Merge tag 'rxrpc-rewrite-20160924' of...

Merge tag 'rxrpc-rewrite-20160924' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs

David Howells says:

====================
rxrpc: Implement slow-start and other bits

This set of patches implements the RxRPC slow-start feature for AF_RXRPC to
improve performance and handling of occasional packet loss.  This is more or
less the same as TCP slow start [RFC 5681].  Firstly, there are some ACK
generation improvements:

 (1) Send ACKs regularly to apprise the peer of our state so that they can do
     congestion management of their own.

 (2) Send an ACK when we fill in a hole in the buffer so that the peer can
     find out that we did this thus forestalling retransmission.

 (3) Note the final DATA packet's serial number in the final ACK for
     correlation purposes.

and a couple of bug fixes:

 (4) Reinitialise the ACK state and clear the ACK and resend timers upon
     entering the client reply reception phase to kill off any pending probe
     ACKs.

 (5) Delay the resend timer to allow for nsec->jiffies conversion errors.

and then there's the slow-start pieces:

 (6) Summarise an ACK.

 (7) Schedule a PING or IDLE ACK if the reply to a client call is overdue to
     try and find out what happened to it.

 (8) Implement the slow start feature.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

include/trace/events/rxrpc.h

@@ -570,6 +570,51 @@ TRACE_EVENT(rxrpc_retransmit,
 		      __entry->expiry)
 	    );
 
+TRACE_EVENT(rxrpc_congest,
+	    TP_PROTO(struct rxrpc_call *call, struct rxrpc_ack_summary *summary,
+		     rxrpc_serial_t ack_serial, enum rxrpc_congest_change change),
+
+	    TP_ARGS(call, summary, ack_serial, change),
+
+	    TP_STRUCT__entry(
+		    __field(struct rxrpc_call *,		call		)
+		    __field(enum rxrpc_congest_change,		change		)
+		    __field(rxrpc_seq_t,			hard_ack	)
+		    __field(rxrpc_seq_t,			top		)
+		    __field(rxrpc_seq_t,			lowest_nak	)
+		    __field(rxrpc_serial_t,			ack_serial	)
+		    __field_struct(struct rxrpc_ack_summary,	sum		)
+			     ),
+
+	    TP_fast_assign(
+		    __entry->call	= call;
+		    __entry->change	= change;
+		    __entry->hard_ack	= call->tx_hard_ack;
+		    __entry->top	= call->tx_top;
+		    __entry->lowest_nak	= call->acks_lowest_nak;
+		    __entry->ack_serial	= ack_serial;
+		    memcpy(&__entry->sum, summary, sizeof(__entry->sum));
+			   ),
+
+	    TP_printk("c=%p %08x %s %08x %s cw=%u ss=%u nr=%u,%u nw=%u,%u r=%u b=%u u=%u d=%u l=%x%s%s%s",
+		      __entry->call,
+		      __entry->ack_serial,
+		      rxrpc_ack_names[__entry->sum.ack_reason],
+		      __entry->hard_ack,
+		      rxrpc_congest_modes[__entry->sum.mode],
+		      __entry->sum.cwnd,
+		      __entry->sum.ssthresh,
+		      __entry->sum.nr_acks, __entry->sum.nr_nacks,
+		      __entry->sum.nr_new_acks, __entry->sum.nr_new_nacks,
+		      __entry->sum.nr_rot_new_acks,
+		      __entry->top - __entry->hard_ack,
+		      __entry->sum.cumulative_acks,
+		      __entry->sum.dup_acks,
+		      __entry->lowest_nak, __entry->sum.new_low_nack ? "!" : "",
+		      rxrpc_congest_changes[__entry->change],
+		      __entry->sum.retrans_timeo ? " rTxTo" : "")
+	    );
+
 #endif /* _TRACE_RXRPC_H */
 
 /* This part must be outside protection */

net/rxrpc/ar-internal.h

@@ -402,6 +402,7 @@ enum rxrpc_call_flag {
 	RXRPC_CALL_RX_LAST,		/* Received the last packet (at rxtx_top) */
 	RXRPC_CALL_TX_LAST,		/* Last packet in Tx buffer (at rxtx_top) */
 	RXRPC_CALL_PINGING,		/* Ping in process */
+	RXRPC_CALL_RETRANS_TIMEOUT,	/* Retransmission due to timeout occurred */
 };
 
 /*
@@ -446,6 +447,17 @@ enum rxrpc_call_completion {
 	NR__RXRPC_CALL_COMPLETIONS
 };
 
