Implement getsockopt(TCP_INFO).

Export some readily-available fields for TCP_INFO and stub out the rest.

PiperOrigin-RevId: 261191548

8 files changed, 206 insertions, 78 deletions

diff --git a/pkg/sentry/socket/epsocket/epsocket.go b/pkg/sentry/socket/epsocket/epsocket.go
index 0f483faa8..586523d3d 100644
--- a/pkg/sentry/socket/epsocket/epsocket.go
+++ b/pkg/sentry/socket/epsocket/epsocket.go
@@ -845,6 +845,68 @@ func getSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, family
 	return nil, syserr.ErrProtocolNotAvailable
 }
 
+func toLinuxTCPInfo(i tcp.InfoOption) linux.TCPInfo {
+	// Unimplemented fields are explicitly initialized to zero below.
+	return linux.TCPInfo{
+		State:                  uint8(translateTCPState(tcp.EndpointState(i.ProtocolState))),
+		CaState:                0,
+		Retransmits:            0,
+		Probes:                 0,
+		Backoff:                0,
+		Options:                0,
+		WindowScale:            uint8((i.Sender.SndWndScale&0xf)<<4 | (i.Receiver.RcvWndScale & 0xf)),
+		DeliveryRateAppLimited: 0,
+
+		RTO:    uint32(i.Sender.RTO / time.Microsecond),
+		ATO:    0,
+		SndMss: uint32(i.Sender.MSS),
+		RcvMss: uint32(i.RcvMSS),
+
+		Unacked: uint32(i.Sender.Outstanding),
+		Sacked:  uint32(i.SACK.Sacked),
+		Lost:    0,
+		Retrans: 0,
+		Fackets: 0,
+
+		LastDataSent: uint32(i.Sender.LastSendTime.UnixNano() / int64(time.Millisecond)),
+		LastAckSent:  0, // Not tracked by Linux.
+		LastDataRecv: uint32(i.RcvLastDataNanos / int64(time.Millisecond)),
+		LastAckRecv:  uint32(i.RcvLastAckNanos / int64(time.Millisecond)),
+
+		PMTU:        uint32(i.SndMTU),
+		RcvSsthresh: 0,
+		RTT:         uint32(i.Sender.SRTT / time.Microsecond),
+		RTTVar:      uint32(i.Sender.RTTVar / time.Microsecond),
+		SndSsthresh: uint32(i.Sender.Ssthresh),
+		SndCwnd:     uint32(i.Sender.SndCwnd),
+		Advmss:      uint32(i.AMSS),
+		Reordering:  0,
+
+		RcvRTT:   uint32(i.RcvAutoParams.RTT / time.Microsecond),
+		RcvSpace: uint32(i.RcvBufSize),
+
+		TotalRetrans: 0,
+
+		PacingRate:    0,
+		MaxPacingRate: 0,
+		BytesAcked:    0,
+		BytesReceived: 0,
+		SegsOut:       0,
+		SegsIn:        0,
+
+		NotSentBytes: 0,
+		MinRTT:       uint32(i.RcvAutoParams.RTT / time.Microsecond),
+		DataSegsIn:   0,
+		DataSegsOut:  0,
+
+		DeliveryRate: 0,
+
+		BusyTime:      0,
+		RwndLimited:   0,
+		SndBufLimited: 0,
+	}
+}
+
 // getSockOptTCP implements GetSockOpt when level is SOL_TCP.
 func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interface{}, *syserr.Error) {
 	switch name {
@@ -924,17 +986,14 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa
 		return int32(time.Duration(v) / time.Second), nil
 
 	case linux.TCP_INFO:
-		var v tcpip.TCPInfoOption
+		var v tcp.InfoOption
 		if err := ep.GetSockOpt(&v); err != nil {
 			return nil, syserr.TranslateNetstackError(err)
 		}
-
-		// TODO(b/64800844): Translate fields once they are added to
-		// tcpip.TCPInfoOption.
-		info := linux.TCPInfo{}
+		info := toLinuxTCPInfo(v)
 
