Add support for ping sockets

PiperOrigin-RevId: 195049322
Change-Id: I09f6dd58cf10a2e50e53d17d2823d540102913c5

10 files changed, 898 insertions, 321 deletions

diff --git a/pkg/tcpip/network/arp/BUILD b/pkg/tcpip/network/arp/BUILD
index e6d0899a9..58d174965 100644
--- a/pkg/tcpip/network/arp/BUILD
+++ b/pkg/tcpip/network/arp/BUILD
@@ -30,5 +30,6 @@ go_test(
         "//pkg/tcpip/link/sniffer",
         "//pkg/tcpip/network/ipv4",
         "//pkg/tcpip/stack",
+        "//pkg/tcpip/transport/ping",
     ],
 )
diff --git a/pkg/tcpip/network/arp/arp_test.go b/pkg/tcpip/network/arp/arp_test.go
index 47b10e64e..6d61ff1d7 100644
--- a/pkg/tcpip/network/arp/arp_test.go
+++ b/pkg/tcpip/network/arp/arp_test.go
@@ -16,6 +16,7 @@ import (
 	"gvisor.googlesource.com/gvisor/pkg/tcpip/network/arp"
 	"gvisor.googlesource.com/gvisor/pkg/tcpip/network/ipv4"
 	"gvisor.googlesource.com/gvisor/pkg/tcpip/stack"
+	"gvisor.googlesource.com/gvisor/pkg/tcpip/transport/ping"
 )
 
 const (
@@ -32,7 +33,7 @@ type testContext struct {
 }
 
 func newTestContext(t *testing.T) *testContext {
-	s := stack.New(&tcpip.StdClock{}, []string{ipv4.ProtocolName, arp.ProtocolName}, []string{ipv4.PingProtocolName})
+	s := stack.New(&tcpip.StdClock{}, []string{ipv4.ProtocolName, arp.ProtocolName}, []string{ping.ProtocolName4})
 
 	const defaultMTU = 65536
 	id, linkEP := channel.New(256, defaultMTU, stackLinkAddr)
diff --git a/pkg/tcpip/network/ipv4/BUILD b/pkg/tcpip/network/ipv4/BUILD
index 9df113df1..02d55355c 100644
--- a/pkg/tcpip/network/ipv4/BUILD
+++ b/pkg/tcpip/network/ipv4/BUILD
@@ -1,6 +1,6 @@
 package(licenses = ["notice"])  # BSD
 
-load("@io_bazel_rules_go//go:def.bzl", "go_library", "go_test")
+load("@io_bazel_rules_go//go:def.bzl", "go_library")
 
 go_library(
     name = "ipv4",
@@ -19,20 +19,5 @@ go_library(
         "//pkg/tcpip/network/fragmentation",
         "//pkg/tcpip/network/hash",
         "//pkg/tcpip/stack",
-        "//pkg/waiter",
-    ],
-)
-
-go_test(
-    name = "ipv4_test",
-    size = "small",
-    srcs = ["icmp_test.go"],
-    deps = [
-        ":ipv4",
-        "//pkg/tcpip",
-        "//pkg/tcpip/buffer",
-        "//pkg/tcpip/link/channel",
-        "//pkg/tcpip/link/sniffer",
-        "//pkg/tcpip/stack",
     ],
 )
diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go
index ffd761350..3c382fdc2 100644
--- a/pkg/tcpip/network/ipv4/icmp.go
+++ b/pkg/tcpip/network/ipv4/icmp.go
@@ -5,26 +5,14 @@
 package ipv4
 
 import (
-	"context"
 	"encoding/binary"
-	"time"
 
 	"gvisor.googlesource.com/gvisor/pkg/tcpip"
 	"gvisor.googlesource.com/gvisor/pkg/tcpip/buffer"
 	"gvisor.googlesource.com/gvisor/pkg/tcpip/header"
 	"gvisor.googlesource.com/gvisor/pkg/tcpip/stack"
-	"gvisor.googlesource.com/gvisor/pkg/waiter"
 )
 
-// PingProtocolName is a pseudo transport protocol used to handle ping replies.
-// Use it when constructing a stack that intends to use ipv4.Ping.
-const PingProtocolName = "icmpv4ping"
-
-// pingProtocolNumber is a fake transport protocol used to
-// deliver incoming ICMP echo replies. The ICMP identifier
-// number is used as a port number for multiplexing.
-const pingProtocolNumber tcpip.TransportProtocolNumber = 256 + 11
-
 // handleControl handles the case when an ICMP packet contains the headers of
 // the original packet that caused the ICMP one to be sent. This information is
 // used to find out which transport endpoint must be notified about the ICMP
@@ -78,7 +66,10 @@ func (e *endpoint) handleICMP(r *stack.Route, vv *buffer.VectorisedView) {
 		}
 
 	case header.ICMPv4EchoReply:
-		e.dispatcher.DeliverTransportPacket(r, pingProtocolNumber, vv)
+		if len(v) < header.ICMPv4EchoMinimumSize {
+			return
+		}
+		e.dispatcher.DeliverTransportPacket(r, header.ICMPv4ProtocolNumber, vv)
 
 	case header.ICMPv4DstUnreachable:
 		if len(v) < header.ICMPv4DstUnreachableMinimumSize {
@@ -104,179 +95,20 @@ type echoRequest struct {
 
 func (e *endpoint) echoReplier() {
 	for req := range e.echoRequests {
-		sendICMPv4(&req.r, header.ICMPv4EchoReply, 0, req.v)
+		sendPing4(&req.r, 0, req.v)
 		req.r.Release()
 	}
 }
 
-func sendICMPv4(r *stack.Route, typ header.ICMPv4Type, code byte, data buffer.View) *tcpip.Error {
-	hdr := buffer.NewPrependable(header.ICMPv4MinimumSize + int(r.MaxHeaderLength()))
+func sendPing4(r *stack.Route, code byte, data buffer.View) *tcpip.Error {
+	hdr := buffer.NewPrependable(header.ICMPv4EchoMinimumSize + int(r.MaxHeaderLength()))
 