+/*
+ * Call Tx congestion management modes.
+ */
+enum rxrpc_congest_mode {
+	RXRPC_CALL_SLOW_START,
+	RXRPC_CALL_CONGEST_AVOIDANCE,
+	RXRPC_CALL_PACKET_LOSS,
+	RXRPC_CALL_FAST_RETRANSMIT,
+	NR__RXRPC_CONGEST_MODES
+};
+
 /*
  * RxRPC call definition
  * - matched by { connection, call_id }
@@ -518,6 +530,20 @@ struct rxrpc_call {
 						 * not hard-ACK'd packet follows this.
 						 */
 	rxrpc_seq_t		tx_top;		/* Highest Tx slot allocated. */
+
+	/* TCP-style slow-start congestion control [RFC5681].  Since the SMSS
+	 * is fixed, we keep these numbers in terms of segments (ie. DATA
+	 * packets) rather than bytes.
+	 */
+#define RXRPC_TX_SMSS		RXRPC_JUMBO_DATALEN
+	u8			cong_cwnd;	/* Congestion window size */
+	u8			cong_extra;	/* Extra to send for congestion management */
+	u8			cong_ssthresh;	/* Slow-start threshold */
+	enum rxrpc_congest_mode	cong_mode:8;	/* Congestion management mode */
+	u8			cong_dup_acks;	/* Count of ACKs showing missing packets */
+	u8			cong_cumul_acks; /* Cumulative ACK count */
+	ktime_t			cong_tstamp;	/* Last time cwnd was changed */
+
 	rxrpc_seq_t		rx_hard_ack;	/* Dead slot in buffer; the first received but not
 						 * consumed packet follows this.
 						 */
@@ -533,11 +559,36 @@ struct rxrpc_call {
 	u16			ackr_skew;	/* skew on packet being ACK'd */
 	rxrpc_serial_t		ackr_serial;	/* serial of packet being ACK'd */
 	rxrpc_seq_t		ackr_prev_seq;	/* previous sequence number received */
+	rxrpc_seq_t		ackr_consumed;	/* Highest packet shown consumed */
+	rxrpc_seq_t		ackr_seen;	/* Highest packet shown seen */
 	rxrpc_serial_t		ackr_ping;	/* Last ping sent */
 	ktime_t			ackr_ping_time;	/* Time last ping sent */
 
 	/* transmission-phase ACK management */
+	ktime_t			acks_latest_ts;	/* Timestamp of latest ACK received */
 	rxrpc_serial_t		acks_latest;	/* serial number of latest ACK received */
+	rxrpc_seq_t		acks_lowest_nak; /* Lowest NACK in the buffer (or ==tx_hard_ack) */
+};
+
+/*
+ * Summary of a new ACK and the changes it made to the Tx buffer packet states.
+ */
+struct rxrpc_ack_summary {
+	u8			ack_reason;
+	u8			nr_acks;		/* Number of ACKs in packet */
+	u8			nr_nacks;		/* Number of NACKs in packet */
+	u8			nr_new_acks;		/* Number of new ACKs in packet */
+	u8			nr_new_nacks;		/* Number of new NACKs in packet */
+	u8			nr_rot_new_acks;	/* Number of rotated new ACKs */
+	bool			new_low_nack;		/* T if new low NACK found */
+	bool			retrans_timeo;		/* T if reTx due to timeout happened */
+	u8			flight_size;		/* Number of unreceived transmissions */
+	/* Place to stash values for tracing */
+	enum rxrpc_congest_mode	mode:8;
+	u8			cwnd;
+	u8			ssthresh;
+	u8			dup_acks;
+	u8			cumulative_acks;
 };
 
 enum rxrpc_skb_trace {
@@ -680,6 +731,7 @@ extern const char rxrpc_rtt_rx_traces[rxrpc_rtt_rx__nr_trace][5];
 
 enum rxrpc_timer_trace {
 	rxrpc_timer_begin,
+	rxrpc_timer_init_for_reply,
 	rxrpc_timer_expired,
 	rxrpc_timer_set_for_ack,
 	rxrpc_timer_set_for_resend,
@@ -690,11 +742,15 @@ enum rxrpc_timer_trace {
 extern const char rxrpc_timer_traces[rxrpc_timer__nr_trace][8];
 
 enum rxrpc_propose_ack_trace {
+	rxrpc_propose_ack_client_tx_end,
 	rxrpc_propose_ack_input_data,
+	rxrpc_propose_ack_ping_for_lost_ack,
+	rxrpc_propose_ack_ping_for_lost_reply,
 	rxrpc_propose_ack_ping_for_params,
 	rxrpc_propose_ack_respond_to_ack,
 	rxrpc_propose_ack_respond_to_ping,
 	rxrpc_propose_ack_retry_tx,
+	rxrpc_propose_ack_rotate_rx,
 	rxrpc_propose_ack_terminal_ack,
 	rxrpc_propose_ack__nr_trace
 };
@@ -709,6 +765,21 @@ enum rxrpc_propose_ack_outcome {
 extern const char rxrpc_propose_ack_traces[rxrpc_propose_ack__nr_trace][8];
 extern const char *const rxrpc_propose_ack_outcomes[rxrpc_propose_ack__nr_outcomes];
 
+enum rxrpc_congest_change {
+	rxrpc_cong_begin_retransmission,
+	rxrpc_cong_cleared_nacks,
+	rxrpc_cong_new_low_nack,
+	rxrpc_cong_no_change,
+	rxrpc_cong_progress,
+	rxrpc_cong_retransmit_again,
+	rxrpc_cong_rtt_window_end,
+	rxrpc_cong_saw_nack,
+	rxrpc_congest__nr_change
+};
+
+extern const char rxrpc_congest_modes[NR__RXRPC_CONGEST_MODES][10];
+extern const char rxrpc_congest_changes[rxrpc_congest__nr_change][9];
+
 extern const char *const rxrpc_pkts[];
 extern const char const rxrpc_ack_names[RXRPC_ACK__INVALID + 1][4];
 

net/rxrpc/call_event.c

@@ -100,6 +100,7 @@ static void __rxrpc_propose_ACK(struct rxrpc_call *call, u8 ack_reason,
 			expiry = rxrpc_soft_ack_delay;
 		break;
 
+	case RXRPC_ACK_PING:
 	case RXRPC_ACK_IDLE:
 		if (rxrpc_idle_ack_delay < expiry)
 			expiry = rxrpc_idle_ack_delay;
@@ -145,6 +146,14 @@ void rxrpc_propose_ACK(struct rxrpc_call *call, u8 ack_reason,
 	spin_unlock_bh(&call->lock);
 }
 
