netstack: Refactor tcpip.Endpoint.Read

Read now takes a destination io.Writer, count, options. Keeping the method name
Read, in contrast to the Write method.

This enables:
* direct transfer of views under VV
* zero copy

It also eliminates the need for sentry to keep a slice of view because
userspace had requested a read that is smaller than the view returned, removing
the complexity there.

Read/Peek/ReadPacket are now consolidated together and some duplicate code is
removed.

PiperOrigin-RevId: 350636322

pkg/sentry/socket/netstack/BUILD

@@ -25,7 +25,6 @@ go_library(
         "//pkg/marshal",
         "//pkg/marshal/primitive",
         "//pkg/metric",
-        "//pkg/safemem",
         "//pkg/sentry/arch",
         "//pkg/sentry/device",
         "//pkg/sentry/fs",

pkg/sentry/socket/netstack/netstack.go

@@ -28,9 +28,9 @@ import (
 	"bytes"
 	"fmt"
 	"io"
+	"io/ioutil"
 	"math"
 	"reflect"
-	"sync/atomic"
 	"syscall"
 	"time"
 
@@ -43,7 +43,6 @@ import (
 	"gvisor.dev/gvisor/pkg/marshal"
 	"gvisor.dev/gvisor/pkg/marshal/primitive"
 	"gvisor.dev/gvisor/pkg/metric"
-	"gvisor.dev/gvisor/pkg/safemem"
 	"gvisor.dev/gvisor/pkg/sentry/arch"
 	"gvisor.dev/gvisor/pkg/sentry/fs"
 	"gvisor.dev/gvisor/pkg/sentry/fs/fsutil"
@@ -308,16 +307,8 @@ type socketOpsCommon struct {
 	skType   linux.SockType
 	protocol int
 
-	// readViewHasData is 1 iff readView has data to be read, 0 otherwise.
-	// Must be accessed using atomic operations. It must only be written
-	// with readMu held but can be read without holding readMu. The latter
-	// is required to avoid deadlocks in epoll Readiness checks.
-	readViewHasData uint32
-
 	// readMu protects access to the below fields.
 	readMu sync.Mutex `state:"nosave"`
-	// readView contains the remaining payload from the last packet.
-	readView buffer.View
 	// readCM holds control message information for the last packet read
 	// from Endpoint.
 	readCM         socket.IPControlMessages
@@ -336,8 +327,8 @@ type socketOpsCommon struct {
 	// valid when timestampValid is true. It is protected by readMu.
 	timestampNS int64
 
-	// sockOptInq corresponds to TCP_INQ. It is implemented at this level
-	// because it takes into account data from readView.
+	// TODO(b/153685824): Move this to SocketOptions.
+	// sockOptInq corresponds to TCP_INQ.
 	sockOptInq bool
 }
 
@@ -377,41 +368,23 @@ func (s *socketOpsCommon) isPacketBased() bool {
 	return s.skType == linux.SOCK_DGRAM || s.skType == linux.SOCK_SEQPACKET || s.skType == linux.SOCK_RDM || s.skType == linux.SOCK_RAW
 }
 
-// fetchReadView updates the readView field of the socket if it's currently
-// empty. It assumes that the socket is locked.
-//
 // Precondition: s.readMu must be held.
-func (s *socketOpsCommon) fetchReadView() *syserr.Error {
-	if len(s.readView) > 0 {
-		return nil
-	}
-	s.readView = nil
-	s.sender = tcpip.FullAddress{}
-	s.linkPacketInfo = tcpip.LinkPacketInfo{}
+func (s *socketOpsCommon) readLocked(dst io.Writer, count int, peek bool) (numRead, numTotal int, serr *syserr.Error) {
+	res, err := s.Endpoint.Read(dst, count, tcpip.ReadOptions{
+		Peek:               peek,
+		NeedRemoteAddr:     true,
+		NeedLinkPacketInfo: true,
+	})
 
-	var v buffer.View
-	var cms tcpip.ControlMessages
-	var err *tcpip.Error
+	// Assign these anyways.
+	s.readCM = socket.NewIPControlMessages(s.family, res.ControlMessages)
+	s.sender = res.RemoteAddr
+	s.linkPacketInfo = res.LinkPacketInfo
 
-	switch e := s.Endpoint.(type) {
-	// The ordering of these interfaces matters. The most specific
-	// interfaces must be specified before the more generic Endpoint
-	// interface.
-	case tcpip.PacketEndpoint:
-		v, cms, err = e.ReadPacket(&s.sender, &s.linkPacketInfo)
-	case tcpip.Endpoint:
-		v, cms, err = e.Read(&s.sender)
-	}
 	if err != nil {
-		atomic.StoreUint32(&s.readViewHasData, 0)
-		return syserr.TranslateNetstackError(err)
+		return 0, 0, syserr.TranslateNetstackError(err)
 	}
-
-	s.readView = v
-	s.readCM = socket.NewIPControlMessages(s.family, cms)
-	atomic.StoreUint32(&s.readViewHasData, 1)
-
-	return nil
+	return res.Count, res.Total, nil
 }
 
 // Release implements fs.FileOperations.Release.
@@ -460,38 +433,14 @@ func (s *SocketOperations) Read(ctx context.Context, _ *fs.File, dst usermem.IOS
 // WriteTo implements fs.FileOperations.WriteTo.
 func (s *SocketOperations) WriteTo(ctx context.Context, _ *fs.File, dst io.Writer, count int64, dup bool) (int64, error) {
 	s.readMu.Lock()
+	defer s.readMu.Unlock()
 
-	// Copy as much data as possible.
-	done := int64(0)
-	for count > 0 {
-		// This may return a blocking error.
-		if err := s.fetchReadView(); err != nil {
-			s.readMu.Unlock()
-			return done, err.ToError()
-		}
-
-		// Write to the underlying file.
-		n, err := dst.Write(s.readView)
-		done += int64(n)
-		count -= int64(n)
-		if dup {
-			// That's all we support for dup. This is generally
-			// supported by any Linux system calls, but the
-			// expectation is that now a caller will call read to
-			// actually remove these bytes from the socket.
-			break
-		}
-
-		// Drop that part of the view.
-		s.readView.TrimFront(n)
-		if err != nil {
-			s.readMu.Unlock()
-			return done, err
-		}
+	// This may return a blocking error.
+	n, _, err := s.readLocked(dst, int(count), dup /* peek */)
+	if err != nil {
+		return 0, err.ToError()
 	}
-
-	s.readMu.Unlock()
-	return done, nil
+	return int64(n), nil
 }
 
