Add support for more fields in netstack for TCP_INFO

This CL adds support for the following fields:
- RTT, RTTVar, RTO
- send congestion window (sndCwnd) and send slow start threshold (sndSsthresh)
- congestion control state(CaState)
- ReorderSeen

PiperOrigin-RevId: 354195361

11 files changed, 265 insertions, 41 deletions

diff --git a/pkg/abi/linux/socket.go b/pkg/abi/linux/socket.go
index 8591acbf2..cb33c37bd 100644
--- a/pkg/abi/linux/socket.go
+++ b/pkg/abi/linux/socket.go
@@ -416,6 +416,18 @@ type TCPInfo struct {
 	RwndLimited uint64
 	// SndBufLimited is the time in microseconds limited by send buffer.
 	SndBufLimited uint64
+
+	Delievered   uint32
+	DelieveredCe uint32
+
+	// BytesSent is RFC4898 tcpEStatsPerfHCDataOctetsOut.
+	BytesSent uint64
+	// BytesRetrans is RFC4898 tcpEStatsPerfOctetsRetrans.
+	BytesRetrans uint64
+	// DSACKDups is RFC4898 tcpEStatsStackDSACKDups.
+	DSACKDups uint32
+	// ReordSeen is the number of reordering events seen.
+	ReordSeen uint32
 }
 
 // SizeOfTCPInfo is the binary size of a TCPInfo struct.
diff --git a/pkg/abi/linux/tcp.go b/pkg/abi/linux/tcp.go
index 2a8d4708b..1a3c0916f 100644
--- a/pkg/abi/linux/tcp.go
+++ b/pkg/abi/linux/tcp.go
@@ -59,3 +59,12 @@ const (
 	MAX_TCP_KEEPINTVL = 32767
 	MAX_TCP_KEEPCNT   = 127
 )
+
+// Congestion control states from include/uapi/linux/tcp.h.
+const (
+	TCP_CA_Open     = 0
+	TCP_CA_Disorder = 1
+	TCP_CA_CWR      = 2
+	TCP_CA_Recovery = 3
+	TCP_CA_Loss     = 4
+)
diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go
index 7065a0e46..3115a227d 100644
--- a/pkg/sentry/socket/netstack/netstack.go
+++ b/pkg/sentry/socket/netstack/netstack.go
@@ -1098,6 +1098,29 @@ func getSockOptTCP(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name,
 		// TODO(b/64800844): Translate fields once they are added to
 		// tcpip.TCPInfoOption.
 		info := linux.TCPInfo{}
+		switch v.CcState {
+		case tcpip.RTORecovery:
+			info.CaState = linux.TCP_CA_Loss
+		case tcpip.FastRecovery, tcpip.SACKRecovery:
+			info.CaState = linux.TCP_CA_Recovery
+		case tcpip.Disorder:
+			info.CaState = linux.TCP_CA_Disorder
+		case tcpip.Open:
+			info.CaState = linux.TCP_CA_Open
+		}
+		info.RTO = uint32(v.RTO / time.Microsecond)
+		info.RTT = uint32(v.RTT / time.Microsecond)
+		info.RTTVar = uint32(v.RTTVar / time.Microsecond)
+		info.SndSsthresh = v.SndSsthresh
+		info.SndCwnd = v.SndCwnd
+
+		// In netstack reorderSeen is updated only when RACK is enabled.
+		// We only track whether the reordering is seen, which is
+		// different than Linux where reorderSeen is not specific to
+		// RACK and is incremented when a reordering event is seen.
+		if v.ReorderSeen {
+			info.ReordSeen = 1
+		}
 
 		// Linux truncates the output binary to outLen.
 		buf := t.CopyScratchBuffer(info.SizeBytes())
diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go
index 812ee36ed..e70ae69ef 100644
--- a/pkg/tcpip/tcpip.go
+++ b/pkg/tcpip/tcpip.go
@@ -993,12 +993,54 @@ type SettableSocketOption interface {
 	isSettableSocketOption()
 }
 
