Add new retransmissions and recovery related metrics.

PiperOrigin-RevId: 236945145
Change-Id: I051760d95154ea5574c8bb6aea526f488af5e07b

6 files changed, 87 insertions, 1 deletions

diff --git a/pkg/sentry/socket/epsocket/epsocket.go b/pkg/sentry/socket/epsocket/epsocket.go
index 8fa108bea..4e547ea33 100644
--- a/pkg/sentry/socket/epsocket/epsocket.go
+++ b/pkg/sentry/socket/epsocket/epsocket.go
@@ -82,6 +82,12 @@ var Metrics = tcpip.Stats{
 		SegmentsSent:              mustCreateMetric("/netstack/tcp/segments_sent", "Number of TCP segments sent."),
 		ResetsSent:                mustCreateMetric("/netstack/tcp/resets_sent", "Number of TCP resets sent."),
 		ResetsReceived:            mustCreateMetric("/netstack/tcp/resets_received", "Number of TCP resets received."),
+		Retransmits:               mustCreateMetric("/netstack/tcp/retransmits", "Number of TCP segments retransmitted."),
+		FastRecovery:              mustCreateMetric("/netstack/tcp/fast_recovery", "Number of times fast recovery was used to recover from packet loss."),
+		SACKRecovery:              mustCreateMetric("/netstack/tcp/sack_recovery", "Number of times SACK recovery was used to recover from packet loss."),
+		SlowStartRetransmits:      mustCreateMetric("/netstack/tcp/slow_start_retransmits", "Number of segments retransmitted in slow start mode."),
+		FastRetransmit:            mustCreateMetric("/netstack/tcp/fast_retransmit", "Number of TCP segments which were fast retransmitted."),
+		Timeouts:                  mustCreateMetric("/netstack/tcp/timeouts", "Number of times RTO expired."),
 	},
 	UDP: tcpip.UDPStats{
 		PacketsReceived:          mustCreateMetric("/netstack/udp/packets_received", "Number of UDP datagrams received via HandlePacket."),
diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go
index 49cc8705a..7010d1b68 100644
--- a/pkg/tcpip/tcpip.go
+++ b/pkg/tcpip/tcpip.go
@@ -628,6 +628,28 @@ type TCPStats struct {
 
 	// ResetsReceived is the number of TCP resets received.
 	ResetsReceived *StatCounter
+
+	// Retransmits is the number of TCP segments retransmitted.
+	Retransmits *StatCounter
+
+	// FastRecovery is the number of times Fast Recovery was used to
+	// recover from packet loss.
+	FastRecovery *StatCounter
+
+	// SACKRecovery is the number of times SACK Recovery was used to
+	// recover from packet loss.
+	SACKRecovery *StatCounter
+
+	// SlowStartRetransmits is the number of segments retransmitted in slow
+	// start.
+	SlowStartRetransmits *StatCounter
+
+	// FastRetransmit is the number of segments retransmitted in fast
+	// recovery.
+	FastRetransmit *StatCounter
+
+	// Timeouts is the number of times the RTO expired.
+	Timeouts *StatCounter
 }
 
 // UDPStats collects UDP-specific stats.
diff --git a/pkg/tcpip/transport/tcp/segment.go b/pkg/tcpip/transport/tcp/segment.go
index bd8017f64..a4c4a115c 100644
--- a/pkg/tcpip/transport/tcp/segment.go
+++ b/pkg/tcpip/transport/tcp/segment.go
@@ -61,6 +61,9 @@ type segment struct {
 	options        []byte `state:".([]byte)"`
 	hasNewSACKInfo bool
 	rcvdTime       time.Time `state:".(unixTime)"`
+	// xmitTime is the last transmit time of this segment. A zero value
+	// indicates that the segment has yet to be transmitted.
+	xmitTime time.Time `state:".(unixTime)"`
 }
 
 func newSegment(r *stack.Route, id stack.TransportEndpointID, vv buffer.VectorisedView) *segment {
diff --git a/pkg/tcpip/transport/tcp/segment_state.go b/pkg/tcpip/transport/tcp/segment_state.go
index 7b98a3ec8..68b049f06 100644
--- a/pkg/tcpip/transport/tcp/segment_state.go
+++ b/pkg/tcpip/transport/tcp/segment_state.go
@@ -70,3 +70,13 @@ func (s *segment) saveRcvdTime() unixTime {
 func (s *segment) loadRcvdTime(unix unixTime) {
 	s.rcvdTime = time.Unix(unix.second, unix.nano)
 }
+
+// saveXmitTime is invoked by stateify.
+func (s *segment) saveXmitTime() unixTime {
+	return unixTime{s.rcvdTime.Unix(), s.rcvdTime.UnixNano()}
+}
+
+// loadXmitTime is invoked by stateify.
+func (s *segment) loadXmitTime(unix unixTime) {
+	s.rcvdTime = time.Unix(unix.second, unix.nano)
+}
diff --git a/pkg/tcpip/transport/tcp/snd.go b/pkg/tcpip/transport/tcp/snd.go
index 8312ae077..2cf12f4a6 100644
--- a/pkg/tcpip/transport/tcp/snd.go
+++ b/pkg/tcpip/transport/tcp/snd.go
@@ -338,6 +338,8 @@ func (s *sender) resendSegment() {
 	// Resend the segment.
 	if seg := s.writeList.Front(); seg != nil {
 		s.sendSegment(seg.data, seg.flags, seg.sequenceNumber)
+		s.ep.stack.Stats().TCP.FastRetransmit.Increment()
+		s.ep.stack.Stats().TCP.Retransmits.Increment()
 	}
 }
 