 // ioSequencePayload implements tcpip.Payload.
@@ -627,17 +576,7 @@ func (s *SocketOperations) ReadFrom(ctx context.Context, _ *fs.File, r io.Reader
 
 // Readiness returns a mask of ready events for socket s.
 func (s *socketOpsCommon) Readiness(mask waiter.EventMask) waiter.EventMask {
-	r := s.Endpoint.Readiness(mask)
-
-	// Check our cached value iff the caller asked for readability and the
-	// endpoint itself is currently not readable.
-	if (mask & ^r & waiter.EventIn) != 0 {
-		if atomic.LoadUint32(&s.readViewHasData) == 1 {
-			r |= waiter.EventIn
-		}
-	}
-
-	return r
+	return s.Endpoint.Readiness(mask)
 }
 
 func (s *socketOpsCommon) checkFamily(family uint16, exact bool) *syserr.Error {
@@ -2618,66 +2557,20 @@ func (s *socketOpsCommon) GetPeerName(t *kernel.Task) (linux.SockAddr, uint32, *
 	return a, l, nil
 }
 
-// coalescingRead is the fast path for non-blocking, non-peek, stream-based
-// case. It coalesces as many packets as possible before returning to the
-// caller.
+// streamRead is the fast path for non-blocking, non-peek, stream-based socket.
 //
 // Precondition: s.readMu must be locked.
-func (s *socketOpsCommon) coalescingRead(ctx context.Context, dst usermem.IOSequence, discard bool) (int, *syserr.Error) {
-	var err *syserr.Error
-	var copied int
-
-	// Copy as many views as possible into the user-provided buffer.
-	for {
-		// Always do at least one fetchReadView, even if the number of bytes to
-		// read is 0.
-		err = s.fetchReadView()
-		if err != nil || len(s.readView) == 0 {
-			break
-		}
-		if dst.NumBytes() == 0 {
-			break
-		}
-
-		var n int
-		var e error
-		if discard {
-			n = len(s.readView)
-			if int64(n) > dst.NumBytes() {
-				n = int(dst.NumBytes())
-			}
-		} else {
-			n, e = dst.CopyOut(ctx, s.readView)
-			// Set the control message, even if 0 bytes were read.
-			if e == nil {
-				s.updateTimestamp()
-			}
-		}
-		copied += n
-		s.readView.TrimFront(n)
-
-		dst = dst.DropFirst(n)
-		if e != nil {
-			err = syserr.FromError(e)
-			break
-		}
-		// If we are done reading requested data then stop.
-		if dst.NumBytes() == 0 {
-			break
-		}
+func (s *socketOpsCommon) streamRead(ctx context.Context, dst io.Writer, count int) (int, *syserr.Error) {
+	// Always do at least one read, even if the number of bytes to read is 0.
+	var n int
+	n, _, err := s.readLocked(dst, count, false /* peek */)
+	if err != nil {
+		return 0, err
 	}
-
-	if len(s.readView) == 0 {
-		atomic.StoreUint32(&s.readViewHasData, 0)
+	if n > 0 {
+		s.Endpoint.ModerateRecvBuf(n)
 	}
-
-	// If we managed to copy something, we must deliver it.
-	if copied > 0 {
-		s.Endpoint.ModerateRecvBuf(copied)
-		return copied, nil
-	}
-
-	return 0, err
+	return n, nil
 }
 
 func (s *socketOpsCommon) fillCmsgInq(cmsg *socket.ControlMessages) {
@@ -2689,7 +2582,7 @@ func (s *socketOpsCommon) fillCmsgInq(cmsg *socket.ControlMessages) {
 		return
 	}
 	cmsg.IP.HasInq = true
-	cmsg.IP.Inq = int32(len(s.readView) + rcvBufUsed)
+	cmsg.IP.Inq = int32(rcvBufUsed)
 }
 
 func toLinuxPacketType(pktType tcpip.PacketType) uint8 {
@@ -2726,7 +2619,21 @@ func (s *socketOpsCommon) nonBlockingRead(ctx context.Context, dst usermem.IOSeq
 		// bytes of data to be discarded, rather than passed back in a
 		// caller-supplied  buffer.
 		s.readMu.Lock()
-		n, err := s.coalescingRead(ctx, dst, trunc)
+
+		var w io.Writer
+		if trunc {
+			w = ioutil.Discard
+		} else {
+			w = dst.Writer(ctx)
+		}
+
+		n, err := s.streamRead(ctx, w, int(dst.NumBytes()))
+
+		if err == nil && !trunc {
+			// Set the control message, even if 0 bytes were read.
+			s.updateTimestamp()
+		}
+
 		cmsg := s.controlMessages()
 		s.fillCmsgInq(&cmsg)
 		s.readMu.Unlock()
@@ -2736,18 +2643,32 @@ func (s *socketOpsCommon) nonBlockingRead(ctx context.Context, dst usermem.IOSeq
 	s.readMu.Lock()
 	defer s.readMu.Unlock()
 
-	if err := s.fetchReadView(); err != nil {
+	// MSG_TRUNC with MSG_PEEK on a TCP socket returns the
+	// amount that could be read, and does not write to buffer.
+	isTCPPeekTrunc := !isPacket && peek && trunc
+
+	var w io.Writer
+	if isTCPPeekTrunc {
+		w = ioutil.Discard
+	} else {
+		w = dst.Writer(ctx)
+	}
+
+	var numRead, numTotal int
+	var err *syserr.Error
+	numRead, numTotal, err = s.readLocked(w, int(dst.NumBytes()), peek)
+	if err != nil {
 		return 0, 0, nil, 0, socket.ControlMessages{}, err
 	}
 
-	if !isPacket && peek && trunc {
-		// MSG_TRUNC with MSG_PEEK on a TCP socket returns the
-		// amount that could be read.
+	if isTCPPeekTrunc {
+		// TCP endpoint does not return the total bytes in buffer as numTotal.
+		// We need to query it from socket option.
 		rql, err := s.Endpoint.GetSockOptInt(tcpip.ReceiveQueueSizeOption)
 		if err != nil {
 			return 0, 0, nil, 0, socket.ControlMessages{}, syserr.TranslateNetstackError(err)
 		}
-		available := len(s.readView) + int(rql)
+		available := int(rql)
 		bufLen := int(dst.NumBytes())
 		if available < bufLen {
 			return available, 0, nil, 0, socket.ControlMessages{}, nil
@@ -2755,11 +2676,9 @@ func (s *socketOpsCommon) nonBlockingRead(ctx context.Context, dst usermem.IOSeq
 		return bufLen, 0, nil, 0, socket.ControlMessages{}, nil
 	}
 