+// CongestionControlState indicates the current congestion control state for
+// TCP sender.
+type CongestionControlState int
+
+const (
+	// Open indicates that the sender is receiving acks in order and
+	// no loss or dupACK's etc have been detected.
+	Open CongestionControlState = iota
+	// RTORecovery indicates that an RTO has occurred and the sender
+	// has entered an RTO based recovery phase.
+	RTORecovery
+	// FastRecovery indicates that the sender has entered FastRecovery
+	// based on receiving nDupAck's. This state is entered only when
+	// SACK is not in use.
+	FastRecovery
+	// SACKRecovery indicates that the sender has entered SACK based
+	// recovery.
+	SACKRecovery
+	// Disorder indicates the sender either received some SACK blocks
+	// or dupACK's.
+	Disorder
+)
+
 // TCPInfoOption is used by GetSockOpt to expose TCP statistics.
 //
 // TODO(b/64800844): Add and populate stat fields.
 type TCPInfoOption struct {
-	RTT    time.Duration
+	// RTT is the smoothed round trip time.
+	RTT time.Duration
+
+	// RTTVar is the round trip time variation.
 	RTTVar time.Duration
+
+	// RTO is the retransmission timeout for the endpoint.
+	RTO time.Duration
+
+	// CcState is the congestion control state.
+	CcState CongestionControlState
+
+	// SndCwnd is the congestion window, in packets.
+	SndCwnd uint32
+
+	// SndSsthresh is the threshold between slow start and congestion
+	// avoidance.
+	SndSsthresh uint32
+
+	// ReorderSeen indicates if reordering is seen in the endpoint.
+	ReorderSeen bool
 }
 
 func (*TCPInfoOption) isGettableSocketOption() {}
diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go
index b6bd6d455..bfa5b01fb 100644
--- a/pkg/tcpip/transport/tcp/endpoint.go
+++ b/pkg/tcpip/transport/tcp/endpoint.go
@@ -2011,20 +2011,34 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) {
 	}
 }
 
+func (e *endpoint) getTCPInfo() tcpip.TCPInfoOption {
+	info := tcpip.TCPInfoOption{}
+	e.LockUser()
+	snd := e.snd
+	if snd != nil {
+		// We do not calculate RTT before sending the data packets. If
+		// the connection did not send and receive data, then RTT will
+		// be zero.
+		snd.rtt.Lock()
+		info.RTT = snd.rtt.srtt
+		info.RTTVar = snd.rtt.rttvar
+		snd.rtt.Unlock()
+
+		info.RTO = snd.rto
+		info.CcState = snd.state
+		info.SndSsthresh = uint32(snd.sndSsthresh)
+		info.SndCwnd = uint32(snd.sndCwnd)
+		info.ReorderSeen = snd.rc.reorderSeen
+	}
+	e.UnlockUser()
+	return info
+}
+
 // GetSockOpt implements tcpip.Endpoint.GetSockOpt.
 func (e *endpoint) GetSockOpt(opt tcpip.GettableSocketOption) *tcpip.Error {
 	switch o := opt.(type) {
 	case *tcpip.TCPInfoOption:
-		*o = tcpip.TCPInfoOption{}
-		e.LockUser()
-		snd := e.snd
-		e.UnlockUser()
-		if snd != nil {
-			snd.rtt.Lock()
-			o.RTT = snd.rtt.srtt
-			o.RTTVar = snd.rtt.rttvar
-			snd.rtt.Unlock()
-		}
+		*o = e.getTCPInfo()
 
 	case *tcpip.KeepaliveIdleOption:
 		e.keepalive.Lock()
diff --git a/pkg/tcpip/transport/tcp/rack.go b/pkg/tcpip/transport/tcp/rack.go
index 307bacca5..fdb7e3dc6 100644
--- a/pkg/tcpip/transport/tcp/rack.go
+++ b/pkg/tcpip/transport/tcp/rack.go
@@ -162,7 +162,7 @@ func (s *sender) shouldSchedulePTO() bool {
 		// The connection supports SACK.
 		s.ep.sackPermitted &&
 		// The connection is not in loss recovery.
-		(s.state != RTORecovery && s.state != SACKRecovery) &&
+		(s.state != tcpip.RTORecovery && s.state != tcpip.SACKRecovery) &&
 		// The connection has no SACKed sequences in the SACK scoreboard.
 		s.ep.scoreboard.Sacked() == 0
 }
diff --git a/pkg/tcpip/transport/tcp/snd.go b/pkg/tcpip/transport/tcp/snd.go
index 079d90848..28ef9f899 100644
--- a/pkg/tcpip/transport/tcp/snd.go
+++ b/pkg/tcpip/transport/tcp/snd.go
@@ -48,28 +48,6 @@ const (
 	MaxRetries = 15
 )
 