-	icmpv4 := header.ICMPv4(hdr.Prepend(header.ICMPv4MinimumSize))
-	icmpv4.SetType(typ)
+	icmpv4 := header.ICMPv4(hdr.Prepend(header.ICMPv4EchoMinimumSize))
+	icmpv4.SetType(header.ICMPv4EchoReply)
 	icmpv4.SetCode(code)
+	copy(icmpv4[header.ICMPv4MinimumSize:], data)
+	data = data[header.ICMPv4EchoMinimumSize-header.ICMPv4MinimumSize:]
 	icmpv4.SetChecksum(^header.Checksum(icmpv4, header.Checksum(data, 0)))
 
 	return r.WritePacket(&hdr, data, header.ICMPv4ProtocolNumber)
 }
-
-// A Pinger can send echo requests to an address.
-type Pinger struct {
-	Stack     *stack.Stack
-	NICID     tcpip.NICID
-	Addr      tcpip.Address
-	LocalAddr tcpip.Address // optional
-	Wait      time.Duration // if zero, defaults to 1 second
-	Count     uint16        // if zero, defaults to MaxUint16
-}
-
-// Ping sends echo requests to an ICMPv4 endpoint.
-// Responses are streamed to the channel ch.
-func (p *Pinger) Ping(ctx context.Context, ch chan<- PingReply) *tcpip.Error {
-	count := p.Count
-	if count == 0 {
-		count = 1<<16 - 1
-	}
-	wait := p.Wait
-	if wait == 0 {
-		wait = 1 * time.Second
-	}
-
-	r, err := p.Stack.FindRoute(p.NICID, p.LocalAddr, p.Addr, ProtocolNumber)
-	if err != nil {
-		return err
-	}
-
-	netProtos := []tcpip.NetworkProtocolNumber{ProtocolNumber}
-	ep := &pingEndpoint{
-		stack: p.Stack,
-		pktCh: make(chan buffer.View, 1),
-	}
-	id := stack.TransportEndpointID{
-		LocalAddress:  r.LocalAddress,
-		RemoteAddress: p.Addr,
-	}
-
-	_, err = p.Stack.PickEphemeralPort(func(port uint16) (bool, *tcpip.Error) {
-		id.LocalPort = port
-		err := p.Stack.RegisterTransportEndpoint(p.NICID, netProtos, pingProtocolNumber, id, ep)
-		switch err {
-		case nil:
-			return true, nil
-		case tcpip.ErrPortInUse:
-			return false, nil
-		default:
-			return false, err
-		}
-	})
-	if err != nil {
-		return err
-	}
-	defer p.Stack.UnregisterTransportEndpoint(p.NICID, netProtos, pingProtocolNumber, id)
-
-	v := buffer.NewView(4)
-	binary.BigEndian.PutUint16(v[0:], id.LocalPort)
-
-	start := time.Now()
-
-	done := make(chan struct{})
-	go func(count int) {
-	loop:
-		for ; count > 0; count-- {
-			select {
-			case v := <-ep.pktCh:
-				seq := binary.BigEndian.Uint16(v[header.ICMPv4MinimumSize+2:])
-				ch <- PingReply{
-					Duration:  time.Since(start) - time.Duration(seq)*wait,
-					SeqNumber: seq,
-				}
-			case <-ctx.Done():
-				break loop
-			}
-		}
-		close(done)
-	}(int(count))
-	defer func() { <-done }()
-
-	t := time.NewTicker(wait)
-	defer t.Stop()
-	for seq := uint16(0); seq < count; seq++ {
-		select {
-		case <-t.C:
-		case <-ctx.Done():
-			return nil
-		}
-		binary.BigEndian.PutUint16(v[2:], seq)
-		sent := time.Now()
-		if err := sendICMPv4(&r, header.ICMPv4Echo, 0, v); err != nil {
-			ch <- PingReply{
-				Error:     err,
-				Duration:  time.Since(sent),
-				SeqNumber: seq,
-			}
-		}
-	}
-	return nil
-}
-
-// PingReply summarizes an ICMP echo reply.
-type PingReply struct {
-	Error     *tcpip.Error // reports any errors sending a ping request
-	Duration  time.Duration
-	SeqNumber uint16
-}
-
-type pingProtocol struct{}
-
-func (*pingProtocol) NewEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) {
-	return nil, tcpip.ErrNotSupported // endpoints are created directly
-}
-
-func (*pingProtocol) Number() tcpip.TransportProtocolNumber { return pingProtocolNumber }
-
-func (*pingProtocol) MinimumPacketSize() int { return header.ICMPv4EchoMinimumSize }
-
-func (*pingProtocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) {
-	ident := binary.BigEndian.Uint16(v[4:])
-	return 0, ident, nil
-}
-
-func (*pingProtocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, *buffer.VectorisedView) bool {
-	return true
-}
-
-// SetOption implements TransportProtocol.SetOption.
-func (p *pingProtocol) SetOption(option interface{}) *tcpip.Error {
-	return tcpip.ErrUnknownProtocolOption
-}
-
-// Option implements TransportProtocol.Option.
-func (p *pingProtocol) Option(option interface{}) *tcpip.Error {
-	return tcpip.ErrUnknownProtocolOption
-}
-
-func init() {
-	stack.RegisterTransportProtocolFactory(PingProtocolName, func() stack.TransportProtocol {
-		return &pingProtocol{}
-	})
-}
-
-type pingEndpoint struct {
-	stack *stack.Stack
-	pktCh chan buffer.View
-}
-
-func (e *pingEndpoint) Close() {
-	close(e.pktCh)
-}
-
-func (e *pingEndpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, vv *buffer.VectorisedView) {
-	select {
-	case e.pktCh <- vv.ToView():
-	default:
-	}
-}
-
-// HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket.
-func (e *pingEndpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, vv *buffer.VectorisedView) {