+/*
+ * Handle congestion being detected by the retransmit timeout.
+ */
+static void rxrpc_congestion_timeout(struct rxrpc_call *call)
+{
+	set_bit(RXRPC_CALL_RETRANS_TIMEOUT, &call->flags);
+}
+
 /*
  * Perform retransmission of NAK'd and unack'd packets.
  */
@@ -153,9 +162,9 @@ static void rxrpc_resend(struct rxrpc_call *call)
 	struct rxrpc_skb_priv *sp;
 	struct sk_buff *skb;
 	rxrpc_seq_t cursor, seq, top;
-	ktime_t now = ktime_get_real(), max_age, oldest,  resend_at;
+	ktime_t now = ktime_get_real(), max_age, oldest, resend_at, ack_ts;
 	int ix;
-	u8 annotation, anno_type;
+	u8 annotation, anno_type, retrans = 0, unacked = 0;
 
 	_enter("{%d,%d}", call->tx_hard_ack, call->tx_top);
 
@@ -192,16 +201,44 @@ static void rxrpc_resend(struct rxrpc_call *call)
 					oldest = skb->tstamp;
 				continue;
 			}
+			if (!(annotation & RXRPC_TX_ANNO_RESENT))
+				unacked++;
 		}
 
 		/* Okay, we need to retransmit a packet. */
 		call->rxtx_annotations[ix] = RXRPC_TX_ANNO_RETRANS | annotation;
+		retrans++;
 		trace_rxrpc_retransmit(call, seq, annotation | anno_type,
 				       ktime_to_ns(ktime_sub(skb->tstamp, max_age)));
 	}
 
-	resend_at = ktime_sub(ktime_add_ms(oldest, rxrpc_resend_timeout), now);
-	call->resend_at = jiffies + nsecs_to_jiffies(ktime_to_ns(resend_at));
+	resend_at = ktime_add_ms(oldest, rxrpc_resend_timeout);
+	call->resend_at = jiffies +
+		nsecs_to_jiffies(ktime_to_ns(ktime_sub(resend_at, now))) +
+		1; /* We have to make sure that the calculated jiffies value
+		    * falls at or after the nsec value, or we shall loop
+		    * ceaselessly because the timer times out, but we haven't
+		    * reached the nsec timeout yet.
+		    */
+
+	if (unacked)
+		rxrpc_congestion_timeout(call);
+
+	/* If there was nothing that needed retransmission then it's likely
+	 * that an ACK got lost somewhere.  Send a ping to find out instead of
+	 * retransmitting data.
+	 */
+	if (!retrans) {
+		rxrpc_set_timer(call, rxrpc_timer_set_for_resend);
+		spin_unlock_bh(&call->lock);
+		ack_ts = ktime_sub(now, call->acks_latest_ts);
+		if (ktime_to_ns(ack_ts) < call->peer->rtt)
+			goto out;
+		rxrpc_propose_ACK(call, RXRPC_ACK_PING, 0, 0, true, false,
+				  rxrpc_propose_ack_ping_for_lost_ack);
+		rxrpc_send_call_packet(call, RXRPC_PACKET_TYPE_ACK);
+		goto out;
+	}
 
 	/* Now go through the Tx window and perform the retransmissions.  We
 	 * have to drop the lock for each send.  If an ACK comes in whilst the
@@ -253,6 +290,7 @@ static void rxrpc_resend(struct rxrpc_call *call)
 
 out_unlock:
 	spin_unlock_bh(&call->lock);
+out:
 	_leave("");
 }
 
@@ -286,6 +324,7 @@ void rxrpc_process_call(struct work_struct *work)
 	if (time_after_eq(now, call->expire_at)) {
 		rxrpc_abort_call("EXP", call, 0, RX_CALL_TIMEOUT, ETIME);
 		set_bit(RXRPC_CALL_EV_ABORT, &call->events);
+		goto recheck_state;
 	}
 
 	if (test_and_clear_bit(RXRPC_CALL_EV_ACK, &call->events) ||

net/rxrpc/call_object.c

@@ -160,6 +160,14 @@ struct rxrpc_call *rxrpc_alloc_call(gfp_t gfp)
 	call->rx_winsize = rxrpc_rx_window_size;
 	call->tx_winsize = 16;
 	call->rx_expect_next = 1;
+
+	if (RXRPC_TX_SMSS > 2190)
+		call->cong_cwnd = 2;
+	else if (RXRPC_TX_SMSS > 1095)
+		call->cong_cwnd = 3;
+	else
+		call->cong_cwnd = 4;
+	call->cong_ssthresh = RXRPC_RXTX_BUFF_SIZE - 1;
 	return call;
 
 nomem_2:
@@ -176,6 +184,7 @@ static struct rxrpc_call *rxrpc_alloc_client_call(struct sockaddr_rxrpc *srx,
 						  gfp_t gfp)
 {
 	struct rxrpc_call *call;
+	ktime_t now;
 
 	_enter("");
 
@@ -185,6 +194,9 @@ static struct rxrpc_call *rxrpc_alloc_client_call(struct sockaddr_rxrpc *srx,
 	call->state = RXRPC_CALL_CLIENT_AWAIT_CONN;
 	call->service_id = srx->srx_service;
 	call->tx_phase = true;
+	now = ktime_get_real();
+	call->acks_latest_ts = now;
+	call->cong_tstamp = now;
 