-// ccState indicates the current congestion control state for this sender.
-type ccState int
-
-const (
-	// Open indicates that the sender is receiving acks in order and
-	// no loss or dupACK's etc have been detected.
-	Open ccState = iota
-	// RTORecovery indicates that an RTO has occurred and the sender
-	// has entered an RTO based recovery phase.
-	RTORecovery
-	// FastRecovery indicates that the sender has entered FastRecovery
-	// based on receiving nDupAck's. This state is entered only when
-	// SACK is not in use.
-	FastRecovery
-	// SACKRecovery indicates that the sender has entered SACK based
-	// recovery.
-	SACKRecovery
-	// Disorder indicates the sender either received some SACK blocks
-	// or dupACK's.
-	Disorder
-)
-
 // congestionControl is an interface that must be implemented by any supported
 // congestion control algorithm.
 type congestionControl interface {
@@ -204,7 +182,7 @@ type sender struct {
 	maxSentAck seqnum.Value
 
 	// state is the current state of congestion control for this endpoint.
-	state ccState
+	state tcpip.CongestionControlState
 
 	// cc is the congestion control algorithm in use for this sender.
 	cc congestionControl
@@ -593,7 +571,7 @@ func (s *sender) retransmitTimerExpired() bool {
 		s.leaveRecovery()
 	}
 
-	s.state = RTORecovery
+	s.state = tcpip.RTORecovery
 	s.cc.HandleRTOExpired()
 
 	// Mark the next segment to be sent as the first unacknowledged one and
@@ -1018,7 +996,7 @@ func (s *sender) sendData() {
 	// "A TCP SHOULD set cwnd to no more than RW before beginning
 	// transmission if the TCP has not sent data in the interval exceeding
 	// the retrasmission timeout."
-	if !s.fr.active && s.state != RTORecovery && time.Now().Sub(s.lastSendTime) > s.rto {
+	if !s.fr.active && s.state != tcpip.RTORecovery && time.Now().Sub(s.lastSendTime) > s.rto {
 		if s.sndCwnd > InitialCwnd {
 			s.sndCwnd = InitialCwnd
 		}
@@ -1062,14 +1040,14 @@ func (s *sender) enterRecovery() {
 	s.fr.highRxt = s.sndUna
 	s.fr.rescueRxt = s.sndUna
 	if s.ep.sackPermitted {
-		s.state = SACKRecovery
+		s.state = tcpip.SACKRecovery
 		s.ep.stack.Stats().TCP.SACKRecovery.Increment()
 		// Set TLPRxtOut to false according to
 		// https://tools.ietf.org/html/draft-ietf-tcpm-rack-08#section-7.6.1.
 		s.rc.tlpRxtOut = false
 		return
 	}
-	s.state = FastRecovery
+	s.state = tcpip.FastRecovery
 	s.ep.stack.Stats().TCP.FastRecovery.Increment()
 }
 
@@ -1166,7 +1144,7 @@ func (s *sender) detectLoss(seg *segment) (fastRetransmit bool) {
 		s.fr.highRxt = s.sndUna - 1
 		// Do run SetPipe() to calculate the outstanding segments.
 		s.SetPipe()
-		s.state = Disorder
+		s.state = tcpip.Disorder
 		return false
 	}
 