-}
diff --git a/pkg/tcpip/network/ipv4/icmp_test.go b/pkg/tcpip/network/ipv4/icmp_test.go
deleted file mode 100644
index c55aa1835..000000000
--- a/pkg/tcpip/network/ipv4/icmp_test.go
+++ /dev/null
@@ -1,124 +0,0 @@
-// Copyright 2016 The Netstack Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-package ipv4_test
-
-import (
-	"context"
-	"testing"
-	"time"
-
-	"gvisor.googlesource.com/gvisor/pkg/tcpip"
-	"gvisor.googlesource.com/gvisor/pkg/tcpip/buffer"
-	"gvisor.googlesource.com/gvisor/pkg/tcpip/link/channel"
-	"gvisor.googlesource.com/gvisor/pkg/tcpip/link/sniffer"
-	"gvisor.googlesource.com/gvisor/pkg/tcpip/network/ipv4"
-	"gvisor.googlesource.com/gvisor/pkg/tcpip/stack"
-)
-
-const stackAddr = "\x0a\x00\x00\x01"
-
-type testContext struct {
-	t      *testing.T
-	linkEP *channel.Endpoint
-	s      *stack.Stack
-}
-
-func newTestContext(t *testing.T) *testContext {
-	s := stack.New(&tcpip.StdClock{}, []string{ipv4.ProtocolName}, []string{ipv4.PingProtocolName})
-
-	const defaultMTU = 65536
-	id, linkEP := channel.New(256, defaultMTU, "")
-	if testing.Verbose() {
-		id = sniffer.New(id)
-	}
-	if err := s.CreateNIC(1, id); err != nil {
-		t.Fatalf("CreateNIC failed: %v", err)
-	}
-
-	if err := s.AddAddress(1, ipv4.ProtocolNumber, stackAddr); err != nil {
-		t.Fatalf("AddAddress failed: %v", err)
-	}
-
-	s.SetRouteTable([]tcpip.Route{{
-		Destination: "\x00\x00\x00\x00",
-		Mask:        "\x00\x00\x00\x00",
-		Gateway:     "",
-		NIC:         1,
-	}})
-
-	return &testContext{
-		t:      t,
-		s:      s,
-		linkEP: linkEP,
-	}
-}
-
-func (c *testContext) cleanup() {
-	close(c.linkEP.C)
-}
-
-func (c *testContext) loopback() {
-	go func() {
-		for pkt := range c.linkEP.C {
-			v := make(buffer.View, len(pkt.Header)+len(pkt.Payload))
-			copy(v, pkt.Header)
-			copy(v[len(pkt.Header):], pkt.Payload)
-			vv := v.ToVectorisedView([1]buffer.View{})
-			c.linkEP.Inject(pkt.Proto, &vv)
-		}
-	}()
-}
-
-func TestEcho(t *testing.T) {
-	c := newTestContext(t)
-	defer c.cleanup()
-	c.loopback()
-
-	ch := make(chan ipv4.PingReply, 1)
-	p := ipv4.Pinger{
-		Stack: c.s,
-		NICID: 1,
-		Addr:  stackAddr,
-		Wait:  10 * time.Millisecond,
-		Count: 1, // one ping only
-	}
-	if err := p.Ping(context.Background(), ch); err != nil {
-		t.Fatalf("icmp.Ping failed: %v", err)
-	}
-
-	ping := <-ch
-	if ping.Error != nil {
-		t.Errorf("bad ping response: %v", ping.Error)
-	}
-}
-
-func TestEchoSequence(t *testing.T) {
-	c := newTestContext(t)
-	defer c.cleanup()
-	c.loopback()
-
-	const numPings = 3
-	ch := make(chan ipv4.PingReply, numPings)
-	p := ipv4.Pinger{
-		Stack: c.s,
-		NICID: 1,
-		Addr:  stackAddr,
-		Wait:  10 * time.Millisecond,
-		Count: numPings,
-	}
-	if err := p.Ping(context.Background(), ch); err != nil {
-		t.Fatalf("icmp.Ping failed: %v", err)
-	}
-
-	for i := uint16(0); i < numPings; i++ {
-		ping := <-ch
-		if ping.Error != nil {
-			t.Errorf("i=%d bad ping response: %v", i, ping.Error)
-		}
-		if ping.SeqNumber != i {
-			t.Errorf("SeqNumber=%d, want %d", ping.SeqNumber, i)
-		}
-	}
-}
diff --git a/pkg/tcpip/transport/ping/BUILD b/pkg/tcpip/transport/ping/BUILD
new file mode 100644
index 000000000..a39a887b6
--- /dev/null
+++ b/pkg/tcpip/transport/ping/BUILD
@@ -0,0 +1,50 @@
+package(licenses = ["notice"])  # BSD
+
+load("@io_bazel_rules_go//go:def.bzl", "go_library")
+load("//tools/go_generics:defs.bzl", "go_template_instance")
+load("//tools/go_stateify:defs.bzl", "go_stateify")
+
+go_stateify(
+    name = "ping_state",
+    srcs = [
+        "endpoint.go",
+        "endpoint_state.go",
+        "ping_packet_list.go",
+    ],
+    out = "ping_state.go",
+    imports = ["gvisor.googlesource.com/gvisor/pkg/tcpip/buffer"],
+    package = "ping",
+)
+
+go_template_instance(
+    name = "ping_packet_list",
+    out = "ping_packet_list.go",
+    package = "ping",
+    prefix = "pingPacket",
+    template = "//pkg/ilist:generic_list",
+    types = {
+        "Linker": "*pingPacket",
+    },
+)
+
+go_library(
+    name = "ping",
+    srcs = [
+        "endpoint.go",
+        "endpoint_state.go",
+        "ping_packet_list.go",
+        "ping_state.go",
+        "protocol.go",
+    ],
+    importpath = "gvisor.googlesource.com/gvisor/pkg/tcpip/transport/ping",
+    visibility = ["//visibility:public"],
+    deps = [
+        "//pkg/sleep",
+        "//pkg/state",
+        "//pkg/tcpip",
+        "//pkg/tcpip/buffer",
+        "//pkg/tcpip/header",
+        "//pkg/tcpip/stack",
+        "//pkg/waiter",
+    ],
+)
diff --git a/pkg/tcpip/transport/ping/endpoint.go b/pkg/tcpip/transport/ping/endpoint.go
new file mode 100644
index 000000000..609e7d947
