Resend packets back to netstack if destined to itself

Add option to redirect packet back to netstack if it's destined to itself.
This fixes the problem where connecting to the local NIC address would
not work, e.g.:
echo bar | nc -l -p 8080 &
echo foo | nc 192.168.0.2 8080

PiperOrigin-RevId: 207995083
Change-Id: I17adc2a04df48bfea711011a5df206326a1fb8ef

7 files changed, 107 insertions, 49 deletions

diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go
index 4e20cfbf8..152d8f0b2 100644
--- a/pkg/tcpip/link/fdbased/endpoint.go
+++ b/pkg/tcpip/link/fdbased/endpoint.go
@@ -55,10 +55,15 @@ type endpoint struct {
 	// its end of the communication pipe.
 	closed func(*tcpip.Error)
 
-	vv       *buffer.VectorisedView
-	iovecs   []syscall.Iovec
-	views    []buffer.View
-	attached bool
+	vv         *buffer.VectorisedView
+	iovecs     []syscall.Iovec
+	views      []buffer.View
+	dispatcher stack.NetworkDispatcher
+
+	// handleLocal indicates whether packets destined to itself should be
+	// handled by the netstack internally (true) or be forwarded to the FD
+	// endpoint (false).
+	handleLocal bool
 }
 
 // Options specify the details about the fd-based endpoint to be created.
@@ -71,6 +76,7 @@ type Options struct {
 	Address         tcpip.LinkAddress
 	SaveRestore     bool
 	DisconnectOk    bool
+	HandleLocal     bool
 }
 
 // New creates a new fd-based endpoint.
@@ -100,14 +106,15 @@ func New(opts *Options) tcpip.LinkEndpointID {
 	}
 
 	e := &endpoint{
-		fd:      opts.FD,
-		mtu:     opts.MTU,
-		caps:    caps,
-		closed:  opts.ClosedFunc,
-		addr:    opts.Address,
-		hdrSize: hdrSize,
-		views:   make([]buffer.View, len(BufConfig)),
-		iovecs:  make([]syscall.Iovec, len(BufConfig)),
+		fd:          opts.FD,
+		mtu:         opts.MTU,
+		caps:        caps,
+		closed:      opts.ClosedFunc,
+		addr:        opts.Address,
+		hdrSize:     hdrSize,
+		views:       make([]buffer.View, len(BufConfig)),
+		iovecs:      make([]syscall.Iovec, len(BufConfig)),
+		handleLocal: opts.HandleLocal,
 	}
 	vv := buffer.NewVectorisedView(0, e.views)
 	e.vv = &vv
@@ -117,16 +124,16 @@ func New(opts *Options) tcpip.LinkEndpointID {
 // Attach launches the goroutine that reads packets from the file descriptor and
 // dispatches them via the provided dispatcher.
 func (e *endpoint) Attach(dispatcher stack.NetworkDispatcher) {
-	e.attached = true
+	e.dispatcher = dispatcher
 	// Link endpoints are not savable. When transportation endpoints are
 	// saved, they stop sending outgoing packets and all incoming packets
 	// are rejected.
-	go e.dispatchLoop(dispatcher) // S/R-SAFE: See above.
+	go e.dispatchLoop() // S/R-SAFE: See above.
 }
 
 // IsAttached implements stack.LinkEndpoint.IsAttached.
 func (e *endpoint) IsAttached() bool {
-	return e.attached
+	return e.dispatcher != nil
 }
 
 // MTU implements stack.LinkEndpoint.MTU. It returns the value initialized
@@ -153,6 +160,12 @@ func (e *endpoint) LinkAddress() tcpip.LinkAddress {
 // WritePacket writes outbound packets to the file descriptor. If it is not
 // currently writable, the packet is dropped.
 func (e *endpoint) WritePacket(r *stack.Route, hdr *buffer.Prependable, payload buffer.View, protocol tcpip.NetworkProtocolNumber) *tcpip.Error {
+	if e.handleLocal && r.LocalAddress != "" && r.LocalAddress == r.RemoteAddress {
+		hdrView := hdr.View()
+		vv := buffer.NewVectorisedView(len(hdrView)+len(payload), []buffer.View{hdrView, payload})
+		e.dispatcher.DeliverNetworkPacket(e, r.RemoteLinkAddress, protocol, &vv)
+		return nil
+	}
 	if e.hdrSize > 0 {
 		// Add ethernet header if needed.
 		eth := header.Ethernet(hdr.Prepend(header.EthernetMinimumSize))
@@ -165,7 +178,6 @@ func (e *endpoint) WritePacket(r *stack.Route, hdr *buffer.Prependable, payload
 
 	if len(payload) == 0 {
 		return rawfile.NonBlockingWrite(e.fd, hdr.UsedBytes())
-
 	}
 
 	return rawfile.NonBlockingWrite2(e.fd, hdr.UsedBytes(), payload)
@@ -198,7 +210,7 @@ func (e *endpoint) allocateViews(bufConfig []int) {
 }
 