 	_leave(" = %p", call);
 	return call;
@@ -325,6 +337,7 @@ void rxrpc_incoming_call(struct rxrpc_sock *rx,
 	call->state		= RXRPC_CALL_SERVER_ACCEPTING;
 	if (sp->hdr.securityIndex > 0)
 		call->state	= RXRPC_CALL_SERVER_SECURING;
+	call->cong_tstamp	= skb->tstamp;
 
 	/* Set the channel for this call.  We don't get channel_lock as we're
 	 * only defending against the data_ready handler (which we're called

net/rxrpc/conn_event.c

@@ -97,6 +97,7 @@ static void rxrpc_conn_retransmit_call(struct rxrpc_connection *conn,
 		pkt.info.maxMTU		= htonl(mtu);
 		pkt.info.rwind		= htonl(rxrpc_rx_window_size);
 		pkt.info.jumbo_max	= htonl(rxrpc_rx_jumbo_max);
+		pkt.whdr.flags		|= RXRPC_SLOW_START_OK;
 		len += sizeof(pkt.ack) + sizeof(pkt.info);
 		break;
 	}

net/rxrpc/input.c

@@ -36,6 +36,166 @@ static void rxrpc_proto_abort(const char *why,
 	}
 }
 
+/*
+ * Do TCP-style congestion management [RFC 5681].
+ */
+static void rxrpc_congestion_management(struct rxrpc_call *call,
+					struct sk_buff *skb,
+					struct rxrpc_ack_summary *summary)
+{
+	enum rxrpc_congest_change change = rxrpc_cong_no_change;
+	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
+	unsigned int cumulative_acks = call->cong_cumul_acks;
+	unsigned int cwnd = call->cong_cwnd;
+	bool resend = false;
+
+	summary->flight_size =
+		(call->tx_top - call->tx_hard_ack) - summary->nr_acks;
+
+	if (test_and_clear_bit(RXRPC_CALL_RETRANS_TIMEOUT, &call->flags)) {
+		summary->retrans_timeo = true;
+		call->cong_ssthresh = max_t(unsigned int,
+					    summary->flight_size / 2, 2);
+		cwnd = 1;
+		if (cwnd > call->cong_ssthresh &&
+		    call->cong_mode == RXRPC_CALL_SLOW_START) {
+			call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
+			call->cong_tstamp = skb->tstamp;
+			cumulative_acks = 0;
+		}
+	}
+
+	cumulative_acks += summary->nr_new_acks;
+	cumulative_acks += summary->nr_rot_new_acks;
+	if (cumulative_acks > 255)
+		cumulative_acks = 255;
+
+	summary->mode = call->cong_mode;
+	summary->cwnd = call->cong_cwnd;
+	summary->ssthresh = call->cong_ssthresh;
+	summary->cumulative_acks = cumulative_acks;
+	summary->dup_acks = call->cong_dup_acks;
+
+	switch (call->cong_mode) {
+	case RXRPC_CALL_SLOW_START:
+		if (summary->nr_nacks > 0)
+			goto packet_loss_detected;
+		if (summary->cumulative_acks > 0)
+			cwnd += 1;
+		if (cwnd > call->cong_ssthresh) {
+			call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
+			call->cong_tstamp = skb->tstamp;
+		}
+		goto out;
+
+	case RXRPC_CALL_CONGEST_AVOIDANCE:
+		if (summary->nr_nacks > 0)
+			goto packet_loss_detected;
+
+		/* We analyse the number of packets that get ACK'd per RTT
+		 * period and increase the window if we managed to fill it.
+		 */
+		if (call->peer->rtt_usage == 0)
+			goto out;
+		if (ktime_before(skb->tstamp,
+				 ktime_add_ns(call->cong_tstamp,
+					      call->peer->rtt)))
+			goto out_no_clear_ca;
+		change = rxrpc_cong_rtt_window_end;
+		call->cong_tstamp = skb->tstamp;
+		if (cumulative_acks >= cwnd)
+			cwnd++;
+		goto out;
+
+	case RXRPC_CALL_PACKET_LOSS:
+		if (summary->nr_nacks == 0)
+			goto resume_normality;
+
+		if (summary->new_low_nack) {
+			change = rxrpc_cong_new_low_nack;
+			call->cong_dup_acks = 1;
+			if (call->cong_extra > 1)
+				call->cong_extra = 1;
+			goto send_extra_data;
+		}
+
+		call->cong_dup_acks++;
+		if (call->cong_dup_acks < 3)
+			goto send_extra_data;
+
+		change = rxrpc_cong_begin_retransmission;
+		call->cong_mode = RXRPC_CALL_FAST_RETRANSMIT;
+		call->cong_ssthresh = max_t(unsigned int,
+					    summary->flight_size / 2, 2);
+		cwnd = call->cong_ssthresh + 3;
+		call->cong_extra = 0;
+		call->cong_dup_acks = 0;
+		resend = true;
+		goto out;
+
+	case RXRPC_CALL_FAST_RETRANSMIT:
+		if (!summary->new_low_nack) {
+			if (summary->nr_new_acks == 0)
+				cwnd += 1;
+			call->cong_dup_acks++;
+			if (call->cong_dup_acks == 2) {
+				change = rxrpc_cong_retransmit_again;
+				call->cong_dup_acks = 0;
+				resend = true;
+			}
+		} else {
+			change = rxrpc_cong_progress;
+			cwnd = call->cong_ssthresh;
+			if (summary->nr_nacks == 0)
+				goto resume_normality;
+		}
+		goto out;
+
+	default:
+		BUG();
+		goto out;
+	}
+
+resume_normality:
+	change = rxrpc_cong_cleared_nacks;
+	call->cong_dup_acks = 0;
+	call->cong_extra = 0;
+	call->cong_tstamp = skb->tstamp;
+	if (cwnd <= call->cong_ssthresh)
+		call->cong_mode = RXRPC_CALL_SLOW_START;
+	else
+		call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
+out:
+	cumulative_acks = 0;
+out_no_clear_ca:
+	if (cwnd >= RXRPC_RXTX_BUFF_SIZE - 1)
+		cwnd = RXRPC_RXTX_BUFF_SIZE - 1;
+	call->cong_cwnd = cwnd;
+	call->cong_cumul_acks = cumulative_acks;
+	trace_rxrpc_congest(call, summary, sp->hdr.serial, change);
+	if (resend && !test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events))
+		rxrpc_queue_call(call);
+	return;
+
+packet_loss_detected:
+	change = rxrpc_cong_saw_nack;
+	call->cong_mode = RXRPC_CALL_PACKET_LOSS;
+	call->cong_dup_acks = 0;
+	goto send_extra_data;
+
+send_extra_data:
+	/* Send some previously unsent DATA if we have some to advance the ACK
+	 * state.
+	 */
+	if (call->rxtx_annotations[call->tx_top & RXRPC_RXTX_BUFF_MASK] &
+	    RXRPC_TX_ANNO_LAST ||
+	    summary->nr_acks != call->tx_top - call->tx_hard_ack) {
+		call->cong_extra++;
+		wake_up(&call->waitq);
+	}
+	goto out_no_clear_ca;
+}
+
 /*
  * Ping the other end to fill our RTT cache and to retrieve the rwind
  * and MTU parameters.
@@ -56,12 +216,20 @@ static void rxrpc_send_ping(struct rxrpc_call *call, struct sk_buff *skb,
 /*
  * Apply a hard ACK by advancing the Tx window.
  */
-static void rxrpc_rotate_tx_window(struct rxrpc_call *call, rxrpc_seq_t to)
+static void rxrpc_rotate_tx_window(struct rxrpc_call *call, rxrpc_seq_t to,
+				   struct rxrpc_ack_summary *summary)
 {
 	struct sk_buff *skb, *list = NULL;
 	int ix;
 	u8 annotation;
 