-	n, err := dst.CopyOut(ctx, s.readView)
 	// Set the control message, even if 0 bytes were read.
-	if err == nil {
-		s.updateTimestamp()
-	}
+	s.updateTimestamp()
+
 	var addr linux.SockAddr
 	var addrLen uint32
 	if isPacket && senderRequested {
@@ -2772,58 +2691,33 @@ func (s *socketOpsCommon) nonBlockingRead(ctx context.Context, dst usermem.IOSeq
 	}
 
 	if peek {
-		if l := len(s.readView); trunc && l > n {
+		if trunc && numTotal > numRead {
 			// isPacket must be true.
-			return l, linux.MSG_TRUNC, addr, addrLen, s.controlMessages(), syserr.FromError(err)
+			return numTotal, linux.MSG_TRUNC, addr, addrLen, s.controlMessages(), nil
 		}
-
-		if isPacket || err != nil {
-			return n, 0, addr, addrLen, s.controlMessages(), syserr.FromError(err)
-		}
-
-		// We need to peek beyond the first message.
-		dst = dst.DropFirst(n)
-		num, err := dst.CopyOutFrom(ctx, safemem.FromVecReaderFunc{func(dsts [][]byte) (int64, error) {
-			n, err := s.Endpoint.Peek(dsts)
-			// TODO(b/78348848): Handle peek timestamp.
-			if err != nil {
-				return int64(n), syserr.TranslateNetstackError(err).ToError()
-			}
-			return int64(n), nil
-		}})
-		n += int(num)
-		if err == syserror.ErrWouldBlock && n > 0 {
-			// We got some data, so no need to return an error.
-			err = nil
-		}
-		return n, 0, nil, 0, s.controlMessages(), syserr.FromError(err)
+		return numRead, 0, nil, 0, s.controlMessages(), nil
 	}
 
 	var msgLen int
 	if isPacket {
-		msgLen = len(s.readView)
-		s.readView = nil
+		msgLen = numTotal
 	} else {
-		msgLen = int(n)
-		s.readView.TrimFront(int(n))
-	}
-
-	if len(s.readView) == 0 {
-		atomic.StoreUint32(&s.readViewHasData, 0)
+		msgLen = numRead
 	}
 
 	var flags int
-	if msgLen > int(n) {
+	if msgLen > numRead {
 		flags |= linux.MSG_TRUNC
 	}
 
+	n := numRead
 	if trunc {
 		n = msgLen
 	}
 
 	cmsg := s.controlMessages()
 	s.fillCmsgInq(&cmsg)
-	return n, flags, addr, addrLen, cmsg, syserr.FromError(err)
+	return n, flags, addr, addrLen, cmsg, nil
 }
 
 func (s *socketOpsCommon) controlMessages() socket.ControlMessages {
@@ -3090,11 +2984,6 @@ func (s *socketOpsCommon) ioctl(ctx context.Context, io usermem.IO, args arch.Sy
 			return 0, syserr.TranslateNetstackError(terr).ToError()
 		}
 
-		// Add bytes removed from the endpoint but not yet sent to the caller.
-		s.readMu.Lock()
-		v += len(s.readView)
-		s.readMu.Unlock()
-
 		if v > math.MaxInt32 {
 			v = math.MaxInt32
 		}

pkg/syserr/netstack.go

@@ -48,6 +48,7 @@ var (
 	ErrInvalidOptionValue    = New(tcpip.ErrInvalidOptionValue.String(), linux.EINVAL)
 	ErrBroadcastDisabled     = New(tcpip.ErrBroadcastDisabled.String(), linux.EACCES)
 	ErrNotPermittedNet       = New(tcpip.ErrNotPermitted.String(), linux.EPERM)
+	ErrBadBuffer             = New(tcpip.ErrBadBuffer.String(), linux.EFAULT)
 )
 
 var netstackErrorTranslations map[string]*Error
@@ -100,6 +101,7 @@ func init() {
 	addErrMapping(tcpip.ErrBroadcastDisabled, ErrBroadcastDisabled)
 	addErrMapping(tcpip.ErrNotPermitted, ErrNotPermittedNet)
 	addErrMapping(tcpip.ErrAddressFamilyNotSupported, ErrAddressFamilyNotSupported)
+	addErrMapping(tcpip.ErrBadBuffer, ErrBadBuffer)
 }
 
 // TranslateNetstackError converts an error from the tcpip package to a sentry

pkg/tcpip/adapters/gonet/gonet.go

@@ -286,45 +286,47 @@ type opErrorer interface {
 
 // commonRead implements the common logic between net.Conn.Read and
 // net.PacketConn.ReadFrom.
-func commonRead(ep tcpip.Endpoint, wq *waiter.Queue, deadline <-chan struct{}, addr *tcpip.FullAddress, errorer opErrorer, dontWait bool) ([]byte, error) {
+func commonRead(b []byte, ep tcpip.Endpoint, wq *waiter.Queue, deadline <-chan struct{}, addr *tcpip.FullAddress, errorer opErrorer) (int, error) {
 	select {
 	case <-deadline:
-		return nil, errorer.newOpError("read", &timeoutError{})
+		return 0, errorer.newOpError("read", &timeoutError{})
 	default:
 	}
 
-	read, _, err := ep.Read(addr)
+	w := tcpip.SliceWriter(b)
+	opts := tcpip.ReadOptions{NeedRemoteAddr: addr != nil}
+	res, err := ep.Read(&w, len(b), opts)
 
 	if err == tcpip.ErrWouldBlock {
-		if dontWait {
-			return nil, errWouldBlock
-		}
 		// Create wait queue entry that notifies a channel.
 		waitEntry, notifyCh := waiter.NewChannelEntry(nil)
 		wq.EventRegister(&waitEntry, waiter.EventIn)
 		defer wq.EventUnregister(&waitEntry)
 		for {
-			read, _, err = ep.Read(addr)
+			res, err = ep.Read(&w, len(b), opts)
 			if err != tcpip.ErrWouldBlock {
 				break
 			}
 			select {
 			case <-deadline:
-				return nil, errorer.newOpError("read", &timeoutError{})
+				return 0, errorer.newOpError("read", &timeoutError{})
 			case <-notifyCh:
 			}
 		}
 	}
 
 	if err == tcpip.ErrClosedForReceive {
-		return nil, io.EOF
+		return 0, io.EOF
 	}
 
 	if err != nil {
-		return nil, errorer.newOpError("read", errors.New(err.String()))
+		return 0, errorer.newOpError("read", errors.New(err.String()))
 	}
 