 // dispatch reads one packet from the file descriptor and dispatches it.
-func (e *endpoint) dispatch(d stack.NetworkDispatcher, largeV buffer.View) (bool, *tcpip.Error) {
+func (e *endpoint) dispatch(largeV buffer.View) (bool, *tcpip.Error) {
 	e.allocateViews(BufConfig)
 
 	n, err := rawfile.BlockingReadv(e.fd, e.iovecs)
@@ -234,7 +246,7 @@ func (e *endpoint) dispatch(d stack.NetworkDispatcher, largeV buffer.View) (bool
 	e.vv.SetSize(n)
 	e.vv.TrimFront(e.hdrSize)
 
-	d.DeliverNetworkPacket(e, addr, p, e.vv)
+	e.dispatcher.DeliverNetworkPacket(e, addr, p, e.vv)
 
 	// Prepare e.views for another packet: release used views.
 	for i := 0; i < used; i++ {
@@ -246,10 +258,10 @@ func (e *endpoint) dispatch(d stack.NetworkDispatcher, largeV buffer.View) (bool
 
 // dispatchLoop reads packets from the file descriptor in a loop and dispatches
 // them to the network stack.
-func (e *endpoint) dispatchLoop(d stack.NetworkDispatcher) *tcpip.Error {
+func (e *endpoint) dispatchLoop() *tcpip.Error {
 	v := buffer.NewView(header.MaxIPPacketSize)
 	for {
-		cont, err := e.dispatch(d, v)
+		cont, err := e.dispatch(v)
 		if err != nil || !cont {
 			if e.closed != nil {
 				e.closed(err)
diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go
index 25c06cba5..c1480f97b 100644
--- a/pkg/tcpip/stack/nic.go
+++ b/pkg/tcpip/stack/nic.go
@@ -327,8 +327,7 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remoteLinkAddr tcpip.Lin
 		return
 	}
 
-	r := makeRoute(protocol, dst, src, ref)
-	r.LocalLinkAddress = linkEP.LinkAddress()
+	r := makeRoute(protocol, dst, src, linkEP.LinkAddress(), ref)
 	r.RemoteLinkAddress = remoteLinkAddr
 	ref.ep.HandlePacket(&r, vv)
 	ref.decRef()
diff --git a/pkg/tcpip/stack/route.go b/pkg/tcpip/stack/route.go
index 200c39289..423f428df 100644
--- a/pkg/tcpip/stack/route.go
+++ b/pkg/tcpip/stack/route.go
@@ -50,12 +50,13 @@ type Route struct {
 
 // makeRoute initializes a new route. It takes ownership of the provided
 // reference to a network endpoint.
-func makeRoute(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr tcpip.Address, ref *referencedNetworkEndpoint) Route {
+func makeRoute(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr tcpip.Address, localLinkAddr tcpip.LinkAddress, ref *referencedNetworkEndpoint) Route {
 	return Route{
-		NetProto:      netProto,
-		LocalAddress:  localAddr,
-		RemoteAddress: remoteAddr,
-		ref:           ref,
+		NetProto:         netProto,
+		LocalAddress:     localAddr,
+		LocalLinkAddress: localLinkAddr,
+		RemoteAddress:    remoteAddr,
+		ref:              ref,
 	}
 }
 