+	if (call->acks_lowest_nak == call->tx_hard_ack) {
+		call->acks_lowest_nak = to;
+	} else if (before_eq(call->acks_lowest_nak, to)) {
+		summary->new_low_nack = true;
+		call->acks_lowest_nak = to;
+	}
+
 	spin_lock(&call->lock);
 
 	while (before(call->tx_hard_ack, to)) {
@@ -77,6 +245,8 @@ static void rxrpc_rotate_tx_window(struct rxrpc_call *call, rxrpc_seq_t to)
 
 		if (annotation & RXRPC_TX_ANNO_LAST)
 			set_bit(RXRPC_CALL_TX_LAST, &call->flags);
+		if ((annotation & RXRPC_TX_ANNO_MASK) != RXRPC_TX_ANNO_ACK)
+			summary->nr_rot_new_acks++;
 	}
 
 	spin_unlock(&call->lock);
@@ -128,6 +298,8 @@ static bool rxrpc_end_tx_phase(struct rxrpc_call *call, bool reply_begun,
 
 	write_unlock(&call->state_lock);
 	if (call->state == RXRPC_CALL_CLIENT_AWAIT_REPLY) {
+		rxrpc_propose_ACK(call, RXRPC_ACK_IDLE, 0, 0, false, true,
+				  rxrpc_propose_ack_client_tx_end);
 		trace_rxrpc_transmit(call, rxrpc_transmit_await_reply);
 	} else {
 		trace_rxrpc_transmit(call, rxrpc_transmit_end);
@@ -147,10 +319,20 @@ static bool rxrpc_end_tx_phase(struct rxrpc_call *call, bool reply_begun,
  */
 static bool rxrpc_receiving_reply(struct rxrpc_call *call)
 {
+	struct rxrpc_ack_summary summary = { 0 };
 	rxrpc_seq_t top = READ_ONCE(call->tx_top);
 
+	if (call->ackr_reason) {
+		spin_lock_bh(&call->lock);
+		call->ackr_reason = 0;
+		call->resend_at = call->expire_at;
+		call->ack_at = call->expire_at;
+		spin_unlock_bh(&call->lock);
+		rxrpc_set_timer(call, rxrpc_timer_init_for_reply);
+	}
+
 	if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags))
-		rxrpc_rotate_tx_window(call, top);
+		rxrpc_rotate_tx_window(call, top, &summary);
 	if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags)) {
 		rxrpc_proto_abort("TXL", call, top);
 		return false;
@@ -331,8 +513,16 @@ static void rxrpc_input_data(struct rxrpc_call *call, struct sk_buff *skb,
 	call->rxtx_annotations[ix] = annotation;
 	smp_wmb();
 	call->rxtx_buffer[ix] = skb;
-	if (after(seq, call->rx_top))
+	if (after(seq, call->rx_top)) {
 		smp_store_release(&call->rx_top, seq);
+	} else if (before(seq, call->rx_top)) {
+		/* Send an immediate ACK if we fill in a hole */
+		if (!ack) {
+			ack = RXRPC_ACK_DELAY;
+			ack_serial = serial;
+		}
+		immediate_ack = true;
+	}
 	if (flags & RXRPC_LAST_PACKET) {
 		set_bit(RXRPC_CALL_RX_LAST, &call->flags);
 		trace_rxrpc_receive(call, rxrpc_receive_queue_last, serial, seq);
@@ -491,9 +681,9 @@ static void rxrpc_input_ackinfo(struct rxrpc_call *call, struct sk_buff *skb,
  * the time the ACK was sent.
  */
 static void rxrpc_input_soft_acks(struct rxrpc_call *call, u8 *acks,
-				  rxrpc_seq_t seq, int nr_acks)
+				  rxrpc_seq_t seq, int nr_acks,
+				  struct rxrpc_ack_summary *summary)
 {
-	bool resend = false;
 	int ix;
 	u8 annotation, anno_type;
 