-	return read, nil
+	if addr != nil {
+		*addr = res.RemoteAddr
+	}
+	return res.Count, nil
 }
 
 // Read implements net.Conn.Read.
@@ -334,31 +336,11 @@ func (c *TCPConn) Read(b []byte) (int, error) {
 
 	deadline := c.readCancel()
 
-	numRead := 0
-	defer func() {
-		if numRead != 0 {
-			c.ep.ModerateRecvBuf(numRead)
-		}
-	}()
-	for numRead != len(b) {
-		if len(c.read) == 0 {
-			var err error
-			c.read, err = commonRead(c.ep, c.wq, deadline, nil, c, numRead != 0)
-			if err != nil {
-				if numRead != 0 {
-					return numRead, nil
-				}
-				return numRead, err
-			}
-		}
-		n := copy(b[numRead:], c.read)
-		c.read.TrimFront(n)
-		numRead += n
-		if len(c.read) == 0 {
-			c.read = nil
-		}
+	n, err := commonRead(b, c.ep, c.wq, deadline, nil, c)
+	if n != 0 {
+		c.ep.ModerateRecvBuf(n)
 	}
-	return numRead, nil
+	return n, err
 }
 
 // Write implements net.Conn.Write.
@@ -652,12 +634,11 @@ func (c *UDPConn) ReadFrom(b []byte) (int, net.Addr, error) {
 	deadline := c.readCancel()
 
 	var addr tcpip.FullAddress
-	read, err := commonRead(c.ep, c.wq, deadline, &addr, c, false)
+	n, err := commonRead(b, c.ep, c.wq, deadline, &addr, c)
 	if err != nil {
 		return 0, nil, err
 	}
-
-	return copy(b, read), fullToUDPAddr(addr), nil
+	return n, fullToUDPAddr(addr), nil
 }
 
 func (c *UDPConn) Write(b []byte) (int, error) {

pkg/tcpip/buffer/view.go

@@ -105,18 +105,18 @@ func (vv *VectorisedView) TrimFront(count int) {
 }
 
 // Read implements io.Reader.
-func (vv *VectorisedView) Read(v View) (copied int, err error) {
-	count := len(v)
+func (vv *VectorisedView) Read(b []byte) (copied int, err error) {
+	count := len(b)
 	for count > 0 && len(vv.views) > 0 {
 		if count < len(vv.views[0]) {
 			vv.size -= count
-			copy(v[copied:], vv.views[0][:count])
+			copy(b[copied:], vv.views[0][:count])
 			vv.views[0].TrimFront(count)
 			copied += count
 			return copied, nil
 		}
 		count -= len(vv.views[0])
-		copy(v[copied:], vv.views[0])
+		copy(b[copied:], vv.views[0])
 		copied += len(vv.views[0])
 		vv.removeFirst()
 	}
@@ -145,6 +145,35 @@ func (vv *VectorisedView) ReadToVV(dstVV *VectorisedView, count int) (copied int
 	return copied
 }
 
+// ReadTo reads up to count bytes from vv to dst. It also removes them from vv
+// unless peek is true.
+func (vv *VectorisedView) ReadTo(dst io.Writer, count int, peek bool) (int, error) {
+	var err error
+	done := 0
+	for _, v := range vv.Views() {
+		remaining := count - done
+		if remaining <= 0 {
+			break
+		}
+		if len(v) > remaining {
+			v = v[:remaining]
+		}
+
+		var n int
+		n, err = dst.Write(v)
+		if n > 0 {
+			done += n
+		}
+		if err != nil {
+			break
+		}
+	}
+	if !peek {
+		vv.TrimFront(done)
+	}
+	return done, err
+}
+
 // CapLength irreversibly reduces the length of the vectorised view.
 func (vv *VectorisedView) CapLength(length int) {
 	if length < 0 {

pkg/tcpip/buffer/view_test.go

@@ -235,14 +235,16 @@ func TestToClone(t *testing.T) {
 	}
 }
 
-func TestVVReadToVV(t *testing.T) {
-	testCases := []struct {
-		comment     string
-		vv          VectorisedView
-		bytesToRead int
-		wantBytes   string
-		leftVV      VectorisedView
-	}{
+type readToTestCases struct {
+	comment     string
+	vv          VectorisedView
+	bytesToRead int
+	wantBytes   string
+	leftVV      VectorisedView
+}
+
+func createReadToTestCases() []readToTestCases {
+	return []readToTestCases{
 		{
 			comment:     "large VV, short read",
 			vv:          vv(30, "012345678901234567890123456789"),
@@ -279,8 +281,10 @@ func TestVVReadToVV(t *testing.T) {
 			leftVV:      vv(0, ""),
 		},
 	}
+}
 
-	for _, tc := range testCases {
+func TestVVReadToVV(t *testing.T) {
+	for _, tc := range createReadToTestCases() {
 		t.Run(tc.comment, func(t *testing.T) {
 			var readTo VectorisedView
 			inSize := tc.vv.Size()
@@ -301,6 +305,52 @@ func TestVVReadToVV(t *testing.T) {
 	}
 }
 