 		// Linux truncates the output binary to outLen.
-		ib := binary.Marshal(nil, usermem.ByteOrder, &info)
+		ib := binary.Marshal(nil, usermem.ByteOrder, info)
 		if len(ib) > outLen {
 			ib = ib[:outLen]
 		}
@@ -2375,6 +2434,38 @@ func nicStateFlagsToLinux(f stack.NICStateFlags) uint32 {
 	return rv
 }
 
+// translateTCPState translates an internal endpoint state to the equivalent
+// state in the Linux ABI.
+func translateTCPState(s tcp.EndpointState) uint32 {
+	switch s {
+	case tcp.StateEstablished:
+		return linux.TCP_ESTABLISHED
+	case tcp.StateSynSent:
+		return linux.TCP_SYN_SENT
+	case tcp.StateSynRecv:
+		return linux.TCP_SYN_RECV
+	case tcp.StateFinWait1:
+		return linux.TCP_FIN_WAIT1
+	case tcp.StateFinWait2:
+		return linux.TCP_FIN_WAIT2
+	case tcp.StateTimeWait:
+		return linux.TCP_TIME_WAIT
+	case tcp.StateClose, tcp.StateInitial, tcp.StateBound, tcp.StateConnecting, tcp.StateError:
+		return linux.TCP_CLOSE
+	case tcp.StateCloseWait:
+		return linux.TCP_CLOSE_WAIT
+	case tcp.StateLastAck:
+		return linux.TCP_LAST_ACK
+	case tcp.StateListen:
+		return linux.TCP_LISTEN
+	case tcp.StateClosing:
+		return linux.TCP_CLOSING
+	default:
+		// Internal or unknown state.
+		return 0
+	}
+}
+
 // State implements socket.Socket.State. State translates the internal state
 // returned by netstack to values defined by Linux.
 func (s *SocketOperations) State() uint32 {
@@ -2385,33 +2476,7 @@ func (s *SocketOperations) State() uint32 {
 
 	if !s.isPacketBased() {
 		// TCP socket.
-		switch tcp.EndpointState(s.Endpoint.State()) {
-		case tcp.StateEstablished:
-			return linux.TCP_ESTABLISHED
-		case tcp.StateSynSent:
-			return linux.TCP_SYN_SENT
-		case tcp.StateSynRecv:
-			return linux.TCP_SYN_RECV
-		case tcp.StateFinWait1:
-			return linux.TCP_FIN_WAIT1
-		case tcp.StateFinWait2:
-			return linux.TCP_FIN_WAIT2
-		case tcp.StateTimeWait:
-			return linux.TCP_TIME_WAIT
-		case tcp.StateClose, tcp.StateInitial, tcp.StateBound, tcp.StateConnecting, tcp.StateError:
-			return linux.TCP_CLOSE
-		case tcp.StateCloseWait:
-			return linux.TCP_CLOSE_WAIT
-		case tcp.StateLastAck:
-			return linux.TCP_LAST_ACK
-		case tcp.StateListen:
-			return linux.TCP_LISTEN
-		case tcp.StateClosing:
-			return linux.TCP_CLOSING
-		default:
-			// Internal or unknown state.
-			return 0
-		}
+		return translateTCPState(tcp.EndpointState(s.Endpoint.State()))
 	}
 
 	// TODO(b/112063468): Export states for UDP, ICMP, and raw sockets.
diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go
index 6156c3f46..7c31cf493 100644
--- a/pkg/tcpip/stack/stack.go
+++ b/pkg/tcpip/stack/stack.go
@@ -208,6 +208,10 @@ type TCPSenderState struct {
 
 	// Cubic holds the state related to CUBIC congestion control.
 	Cubic TCPCubicState
+
+	// MSS is the size of the largest segment that can be sent without
+	// fragmentation.
+	MSS int
 }
 
 // TCPSACKInfo holds TCP SACK related information for a given TCP endpoint.
@@ -220,6 +224,9 @@ type TCPSACKInfo struct {
 	// from the peer endpoint.
 	ReceivedBlocks []header.SACKBlock
 