@@ -504,28 +694,32 @@ static void rxrpc_input_soft_acks(struct rxrpc_call *call, u8 *acks,
 		annotation &= ~RXRPC_TX_ANNO_MASK;
 		switch (*acks++) {
 		case RXRPC_ACK_TYPE_ACK:
+			summary->nr_acks++;
 			if (anno_type == RXRPC_TX_ANNO_ACK)
 				continue;
+			summary->nr_new_acks++;
 			call->rxtx_annotations[ix] =
 				RXRPC_TX_ANNO_ACK | annotation;
 			break;
 		case RXRPC_ACK_TYPE_NACK:
+			if (!summary->nr_nacks &&
+			    call->acks_lowest_nak != seq) {
+				call->acks_lowest_nak = seq;
+				summary->new_low_nack = true;
+			}
+			summary->nr_nacks++;
 			if (anno_type == RXRPC_TX_ANNO_NAK)
 				continue;
+			summary->nr_new_nacks++;
 			if (anno_type == RXRPC_TX_ANNO_RETRANS)
 				continue;
 			call->rxtx_annotations[ix] =
 				RXRPC_TX_ANNO_NAK | annotation;
-			resend = true;
 			break;
 		default:
 			return rxrpc_proto_abort("SFT", call, 0);
 		}
 	}
-
-	if (resend &&
-	    !test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events))
-		rxrpc_queue_call(call);
 }
 
 /*
@@ -541,7 +735,7 @@ static void rxrpc_input_soft_acks(struct rxrpc_call *call, u8 *acks,
 static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb,
 			    u16 skew)
 {
-	u8 ack_reason;
+	struct rxrpc_ack_summary summary = { 0 };
 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
 	union {
 		struct rxrpc_ackpacket ack;
@@ -564,10 +758,10 @@ static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb,
 	first_soft_ack = ntohl(buf.ack.firstPacket);
 	hard_ack = first_soft_ack - 1;
 	nr_acks = buf.ack.nAcks;
-	ack_reason = (buf.ack.reason < RXRPC_ACK__INVALID ?
-		      buf.ack.reason : RXRPC_ACK__INVALID);
+	summary.ack_reason = (buf.ack.reason < RXRPC_ACK__INVALID ?
+			      buf.ack.reason : RXRPC_ACK__INVALID);
 
-	trace_rxrpc_rx_ack(call, first_soft_ack, ack_reason, nr_acks);
+	trace_rxrpc_rx_ack(call, first_soft_ack, summary.ack_reason, nr_acks);
 
 	_proto("Rx ACK %%%u { m=%hu f=#%u p=#%u s=%%%u r=%s n=%u }",
 	       sp->hdr.serial,
@@ -575,7 +769,7 @@ static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb,
 	       first_soft_ack,
 	       ntohl(buf.ack.previousPacket),
 	       acked_serial,
-	       rxrpc_ack_names[ack_reason],
+	       rxrpc_ack_names[summary.ack_reason],
 	       buf.ack.nAcks);
 
 	if (buf.ack.reason == RXRPC_ACK_PING_RESPONSE)
@@ -623,6 +817,7 @@ static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb,
 		       sp->hdr.serial, call->acks_latest);
 		return;
 	}
+	call->acks_latest_ts = skb->tstamp;
 	call->acks_latest = sp->hdr.serial;
 
 	if (before(hard_ack, call->tx_hard_ack) ||
@@ -632,12 +827,13 @@ static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb,
 		return rxrpc_proto_abort("AKN", call, 0);
 
 	if (after(hard_ack, call->tx_hard_ack))
-		rxrpc_rotate_tx_window(call, hard_ack);
+		rxrpc_rotate_tx_window(call, hard_ack, &summary);
 
 	if (nr_acks > 0) {
 		if (skb_copy_bits(skb, sp->offset, buf.acks, nr_acks) < 0)
 			return rxrpc_proto_abort("XSA", call, 0);
-		rxrpc_input_soft_acks(call, buf.acks, first_soft_ack, nr_acks);
+		rxrpc_input_soft_acks(call, buf.acks, first_soft_ack, nr_acks,
+				      &summary);
 	}
 
 	if (test_bit(RXRPC_CALL_TX_LAST, &call->flags)) {
@@ -645,6 +841,14 @@ static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb,
 		return;
 	}
 
+	if (call->rxtx_annotations[call->tx_top & RXRPC_RXTX_BUFF_MASK] &
+	    RXRPC_TX_ANNO_LAST &&
+	    summary.nr_acks == call->tx_top - hard_ack)
+		rxrpc_propose_ACK(call, RXRPC_ACK_PING, skew, sp->hdr.serial,
+				  false, true,
+				  rxrpc_propose_ack_ping_for_lost_reply);
+
+	return rxrpc_congestion_management(call, skb, &summary);
 }
 
 /*
@@ -652,11 +856,12 @@ static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb,
  */
 static void rxrpc_input_ackall(struct rxrpc_call *call, struct sk_buff *skb)
 {
+	struct rxrpc_ack_summary summary = { 0 };
 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
 