+func TestVVReadTo(t *testing.T) {
+	for _, tc := range createReadToTestCases() {
+		t.Run(tc.comment, func(t *testing.T) {
+			var dst bytes.Buffer
+			origSize := tc.vv.Size()
+			copied, err := tc.vv.ReadTo(&dst, tc.bytesToRead, false /* peek */)
+			if got, want := copied, len(tc.wantBytes); err != nil || got != want {
+				t.Errorf("got ReadTo(&dst, %d, false) = %d, %v; want %d, nil", tc.bytesToRead, got, err, want)
+			}
+			if got, want := string(dst.Bytes()), tc.wantBytes; got != want {
+				t.Errorf("got dst = %q, want %q", got, want)
+			}
+			if got, want := tc.vv.Size(), origSize-copied; got != want {
+				t.Errorf("got after-read tc.vv.Size() = %d, want %d", got, want)
+			}
+			if got, want := string(tc.vv.ToView()), string(tc.leftVV.ToView()); got != want {
+				t.Errorf("got after-read data in tc.vv = %q, want %q", got, want)
+			}
+		})
+	}
+}
+
+func TestVVReadToPeek(t *testing.T) {
+	for _, tc := range createReadToTestCases() {
+		t.Run(tc.comment, func(t *testing.T) {
+			var dst bytes.Buffer
+			origSize := tc.vv.Size()
+			origData := string(tc.vv.ToView())
+			copied, err := tc.vv.ReadTo(&dst, tc.bytesToRead, true /* peek */)
+			if got, want := copied, len(tc.wantBytes); err != nil || got != want {
+				t.Errorf("got ReadTo(&dst, %d, false) = %d, %v; want %d, nil", tc.bytesToRead, got, err, want)
+			}
+			if got, want := string(dst.Bytes()), tc.wantBytes; got != want {
+				t.Errorf("got dst = %q, want %q", got, want)
+			}
+			// Expect tc.vv is unchanged.
+			if got, want := tc.vv.Size(), origSize; got != want {
+				t.Errorf("got after-read tc.vv.Size() = %d, want %d", got, want)
+			}
+			if got, want := string(tc.vv.ToView()), origData; got != want {
+				t.Errorf("got after-read data in tc.vv = %q, want %q", got, want)
+			}
+		})
+	}
+}
+
 func TestVVRead(t *testing.T) {
 	testCases := []struct {
 		comment     string

pkg/tcpip/checker/checker.go

@@ -1603,3 +1603,15 @@ func IPv6RouterAlert(want header.IPv6RouterAlertValue) IPv6ExtHdrOptionChecker {
 		}
 	}
 }
+
+// IgnoreCmpPath returns a cmp.Option that ignores listed field paths.
+func IgnoreCmpPath(paths ...string) cmp.Option {
+	ignores := map[string]struct{}{}
+	for _, path := range paths {
+		ignores[path] = struct{}{}
+	}
+	return cmp.FilterPath(func(path cmp.Path) bool {
+		_, ok := ignores[path.String()]
+		return ok
+	}, cmp.Ignore())
+}

pkg/tcpip/network/ipv4/ipv4_test.go

@@ -15,9 +15,11 @@
 package ipv4_test
 
 import (
+	"bytes"
 	"context"
 	"encoding/hex"
 	"fmt"
+	"io/ioutil"
 	"math"
 	"net"
 	"testing"
@@ -2408,18 +2410,26 @@ func TestReceiveFragments(t *testing.T) {
 				t.Errorf("got UDP Rx Packets = %d, want = %d", got, want)
 			}
 
+			const rcvSize = 65536 // Account for reassembled packets.
 			for i, expectedPayload := range test.expectedPayloads {
-				gotPayload, _, err := ep.Read(nil)
+				var buf bytes.Buffer
+				result, err := ep.Read(&buf, rcvSize, tcpip.ReadOptions{})
 				if err != nil {
-					t.Fatalf("(i=%d) Read(nil): %s", i, err)
+					t.Fatalf("(i=%d) Read: %s", i, err)
 				}
-				if diff := cmp.Diff(buffer.View(expectedPayload), gotPayload); diff != "" {
+				if diff := cmp.Diff(tcpip.ReadResult{
+					Count: len(expectedPayload),
+					Total: len(expectedPayload),
+				}, result, checker.IgnoreCmpPath("ControlMessages")); diff != "" {
+					t.Errorf("(i=%d) ep.Read: unexpected result (-want +got):\n%s", i, diff)
+				}
+				if diff := cmp.Diff(expectedPayload, buf.Bytes()); diff != "" {
 					t.Errorf("(i=%d) got UDP payload mismatch (-want +got):\n%s", i, diff)
 				}
 			}
 
-			if gotPayload, _, err := ep.Read(nil); err != tcpip.ErrWouldBlock {
-				t.Fatalf("(last) got Read(nil) = (%x, _, %v), want = (_, _, %s)", gotPayload, err, tcpip.ErrWouldBlock)
+			if res, err := ep.Read(ioutil.Discard, rcvSize, tcpip.ReadOptions{}); err != tcpip.ErrWouldBlock {
+				t.Fatalf("(last) got Read = (%v, %v), want = (_, %s)", res, err, tcpip.ErrWouldBlock)
 			}
 		})
 	}

pkg/tcpip/network/ipv6/ipv6_test.go

@@ -15,8 +15,10 @@
 package ipv6
 
 import (
+	"bytes"
 	"encoding/hex"
 	"fmt"
+	"io/ioutil"
 	"math"
 	"net"
 	"testing"
@@ -844,13 +846,14 @@ func TestReceiveIPv6ExtHdrs(t *testing.T) {
 		},
 	}
 
+	const mtu = header.IPv6MinimumMTU
 	for _, test := range tests {
 		t.Run(test.name, func(t *testing.T) {
 			s := stack.New(stack.Options{
 				NetworkProtocols:   []stack.NetworkProtocolFactory{NewProtocol},
 				TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol},
 			})
-			e := channel.New(1, header.IPv6MinimumMTU, linkAddr1)
+			e := channel.New(1, mtu, linkAddr1)
 			if err := s.CreateNIC(nicID, e); err != nil {
 				t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
 			}
@@ -979,17 +982,24 @@ func TestReceiveIPv6ExtHdrs(t *testing.T) {
 			if got := stats.Value(); got != 1 {
 				t.Errorf("got UDP Rx Packets = %d, want = 1", got)
 			}
-			gotPayload, _, err := ep.Read(nil)
+			var buf bytes.Buffer
+			result, err := ep.Read(&buf, mtu, tcpip.ReadOptions{})
 			if err != nil {
-				t.Fatalf("Read(nil): %s", err)
+				t.Fatalf("Read: %s", err)
 			}
-			if diff := cmp.Diff(buffer.View(udpPayload), gotPayload); diff != "" {
+			if diff := cmp.Diff(tcpip.ReadResult{
+				Count: len(udpPayload),
+				Total: len(udpPayload),
+			}, result, checker.IgnoreCmpPath("ControlMessages")); diff != "" {
+				t.Errorf("Read: unexpected result (-want +got):\n%s", diff)
+			}
+			if diff := cmp.Diff(udpPayload, buf.Bytes()); diff != "" {
 				t.Errorf("got UDP payload mismatch (-want +got):\n%s", diff)
 			}
 
