7 files changed, 97 insertions, 51 deletions
diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go
index 693ea064a..41a489047 100644
--- a/pkg/tcpip/stack/nic.go
+++ b/pkg/tcpip/stack/nic.go
@@ -911,9 +911,8 @@ func (n *NIC) DeliverTransportPacket(protocol tcpip.TransportProtocolNumber, pkt
 	}
 }
 
-// DeliverTransportControlPacket delivers control packets to the appropriate
-// transport protocol endpoint.
-func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, pkt *PacketBuffer) {
+// DeliverTransportError implements TransportDispatcher.
+func (n *NIC) DeliverTransportError(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, transErr TransportError, pkt *PacketBuffer) {
 	state, ok := n.stack.transportProtocols[trans]
 	if !ok {
 		return
@@ -935,7 +934,7 @@ func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcp
 	}
 
 	id := TransportEndpointID{srcPort, local, dstPort, remote}
-	if n.stack.demux.deliverControlPacket(n, net, trans, typ, extra, pkt, id) {
+	if n.stack.demux.deliverError(n, net, trans, transErr, pkt, id) {
 		return
 	}
 }
diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go
index e02f7190c..d589f798d 100644
--- a/pkg/tcpip/stack/registration.go
+++ b/pkg/tcpip/stack/registration.go
@@ -49,31 +49,6 @@ type TransportEndpointID struct {
 	RemoteAddress tcpip.Address
 }
 
-// ControlType is the type of network control message.
-type ControlType int
-
-// The following are the allowed values for ControlType values.
-// TODO(http://gvisor.dev/issue/3210): Support time exceeded messages.
-const (
-	// ControlAddressUnreachable indicates that an IPv6 packet did not reach its
-	// destination as the destination address was unreachable.
-	//
-	// This maps to the ICMPv6 Destination Ureachable Code 3 error; see
-	// RFC 4443 section 3.1 for more details.
-	ControlAddressUnreachable ControlType = iota
-	ControlNetworkUnreachable
-	// ControlNoRoute indicates that an IPv4 packet did not reach its destination
-	// because the destination host was unreachable.
-	//
-	// This maps to the ICMPv4 Destination Ureachable Code 1 error; see
-	// RFC 791's Destination Unreachable Message section (page 4) for more
-	// details.
-	ControlNoRoute
-	ControlPacketTooBig
-	ControlPortUnreachable
-	ControlUnknown
-)
-
 // NetworkPacketInfo holds information about a network layer packet.
 type NetworkPacketInfo struct {
 	// LocalAddressBroadcast is true if the packet's local address is a broadcast
@@ -81,6 +56,39 @@ type NetworkPacketInfo struct {
 	LocalAddressBroadcast bool
 }
 
+// TransportErrorKind enumerates error types that are handled by the transport
+// layer.
+type TransportErrorKind int
+
+const (
+	// PacketTooBigTransportError indicates that a packet did not reach its
+	// destination because a link on the path to the destination had an MTU that
+	// was too small to carry the packet.
+	PacketTooBigTransportError TransportErrorKind = iota
+
+	// DestinationHostUnreachableTransportError indicates that the destination
+	// host was unreachable.
+	DestinationHostUnreachableTransportError
+
+	// DestinationPortUnreachableTransportError indicates that a packet reached
+	// the destination host, but the transport protocol was not active on the
+	// destination port.
+	DestinationPortUnreachableTransportError
+
+	// DestinationNetworkUnreachableTransportError indicates that the destination
+	// network was unreachable.
+	DestinationNetworkUnreachableTransportError
+)
+
+// TransportError is a marker interface for errors that may be handled by the
+// transport layer.
+type TransportError interface {
+	tcpip.SockErrorCause
+
+	// Kind returns the type of the transport error.
+	Kind() TransportErrorKind
+}
+
 // TransportEndpoint is the interface that needs to be implemented by transport
 // protocol (e.g., tcp, udp) endpoints that can handle packets.
 type TransportEndpoint interface {
@@ -93,10 +101,10 @@ type TransportEndpoint interface {
 	// HandlePacket takes ownership of the packet.
 	HandlePacket(TransportEndpointID, *PacketBuffer)
 
-	// HandleControlPacket is called by the stack when new control (e.g.
-	// ICMP) packets arrive to this transport endpoint.
-	// HandleControlPacket takes ownership of pkt.
-	HandleControlPacket(typ ControlType, extra uint32, pkt *PacketBuffer)
+	// HandleError is called when the transport endpoint receives an error.
+	//
+	// HandleError takes ownership of the packet buffer.
+	HandleError(TransportError, *PacketBuffer)
 