@@ -92,6 +93,11 @@ func (r *Route) Resolve(waker *sleep.Waker) *tcpip.Error {
 
 	nextAddr := r.NextHop
 	if nextAddr == "" {
+		// Local link address is already known.
+		if r.RemoteAddress == r.LocalAddress {
+			r.RemoteLinkAddress = r.LocalLinkAddress
+			return nil
+		}
 		nextAddr = r.RemoteAddress
 	}
 	linkAddr, err := r.ref.linkCache.GetLinkAddress(r.ref.nic.ID(), nextAddr, r.LocalAddress, r.NetProto, waker)
diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go
index fa7aeb051..6c4aa7cc5 100644
--- a/pkg/tcpip/stack/stack.go
+++ b/pkg/tcpip/stack/stack.go
@@ -687,7 +687,7 @@ func (s *Stack) FindRoute(id tcpip.NICID, localAddr, remoteAddr tcpip.Address, n
 			remoteAddr = ref.ep.ID().LocalAddress
 		}
 
-		r := makeRoute(netProto, ref.ep.ID().LocalAddress, remoteAddr, ref)
+		r := makeRoute(netProto, ref.ep.ID().LocalAddress, remoteAddr, nic.linkEP.LinkAddress(), ref)
 		r.NextHop = s.routeTable[i].Gateway
 		return r, nil
 	}
diff --git a/runsc/boot/network.go b/runsc/boot/network.go
index d2b52c823..d702ae74e 100644
--- a/runsc/boot/network.go
+++ b/runsc/boot/network.go
@@ -134,11 +134,11 @@ func (n *Network) CreateLinksAndRoutes(args *CreateLinksAndRoutesArgs, _ *struct
 		}
 
 		linkEP := fdbased.New(&fdbased.Options{
-			FD:              newFD,
-			MTU:             uint32(link.MTU),
-			ChecksumOffload: false,
-			EthernetHeader:  true,
-			Address:         tcpip.LinkAddress(generateRndMac()),
+			FD:             newFD,
+			MTU:            uint32(link.MTU),
+			EthernetHeader: true,
+			HandleLocal:    true,
+			Address:        tcpip.LinkAddress(generateRndMac()),
 		})
 
 		log.Infof("Enabling interface %q with id %d on addresses %+v", link.Name, nicID, link.Addresses)
diff --git a/runsc/test/integration/integration_test.go b/runsc/test/integration/integration_test.go
index 67b58523d..c286e48d2 100644
--- a/runsc/test/integration/integration_test.go
+++ b/runsc/test/integration/integration_test.go
@@ -31,6 +31,7 @@ import (
 	"net"
 	"net/http"
 	"os"
+	"strings"
 	"testing"
 	"time"
 
@@ -54,36 +55,36 @@ func httpRequestSucceeds(client http.Client, server string, port int) error {
 // TestLifeCycle tests a basic Create/Start/Stop docker container life cycle.
 func TestLifeCycle(t *testing.T) {
 	if err := testutil.Pull("nginx"); err != nil {
-		t.Fatalf("docker pull failed: %v", err)
+		t.Fatal("docker pull failed:", err)
 	}
 	d := testutil.MakeDocker("lifecycle-test")
 	if err := d.Create("-p", "80", "nginx"); err != nil {
-		t.Fatalf("docker create failed: %v", err)
+		t.Fatal("docker create failed:", err)
 	}
 	if err := d.Start(); err != nil {
 		d.CleanUp()
-		t.Fatalf("docker start failed: %v", err)
+		t.Fatal("docker start failed:", err)
 	}
 
 	// Test that container is working
 	port, err := d.FindPort(80)
 	if err != nil {
-		t.Fatalf("docker.FindPort(80) failed: %v", err)
+		t.Fatal("docker.FindPort(80) failed: ", err)
 	}
 	if err := testutil.WaitForHTTP(port, 5*time.Second); err != nil {
-		t.Fatalf("WaitForHTTP() timeout: %v", err)
+		t.Fatal("WaitForHTTP() timeout:", err)
 	}
 	client := http.Client{Timeout: time.Duration(2 * time.Second)}
 	if err := httpRequestSucceeds(client, "localhost", port); err != nil {
-		t.Errorf("http request failed: %v", err)
+		t.Error("http request failed:", err)
 	}
 
 	if err := d.Stop(); err != nil {
 		d.CleanUp()
-		t.Fatalf("docker stop failed: %v", err)
+		t.Fatal("docker stop failed:", err)
 	}
 	if err := d.Remove(); err != nil {
-		t.Fatalf("docker rm failed: %v", err)
+		t.Fatal("docker rm failed:", err)
 	}
 }
 