 			// Should not have any more UDP packets.
-			if gotPayload, _, err := ep.Read(nil); err != tcpip.ErrWouldBlock {
-				t.Fatalf("got Read(nil) = (%x, _, %v), want = (_, _, %s)", gotPayload, err, tcpip.ErrWouldBlock)
+			if res, err := ep.Read(ioutil.Discard, mtu, tcpip.ReadOptions{}); err != tcpip.ErrWouldBlock {
+				t.Fatalf("got Read = (%v, %v), want = (_, %s)", res, err, tcpip.ErrWouldBlock)
 			}
 		})
 	}
@@ -1969,18 +1979,20 @@ func TestReceiveIPv6Fragments(t *testing.T) {
 				t.Errorf("got UDP Rx Packets = %d, want = %d", got, want)
 			}
 
+			const rcvSize = 65536 // Account for reassembled packets.
 			for i, p := range test.expectedPayloads {
-				gotPayload, _, err := ep.Read(nil)
+				var buf bytes.Buffer
+				_, err := ep.Read(&buf, rcvSize, tcpip.ReadOptions{})
 				if err != nil {
-					t.Fatalf("(i=%d) Read(nil): %s", i, err)
+					t.Fatalf("(i=%d) Read: %s", i, err)
 				}
-				if diff := cmp.Diff(buffer.View(p), gotPayload); diff != "" {
+				if diff := cmp.Diff(p, buf.Bytes()); diff != "" {
 					t.Errorf("(i=%d) got UDP payload mismatch (-want +got):\n%s", i, diff)
 				}
 			}
 
-			if gotPayload, _, err := ep.Read(nil); err != tcpip.ErrWouldBlock {
-				t.Fatalf("(last) got Read(nil) = (%x, _, %v), want = (_, _, %s)", gotPayload, err, tcpip.ErrWouldBlock)
+			if res, err := ep.Read(ioutil.Discard, rcvSize, tcpip.ReadOptions{}); err != tcpip.ErrWouldBlock {
+				t.Fatalf("(last) got Read = (%v, %v), want = (_, %s)", res, err, tcpip.ErrWouldBlock)
 			}
 		})
 	}

pkg/tcpip/sample/tun_tcp_connect/main.go

@@ -44,6 +44,7 @@ import (
 	"bufio"
 	"fmt"
 	"log"
+	"math"
 	"math/rand"
 	"net"
 	"os"
@@ -200,7 +201,7 @@ func main() {
 	// connection from its side.
 	wq.EventRegister(&waitEntry, waiter.EventIn)
 	for {
-		v, _, err := ep.Read(nil)
+		_, err := ep.Read(os.Stdout, math.MaxUint16, tcpip.ReadOptions{})
 		if err != nil {
 			if err == tcpip.ErrClosedForReceive {
 				break
@@ -213,8 +214,6 @@ func main() {
 
 			log.Fatal("Read() failed:", err)
 		}
-
-		os.Stdout.Write(v)
 	}
 	wq.EventUnregister(&waitEntry)
 

pkg/tcpip/sample/tun_tcp_echo/main.go

@@ -20,8 +20,10 @@
 package main
 
 import (
+	"bytes"
 	"flag"
 	"log"
+	"math"
 	"math/rand"
 	"net"
 	"os"
@@ -54,7 +56,8 @@ func echo(wq *waiter.Queue, ep tcpip.Endpoint) {
 	defer wq.EventUnregister(&waitEntry)
 
 	for {
-		v, _, err := ep.Read(nil)
+		var buf bytes.Buffer
+		_, err := ep.Read(&buf, math.MaxUint16, tcpip.ReadOptions{})
 		if err != nil {
 			if err == tcpip.ErrWouldBlock {
 				<-notifyCh
@@ -64,7 +67,7 @@ func echo(wq *waiter.Queue, ep tcpip.Endpoint) {
 			return
 		}
 
-		ep.Write(tcpip.SlicePayload(v), tcpip.WriteOptions{})
+		ep.Write(tcpip.SlicePayload(buf.Bytes()), tcpip.WriteOptions{})
 	}
 }
 

pkg/tcpip/stack/transport_demuxer_test.go

@@ -15,6 +15,7 @@
 package stack_test
 
 import (
+	"io/ioutil"
 	"math"
 	"math/rand"
 	"testing"
@@ -351,7 +352,7 @@ func TestBindToDeviceDistribution(t *testing.T) {
 						}
 
 						ep := <-pollChannel
-						if _, _, err := ep.Read(nil); err != nil {
+						if _, err := ep.Read(ioutil.Discard, math.MaxUint16, tcpip.ReadOptions{}); err != nil {
 							t.Fatalf("Read on endpoint %d failed: %s", eps[ep], err)
 						}
 						stats[ep]++

pkg/tcpip/stack/transport_test.go

@@ -15,6 +15,7 @@
 package stack_test
 
 import (
+	"io"
 	"testing"
 
 	"gvisor.dev/gvisor/pkg/tcpip"
@@ -85,8 +86,8 @@ func (*fakeTransportEndpoint) Readiness(mask waiter.EventMask) waiter.EventMask
 	return mask
 }
 
-func (*fakeTransportEndpoint) Read(*tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) {
-	return buffer.View{}, tcpip.ControlMessages{}, nil
+func (*fakeTransportEndpoint) Read(io.Writer, int, tcpip.ReadOptions) (tcpip.ReadResult, *tcpip.Error) {
+	return tcpip.ReadResult{}, nil
 }
 
 func (f *fakeTransportEndpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) {
@@ -110,10 +111,6 @@ func (f *fakeTransportEndpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions
 	return int64(len(v)), nil, nil
 }
 
-func (*fakeTransportEndpoint) Peek([][]byte) (int64, *tcpip.Error) {
-	return 0, nil
-}
-
 // SetSockOpt sets a socket option. Currently not supported.
 func (*fakeTransportEndpoint) SetSockOpt(tcpip.SettableSocketOption) *tcpip.Error {
 	return tcpip.ErrInvalidEndpointState

pkg/tcpip/tcpip.go

@@ -31,6 +31,7 @@ package tcpip
 import (
 	"errors"
 	"fmt"
+	"io"
 	"math/bits"
 	"reflect"
 	"strconv"
@@ -39,7 +40,6 @@ import (
 	"time"
 
 	"gvisor.dev/gvisor/pkg/sync"
-	"gvisor.dev/gvisor/pkg/tcpip/buffer"
 	"gvisor.dev/gvisor/pkg/waiter"
 )
 