+	// Sacked is the current number of bytes held in the SACK scoreboard.
+	Sacked seqnum.Size
+
 	// MaxSACKED is the highest sequence number that has been SACKED
 	// by the peer.
 	MaxSACKED seqnum.Value
@@ -269,6 +276,14 @@ type TCPEndpointState struct {
 	// ID is a copy of the TransportEndpointID for the endpoint.
 	ID TCPEndpointID
 
+	// ProtocolState denotes the TCP state the endpoint is currently
+	// in, encoded in a netstack-specific manner. Should be translated
+	// to the Linux ABI before exposing to userspace.
+	ProtocolState uint32
+
+	// AMSS is the MSS advertised to the peer by this endpoint.
+	AMSS uint16
+
 	// SegTime denotes the absolute time when this segment was received.
 	SegTime time.Time
 
@@ -286,6 +301,18 @@ type TCPEndpointState struct {
 	// RcvClosed if true, indicates the endpoint has been closed for reading.
 	RcvClosed bool
 
+	// RcvLastAck is the time of receipt of the last packet with the
+	// ACK flag set.
+	RcvLastAckNanos int64
+
+	// RcvLastData is the time of reciept of the last packet
+	// containing data.
+	RcvLastDataNanos int64
+
+	// RcvMSS is the size of the largest segment the receiver is willing to
+	// accept, not including TCP headers and options.
+	RcvMSS int
+
 	// SendTSOk is used to indicate when the TS Option has been negotiated.
 	// When sendTSOk is true every non-RST segment should carry a TS as per
 	// RFC7323#section-1.1.
@@ -326,6 +353,9 @@ type TCPEndpointState struct {
 	// SndMTU is the smallest MTU seen in the control packets received.
 	SndMTU int
 
+	// MaxOptionSize is the maximum size of TCP options.
+	MaxOptionSize int
+
 	// Receiver holds variables related to the TCP receiver for the endpoint.
 	Receiver TCPReceiverState
 
diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go
index 4208c0303..690c00edb 100644
--- a/pkg/tcpip/tcpip.go
+++ b/pkg/tcpip/tcpip.go
@@ -476,14 +476,6 @@ type QuickAckOption int
 // Only supported on Unix sockets.
 type PasscredOption int
 
-// TCPInfoOption is used by GetSockOpt to expose TCP statistics.
-//
-// TODO(b/64800844): Add and populate stat fields.
-type TCPInfoOption struct {
-	RTT    time.Duration
-	RTTVar time.Duration
-}
-
 // KeepaliveEnabledOption is used by SetSockOpt/GetSockOpt to specify whether
 // TCP keepalive is enabled for this socket.
 type KeepaliveEnabledOption int
diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go
index cc49c8272..e94307bd5 100644
--- a/pkg/tcpip/transport/tcp/endpoint.go
+++ b/pkg/tcpip/transport/tcp/endpoint.go
@@ -108,6 +108,9 @@ func (s EndpointState) String() string {
 	}
 }
 
+// InfoOption is used by GetSockOpt to expose TCP endpoint state.
+type InfoOption stack.TCPEndpointState
+
 // Reasons for notifying the protocol goroutine.
 const (
 	notifyNonZeroReceiveWindow = 1 << iota
@@ -202,12 +205,14 @@ type endpoint struct {
 	// to indicate to users that no more data is coming.
 	//
 	// rcvListMu can be taken after the endpoint mu below.
-	rcvListMu     sync.Mutex  `state:"nosave"`
-	rcvList       segmentList `state:"wait"`
-	rcvClosed     bool
-	rcvBufSize    int
-	rcvBufUsed    int
-	rcvAutoParams rcvBufAutoTuneParams
+	rcvListMu        sync.Mutex  `state:"nosave"`
+	rcvList          segmentList `state:"wait"`
+	rcvClosed        bool
+	rcvBufSize       int
+	rcvBufUsed       int
+	rcvAutoParams    rcvBufAutoTuneParams
+	rcvLastAckNanos  int64 // timestamp
+	rcvLastDataNanos int64 // timestamp
 	// zeroWindow indicates that the window was closed due to receive buffer
 	// space being filled up. This is set by the worker goroutine before
 	// moving a segment to the rcvList. This setting is cleared by the
@@ -1198,17 +1203,10 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error {
 		}
 		return nil
 
