Test TCP sender behavior against window shrinking

RFC 1122 Section 3.7: A sending TCP MUST be robust against window shrinking,
which may cause the "useable window" to become negative.

PiperOrigin-RevId: 305377072

6 files changed, 220 insertions, 7 deletions

diff --git a/test/packetimpact/dut/posix_server.cc b/test/packetimpact/dut/posix_server.cc
index b8177f5b1..86e580c6f 100644
--- a/test/packetimpact/dut/posix_server.cc
+++ b/test/packetimpact/dut/posix_server.cc
@@ -149,6 +149,15 @@ class PosixImpl final : public posix_server::Posix::Service {
     return ::grpc::Status::OK;
   }
 
+  ::grpc::Status Send(::grpc::ServerContext *context,
+                      const ::posix_server::SendRequest *request,
+                      ::posix_server::SendResponse *response) override {
+    response->set_ret(::send(request->sockfd(), request->buf().data(),
+                             request->buf().size(), request->flags()));
+    response->set_errno_(errno);
+    return ::grpc::Status::OK;
+  }
+
   ::grpc::Status SetSockOpt(
       grpc_impl::ServerContext *context,
       const ::posix_server::SetSockOptRequest *request,
@@ -160,6 +169,17 @@ class PosixImpl final : public posix_server::Posix::Service {
     return ::grpc::Status::OK;
   }
 
+  ::grpc::Status SetSockOptInt(
+      ::grpc::ServerContext *context,
+      const ::posix_server::SetSockOptIntRequest *request,
+      ::posix_server::SetSockOptIntResponse *response) override {
+    int opt = request->intval();
+    response->set_ret(::setsockopt(request->sockfd(), request->level(),
+                                   request->optname(), &opt, sizeof(opt)));
+    response->set_errno_(errno);
+    return ::grpc::Status::OK;
+  }
+
   ::grpc::Status SetSockOptTimeval(
       ::grpc::ServerContext *context,
       const ::posix_server::SetSockOptTimevalRequest *request,
diff --git a/test/packetimpact/proto/posix_server.proto b/test/packetimpact/proto/posix_server.proto
index 1565f31fa..4035e1ee6 100644
--- a/test/packetimpact/proto/posix_server.proto
+++ b/test/packetimpact/proto/posix_server.proto
@@ -93,6 +93,17 @@ message ListenResponse {
   int32 errno_ = 2;  // "errno" may fail to compile in c++.
 }
 
+message SendRequest {
+  int32 sockfd = 1;
+  bytes buf = 2;
+  int32 flags = 3;
+}
+
+message SendResponse {
+  int32 ret = 1;
+  int32 errno_ = 2;
+}
+
 message SetSockOptRequest {
   int32 sockfd = 1;
   int32 level = 2;
@@ -105,6 +116,18 @@ message SetSockOptResponse {
   int32 errno_ = 2;  // "errno" may fail to compile in c++.
 }
 
+message SetSockOptIntRequest {
+  int32 sockfd = 1;
+  int32 level = 2;
+  int32 optname = 3;
+  int32 intval = 4;
+}
+
+message SetSockOptIntResponse {
+  int32 ret = 1;
+  int32 errno_ = 2;
+}
+
 message SetSockOptTimevalRequest {
   int32 sockfd = 1;
   int32 level = 2;
@@ -151,11 +174,15 @@ service Posix {
   rpc GetSockName(GetSockNameRequest) returns (GetSockNameResponse);
   // Call listen() on the DUT.
   rpc Listen(ListenRequest) returns (ListenResponse);
+  // Call send() on the DUT.
+  rpc Send(SendRequest) returns (SendResponse);
   // Call setsockopt() on the DUT.  You should prefer one of the other
   // SetSockOpt* functions with a more structured optval or else you may get the
   // encoding wrong, such as making a bad assumption about the server's word
   // sizes or endianness.
   rpc SetSockOpt(SetSockOptRequest) returns (SetSockOptResponse);
+  // Call setsockopt() on the DUT with an int optval.
+  rpc SetSockOptInt(SetSockOptIntRequest) returns (SetSockOptIntResponse);
   // Call setsockopt() on the DUT with a Timeval optval.
   rpc SetSockOptTimeval(SetSockOptTimevalRequest)
       returns (SetSockOptTimevalResponse);
diff --git a/test/packetimpact/testbench/connections.go b/test/packetimpact/testbench/connections.go
index 8d1f562ee..579da59c3 100644
--- a/test/packetimpact/testbench/connections.go
+++ b/test/packetimpact/testbench/connections.go
@@ -187,9 +187,19 @@ func (conn *TCPIPv4) Send(tcp TCP, additionalLayers ...Layer) {
 	conn.SendFrame(conn.CreateFrame(tcp, additionalLayers...))
 }
 