@@ -113,6 +113,7 @@ var (
 	ErrNotPermitted              = &Error{msg: "operation not permitted"}
 	ErrAddressFamilyNotSupported = &Error{msg: "address family not supported by protocol"}
 	ErrMalformedHeader           = &Error{msg: "header is malformed"}
+	ErrBadBuffer                 = &Error{msg: "bad buffer"}
 )
 
 var messageToError map[string]*Error
@@ -162,6 +163,7 @@ func StringToError(s string) *Error {
 			ErrNotPermitted,
 			ErrAddressFamilyNotSupported,
 			ErrMalformedHeader,
+			ErrBadBuffer,
 		}
 
 		messageToError = make(map[string]*Error)
@@ -496,6 +498,21 @@ func (s SlicePayload) Payload(size int) ([]byte, *Error) {
 	return s[:size], nil
 }
 
+var _ io.Writer = (*SliceWriter)(nil)
+
+// SliceWriter implements io.Writer for slices.
+type SliceWriter []byte
+
+// Write implements io.Writer.Write.
+func (s *SliceWriter) Write(b []byte) (int, error) {
+	n := copy(*s, b)
+	*s = (*s)[n:]
+	if n < len(b) {
+		return n, io.ErrShortWrite
+	}
+	return n, nil
+}
+
 // A ControlMessages contains socket control messages for IP sockets.
 //
 // +stateify savable
@@ -552,6 +569,40 @@ type PacketOwner interface {
 	GID() uint32
 }
 
+// ReadOptions contains options for Endpoint.Read.
+type ReadOptions struct {
+	// Peek indicates whether this read is a peek.
+	Peek bool
+
+	// NeedRemoteAddr indicates whether to return the remote address, if
+	// supported.
+	NeedRemoteAddr bool
+
+	// NeedLinkPacketInfo indicates whether to return the link-layer information,
+	// if supported.
+	NeedLinkPacketInfo bool
+}
+
+// ReadResult represents result for a successful Endpoint.Read.
+type ReadResult struct {
+	// Count is the number of bytes received and written to the buffer.
+	Count int
+
+	// Total is the number of bytes of the received packet. This can be used to
+	// determine whether the read is truncated.
+	Total int
+
+	// ControlMessages is the control messages received.
+	ControlMessages ControlMessages
+
+	// RemoteAddr is the remote address if ReadOptions.NeedAddr is true.
+	RemoteAddr FullAddress
+
+	// LinkPacketInfo is the link-layer information of the received packet if
+	// ReadOptions.NeedLinkPacketInfo is true.
+	LinkPacketInfo LinkPacketInfo
+}
+
 // Endpoint is the interface implemented by transport protocols (e.g., tcp, udp)
 // that exposes functionality like read, write, connect, etc. to users of the
 // networking stack.
@@ -566,11 +617,15 @@ type Endpoint interface {
 	// Abort is best effort; implementing Abort with Close is acceptable.
 	Abort()
 
-	// Read reads data from the endpoint and optionally returns the sender.
+	// Read reads data from the endpoint and optionally writes to dst.
 	//
-	// This method does not block if there is no data pending. It will also
-	// either return an error or data, never both.
-	Read(*FullAddress) (buffer.View, ControlMessages, *Error)
+	// This method does not block if there is no data pending; in this case,
+	// ErrWouldBlock is returned.
+	//
+	// If non-zero number of bytes are successfully read and written to dst, err
+	// must be nil. Otherwise, if dst failed to write anything, ErrBadBuffer
+	// should be returned.
+	Read(dst io.Writer, count int, opts ReadOptions) (res ReadResult, err *Error)
 
 	// Write writes data to the endpoint's peer. This method does not block if
 	// the data cannot be written.
@@ -592,11 +647,6 @@ type Endpoint interface {
 	// not). The channel is only non-nil in this case.
 	Write(Payloader, WriteOptions) (int64, <-chan struct{}, *Error)
 
-	// Peek reads data without consuming it from the endpoint.
-	//
-	// This method does not block if there is no data pending.
-	Peek([][]byte) (int64, *Error)
-
 	// Connect connects the endpoint to its peer. Specifying a NIC is
 	// optional.
 	//
@@ -703,17 +753,6 @@ type LinkPacketInfo struct {
 	PktType PacketType
 }
 
-// PacketEndpoint are additional methods that are only implemented by Packet
-// endpoints.
-type PacketEndpoint interface {
-	// ReadPacket reads a datagram/packet from the endpoint and optionally
-	// returns the sender and additional LinkPacketInfo.
-	//
-	// This method does not block if there is no data pending. It will also
-	// either return an error or data, never both.
-	ReadPacket(*FullAddress, *LinkPacketInfo) (buffer.View, ControlMessages, *Error)
-}
-
 // EndpointInfo is the interface implemented by each endpoint info struct.
 type EndpointInfo interface {
 	// IsEndpointInfo is an empty method to implement the tcpip.EndpointInfo

pkg/tcpip/tests/integration/BUILD

@@ -15,6 +15,7 @@ go_test(
     deps = [
         "//pkg/tcpip",
         "//pkg/tcpip/buffer",
+        "//pkg/tcpip/checker",
         "//pkg/tcpip/header",
         "//pkg/tcpip/link/channel",
         "//pkg/tcpip/link/ethernet",

pkg/tcpip/tests/integration/forward_test.go

@@ -15,12 +15,13 @@
 package integration_test
 
 import (
+	"bytes"
 	"net"
 	"testing"
 
 	"github.com/google/go-cmp/cmp"
 	"gvisor.dev/gvisor/pkg/tcpip"
-	"gvisor.dev/gvisor/pkg/tcpip/buffer"
+	"gvisor.dev/gvisor/pkg/tcpip/checker"
 	"gvisor.dev/gvisor/pkg/tcpip/header"
 	"gvisor.dev/gvisor/pkg/tcpip/link/ethernet"
 	"gvisor.dev/gvisor/pkg/tcpip/link/nested"
@@ -382,24 +383,33 @@ func TestForwarding(t *testing.T) {
 
 				// Wait for the endpoint to be readable.
 				<-ch
-				var addr tcpip.FullAddress
-				v, _, err := ep.Read(&addr)
+				var buf bytes.Buffer
+				opts := tcpip.ReadOptions{NeedRemoteAddr: true}
+				res, err := ep.Read(&buf, len(data), opts)
 				if err != nil {
-					t.Fatalf("ep.Read(_): %s", err)
+					t.Fatalf("ep.Read(_, %d, %#v): %s", len(data), opts, err)
 				}
 
