phase */
+	rs->interval_us = max(snd_us, ack_us);
+
+	/* Normally we expect interval_us >= min-rtt.
+	 * Note that rate may still be over-estimated when a spuriously
+	 * retransmistted skb was first (s)acked because "interval_us"
+	 * is under-estimated (up to an RTT). However continuously
+	 * measuring the delivery rate during loss recovery is crucial
+	 * for connections suffer heavy or prolonged losses.
+	 */
+	if (unlikely(rs->interval_us < tcp_min_rtt(tp))) {
+		if (!rs->is_retrans)
+			pr_debug("tcp rate: %ld %d %u %u %u\n",
+				 rs->interval_us, rs->delivered,
+				 inet_csk(sk)->icsk_ca_state,
+				 tp->rx_opt.sack_ok, tcp_min_rtt(tp));
+		rs->interval_us = -1;
+		return;
+	}
+
+	/* Record the last non-app-limited or the highest app-limited bw */
+	if (!rs->is_app_limited ||
+	    ((u64)rs->delivered * tp->rate_interval_us >=
+	     (u64)tp->rate_delivered * rs->interval_us)) {
+		tp->rate_delivered = rs->delivered;
+		tp->rate_interval_us = rs->interval_us;
+		tp->rate_app_limited = rs->is_app_limited;
+	}
+}
+
+/* If a gap is detected between sends, mark the socket application-limited. */
+void tcp_rate_check_app_limited(struct sock *sk)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+
+	if (/* We have less than one packet to send. */
+	    tp->write_seq - tp->snd_nxt < tp->mss_cache &&
+	    /* Nothing in sending host's qdisc queues or NIC tx queue. */
+	    sk_wmem_alloc_get(sk) < SKB_TRUESIZE(1) &&
+	    /* We are not limited by CWND. */
+	    tcp_packets_in_flight(tp) < tp->snd_cwnd &&
+	    /* All lost packets have been retransmitted. */
+	    tp->lost_out <= tp->retrans_out)
+		tp->app_limited =
+			(tp->delivered + tcp_packets_in_flight(tp)) ? : 1;
+}

net/sched/sch_fq.c

@@ -94,6 +94,7 @@ struct fq_sched_data {
 	u32		flow_max_rate;	/* optional max rate per flow */
 	u32		flow_plimit;	/* max packets per flow */
 	u32		orphan_mask;	/* mask for orphaned skb */
+	u32		low_rate_threshold;
 	struct rb_root	*fq_root;
 	u8		rate_enable;
 	u8		fq_trees_log;
@@ -433,7 +434,7 @@ static struct sk_buff *fq_dequeue(struct Qdisc *sch)
 	struct fq_flow_head *head;
 	struct sk_buff *skb;
 	struct fq_flow *f;
-	u32 rate;
+	u32 rate, plen;
 
 	skb = fq_dequeue_head(sch, &q->internal);
 	if (skb)
@@ -482,7 +483,7 @@ static struct sk_buff *fq_dequeue(struct Qdisc *sch)
 	prefetch(&skb->end);
 	f->credit -= qdisc_pkt_len(skb);
 
-	if (f->credit > 0 || !q->rate_enable)
+	if (!q->rate_enable)
 		goto out;
 
 	/* Do not pace locally generated ack packets */
@@ -493,8 +494,15 @@ static struct sk_buff *fq_dequeue(struct Qdisc *sch)
 	if (skb->sk)
 		rate = min(skb->sk->sk_pacing_rate, rate);
 
+	if (rate <= q->low_rate_threshold) {
+		f->credit = 0;
+		plen = qdisc_pkt_len(skb);
+	} else {
+		plen = max(qdisc_pkt_len(skb), q->quantum);
+		if (f->credit > 0)
+			goto out;
+	}
 	if (rate != ~0U) {
-		u32 plen = max(qdisc_pkt_len(skb), q->quantum);
 		u64 len = (u64)plen * NSEC_PER_SEC;
 
 		if (likely(rate))
@@ -662,6 +670,7 @@ static const struct nla_policy fq_policy[TCA_FQ_MAX + 1] = {
 	[TCA_FQ_FLOW_MAX_RATE]		= { .type = NLA_U32 },
 	[TCA_FQ_BUCKETS_LOG]		= { .type = NLA_U32 },
 	[TCA_FQ_FLOW_REFILL_DELAY]	= { .type = NLA_U32 },
+	[TCA_FQ_LOW_RATE_THRESHOLD]	= { .type = NLA_U32 },
 };
 
 static int fq_change(struct Qdisc *sch, struct nlattr *opt)
@@ -716,6 +725,10 @@ static int fq_change(struct Qdisc *sch, struct nlattr *opt)
 	if (tb[TCA_FQ_FLOW_MAX_RATE])
 		q->flow_max_rate = nla_get_u32(tb[TCA_FQ_FLOW_MAX_RATE]);
 
+	if (tb[TCA_FQ_LOW_RATE_THRESHOLD])
+		q->low_rate_threshold =
+			nla_get_u32(tb[TCA_FQ_LOW_RATE_THRESHOLD]);
+
 	if (tb[TCA_FQ_RATE_ENABLE]) {
 		u32 enable = nla_get_u32(tb[TCA_FQ_RATE_ENABLE]);
 
@@ -781,6 +794,7 @@ static int fq_init(struct Qdisc *sch, struct nlattr *opt)
 	q->fq_root		= NULL;
 	q->fq_trees_log		= ilog2(1024);
 	q->orphan_mask		= 1024 - 1;
+	q->low_rate_threshold	= 550000 / 8;
 	qdisc_watchdog_init(&q->watchdog, sch);
 
 	if (opt)
@@ -811,6 +825,8 @@ static int fq_dump(struct Qdisc *sch, struct sk_buff *skb)
 	    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) ||
+	    nla_put_u32(skb, TCA_FQ_LOW_RATE_THRESHOLD,
+			q->low_rate_threshold) ||
 	    nla_put_u32(skb, TCA_FQ_BUCKETS_LOG, q->fq_trees_log))
 		goto nla_put_failure;