--- /dev/null
+++ b/pkg/tcpip/transport/ping/endpoint.go
@@ -0,0 +1,665 @@
+// Copyright 2016 The Netstack Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package ping
+
+import (
+	"encoding/binary"
+	"sync"
+
+	"gvisor.googlesource.com/gvisor/pkg/sleep"
+	"gvisor.googlesource.com/gvisor/pkg/tcpip"
+	"gvisor.googlesource.com/gvisor/pkg/tcpip/buffer"
+	"gvisor.googlesource.com/gvisor/pkg/tcpip/header"
+	"gvisor.googlesource.com/gvisor/pkg/tcpip/stack"
+	"gvisor.googlesource.com/gvisor/pkg/waiter"
+)
+
+type pingPacket struct {
+	pingPacketEntry
+	senderAddress tcpip.FullAddress
+	data          buffer.VectorisedView `state:".(buffer.VectorisedView)"`
+	timestamp     int64
+	hasTimestamp  bool
+	// views is used as buffer for data when its length is large
+	// enough to store a VectorisedView.
+	views [8]buffer.View `state:"nosave"`
+}
+
+type endpointState int
+
+const (
+	stateInitial endpointState = iota
+	stateBound
+	stateConnected
+	stateClosed
+)
+
+// endpoint represents a ping endpoint. This struct serves as the interface
+// between users of the endpoint and the protocol implementation; it is legal to
+// have concurrent goroutines make calls into the endpoint, they are properly
+// synchronized.
+type endpoint struct {
+	// The following fields are initialized at creation time and do not
+	// change throughout the lifetime of the endpoint.
+	stack       *stack.Stack `state:"manual"`
+	netProto    tcpip.NetworkProtocolNumber
+	waiterQueue *waiter.Queue
+
+	// The following fields are used to manage the receive queue, and are
+	// protected by rcvMu.
+	rcvMu         sync.Mutex `state:"nosave"`
+	rcvReady      bool
+	rcvList       pingPacketList
+	rcvBufSizeMax int
+	rcvBufSize    int
+	rcvClosed     bool
+	rcvTimestamp  bool
+
+	// The following fields are protected by the mu mutex.
+	mu         sync.RWMutex `state:"nosave"`
+	sndBufSize int
+	id         stack.TransportEndpointID
+	state      endpointState
+	bindNICID  tcpip.NICID
+	bindAddr   tcpip.Address
+	regNICID   tcpip.NICID
+	route      stack.Route `state:"manual"`
+}
+
+func newEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) *endpoint {
+	return &endpoint{
+		stack:         stack,
+		netProto:      netProto,
+		waiterQueue:   waiterQueue,
+		rcvBufSizeMax: 32 * 1024,
+		sndBufSize:    32 * 1024,
+	}
+}
+
+// Close puts the endpoint in a closed state and frees all resources
+// associated with it.
+func (e *endpoint) Close() {
+	e.mu.Lock()
+	defer e.mu.Unlock()
+
+	switch e.state {
+	case stateBound, stateConnected:
+		e.stack.UnregisterTransportEndpoint(e.regNICID, []tcpip.NetworkProtocolNumber{e.netProto}, ProtocolNumber4, e.id)
+	}
+
+	// Close the receive list and drain it.
+	e.rcvMu.Lock()
+	e.rcvClosed = true
+	e.rcvBufSize = 0
+	for !e.rcvList.Empty() {
+		p := e.rcvList.Front()
+		e.rcvList.Remove(p)
+	}
+	e.rcvMu.Unlock()
+
+	e.route.Release()
+
+	// Update the state.
+	e.state = stateClosed
+}
+
+// Read reads data from the endpoint. This method does not block if
+// there is no data pending.
+func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) {
+	e.rcvMu.Lock()
+
+	if e.rcvList.Empty() {
+		err := tcpip.ErrWouldBlock
+		if e.rcvClosed {
+			err = tcpip.ErrClosedForReceive
+		}
+		e.rcvMu.Unlock()
+		return buffer.View{}, tcpip.ControlMessages{}, err
+	}
+
+	p := e.rcvList.Front()
+	e.rcvList.Remove(p)
+	e.rcvBufSize -= p.data.Size()
+	ts := e.rcvTimestamp
+
+	e.rcvMu.Unlock()
+
+	if addr != nil {
+		*addr = p.senderAddress
+	}
+
+	if ts && !p.hasTimestamp {
+		// Linux uses the current time.
+		p.timestamp = e.stack.NowNanoseconds()
+	}
+
+	return p.data.ToView(), tcpip.ControlMessages{HasTimestamp: ts, Timestamp: p.timestamp}, nil
+}
+
+// prepareForWrite prepares the endpoint for sending data. In particular, it
+// binds it if it's still in the initial state. To do so, it must first
+// reacquire the mutex in exclusive mode.
+//
+// Returns true for retry if preparation should be retried.
+func (e *endpoint) prepareForWrite(to *tcpip.FullAddress) (retry bool, err *tcpip.Error) {
+	switch e.state {
+	case stateInitial:
+	case stateConnected:
+		return false, nil
+
+	case stateBound:
+		if to == nil {
+			return false, tcpip.ErrDestinationRequired
+		}
+		return false, nil
+	default:
+		return false, tcpip.ErrInvalidEndpointState
+	}
+
+	e.mu.RUnlock()
+	defer e.mu.RLock()
+
+	e.mu.Lock()
+	defer e.mu.Unlock()
+
+	// The state changed when we released the shared locked and re-acquired
+	// it in exclusive mode. Try again.
+	if e.state != stateInitial {
+		return true, nil
+	}
+
+	// The state is still 'initial', so try to bind the endpoint.