-	case *tcpip.TCPInfoOption:
-		*o = tcpip.TCPInfoOption{}
-		e.mu.RLock()
-		snd := e.snd
-		e.mu.RUnlock()
-		if snd != nil {
-			snd.rtt.Lock()
-			o.RTT = snd.rtt.srtt
-			o.RTTVar = snd.rtt.rttvar
-			snd.rtt.Unlock()
-		}
+	case *InfoOption:
+		e.workMu.Lock()
+		*o = InfoOption(e.completeState())
+		e.workMu.Unlock()
 		return nil
 
 	case *tcpip.KeepaliveEnabledOption:
@@ -1933,22 +1931,27 @@ func (e *endpoint) maxOptionSize() (size int) {
 }
 
 // completeState makes a full copy of the endpoint and returns it. This is used
-// before invoking the probe. The state returned may not be fully consistent if
-// there are intervening syscalls when the state is being copied.
+// before invoking the probe and for getsockopt(TCP_INFO). The state returned
+// may not be fully consistent if there are intervening syscalls when the state
+// is being copied.
 func (e *endpoint) completeState() stack.TCPEndpointState {
 	var s stack.TCPEndpointState
 	s.SegTime = time.Now()
 
-	// Copy EndpointID.
-	e.mu.Lock()
+	e.mu.RLock()
 	s.ID = stack.TCPEndpointID(e.id)
-	e.mu.Unlock()
+	s.ProtocolState = uint32(e.state)
+	s.AMSS = e.amss
+	s.RcvMSS = int(e.amss) - e.maxOptionSize()
+	e.mu.RUnlock()
 
 	// Copy endpoint rcv state.
 	e.rcvListMu.Lock()
 	s.RcvBufSize = e.rcvBufSize
 	s.RcvBufUsed = e.rcvBufUsed
 	s.RcvClosed = e.rcvClosed
+	s.RcvLastAckNanos = e.rcvLastAckNanos
+	s.RcvLastDataNanos = e.rcvLastDataNanos
 	s.RcvAutoParams.MeasureTime = e.rcvAutoParams.measureTime
 	s.RcvAutoParams.CopiedBytes = e.rcvAutoParams.copied
 	s.RcvAutoParams.PrevCopiedBytes = e.rcvAutoParams.prevCopied
@@ -1956,6 +1959,7 @@ func (e *endpoint) completeState() stack.TCPEndpointState {
 	s.RcvAutoParams.RTTMeasureSeqNumber = e.rcvAutoParams.rttMeasureSeqNumber
 	s.RcvAutoParams.RTTMeasureTime = e.rcvAutoParams.rttMeasureTime
 	s.RcvAutoParams.Disabled = e.rcvAutoParams.disabled
+
 	e.rcvListMu.Unlock()
 
 	// Endpoint TCP Option state.
@@ -1965,7 +1969,7 @@ func (e *endpoint) completeState() stack.TCPEndpointState {
 	s.SACKPermitted = e.sackPermitted
 	s.SACK.Blocks = make([]header.SACKBlock, e.sack.NumBlocks)
 	copy(s.SACK.Blocks, e.sack.Blocks[:e.sack.NumBlocks])
-	s.SACK.ReceivedBlocks, s.SACK.MaxSACKED = e.scoreboard.Copy()
+	s.SACK.ReceivedBlocks, s.SACK.Sacked, s.SACK.MaxSACKED = e.scoreboard.Copy()
 
 	// Copy endpoint send state.
 	e.sndBufMu.Lock()
@@ -2009,12 +2013,14 @@ func (e *endpoint) completeState() stack.TCPEndpointState {
 		RTTMeasureTime:   e.snd.rttMeasureTime,
 		Closed:           e.snd.closed,
 		RTO:              e.snd.rto,
+		MSS:              e.snd.mss,
 		MaxPayloadSize:   e.snd.maxPayloadSize,
 		SndWndScale:      e.snd.sndWndScale,
 		MaxSentAck:       e.snd.maxSentAck,
 	}
 	e.snd.rtt.Lock()
 	s.Sender.SRTT = e.snd.rtt.srtt
+	s.Sender.RTTVar = e.snd.rtt.rttvar
 	s.Sender.SRTTInited = e.snd.rtt.srttInited
 	e.snd.rtt.Unlock()
 