@@ -352,6 +354,8 @@ func (s *sender) retransmitTimerExpired() bool {
 		return true
 	}
 
+	s.ep.stack.Stats().TCP.Timeouts.Increment()
+
 	// Give up if we've waited more than a minute since the last resend.
 	if s.rto >= 60*time.Second {
 		return false
@@ -422,7 +426,6 @@ func (s *sender) sendData() {
 	end := s.sndUna.Add(s.sndWnd)
 	var dataSent bool
 	for ; seg != nil && s.outstanding < s.sndCwnd; seg = seg.Next() {
-
 		// We abuse the flags field to determine if we have already
 		// assigned a sequence number to this segment.
 		if seg.flags == 0 {
@@ -524,6 +527,15 @@ func (s *sender) sendData() {
 			// ensure that no keepalives are sent while there is pending data.
 			s.ep.disableKeepaliveTimer()
 		}
+
+		if !seg.xmitTime.IsZero() {
+			s.ep.stack.Stats().TCP.Retransmits.Increment()
+			if s.sndCwnd < s.sndSsthresh {
+				s.ep.stack.Stats().TCP.SlowStartRetransmits.Increment()
+			}
+		}
+
+		seg.xmitTime = time.Now()
 		s.sendSegment(seg.data, seg.flags, seg.sequenceNumber)
 
 		// Update sndNxt if we actually sent new data (as opposed to
@@ -556,6 +568,7 @@ func (s *sender) enterFastRecovery() {
 	s.fr.first = s.sndUna
 	s.fr.last = s.sndNxt - 1
 	s.fr.maxCwnd = s.sndCwnd + s.outstanding
+	s.ep.stack.Stats().TCP.FastRecovery.Increment()
 }
 
 func (s *sender) leaveFastRecovery() {
diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go
index 2011189b7..7f2615ca9 100644
--- a/pkg/tcpip/transport/tcp/tcp_test.go
+++ b/pkg/tcpip/transport/tcp/tcp_test.go
@@ -2669,6 +2669,18 @@ func TestFastRecovery(t *testing.T) {
 	// Receive the retransmitted packet.
 	c.ReceiveAndCheckPacket(data, rtxOffset, maxPayload)
 
+	if got, want := c.Stack().Stats().TCP.FastRetransmit.Value(), uint64(1); got != want {
+		t.Errorf("got stats.TCP.FastRetransmit.Value = %v, want = %v", got, want)
+	}
+
+	if got, want := c.Stack().Stats().TCP.Retransmits.Value(), uint64(1); got != want {
+		t.Errorf("got stats.TCP.Retransmit.Value = %v, want = %v", got, want)
+	}
+
+	if got, want := c.Stack().Stats().TCP.FastRecovery.Value(), uint64(1); got != want {
+		t.Errorf("got stats.TCP.FastRecovery.Value = %v, want = %v", got, want)
+	}
+
 	// Now send 7 mode duplicate acks. Each of these should cause a window
 	// inflation by 1 and cause the sender to send an extra packet.
 	for i := 0; i < 7; i++ {
@@ -2688,6 +2700,14 @@ func TestFastRecovery(t *testing.T) {
 	// Receive the retransmit due to partial ack.
 	c.ReceiveAndCheckPacket(data, rtxOffset, maxPayload)
 
+	if got, want := c.Stack().Stats().TCP.FastRetransmit.Value(), uint64(2); got != want {
+		t.Errorf("got stats.TCP.FastRetransmit.Value = %v, want = %v", got, want)
+	}
+
+	if got, want := c.Stack().Stats().TCP.Retransmits.Value(), uint64(2); got != want {
+		t.Errorf("got stats.TCP.Retransmit.Value = %v, want = %v", got, want)
+	}
+
 	// Receive the 10 extra packets that should have been released due to
 	// the congestion window inflation in recovery.
 	for i := 0; i < 10; i++ {
@@ -2799,6 +2819,18 @@ func TestRetransmit(t *testing.T) {
 	rtxOffset := bytesRead - maxPayload*expected
 	c.ReceiveAndCheckPacket(data, rtxOffset, maxPayload)
 
+	if got, want := c.Stack().Stats().TCP.Timeouts.Value(), uint64(1); got != want {
+		t.Errorf("got stats.TCP.Timeouts.Value = %v, want = %v", got, want)
+	}
+
+	if got, want := c.Stack().Stats().TCP.Retransmits.Value(), uint64(1); got != want {
+		t.Errorf("got stats.TCP.Retransmit.Value = %v, want = %v", got, want)
+	}
+
+	if got, want := c.Stack().Stats().TCP.SlowStartRetransmits.Value(), uint64(1); got != want {
+		t.Errorf("got stats.TCP.SlowStartRetransmits.Value = %v, want = %v", got, want)
+	}
+
 	// Acknowledge half of the pending data.
 	rtxOffset = bytesRead - expected*maxPayload/2
 	c.SendAck(790, rtxOffset)