+	if err := e.bindLocked(tcpip.FullAddress{}, nil); err != nil {
+		return false, err
+	}
+
+	return true, nil
+}
+
+// Write writes data to the endpoint's peer. This method does not block
+// if the data cannot be written.
+func (e *endpoint) Write(p tcpip.Payload, opts tcpip.WriteOptions) (uintptr, *tcpip.Error) {
+	// MSG_MORE is unimplemented. (This also means that MSG_EOR is a no-op.)
+	if opts.More {
+		return 0, tcpip.ErrInvalidOptionValue
+	}
+
+	to := opts.To
+
+	e.mu.RLock()
+	defer e.mu.RUnlock()
+
+	// Prepare for write.
+	for {
+		retry, err := e.prepareForWrite(to)
+		if err != nil {
+			return 0, err
+		}
+
+		if !retry {
+			break
+		}
+	}
+
+	var route *stack.Route
+	if to == nil {
+		route = &e.route
+
+		if route.IsResolutionRequired() {
+			// Promote lock to exclusive if using a shared route, given that it may
+			// need to change in Route.Resolve() call below.
+			e.mu.RUnlock()
+			defer e.mu.RLock()
+
+			e.mu.Lock()
+			defer e.mu.Unlock()
+
+			// Recheck state after lock was re-acquired.
+			if e.state != stateConnected {
+				return 0, tcpip.ErrInvalidEndpointState
+			}
+		}
+	} else {
+		// Reject destination address if it goes through a different
+		// NIC than the endpoint was bound to.
+		nicid := to.NIC
+		if e.bindNICID != 0 {
+			if nicid != 0 && nicid != e.bindNICID {
+				return 0, tcpip.ErrNoRoute
+			}
+
+			nicid = e.bindNICID
+		}
+
+		toCopy := *to
+		to = &toCopy
+		netProto, err := e.checkV4Mapped(to, true)
+		if err != nil {
+			return 0, err
+		}
+
+		// Find the enpoint.
+		r, err := e.stack.FindRoute(nicid, e.bindAddr, to.Addr, netProto)
+		if err != nil {
+			return 0, err
+		}
+		defer r.Release()
+
+		route = &r
+	}
+
+	if route.IsResolutionRequired() {
+		waker := &sleep.Waker{}
+		if err := route.Resolve(waker); err != nil {
+			if err == tcpip.ErrWouldBlock {
+				// Link address needs to be resolved. Resolution was triggered the
+				// background. Better luck next time.
+				//
+				// TODO: queue up the request and send after link address
+				// is resolved.
+				route.RemoveWaker(waker)
+				return 0, tcpip.ErrNoLinkAddress
+			}
+			return 0, err
+		}
+	}
+
+	v, err := p.Get(p.Size())
+	if err != nil {
+		return 0, err
+	}
+
+	switch e.netProto {
+	case header.IPv4ProtocolNumber:
+		err = sendPing4(route, e.id.LocalPort, v)
+
+	case header.IPv6ProtocolNumber:
+		// TODO: Support IPv6.
+	}
+
+	return uintptr(len(v)), err
+}
+
+// Peek only returns data from a single datagram, so do nothing here.
+func (e *endpoint) Peek([][]byte) (uintptr, tcpip.ControlMessages, *tcpip.Error) {
+	return 0, tcpip.ControlMessages{}, nil
+}
+
+// SetSockOpt sets a socket option. Currently not supported.
+func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error {
+	switch v := opt.(type) {
+	case tcpip.TimestampOption:
+		e.rcvMu.Lock()
+		e.rcvTimestamp = v != 0
+		e.rcvMu.Unlock()
+	}
+	return nil
+}
+
+// GetSockOpt implements tcpip.Endpoint.GetSockOpt.
+func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error {
+	switch o := opt.(type) {
+	case tcpip.ErrorOption:
+		return nil
+
+	case *tcpip.SendBufferSizeOption:
+		e.mu.Lock()
+		*o = tcpip.SendBufferSizeOption(e.sndBufSize)
+		e.mu.Unlock()
+		return nil
+
+	case *tcpip.ReceiveBufferSizeOption:
+		e.rcvMu.Lock()
+		*o = tcpip.ReceiveBufferSizeOption(e.rcvBufSizeMax)
+		e.rcvMu.Unlock()
+		return nil
+
+	case *tcpip.ReceiveQueueSizeOption:
+		e.rcvMu.Lock()
+		if e.rcvList.Empty() {
+			*o = 0
+		} else {
+			p := e.rcvList.Front()
+			*o = tcpip.ReceiveQueueSizeOption(p.data.Size())
+		}
+		e.rcvMu.Unlock()
+		return nil
+
+	case *tcpip.TimestampOption:
+		e.rcvMu.Lock()
+		*o = 0
+		if e.rcvTimestamp {
+			*o = 1
+		}
+		e.rcvMu.Unlock()
+	}
+
+	return tcpip.ErrUnknownProtocolOption
+}
+
+func sendPing4(r *stack.Route, ident uint16, data buffer.View) *tcpip.Error {
+	if len(data) < header.ICMPv4EchoMinimumSize {
+		return tcpip.ErrInvalidEndpointState
+	}
+
+	// Set the ident. Sequence number is provided by the user.
+	binary.BigEndian.PutUint16(data[header.ICMPv4MinimumSize:], ident)
+
+	hdr := buffer.NewPrependable(header.ICMPv4EchoMinimumSize + int(r.MaxHeaderLength()))
+
+	icmpv4 := header.ICMPv4(hdr.Prepend(header.ICMPv4EchoMinimumSize))
+	copy(icmpv4, data)
+	data = data[header.ICMPv4EchoMinimumSize:]
+
+	// Linux performs these basic checks.
+	if icmpv4.Type() != header.ICMPv4Echo || icmpv4.Code() != 0 {
+		return tcpip.ErrInvalidEndpointState
+	}
+
+	icmpv4.SetChecksum(0)
+	icmpv4.SetChecksum(^header.Checksum(icmpv4, header.Checksum(data, 0)))
+
+	return r.WritePacket(&hdr, data, header.ICMPv4ProtocolNumber)