@@ -2059,8 +2065,8 @@ func (e *endpoint) initGSO() {
 // State implements tcpip.Endpoint.State. It exports the endpoint's protocol
 // state for diagnostics.
 func (e *endpoint) State() uint32 {
-	e.mu.Lock()
-	defer e.mu.Unlock()
+	e.mu.RLock()
+	defer e.mu.RUnlock()
 	return uint32(e.state)
 }
 
diff --git a/pkg/tcpip/transport/tcp/rcv.go b/pkg/tcpip/transport/tcp/rcv.go
index e90f9a7d9..a8f490c4a 100644
--- a/pkg/tcpip/transport/tcp/rcv.go
+++ b/pkg/tcpip/transport/tcp/rcv.go
@@ -220,25 +220,24 @@ func (r *receiver) consumeSegment(s *segment, segSeq seqnum.Value, segLen seqnum
 	return true
 }
 
-// updateRTT updates the receiver RTT measurement based on the sequence number
-// of the received segment.
-func (r *receiver) updateRTT() {
+// updateRTTLocked updates the receiver RTT measurement based on the sequence
+// number of the received segment.
+//
+// Precondition: Caller must hold r.ep.rcvListMu.
+func (r *receiver) updateRTTLocked() {
 	// From: https://public.lanl.gov/radiant/pubs/drs/sc2001-poster.pdf
 	//
 	// A system that is only transmitting acknowledgements can still
 	// estimate the round-trip time by observing the time between when a byte
 	// is first acknowledged and the receipt of data that is at least one
 	// window beyond the sequence number that was acknowledged.
-	r.ep.rcvListMu.Lock()
 	if r.ep.rcvAutoParams.rttMeasureTime.IsZero() {
 		// New measurement.
 		r.ep.rcvAutoParams.rttMeasureTime = time.Now()
 		r.ep.rcvAutoParams.rttMeasureSeqNumber = r.rcvNxt.Add(r.rcvWnd)
-		r.ep.rcvListMu.Unlock()
 		return
 	}
 	if r.rcvNxt.LessThan(r.ep.rcvAutoParams.rttMeasureSeqNumber) {
-		r.ep.rcvListMu.Unlock()
 		return
 	}
 	rtt := time.Since(r.ep.rcvAutoParams.rttMeasureTime)
@@ -250,7 +249,6 @@ func (r *receiver) updateRTT() {
 	}
 	r.ep.rcvAutoParams.rttMeasureTime = time.Now()
 	r.ep.rcvAutoParams.rttMeasureSeqNumber = r.rcvNxt.Add(r.rcvWnd)
-	r.ep.rcvListMu.Unlock()
 }
 
 // handleRcvdSegment handles TCP segments directed at the connection managed by
@@ -291,11 +289,20 @@ func (r *receiver) handleRcvdSegment(s *segment) {
 		return
 	}
 
-	// Since we consumed a segment update the receiver's RTT estimate
-	// if required.
+	r.ep.rcvListMu.Lock()
+	// FIXME(b/137581805): Using the runtime clock here is incorrect as it
+	// doesn't account for potentially virtualized time.
+	now := time.Now().UnixNano()
+	if s.flagIsSet(header.TCPFlagAck) {
+		r.ep.rcvLastAckNanos = now
+	}
 	if segLen > 0 {
-		r.updateRTT()
+		// Since we consumed a segment update the receiver's RTT estimate if
+		// required.
+		r.ep.rcvLastDataNanos = now
+		r.updateRTTLocked()
 	}
+	r.ep.rcvListMu.Unlock()
 