 	// Abort initiates an expedited endpoint teardown. It puts the endpoint
 	// in a closed state and frees all resources associated with it. This
@@ -248,14 +256,11 @@ type TransportDispatcher interface {
 	// DeliverTransportPacket takes ownership of the packet.
 	DeliverTransportPacket(tcpip.TransportProtocolNumber, *PacketBuffer) TransportPacketDisposition
 
-	// DeliverTransportControlPacket delivers control packets to the
-	// appropriate transport protocol endpoint.
-	//
-	// pkt.NetworkHeader must be set before calling
-	// DeliverTransportControlPacket.
+	// DeliverTransportError delivers an error to the appropriate transport
+	// endpoint.
 	//
-	// DeliverTransportControlPacket takes ownership of pkt.
-	DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, pkt *PacketBuffer)
+	// DeliverTransportError takes ownership of the packet buffer.
+	DeliverTransportError(local, remote tcpip.Address, _ tcpip.NetworkProtocolNumber, _ tcpip.TransportProtocolNumber, _ TransportError, _ *PacketBuffer)
 }
 
 // PacketLooping specifies where an outbound packet should be sent.
diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go
index 57ad412a1..a51d758d0 100644
--- a/pkg/tcpip/stack/stack.go
+++ b/pkg/tcpip/stack/stack.go
@@ -458,6 +458,18 @@ type Stack struct {
 	// receiveBufferSize holds the min/default/max receive buffer sizes for
 	// endpoints other than TCP.
 	receiveBufferSize ReceiveBufferSizeOption
+
+	// tcpInvalidRateLimit is the maximal rate for sending duplicate
+	// acknowledgements in response to incoming TCP packets that are for an existing
+	// connection but that are invalid due to any of the following reasons:
+	//
+	//   a) out-of-window sequence number.
+	//   b) out-of-window acknowledgement number.
+	//   c) PAWS check failure (when implemented).
+	//
+	// This is required to prevent potential ACK loops.
+	// Setting this to 0 will disable all rate limiting.
+	tcpInvalidRateLimit time.Duration
 }
 
 // UniqueID is an abstract generator of unique identifiers.
@@ -668,6 +680,7 @@ func New(opts Options) *Stack {
 			Default: DefaultBufferSize,
 			Max:     DefaultMaxBufferSize,
 		},
+		tcpInvalidRateLimit: defaultTCPInvalidRateLimit,
 	}
 
 	// Add specified network protocols.
diff --git a/pkg/tcpip/stack/stack_options.go b/pkg/tcpip/stack/stack_options.go
index 8d9b20b7e..3066f4ffd 100644
--- a/pkg/tcpip/stack/stack_options.go
+++ b/pkg/tcpip/stack/stack_options.go
@@ -15,6 +15,8 @@
 package stack
 
 import (
+	"time"
+
 	"gvisor.dev/gvisor/pkg/tcpip"
 )
 
@@ -29,6 +31,10 @@ const (
 	// DefaultMaxBufferSize is the default maximum permitted size of a
 	// send/receive buffer.
 	DefaultMaxBufferSize = 4 << 20 // 4 MiB
+
+	// defaultTCPInvalidRateLimit is the default value for
+	// stack.TCPInvalidRateLimit.
+	defaultTCPInvalidRateLimit = 500 * time.Millisecond
 )
 
 // ReceiveBufferSizeOption is used by stack.(Stack*).Option/SetOption to
@@ -39,6 +45,10 @@ type ReceiveBufferSizeOption struct {
 	Max     int
 }
 
+// TCPInvalidRateLimitOption is used by stack.(Stack*).Option/SetOption to get/set
+// stack.tcpInvalidRateLimit.
+type TCPInvalidRateLimitOption time.Duration
+
 // SetOption allows setting stack wide options.
 func (s *Stack) SetOption(option interface{}) tcpip.Error {
 	switch v := option.(type) {
@@ -74,6 +84,15 @@ func (s *Stack) SetOption(option interface{}) tcpip.Error {
 		s.mu.Unlock()
 		return nil
 
+	case TCPInvalidRateLimitOption:
+		if v < 0 {
+			return &tcpip.ErrInvalidOptionValue{}
+		}
+		s.mu.Lock()
+		s.tcpInvalidRateLimit = time.Duration(v)
+		s.mu.Unlock()
+		return nil
+
 	default:
 		return &tcpip.ErrUnknownProtocolOption{}
 	}
@@ -94,6 +113,12 @@ func (s *Stack) Option(option interface{}) tcpip.Error {
 		s.mu.RUnlock()
 		return nil
 