+}
+
+func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress, allowMismatch bool) (tcpip.NetworkProtocolNumber, *tcpip.Error) {
+	netProto := e.netProto
+	if header.IsV4MappedAddress(addr.Addr) {
+		return 0, tcpip.ErrNoRoute
+	}
+
+	// Fail if we're bound to an address length different from the one we're
+	// checking.
+	if l := len(e.id.LocalAddress); !allowMismatch && l != 0 && l != len(addr.Addr) {
+		return 0, tcpip.ErrInvalidEndpointState
+	}
+
+	return netProto, nil
+}
+
+// Connect connects the endpoint to its peer. Specifying a NIC is optional.
+func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error {
+	e.mu.Lock()
+	defer e.mu.Unlock()
+
+	nicid := addr.NIC
+	localPort := uint16(0)
+	switch e.state {
+	case stateBound, stateConnected:
+		localPort = e.id.LocalPort
+		if e.bindNICID == 0 {
+			break
+		}
+
+		if nicid != 0 && nicid != e.bindNICID {
+			return tcpip.ErrInvalidEndpointState
+		}
+
+		nicid = e.bindNICID
+	default:
+		return tcpip.ErrInvalidEndpointState
+	}
+
+	netProto, err := e.checkV4Mapped(&addr, false)
+	if err != nil {
+		return err
+	}
+
+	// Find a route to the desired destination.
+	r, err := e.stack.FindRoute(nicid, e.bindAddr, addr.Addr, netProto)
+	if err != nil {
+		return err
+	}
+	defer r.Release()
+
+	id := stack.TransportEndpointID{
+		LocalAddress:  r.LocalAddress,
+		LocalPort:     localPort,
+		RemoteAddress: r.RemoteAddress,
+	}
+
+	// Even if we're connected, this endpoint can still be used to send
+	// packets on a different network protocol, so we register both even if
+	// v6only is set to false and this is an ipv6 endpoint.
+	netProtos := []tcpip.NetworkProtocolNumber{netProto}
+
+	id, err = e.registerWithStack(nicid, netProtos, id)
+	if err != nil {
+		return err
+	}
+
+	e.id = id
+	e.route = r.Clone()
+	e.regNICID = nicid
+
+	e.state = stateConnected
+
+	e.rcvMu.Lock()
+	e.rcvReady = true
+	e.rcvMu.Unlock()
+
+	return nil
+}
+
+// ConnectEndpoint is not supported.
+func (*endpoint) ConnectEndpoint(tcpip.Endpoint) *tcpip.Error {
+	return tcpip.ErrInvalidEndpointState
+}
+
+// Shutdown closes the read and/or write end of the endpoint connection
+// to its peer.
+func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error {
+	e.mu.RLock()
+	defer e.mu.RUnlock()
+
+	if e.state != stateConnected {
+		return tcpip.ErrNotConnected
+	}
+
+	if flags&tcpip.ShutdownRead != 0 {
+		e.rcvMu.Lock()
+		wasClosed := e.rcvClosed
+		e.rcvClosed = true
+		e.rcvMu.Unlock()
+
+		if !wasClosed {
+			e.waiterQueue.Notify(waiter.EventIn)
+		}
+	}
+
+	return nil
+}
+
+// Listen is not supported by UDP, it just fails.
+func (*endpoint) Listen(int) *tcpip.Error {
+	return tcpip.ErrNotSupported
+}
+
+// Accept is not supported by UDP, it just fails.
+func (*endpoint) Accept() (tcpip.Endpoint, *waiter.Queue, *tcpip.Error) {
+	return nil, nil, tcpip.ErrNotSupported
+}
+
+func (e *endpoint) registerWithStack(nicid tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, id stack.TransportEndpointID) (stack.TransportEndpointID, *tcpip.Error) {
+	if id.LocalPort != 0 {
+		// The endpoint already has a local port, just attempt to
+		// register it.
+		err := e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber4, id, e)
+		return id, err
+	}
+
+	// We need to find a port for the endpoint.
+	_, err := e.stack.PickEphemeralPort(func(p uint16) (bool, *tcpip.Error) {
+		id.LocalPort = p
+		err := e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber4, id, e)
+		switch err {
+		case nil:
+			return true, nil
+		case tcpip.ErrPortInUse:
+			return false, nil
+		default:
+			return false, err
+		}
+	})
+
+	return id, err
+}
+
+func (e *endpoint) bindLocked(addr tcpip.FullAddress, commit func() *tcpip.Error) *tcpip.Error {
+	// Don't allow binding once endpoint is not in the initial state
+	// anymore.
+	if e.state != stateInitial {
+		return tcpip.ErrInvalidEndpointState
+	}
+
+	netProto, err := e.checkV4Mapped(&addr, false)
+	if err != nil {
+		return err
+	}
+
+	// Expand netProtos to include v4 and v6 if the caller is binding to a
+	// wildcard (empty) address, and this is an IPv6 endpoint with v6only
+	// set to false.
+	netProtos := []tcpip.NetworkProtocolNumber{netProto}
+
+	if len(addr.Addr) != 0 {
+		// A local address was specified, verify that it's valid.
+		if e.stack.CheckLocalAddress(addr.NIC, netProto, addr.Addr) == 0 {
+			return tcpip.ErrBadLocalAddress
+		}
+	}
+
+	id := stack.TransportEndpointID{
+		LocalPort:    addr.Port,
+		LocalAddress: addr.Addr,
+	}
+	id, err = e.registerWithStack(addr.NIC, netProtos, id)
+	if err != nil {
+		return err
+	}
+	if commit != nil {