@@ -1464,7 +1442,7 @@ func (s *sender) handleRcvdSegment(rcvdSeg *segment) {
 		if !s.fr.active {
 			s.cc.Update(originalOutstanding - s.outstanding)
 			if s.fr.last.LessThan(s.sndUna) {
-				s.state = Open
+				s.state = tcpip.Open
 			}
 		}
 
diff --git a/test/packetimpact/runner/defs.bzl b/test/packetimpact/runner/defs.bzl
index 5c3c569de..2b9bfac76 100644
--- a/test/packetimpact/runner/defs.bzl
+++ b/test/packetimpact/runner/defs.bzl
@@ -281,6 +281,9 @@ ALL_TESTS = [
         name = "tcp_rack",
         expect_netstack_failure = True,
     ),
+    PacketimpactTestInfo(
+        name = "tcp_info",
+    ),
 ]
 
 def validate_all_tests():
diff --git a/test/packetimpact/tests/BUILD b/test/packetimpact/tests/BUILD
index 6c6f2bdf7..42aad541f 100644
--- a/test/packetimpact/tests/BUILD
+++ b/test/packetimpact/tests/BUILD
@@ -390,6 +390,19 @@ packetimpact_testbench(
     ],
 )
 
+packetimpact_testbench(
+    name = "tcp_info",
+    srcs = ["tcp_info_test.go"],
+    deps = [
+        "//pkg/abi/linux",
+        "//pkg/binary",
+        "//pkg/tcpip/header",
+        "//pkg/usermem",
+        "//test/packetimpact/testbench",
+        "@org_golang_x_sys//unix:go_default_library",
+    ],
+)
+
 validate_all_tests()
 
 [packetimpact_go_test(
diff --git a/test/packetimpact/tests/tcp_info_test.go b/test/packetimpact/tests/tcp_info_test.go
new file mode 100644
index 000000000..b66e8f609
--- /dev/null
+++ b/test/packetimpact/tests/tcp_info_test.go
@@ -0,0 +1,103 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package tcp_info_test
+
+import (
+	"flag"
+	"testing"
+	"time"
+
+	"golang.org/x/sys/unix"
+	"gvisor.dev/gvisor/pkg/abi/linux"
+	"gvisor.dev/gvisor/pkg/binary"
+	"gvisor.dev/gvisor/pkg/tcpip/header"
+	"gvisor.dev/gvisor/pkg/usermem"
+	"gvisor.dev/gvisor/test/packetimpact/testbench"
+)
+
+func init() {
+	testbench.Initialize(flag.CommandLine)
+}
+
+func TestTCPInfo(t *testing.T) {
+	// Create a socket, listen, TCP connect, and accept.
+	dut := testbench.NewDUT(t)
+	listenFD, remotePort := dut.CreateListener(t, unix.SOCK_STREAM, unix.IPPROTO_TCP, 1)
+	defer dut.Close(t, listenFD)
+
+	conn := dut.Net.NewTCPIPv4(t, testbench.TCP{DstPort: &remotePort}, testbench.TCP{SrcPort: &remotePort})
+	defer conn.Close(t)
+	conn.Connect(t)
+
+	acceptFD, _ := dut.Accept(t, listenFD)
+	defer dut.Close(t, acceptFD)
+
+	// Send and receive sample data.
+	sampleData := []byte("Sample Data")
+	samplePayload := &testbench.Payload{Bytes: sampleData}
+	dut.Send(t, acceptFD, sampleData, 0)
+	if _, err := conn.ExpectData(t, &testbench.TCP{}, samplePayload, time.Second); err != nil {
+		t.Fatalf("expected a packet with payload %v: %s", samplePayload, err)
+	}
+	conn.Send(t, testbench.TCP{Flags: testbench.Uint8(header.TCPFlagAck)})
+
+	info := linux.TCPInfo{}
+	infoBytes := dut.GetSockOpt(t, acceptFD, unix.SOL_TCP, unix.TCP_INFO, int32(linux.SizeOfTCPInfo))
+	binary.Unmarshal(infoBytes, usermem.ByteOrder, &info)
+
+	rtt := time.Duration(info.RTT) * time.Microsecond
+	rttvar := time.Duration(info.RTTVar) * time.Microsecond
+	rto := time.Duration(info.RTO) * time.Microsecond
+	if rtt == 0 || rttvar == 0 || rto == 0 {
+		t.Errorf("expected rtt(%v), rttvar(%v) and rto(%v) to be greater than zero", rtt, rttvar, rto)
+	}
+	if info.ReordSeen != 0 {
+		t.Errorf("expected the connection to not have any reordering, got: %v want: 0", info.ReordSeen)
+	}
+	if info.SndCwnd == 0 {
+		t.Errorf("expected send congestion window to be greater than zero")
+	}
+	if info.CaState != linux.TCP_CA_Open {
+		t.Errorf("expected the connection to be in open state, got: %v want: %v", info.CaState, linux.TCP_CA_Open)
+	}
+
+	if t.Failed() {
+		t.FailNow()
+	}
+
+	// Check the congestion control state and send congestion window after
+	// retransmission timeout.
+	seq := testbench.Uint32(uint32(*conn.RemoteSeqNum(t)))
+	dut.Send(t, acceptFD, sampleData, 0)
+	if _, err := conn.ExpectData(t, &testbench.TCP{}, samplePayload, time.Second); err != nil {
+		t.Fatalf("expected a packet with payload %v: %s", samplePayload, err)
+	}
+
+	// Expect retransmission of the packet within 1.5*RTO.
+	timeout := time.Duration(float64(info.RTO)*1.5) * time.Microsecond
+	if _, err := conn.ExpectData(t, &testbench.TCP{SeqNum: seq}, samplePayload, timeout); err != nil {
+		t.Fatalf("expected a packet with payload %v: %s", samplePayload, err)
+	}
+
+	info = linux.TCPInfo{}
+	infoBytes = dut.GetSockOpt(t, acceptFD, unix.SOL_TCP, unix.TCP_INFO, int32(linux.SizeOfTCPInfo))
+	binary.Unmarshal(infoBytes, usermem.ByteOrder, &info)
+	if info.CaState != linux.TCP_CA_Loss {
+		t.Errorf("expected the connection to be in loss recovery, got: %v want: %v", info.CaState, linux.TCP_CA_Loss)
+	}
+	if info.SndCwnd != 1 {
+		t.Errorf("expected send congestion window to be 1, got: %v %v", info.SndCwnd)
+	}
+}
diff --git a/test/syscalls/linux/socket_ip_tcp_generic.cc b/test/syscalls/linux/socket_ip_tcp_generic.cc
index 831d96262..a73987a7e 100644
--- a/test/syscalls/linux/socket_ip_tcp_generic.cc
+++ b/test/syscalls/linux/socket_ip_tcp_generic.cc
@@ -65,6 +65,33 @@ TEST_P(TCPSocketPairTest, ZeroTcpInfoSucceeds) {
               SyscallSucceeds());
 }
 