 	_proto("Rx ACKALL %%%u", sp->hdr.serial);
 
-	rxrpc_rotate_tx_window(call, call->tx_top);
+	rxrpc_rotate_tx_window(call, call->tx_top, &summary);
 	if (test_bit(RXRPC_CALL_TX_LAST, &call->flags))
 		rxrpc_end_tx_phase(call, false, "ETL");
 }

net/rxrpc/misc.c

@@ -191,17 +191,22 @@ const char rxrpc_rtt_rx_traces[rxrpc_rtt_rx__nr_trace][5] = {
 const char rxrpc_timer_traces[rxrpc_timer__nr_trace][8] = {
 	[rxrpc_timer_begin]			= "Begin ",
 	[rxrpc_timer_expired]			= "*EXPR*",
+	[rxrpc_timer_init_for_reply]		= "IniRpl",
 	[rxrpc_timer_set_for_ack]		= "SetAck",
 	[rxrpc_timer_set_for_send]		= "SetTx ",
 	[rxrpc_timer_set_for_resend]		= "SetRTx",
 };
 
 const char rxrpc_propose_ack_traces[rxrpc_propose_ack__nr_trace][8] = {
+	[rxrpc_propose_ack_client_tx_end]	= "ClTxEnd",
 	[rxrpc_propose_ack_input_data]		= "DataIn ",
+	[rxrpc_propose_ack_ping_for_lost_ack]	= "LostAck",
+	[rxrpc_propose_ack_ping_for_lost_reply]	= "LostRpl",
 	[rxrpc_propose_ack_ping_for_params]	= "Params ",
 	[rxrpc_propose_ack_respond_to_ack]	= "Rsp2Ack",
 	[rxrpc_propose_ack_respond_to_ping]	= "Rsp2Png",
 	[rxrpc_propose_ack_retry_tx]		= "RetryTx",
+	[rxrpc_propose_ack_rotate_rx]		= "RxAck  ",
 	[rxrpc_propose_ack_terminal_ack]	= "ClTerm ",
 };
 
@@ -210,3 +215,21 @@ const char *const rxrpc_propose_ack_outcomes[rxrpc_propose_ack__nr_outcomes] = {
 	[rxrpc_propose_ack_update]		= " Update",
 	[rxrpc_propose_ack_subsume]		= " Subsume",
 };
+
+const char rxrpc_congest_modes[NR__RXRPC_CONGEST_MODES][10] = {
+	[RXRPC_CALL_SLOW_START]		= "SlowStart",
+	[RXRPC_CALL_CONGEST_AVOIDANCE]	= "CongAvoid",
+	[RXRPC_CALL_PACKET_LOSS]	= "PktLoss  ",
+	[RXRPC_CALL_FAST_RETRANSMIT]	= "FastReTx ",
+};
+
+const char rxrpc_congest_changes[rxrpc_congest__nr_change][9] = {
+	[rxrpc_cong_begin_retransmission]	= " Retrans",
+	[rxrpc_cong_cleared_nacks]		= " Cleared",
+	[rxrpc_cong_new_low_nack]		= " NewLowN",
+	[rxrpc_cong_no_change]			= "",
+	[rxrpc_cong_progress]			= " Progres",
+	[rxrpc_cong_retransmit_again]		= " ReTxAgn",
+	[rxrpc_cong_rtt_window_end]		= " RttWinE",
+	[rxrpc_cong_saw_nack]			= " SawNack",
+};

net/rxrpc/output.c

@@ -36,7 +36,9 @@ struct rxrpc_pkt_buffer {
  * Fill out an ACK packet.
  */
 static size_t rxrpc_fill_out_ack(struct rxrpc_call *call,
-				 struct rxrpc_pkt_buffer *pkt)
+				 struct rxrpc_pkt_buffer *pkt,
+				 rxrpc_seq_t *_hard_ack,
+				 rxrpc_seq_t *_top)
 {
 	rxrpc_serial_t serial;
 	rxrpc_seq_t hard_ack, top, seq;
@@ -48,6 +50,8 @@ static size_t rxrpc_fill_out_ack(struct rxrpc_call *call,
 	serial = call->ackr_serial;
 	hard_ack = READ_ONCE(call->rx_hard_ack);
 	top = smp_load_acquire(&call->rx_top);
+	*_hard_ack = hard_ack;
+	*_top = top;
 
 	pkt->ack.bufferSpace	= htons(8);
 	pkt->ack.maxSkew	= htons(call->ackr_skew);
@@ -96,6 +100,7 @@ int rxrpc_send_call_packet(struct rxrpc_call *call, u8 type)
 	struct msghdr msg;
 	struct kvec iov[2];
 	rxrpc_serial_t serial;
+	rxrpc_seq_t hard_ack, top;
 	size_t len, n;
 	bool ping = false;
 	int ioc, ret;
@@ -146,12 +151,14 @@ int rxrpc_send_call_packet(struct rxrpc_call *call, u8 type)
 			goto out;
 		}
 		ping = (call->ackr_reason == RXRPC_ACK_PING);
-		n = rxrpc_fill_out_ack(call, pkt);
+		n = rxrpc_fill_out_ack(call, pkt, &hard_ack, &top);
 		call->ackr_reason = 0;
 
 		spin_unlock_bh(&call->lock);
 