 	// By consuming the current segment, we may have filled a gap in the
 	// sequence number domain that allows pending segments to be consumed
diff --git a/pkg/tcpip/transport/tcp/sack_scoreboard.go b/pkg/tcpip/transport/tcp/sack_scoreboard.go
index 7ef2df377..02e52a63b 100644
--- a/pkg/tcpip/transport/tcp/sack_scoreboard.go
+++ b/pkg/tcpip/transport/tcp/sack_scoreboard.go
@@ -208,12 +208,12 @@ func (s *SACKScoreboard) Delete(seq seqnum.Value) {
 }
 
 // Copy provides a copy of the SACK scoreboard.
-func (s *SACKScoreboard) Copy() (sackBlocks []header.SACKBlock, maxSACKED seqnum.Value) {
+func (s *SACKScoreboard) Copy() (sackBlocks []header.SACKBlock, sacked seqnum.Size, maxSACKED seqnum.Value) {
 	s.ranges.Ascend(func(i btree.Item) bool {
 		sackBlocks = append(sackBlocks, i.(header.SACKBlock))
 		return true
 	})
-	return sackBlocks, s.maxSACKED
+	return sackBlocks, s.sacked, s.maxSACKED
 }
 
 // IsRangeLost implements the IsLost(SeqNum) operation defined in RFC 6675
diff --git a/pkg/tcpip/transport/tcp/snd.go b/pkg/tcpip/transport/tcp/snd.go
index 0fee7ab72..daf28a49a 100644
--- a/pkg/tcpip/transport/tcp/snd.go
+++ b/pkg/tcpip/transport/tcp/snd.go
@@ -124,6 +124,10 @@ type sender struct {
 	rtt rtt
 	rto time.Duration
 
+	// mss is the largest segment that can be sent without fragmentation.
+	// Initialized when then sender is created, read-only afterwards.
+	mss int
+
 	// maxPayloadSize is the maximum size of the payload of a given segment.
 	// It is initialized on demand.
 	maxPayloadSize int
@@ -201,6 +205,7 @@ func newSender(ep *endpoint, iss, irs seqnum.Value, sndWnd seqnum.Size, mss uint
 		rto:              1 * time.Second,
 		rttMeasureSeqNum: iss + 1,
 		lastSendTime:     time.Now(),
+		mss:              int(mss),
 		maxPayloadSize:   maxPayloadSize,
 		maxSentAck:       irs + 1,
 		fr: fastRecovery{
diff --git a/test/syscalls/linux/socket_ip_tcp_generic.cc b/test/syscalls/linux/socket_ip_tcp_generic.cc
index a43cf9bce..01987d2f0 100644
--- a/test/syscalls/linux/socket_ip_tcp_generic.cc
+++ b/test/syscalls/linux/socket_ip_tcp_generic.cc
@@ -697,5 +697,28 @@ TEST_P(TCPSocketPairTest, SetCongestionControlFailsForUnsupported) {
   EXPECT_EQ(0, memcmp(got_cc, old_cc, sizeof(old_cc)));
 }
 
+TEST_P(TCPSocketPairTest, GetSockOptTCPInfo) {
+  auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair());
+  struct tcp_info info;
+  socklen_t optlen = sizeof(info);
+  ASSERT_THAT(
+      getsockopt(sockets->first_fd(), SOL_TCP, TCP_INFO, &info, &optlen),
+      SyscallSucceedsWithValue(0));
+  EXPECT_EQ(optlen, sizeof(info));
+
+  EXPECT_EQ(info.tcpi_state, TCP_ESTABLISHED);
+
+  EXPECT_GT(info.tcpi_rto, 0);
+
+  // IPv4 MSS is 536 bytes by default, and IPv6 is 1220 bytes.
+  EXPECT_GE(info.tcpi_snd_mss, 536);
+  EXPECT_GE(info.tcpi_rcv_mss, 536);
+  EXPECT_GE(info.tcpi_advmss, 536);
+
+  // MTU is typically 1500 for ethernet, but this is highly protocol
+  // dependent. Opt for a safe lower bound.
+  EXPECT_GT(info.tcpi_pmtu, 500);
+}
+
 }  // namespace testing
 }  // namespace gvisor