+TEST_P(TCPSocketPairTest, CheckTcpInfoFields) {
+  auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair());
+
+  char buf[10] = {};
+  ASSERT_THAT(RetryEINTR(send)(sockets->first_fd(), buf, sizeof(buf), 0),
+              SyscallSucceedsWithValue(sizeof(buf)));
+
+  // Wait until second_fd sees the data and then recv it.
+  struct pollfd poll_fd = {sockets->second_fd(), POLLIN, 0};
+  constexpr int kPollTimeoutMs = 2000;  // Wait up to 2 seconds for the data.
+  ASSERT_THAT(RetryEINTR(poll)(&poll_fd, 1, kPollTimeoutMs),
+              SyscallSucceedsWithValue(1));
+
+  ASSERT_THAT(RetryEINTR(recv)(sockets->second_fd(), buf, sizeof(buf), 0),
+              SyscallSucceedsWithValue(sizeof(buf)));
+
+  struct tcp_info opt = {};
+  socklen_t optLen = sizeof(opt);
+  ASSERT_THAT(getsockopt(sockets->first_fd(), SOL_TCP, TCP_INFO, &opt, &optLen),
+              SyscallSucceeds());
+
+  // Validates the received tcp_info fields.
+  EXPECT_EQ(opt.tcpi_ca_state, 0);
+  EXPECT_GT(opt.tcpi_snd_cwnd, 0);
+  EXPECT_GT(opt.tcpi_rto, 0);
+}
+
 // This test validates that an RST is sent instead of a FIN when data is
 // unread on calls to close(2).
 TEST_P(TCPSocketPairTest, RSTSentOnCloseWithUnreadData) {