-// Recv gets a packet from the sniffer within the timeout provided. If no packet
-// arrives before the timeout, it returns nil.
+// Recv gets a packet from the sniffer within the timeout provided.
+// If no packet arrives before the timeout, it returns nil.
 func (conn *TCPIPv4) Recv(timeout time.Duration) *TCP {
+	layers := conn.RecvFrame(timeout)
+	if tcpLayerIndex < len(layers) {
+		return layers[tcpLayerIndex].(*TCP)
+	}
+	return nil
+}
+
+// RecvFrame gets a frame (of type Layers) within the timeout provided.
+// If no frame arrives before the timeout, it returns nil.
+func (conn *TCPIPv4) RecvFrame(timeout time.Duration) Layers {
 	deadline := time.Now().Add(timeout)
 	for {
 		timeout = time.Until(deadline)
@@ -216,14 +226,16 @@ func (conn *TCPIPv4) Recv(timeout time.Duration) *TCP {
 		for i := tcpLayerIndex + 1; i < len(layers); i++ {
 			conn.RemoteSeqNum.UpdateForward(seqnum.Size(layers[i].length()))
 		}
-		return tcpHeader
+		return layers
 	}
 	return nil
 }
 
 // Expect a packet that matches the provided tcp within the timeout specified.
-// If it doesn't arrive in time, the test fails.
+// If it doesn't arrive in time, it returns nil.
 func (conn *TCPIPv4) Expect(tcp TCP, timeout time.Duration) *TCP {
+	// We cannot implement this directly using ExpectFrame as we cannot specify
+	// the Payload part.
 	deadline := time.Now().Add(timeout)
 	for {
 		timeout = time.Until(deadline)
@@ -231,15 +243,41 @@ func (conn *TCPIPv4) Expect(tcp TCP, timeout time.Duration) *TCP {
 			return nil
 		}
 		gotTCP := conn.Recv(timeout)
-		if gotTCP == nil {
-			return nil
-		}
 		if tcp.match(gotTCP) {
 			return gotTCP
 		}
 	}
 }
 
+// ExpectFrame expects a frame that matches the specified layers within the
+// timeout specified. If it doesn't arrive in time, it returns nil.
+func (conn *TCPIPv4) ExpectFrame(layers Layers, timeout time.Duration) Layers {
+	deadline := time.Now().Add(timeout)
+	for {
+		timeout = time.Until(deadline)
+		if timeout <= 0 {
+			return nil
+		}
+		gotLayers := conn.RecvFrame(timeout)
+		if layers.match(gotLayers) {
+			return gotLayers
+		}
+	}
+}
+
+// ExpectData is a convenient method that expects a TCP packet along with
+// the payload to arrive within the timeout specified. If it doesn't arrive
+// in time, it causes a fatal test failure.
+func (conn *TCPIPv4) ExpectData(tcp TCP, data []byte, timeout time.Duration) {
+	expected := []Layer{&Ether{}, &IPv4{}, &tcp}
+	if len(data) > 0 {
+		expected = append(expected, &Payload{Bytes: data})
+	}
+	if conn.ExpectFrame(expected, timeout) == nil {
+		conn.t.Fatalf("expected to get a TCP frame %s with payload %x", &tcp, data)
+	}
+}
+
 // Handshake performs a TCP 3-way handshake.
 func (conn *TCPIPv4) Handshake() {
 	// Send the SYN.
diff --git a/test/packetimpact/testbench/dut.go b/test/packetimpact/testbench/dut.go
index f342aee01..9335909c0 100644
--- a/test/packetimpact/testbench/dut.go
+++ b/test/packetimpact/testbench/dut.go
@@ -291,6 +291,35 @@ func (dut *DUT) ListenWithErrno(ctx context.Context, sockfd, backlog int32) (int
 	return resp.GetRet(), syscall.Errno(resp.GetErrno_())
 }
 
+// Send calls send on the DUT and causes a fatal test failure if it doesn't
+// succeed. If more control over the timeout or error handling is needed, use
+// SendWithErrno.
+func (dut *DUT) Send(sockfd int32, buf []byte, flags int32) int32 {
+	dut.t.Helper()
+	ctx, cancel := context.WithTimeout(context.Background(), *rpcTimeout)
+	defer cancel()
+	ret, err := dut.SendWithErrno(ctx, sockfd, buf, flags)
+	if ret == -1 {
+		dut.t.Fatalf("failed to send: %s", err)
+	}
+	return ret
+}
+
+// SendWithErrno calls send on the DUT.
+func (dut *DUT) SendWithErrno(ctx context.Context, sockfd int32, buf []byte, flags int32) (int32, error) {
+	dut.t.Helper()
+	req := pb.SendRequest{
+		Sockfd: sockfd,
+		Buf:    buf,
+		Flags:  flags,
+	}
+	resp, err := dut.posixServer.Send(ctx, &req)
+	if err != nil {
+		dut.t.Fatalf("failed to call Send: %s", err)
+	}
+	return resp.GetRet(), syscall.Errno(resp.GetErrno_())
+}
+
 // SetSockOpt calls setsockopt on the DUT and causes a fatal test failure if it
 // doesn't succeed. If more control over the timeout or error handling is
 // needed, use SetSockOptWithErrno. Because endianess and the width of values
@@ -324,6 +353,35 @@ func (dut *DUT) SetSockOptWithErrno(ctx context.Context, sockfd, level, optname
 	return resp.GetRet(), syscall.Errno(resp.GetErrno_())
 }
 