+	case *TCPInvalidRateLimitOption:
+		s.mu.RLock()
+		*v = TCPInvalidRateLimitOption(s.tcpInvalidRateLimit)
+		s.mu.RUnlock()
+		return nil
+
 	default:
 		return &tcpip.ErrUnknownProtocolOption{}
 	}
diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go
index 375cd3080..b641a4aaa 100644
--- a/pkg/tcpip/stack/stack_test.go
+++ b/pkg/tcpip/stack/stack_test.go
@@ -138,12 +138,15 @@ func (f *fakeNetworkEndpoint) HandlePacket(pkt *stack.PacketBuffer) {
 			return
 		}
 		pkt.Data.TrimFront(fakeNetHeaderLen)
-		f.dispatcher.DeliverTransportControlPacket(
+		f.dispatcher.DeliverTransportError(
 			tcpip.Address(nb[srcAddrOffset:srcAddrOffset+1]),
 			tcpip.Address(nb[dstAddrOffset:dstAddrOffset+1]),
 			fakeNetNumber,
 			tcpip.TransportProtocolNumber(nb[protocolNumberOffset]),
-			stack.ControlPortUnreachable, 0, pkt)
+			// Nothing checks the error.
+			nil, /* transport error */
+			pkt,
+		)
 		return
 	}
 
diff --git a/pkg/tcpip/stack/transport_demuxer.go b/pkg/tcpip/stack/transport_demuxer.go
index 26eceb804..7d8d0851e 100644
--- a/pkg/tcpip/stack/transport_demuxer.go
+++ b/pkg/tcpip/stack/transport_demuxer.go
@@ -182,9 +182,8 @@ func (epsByNIC *endpointsByNIC) handlePacket(id TransportEndpointID, pkt *Packet
 	epsByNIC.mu.RUnlock() // Don't use defer for performance reasons.
 }
 
-// handleControlPacket delivers a control packet to the transport endpoint
-// identified by id.
-func (epsByNIC *endpointsByNIC) handleControlPacket(n *NIC, id TransportEndpointID, typ ControlType, extra uint32, pkt *PacketBuffer) {
+// handleError delivers an error to the transport endpoint identified by id.
+func (epsByNIC *endpointsByNIC) handleError(n *NIC, id TransportEndpointID, transErr TransportError, pkt *PacketBuffer) {
 	epsByNIC.mu.RLock()
 	defer epsByNIC.mu.RUnlock()
 
@@ -200,7 +199,7 @@ func (epsByNIC *endpointsByNIC) handleControlPacket(n *NIC, id TransportEndpoint
 	// broadcast like we are doing with handlePacket above?
 
 	// multiPortEndpoints are guaranteed to have at least one element.
-	selectEndpoint(id, mpep, epsByNIC.seed).HandleControlPacket(typ, extra, pkt)
+	selectEndpoint(id, mpep, epsByNIC.seed).HandleError(transErr, pkt)
 }
 
 // registerEndpoint returns true if it succeeds. It fails and returns
@@ -596,9 +595,11 @@ func (d *transportDemuxer) deliverRawPacket(protocol tcpip.TransportProtocolNumb
 	return foundRaw
 }
 
-// deliverControlPacket attempts to deliver the given control packet. Returns
-// true if it found an endpoint, false otherwise.
-func (d *transportDemuxer) deliverControlPacket(n *NIC, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, pkt *PacketBuffer, id TransportEndpointID) bool {
+// deliverError attempts to deliver the given error to the appropriate transport
+// endpoint.
+//
+// Returns true if the error was delivered.
+func (d *transportDemuxer) deliverError(n *NIC, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, transErr TransportError, pkt *PacketBuffer, id TransportEndpointID) bool {
 	eps, ok := d.protocol[protocolIDs{net, trans}]
 	if !ok {
 		return false
@@ -611,7 +612,7 @@ func (d *transportDemuxer) deliverControlPacket(n *NIC, net tcpip.NetworkProtoco
 		return false
 	}
 
-	ep.handleControlPacket(n, id, typ, extra, pkt)
+	ep.handleError(n, id, transErr, pkt)
 	return true
 }
 
diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go
index cf5de747b..bebf4e6b5 100644
--- a/pkg/tcpip/stack/transport_test.go
+++ b/pkg/tcpip/stack/transport_test.go
@@ -237,7 +237,7 @@ func (f *fakeTransportEndpoint) HandlePacket(id stack.TransportEndpointID, pkt *
 	f.acceptQueue = append(f.acceptQueue, ep)
 }
 
-func (f *fakeTransportEndpoint) HandleControlPacket(stack.ControlType, uint32, *stack.PacketBuffer) {
+func (f *fakeTransportEndpoint) HandleError(stack.TransportError, *stack.PacketBuffer) {
 	// Increment the number of received control packets.
 	f.proto.controlCount++
 }