+		if err := commit(); err != nil {
+			// Unregister, the commit failed.
+			e.stack.UnregisterTransportEndpoint(addr.NIC, netProtos, ProtocolNumber4, id)
+			return err
+		}
+	}
+
+	e.id = id
+	e.regNICID = addr.NIC
+
+	// Mark endpoint as bound.
+	e.state = stateBound
+
+	e.rcvMu.Lock()
+	e.rcvReady = true
+	e.rcvMu.Unlock()
+
+	return nil
+}
+
+// Bind binds the endpoint to a specific local address and port.
+// Specifying a NIC is optional.
+func (e *endpoint) Bind(addr tcpip.FullAddress, commit func() *tcpip.Error) *tcpip.Error {
+	e.mu.Lock()
+	defer e.mu.Unlock()
+
+	err := e.bindLocked(addr, commit)
+	if err != nil {
+		return err
+	}
+
+	e.bindNICID = addr.NIC
+	e.bindAddr = addr.Addr
+
+	return nil
+}
+
+// GetLocalAddress returns the address to which the endpoint is bound.
+func (e *endpoint) GetLocalAddress() (tcpip.FullAddress, *tcpip.Error) {
+	e.mu.RLock()
+	defer e.mu.RUnlock()
+
+	return tcpip.FullAddress{
+		NIC:  e.regNICID,
+		Addr: e.id.LocalAddress,
+		Port: e.id.LocalPort,
+	}, nil
+}
+
+// GetRemoteAddress returns the address to which the endpoint is connected.
+func (e *endpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Error) {
+	e.mu.RLock()
+	defer e.mu.RUnlock()
+
+	if e.state != stateConnected {
+		return tcpip.FullAddress{}, tcpip.ErrNotConnected
+	}
+
+	return tcpip.FullAddress{
+		NIC:  e.regNICID,
+		Addr: e.id.RemoteAddress,
+		Port: e.id.RemotePort,
+	}, nil
+}
+
+// Readiness returns the current readiness of the endpoint. For example, if
+// waiter.EventIn is set, the endpoint is immediately readable.
+func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask {
+	// The endpoint is always writable.
+	result := waiter.EventOut & mask
+
+	// Determine if the endpoint is readable if requested.
+	if (mask & waiter.EventIn) != 0 {
+		e.rcvMu.Lock()
+		if !e.rcvList.Empty() || e.rcvClosed {
+			result |= waiter.EventIn
+		}
+		e.rcvMu.Unlock()
+	}
+
+	return result
+}
+
+// HandlePacket is called by the stack when new packets arrive to this transport
+// endpoint.
+func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, vv *buffer.VectorisedView) {
+	e.rcvMu.Lock()
+
+	// Drop the packet if our buffer is currently full.
+	if !e.rcvReady || e.rcvClosed || e.rcvBufSize >= e.rcvBufSizeMax {
+		e.rcvMu.Unlock()
+		return
+	}
+
+	wasEmpty := e.rcvBufSize == 0
+
+	// Push new packet into receive list and increment the buffer size.
+	pkt := &pingPacket{
+		senderAddress: tcpip.FullAddress{
+			NIC:  r.NICID(),
+			Addr: id.RemoteAddress,
+		},
+	}
+	pkt.data = vv.Clone(pkt.views[:])
+	e.rcvList.PushBack(pkt)
+	e.rcvBufSize += vv.Size()
+
+	if e.rcvTimestamp {
+		pkt.timestamp = e.stack.NowNanoseconds()
+		pkt.hasTimestamp = true
+	}
+
+	e.rcvMu.Unlock()
+
+	// Notify any waiters that there's data to be read now.
+	if wasEmpty {
+		e.waiterQueue.Notify(waiter.EventIn)
+	}
+}
+
+// HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket.
+func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, vv *buffer.VectorisedView) {
+}
diff --git a/pkg/tcpip/transport/ping/endpoint_state.go b/pkg/tcpip/transport/ping/endpoint_state.go
new file mode 100644
index 000000000..e1664f049
--- /dev/null
+++ b/pkg/tcpip/transport/ping/endpoint_state.go
@@ -0,0 +1,61 @@
+// Copyright 2016 The Netstack Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package ping
+
+import (
+	"gvisor.googlesource.com/gvisor/pkg/tcpip"
+	"gvisor.googlesource.com/gvisor/pkg/tcpip/buffer"
+	"gvisor.googlesource.com/gvisor/pkg/tcpip/stack"
+)
+
+// saveData saves pingPacket.data field.
+func (p *pingPacket) saveData() buffer.VectorisedView {
+	// We cannot save p.data directly as p.data.views may alias to p.views,
+	// which is not allowed by state framework (in-struct pointer).
+	return p.data.Clone(nil)
+}
+
+// loadData loads pingPacket.data field.
+func (p *pingPacket) loadData(data buffer.VectorisedView) {
+	// NOTE: We cannot do the p.data = data.Clone(p.views[:]) optimization
+	// here because data.views is not guaranteed to be loaded by now. Plus,
+	// data.views will be allocated anyway so there really is little point
+	// of utilizing p.views for data.views.
+	p.data = data
+}
+
+// beforeSave is invoked by stateify.
+func (e *endpoint) beforeSave() {
+	// Stop incoming packets from being handled (and mutate endpoint state).
+	e.rcvMu.Lock()
+}
+
+// afterLoad is invoked by stateify.
+func (e *endpoint) afterLoad() {
+	e.stack = stack.StackFromEnv
+
+	if e.state != stateBound && e.state != stateConnected {
+		return
+	}
+
+	var err *tcpip.Error
+	if e.state == stateConnected {
+		e.route, err = e.stack.FindRoute(e.regNICID, e.bindAddr, e.id.RemoteAddress, e.netProto)