@@ -94,7 +95,7 @@ func TestPauseResume(t *testing.T) {
 	}
 
 	if err := testutil.Pull("google/python-hello"); err != nil {
-		t.Fatalf("docker pull failed: %v", err)
+		t.Fatal("docker pull failed:", err)
 	}
 	d := testutil.MakeDocker("pause-resume-test")
 	if out, err := d.Run("-p", "8080", "google/python-hello"); err != nil {
@@ -105,22 +106,22 @@ func TestPauseResume(t *testing.T) {
 	// Find where port 8080 is mapped to.
 	port, err := d.FindPort(8080)
 	if err != nil {
-		t.Fatalf("docker.FindPort(8080) failed: %v", err)
+		t.Fatal("docker.FindPort(8080) failed:", err)
 	}
 
 	// Wait until it's up and running.
 	if err := testutil.WaitForHTTP(port, 20*time.Second); err != nil {
-		t.Fatalf("WaitForHTTP() timeout: %v", err)
+		t.Fatal("WaitForHTTP() timeout:", err)
 	}
 
 	// Check that container is working.
 	client := http.Client{Timeout: time.Duration(2 * time.Second)}
 	if err := httpRequestSucceeds(client, "localhost", port); err != nil {
-		t.Errorf("http request failed: %v", err)
+		t.Error("http request failed:", err)
 	}
 
 	if err := d.Pause(); err != nil {
-		t.Fatalf("docker pause failed: %v", err)
+		t.Fatal("docker pause failed:", err)
 	}
 
 	// Check if container is paused.
@@ -136,17 +137,50 @@ func TestPauseResume(t *testing.T) {
 	}
 
 	if err := d.Unpause(); err != nil {
-		t.Fatalf("docker unpause failed: %v", err)
+		t.Fatal("docker unpause failed:", err)
 	}
 
 	// Wait until it's up and running.
 	if err := testutil.WaitForHTTP(port, 20*time.Second); err != nil {
-		t.Fatalf("WaitForHTTP() timeout: %v", err)
+		t.Fatal("WaitForHTTP() timeout:", err)
 	}
 
 	// Check if container is working again.
 	if err := httpRequestSucceeds(client, "localhost", port); err != nil {
-		t.Errorf("http request failed: %v", err)
+		t.Error("http request failed:", err)
+	}
+}
+
+// Create client and server that talk to each other using the local IP.
+func TestConnectToSelf(t *testing.T) {
+	d := testutil.MakeDocker("connect-to-self-test")
+
+	// Creates server that replies "server" and exists. Sleeps at the end because
+	// 'docker exec' gets killed if the init process exists before it can finish.
+	if _, err := d.Run("ubuntu:trusty", "/bin/sh", "-c", "echo server | nc -l -p 8080 && sleep 1"); err != nil {
+		t.Fatal("docker run failed:", err)
+	}
+	defer d.CleanUp()
+
+	// Finds IP address for eth0.
+	ip, err := d.Exec("/bin/sh", "-c", "ifconfig eth0 | grep -E -o \".*inet [^ ]+\" | cut -d: -f2")
+	if err != nil {
+		t.Fatal("docker exec failed:", err)
+	}
+	ip = strings.TrimRight(ip, "\n")
+
+	// Runs client that sends "client" to the server and exits.
+	reply, err := d.Exec("/bin/sh", "-c", fmt.Sprintf("echo client | nc %s 8080", ip))
+	if err != nil {
+		t.Fatal("docker exec failed:", err)
+	}
+
+	// Ensure both client and server got the message from each other.
+	if want := "server\n"; reply != want {
+		t.Errorf("Error on server, want: %q, got: %q", want, reply)
+	}
+	if err := d.WaitForOutput("^client\n$", 1*time.Second); err != nil {
+		t.Fatal("docker.WaitForOutput(client) timeout:", err)
 	}
 }
 
diff --git a/runsc/test/testutil/docker.go b/runsc/test/testutil/docker.go
index 6825ed9ec..b7d60e712 100644
--- a/runsc/test/testutil/docker.go
+++ b/runsc/test/testutil/docker.go
@@ -162,6 +162,13 @@ func (d *Docker) Run(args ...string) (string, error) {
 	return do(a...)
 }
 
+// Exec calls 'docker exec' with the arguments provided.
+func (d *Docker) Exec(args ...string) (string, error) {
+	a := []string{"exec", d.Name}
+	a = append(a, args...)
+	return do(a...)
+}
+
 // Pause calls 'docker pause'.
 func (d *Docker) Pause() error {
 	if _, err := do("pause", d.Name); err != nil {