 
+		pkt->whdr.flags |= RXRPC_SLOW_START_OK;
+
 		iov[0].iov_len += sizeof(pkt->ack) + n;
 		iov[1].iov_base = &pkt->ackinfo;
 		iov[1].iov_len	= sizeof(pkt->ackinfo);
@@ -203,18 +210,22 @@ int rxrpc_send_call_packet(struct rxrpc_call *call, u8 type)
 	if (ping)
 		call->ackr_ping_time = ktime_get_real();
 
-	if (ret < 0 && call->state < RXRPC_CALL_COMPLETE) {
-		switch (type) {
-		case RXRPC_PACKET_TYPE_ACK:
+	if (type == RXRPC_PACKET_TYPE_ACK &&
+	    call->state < RXRPC_CALL_COMPLETE) {
+		if (ret < 0) {
 			clear_bit(RXRPC_CALL_PINGING, &call->flags);
 			rxrpc_propose_ACK(call, pkt->ack.reason,
 					  ntohs(pkt->ack.maxSkew),
 					  ntohl(pkt->ack.serial),
 					  true, true,
 					  rxrpc_propose_ack_retry_tx);
-			break;
-		case RXRPC_PACKET_TYPE_ABORT:
-			break;
+		} else {
+			spin_lock_bh(&call->lock);
+			if (after(hard_ack, call->ackr_consumed))
+				call->ackr_consumed = hard_ack;
+			if (after(top, call->ackr_seen))
+				call->ackr_seen = top;
+			spin_unlock_bh(&call->lock);
 		}
 	}
 
@@ -267,8 +278,11 @@ int rxrpc_send_data_packet(struct rxrpc_call *call, struct sk_buff *skb)
 	msg.msg_controllen = 0;
 	msg.msg_flags = 0;
 
-	/* If our RTT cache needs working on, request an ACK. */
-	if ((call->peer->rtt_usage < 3 && sp->hdr.seq & 1) ||
+	/* If our RTT cache needs working on, request an ACK.  Also request
+	 * ACKs if a DATA packet appears to have been lost.
+	 */
+	if (call->cong_mode == RXRPC_CALL_FAST_RETRANSMIT ||
+	    (call->peer->rtt_usage < 3 && sp->hdr.seq & 1) ||
 	    ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000),
 			 ktime_get_real()))
 		whdr.flags |= RXRPC_REQUEST_ACK;

net/rxrpc/recvmsg.c

@@ -133,7 +133,7 @@ static int rxrpc_recvmsg_new_call(struct rxrpc_sock *rx,
 /*
  * End the packet reception phase.
  */
-static void rxrpc_end_rx_phase(struct rxrpc_call *call)
+static void rxrpc_end_rx_phase(struct rxrpc_call *call, rxrpc_serial_t serial)
 {
 	_enter("%d,%s", call->debug_id, rxrpc_call_states[call->state]);
 
@@ -141,7 +141,7 @@ static void rxrpc_end_rx_phase(struct rxrpc_call *call)
 	ASSERTCMP(call->rx_hard_ack, ==, call->rx_top);
 
 	if (call->state == RXRPC_CALL_CLIENT_RECV_REPLY) {
-		rxrpc_propose_ACK(call, RXRPC_ACK_IDLE, 0, 0, true, false,
+		rxrpc_propose_ACK(call, RXRPC_ACK_IDLE, 0, serial, true, false,
 				  rxrpc_propose_ack_terminal_ack);
 		rxrpc_send_call_packet(call, RXRPC_PACKET_TYPE_ACK);
 	}
@@ -201,8 +201,19 @@ static void rxrpc_rotate_rx_window(struct rxrpc_call *call)
 
 	_debug("%u,%u,%02x", hard_ack, top, flags);
 	trace_rxrpc_receive(call, rxrpc_receive_rotate, serial, hard_ack);
-	if (flags & RXRPC_LAST_PACKET)
-		rxrpc_end_rx_phase(call);
+	if (flags & RXRPC_LAST_PACKET) {
+		rxrpc_end_rx_phase(call, serial);
+	} else {
+		/* Check to see if there's an ACK that needs sending. */
+		if (after_eq(hard_ack, call->ackr_consumed + 2) ||
+		    after_eq(top, call->ackr_seen + 2) ||
+		    (hard_ack == top && after(hard_ack, call->ackr_consumed)))
+			rxrpc_propose_ACK(call, RXRPC_ACK_DELAY, 0, serial,
+					  true, false,
+					  rxrpc_propose_ack_rotate_rx);
+		if (call->ackr_reason)
+			rxrpc_send_call_packet(call, RXRPC_PACKET_TYPE_ACK);
+	}
 }
 
 /*

net/rxrpc/sendmsg.c

@@ -45,7 +45,9 @@ static int rxrpc_wait_for_tx_window(struct rxrpc_sock *rx,
 	for (;;) {
 		set_current_state(TASK_INTERRUPTIBLE);
 		ret = 0;
-		if (call->tx_top - call->tx_hard_ack < call->tx_winsize)
+		if (call->tx_top - call->tx_hard_ack <
+		    min_t(unsigned int, call->tx_winsize,
+			  call->cong_cwnd + call->cong_extra))
 			break;
 		if (call->state >= RXRPC_CALL_COMPLETE) {
 			ret = -call->error;
@@ -203,7 +205,8 @@ static int rxrpc_send_data(struct rxrpc_sock *rx,
 			_debug("alloc");
 
 			if (call->tx_top - call->tx_hard_ack >=
-			    call->tx_winsize) {
+			    min_t(unsigned int, call->tx_winsize,
+				  call->cong_cwnd + call->cong_extra)) {
 				ret = -EAGAIN;
 				if (msg->msg_flags & MSG_DONTWAIT)
 					goto maybe_error;