+		if err != nil {
+			panic(*err)
+		}
+
+		e.id.LocalAddress = e.route.LocalAddress
+	} else if len(e.id.LocalAddress) != 0 { // stateBound
+		if e.stack.CheckLocalAddress(e.regNICID, e.netProto, e.id.LocalAddress) == 0 {
+			panic(tcpip.ErrBadLocalAddress)
+		}
+	}
+
+	e.id, err = e.registerWithStack(e.regNICID, []tcpip.NetworkProtocolNumber{e.netProto}, e.id)
+	if err != nil {
+		panic(*err)
+	}
+}
diff --git a/pkg/tcpip/transport/ping/protocol.go b/pkg/tcpip/transport/ping/protocol.go
new file mode 100644
index 000000000..1459b4d60
--- /dev/null
+++ b/pkg/tcpip/transport/ping/protocol.go
@@ -0,0 +1,106 @@
+// Copyright 2016 The Netstack Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package ping contains the implementation of the ICMP and IPv6-ICMP transport
+// protocols for use in ping. To use it in the networking stack, this package
+// must be added to the project, and
+// activated on the stack by passing ping.ProtocolName (or "ping") and/or
+// ping.ProtocolName6 (or "ping6") as one of the transport protocols when
+// calling stack.New(). Then endpoints can be created by passing
+// ping.ProtocolNumber or ping.ProtocolNumber6 as the transport protocol number
+// when calling Stack.NewEndpoint().
+package ping
+
+import (
+	"encoding/binary"
+	"fmt"
+
+	"gvisor.googlesource.com/gvisor/pkg/tcpip"
+	"gvisor.googlesource.com/gvisor/pkg/tcpip/buffer"
+	"gvisor.googlesource.com/gvisor/pkg/tcpip/header"
+	"gvisor.googlesource.com/gvisor/pkg/tcpip/stack"
+	"gvisor.googlesource.com/gvisor/pkg/waiter"
+)
+
+const (
+	// ProtocolName4 is the string representation of the ping protocol name.
+	ProtocolName4 = "ping4"
+
+	// ProtocolNumber4 is the ICMP protocol number.
+	ProtocolNumber4 = header.ICMPv4ProtocolNumber
+
+	// ProtocolName6 is the string representation of the ping protocol name.
+	ProtocolName6 = "ping6"
+
+	// ProtocolNumber6 is the IPv6-ICMP protocol number.
+	ProtocolNumber6 = header.ICMPv6ProtocolNumber
+)
+
+type protocol struct {
+	number tcpip.TransportProtocolNumber
+}
+
+// Number returns the ICMP protocol number.
+func (p *protocol) Number() tcpip.TransportProtocolNumber {
+	return p.number
+}
+
+func (p *protocol) netProto() tcpip.NetworkProtocolNumber {
+	switch p.number {
+	case ProtocolNumber4:
+		return header.IPv4ProtocolNumber
+	case ProtocolNumber6:
+		return header.IPv6ProtocolNumber
+	}
+	panic(fmt.Sprint("unknown protocol number: ", p.number))
+}
+
+// NewEndpoint creates a new ping endpoint.
+func (p *protocol) NewEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) {
+	if netProto != p.netProto() {
+		return nil, tcpip.ErrUnknownProtocol
+	}
+	return newEndpoint(stack, netProto, waiterQueue), nil
+}
+
+// MinimumPacketSize returns the minimum valid ping packet size.
+func (p *protocol) MinimumPacketSize() int {
+	switch p.number {
+	case ProtocolNumber4:
+		return header.ICMPv4EchoMinimumSize
+	case ProtocolNumber6:
+		return header.ICMPv6EchoMinimumSize
+	}
+	panic(fmt.Sprint("unknown protocol number: ", p.number))
+}
+
+// ParsePorts returns the source and destination ports stored in the given udp
+// packet.
+func (*protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) {
+	return 0, binary.BigEndian.Uint16(v[header.ICMPv4MinimumSize:]), nil
+}
+
+// HandleUnknownDestinationPacket handles packets targeted at this protocol but
+// that don't match any existing endpoint.
+func (p *protocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, *buffer.VectorisedView) bool {
+	return true
+}
+
+// SetOption implements TransportProtocol.SetOption.
+func (p *protocol) SetOption(option interface{}) *tcpip.Error {
+	return tcpip.ErrUnknownProtocolOption
+}
+
+// Option implements TransportProtocol.Option.
+func (p *protocol) Option(option interface{}) *tcpip.Error {
+	return tcpip.ErrUnknownProtocolOption
+}
+
+func init() {
+	stack.RegisterTransportProtocolFactory(ProtocolName4, func() stack.TransportProtocol {
+		return &protocol{ProtocolNumber4}
+	})
+
+	// TODO: Support IPv6.
+}
diff --git a/pkg/tcpip/transport/udp/endpoint_state.go b/pkg/tcpip/transport/udp/endpoint_state.go
index 41b98424a..e20d59ca3 100644
--- a/pkg/tcpip/transport/udp/endpoint_state.go
+++ b/pkg/tcpip/transport/udp/endpoint_state.go
@@ -13,7 +13,7 @@ import (
 
 // saveData saves udpPacket.data field.
 func (u *udpPacket) saveData() buffer.VectorisedView {
-	// We canoot save u.data directly as u.data.views may alias to u.views,
+	// We cannot save u.data directly as u.data.views may alias to u.views,
 	// which is not allowed by state framework (in-struct pointer).
 	return u.data.Clone(nil)
 }