+// SetSockOptInt calls setsockopt on the DUT and causes a fatal test failure
+// if it doesn't succeed. If more control over the int optval or error handling
+// is needed, use SetSockOptIntWithErrno.
+func (dut *DUT) SetSockOptInt(sockfd, level, optname, optval int32) {
+	dut.t.Helper()
+	ctx, cancel := context.WithTimeout(context.Background(), *rpcTimeout)
+	defer cancel()
+	ret, err := dut.SetSockOptIntWithErrno(ctx, sockfd, level, optname, optval)
+	if ret != 0 {
+		dut.t.Fatalf("failed to SetSockOptInt: %s", err)
+	}
+}
+
+// SetSockOptIntWithErrno calls setsockopt with an integer optval.
+func (dut *DUT) SetSockOptIntWithErrno(ctx context.Context, sockfd, level, optname, optval int32) (int32, error) {
+	dut.t.Helper()
+	req := pb.SetSockOptIntRequest{
+		Sockfd:  sockfd,
+		Level:   level,
+		Optname: optname,
+		Intval:  optval,
+	}
+	resp, err := dut.posixServer.SetSockOptInt(ctx, &req)
+	if err != nil {
+		dut.t.Fatalf("failed to call SetSockOptInt: %s", err)
+	}
+	return resp.GetRet(), syscall.Errno(resp.GetErrno_())
+}
+
 // SetSockOptTimeval calls setsockopt on the DUT and causes a fatal test failure
 // if it doesn't succeed. If more control over the timeout or error handling is
 // needed, use SetSockOptTimevalWithErrno.
diff --git a/test/packetimpact/tests/BUILD b/test/packetimpact/tests/BUILD
index 9a4d66ea9..a9b2de9b9 100644
--- a/test/packetimpact/tests/BUILD
+++ b/test/packetimpact/tests/BUILD
@@ -28,6 +28,18 @@ packetimpact_go_test(
     ],
 )
 
+packetimpact_go_test(
+    name = "tcp_window_shrink",
+    srcs = ["tcp_window_shrink_test.go"],
+    # TODO(b/153202472): Fix netstack then remove the line below.
+    netstack = False,
+    deps = [
+        "//pkg/tcpip/header",
+        "//test/packetimpact/testbench",
+        "@org_golang_x_sys//unix:go_default_library",
+    ],
+)
+
 sh_binary(
     name = "test_runner",
     srcs = ["test_runner.sh"],
diff --git a/test/packetimpact/tests/tcp_window_shrink_test.go b/test/packetimpact/tests/tcp_window_shrink_test.go
new file mode 100644
index 000000000..b48cc6491
--- /dev/null
+++ b/test/packetimpact/tests/tcp_window_shrink_test.go
@@ -0,0 +1,58 @@
+// 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_window_shrink_test
+
+import (
+	"testing"
+	"time"
+
+	"golang.org/x/sys/unix"
+	"gvisor.dev/gvisor/pkg/tcpip/header"
+	tb "gvisor.dev/gvisor/test/packetimpact/testbench"
+)
+
+func TestWindowShrink(t *testing.T) {
+	dut := tb.NewDUT(t)
+	defer dut.TearDown()
+	listenFd, remotePort := dut.CreateListener(unix.SOCK_STREAM, unix.IPPROTO_TCP, 1)
+	defer dut.Close(listenFd)
+	conn := tb.NewTCPIPv4(t, tb.TCP{DstPort: &remotePort}, tb.TCP{SrcPort: &remotePort})
+	defer conn.Close()
+
+	conn.Handshake()
+	acceptFd, _ := dut.Accept(listenFd)
+	defer dut.Close(acceptFd)
+
+	dut.SetSockOptInt(acceptFd, unix.IPPROTO_TCP, unix.TCP_NODELAY, 1)
+
+	sampleData := []byte("Sample Data")
+
+	dut.Send(acceptFd, sampleData, 0)
+	conn.ExpectData(tb.TCP{}, sampleData, time.Second)
+	conn.Send(tb.TCP{Flags: tb.Uint8(header.TCPFlagAck)})
+
+	dut.Send(acceptFd, sampleData, 0)
+	dut.Send(acceptFd, sampleData, 0)
+	conn.ExpectData(tb.TCP{}, sampleData, time.Second)
+	conn.ExpectData(tb.TCP{}, sampleData, time.Second)
+	// We close our receiving window here
+	conn.Send(tb.TCP{Flags: tb.Uint8(header.TCPFlagAck), WindowSize: tb.Uint16(0)})
+
+	dut.Send(acceptFd, []byte("Sample Data"), 0)
+	// Note: There is another kind of zero-window probing which Windows uses (by sending one
+	// new byte at `RemoteSeqNum`), if netstack wants to go that way, we may want to change
+	// the following lines.
+	conn.ExpectData(tb.TCP{SeqNum: tb.Uint32(uint32(conn.RemoteSeqNum - 1))}, nil, time.Second)
+}