-				if diff := cmp.Diff(v, buffer.View(data)); diff != "" {
-					t.Errorf("received data mismatch (-want +got):\n%s", diff)
+				if diff := cmp.Diff(tcpip.ReadResult{
+					Count:      len(data),
+					Total:      len(data),
+					RemoteAddr: tcpip.FullAddress{Addr: expectedFrom},
+				}, res, checker.IgnoreCmpPath(
+					"ControlMessages",
+					"RemoteAddr.NIC",
+					"RemoteAddr.Port",
+				)); diff != "" {
+					t.Errorf("ep.Read: unexpected result (-want +got):\n%s", diff)
 				}
-				if addr.Addr != expectedFrom {
-					t.Errorf("got addr.Addr = %s, want = %s", addr.Addr, expectedFrom)
+				if diff := cmp.Diff(buf.Bytes(), data); diff != "" {
+					t.Errorf("received data mismatch (-want +got):\n%s", diff)
 				}
 
 				if t.Failed() {
 					t.FailNow()
 				}
 
-				return addr
+				return res.RemoteAddr
 			}
 
 			addr := read(epsAndAddrs.serverReadableCH, epsAndAddrs.serverEP, data, epsAndAddrs.clientAddr)

pkg/tcpip/tests/integration/link_resolution_test.go

@@ -15,12 +15,13 @@
 package integration_test
 
 import (
+	"bytes"
 	"net"
 	"testing"
 
 	"github.com/google/go-cmp/cmp"
 	"gvisor.dev/gvisor/pkg/tcpip"
-	"gvisor.dev/gvisor/pkg/tcpip/buffer"
+	"gvisor.dev/gvisor/pkg/tcpip/checker"
 	"gvisor.dev/gvisor/pkg/tcpip/header"
 	"gvisor.dev/gvisor/pkg/tcpip/link/pipe"
 	"gvisor.dev/gvisor/pkg/tcpip/network/arp"
@@ -86,21 +87,21 @@ func TestPing(t *testing.T) {
 		transProto tcpip.TransportProtocolNumber
 		netProto   tcpip.NetworkProtocolNumber
 		remoteAddr tcpip.Address
-		icmpBuf    func(*testing.T) buffer.View
+		icmpBuf    func(*testing.T) []byte
 	}{
 		{
 			name:       "IPv4 Ping",
 			transProto: icmp.ProtocolNumber4,
 			netProto:   ipv4.ProtocolNumber,
 			remoteAddr: ipv4Addr2.AddressWithPrefix.Address,
-			icmpBuf: func(t *testing.T) buffer.View {
+			icmpBuf: func(t *testing.T) []byte {
 				data := [8]byte{1, 2, 3, 4, 5, 6, 7, 8}
 				hdr := header.ICMPv4(make([]byte, header.ICMPv4MinimumSize+len(data)))
 				hdr.SetType(header.ICMPv4Echo)
 				if n := copy(hdr.Payload(), data[:]); n != len(data) {
 					t.Fatalf("copied %d bytes but expected to copy %d bytes", n, len(data))
 				}
-				return buffer.View(hdr)
+				return hdr
 			},
 		},
 		{
@@ -108,14 +109,14 @@ func TestPing(t *testing.T) {
 			transProto: icmp.ProtocolNumber6,
 			netProto:   ipv6.ProtocolNumber,
 			remoteAddr: ipv6Addr2.AddressWithPrefix.Address,
-			icmpBuf: func(t *testing.T) buffer.View {
+			icmpBuf: func(t *testing.T) []byte {
 				data := [8]byte{1, 2, 3, 4, 5, 6, 7, 8}
 				hdr := header.ICMPv6(make([]byte, header.ICMPv6MinimumSize+len(data)))
 				hdr.SetType(header.ICMPv6EchoRequest)
 				if n := copy(hdr.Payload(), data[:]); n != len(data) {
 					t.Fatalf("copied %d bytes but expected to copy %d bytes", n, len(data))
 				}
-				return buffer.View(hdr)
+				return hdr
 			},
 		},
 	}
@@ -200,17 +201,26 @@ func TestPing(t *testing.T) {
 			// Wait for the endpoint to be readable.
 			<-waiterCH
 
-			var addr tcpip.FullAddress
-			v, _, err := ep.Read(&addr)
+			var buf bytes.Buffer
+			opts := tcpip.ReadOptions{NeedRemoteAddr: true}
+			res, err := ep.Read(&buf, len(icmpBuf), opts)
 			if err != nil {
-				t.Fatalf("ep.Read(_): %s", err)
+				t.Fatalf("ep.Read(_, %d, %#v): %s", len(icmpBuf), opts, err)
 			}
-			if diff := cmp.Diff(v[icmpDataOffset:], icmpBuf[icmpDataOffset:]); diff != "" {
+			if diff := cmp.Diff(tcpip.ReadResult{
+				Count:      buf.Len(),
+				Total:      buf.Len(),
+				RemoteAddr: tcpip.FullAddress{Addr: test.remoteAddr},
+			}, res, checker.IgnoreCmpPath(
+				"ControlMessages",
+				"RemoteAddr.NIC",
+				"RemoteAddr.Port",
+			)); diff != "" {
+				t.Errorf("ep.Read: unexpected result (-want +got):\n%s", diff)
+			}
+			if diff := cmp.Diff(buf.Bytes()[icmpDataOffset:], icmpBuf[icmpDataOffset:]); diff != "" {
 				t.Errorf("received data mismatch (-want +got):\n%s", diff)
 			}
-			if addr.Addr != test.remoteAddr {
-				t.Errorf("got addr.Addr = %s, want = %s", addr.Addr, test.remoteAddr)
-			}
 		})
 	}
 }

pkg/tcpip/tests/integration/loopback_test.go

@@ -15,12 +15,14 @@
 package integration_test
 
 import (
+	"bytes"
 	"testing"
 	"time"
 
 	"github.com/google/go-cmp/cmp"
 	"gvisor.dev/gvisor/pkg/tcpip"
 	"gvisor.dev/gvisor/pkg/tcpip/buffer"
+	"gvisor.dev/gvisor/pkg/tcpip/checker"
 	"gvisor.dev/gvisor/pkg/tcpip/header"
 	"gvisor.dev/gvisor/pkg/tcpip/link/loopback"
 	"gvisor.dev/gvisor/pkg/tcpip/network/ipv4"
@@ -238,21 +240,28 @@ func TestLoopbackAcceptAllInSubnetUDP(t *testing.T) {
 				t.Fatalf("got sep.Write(_, _) = (%d, _, nil), want = (%d, _, nil)", n, want)
 			}
 
-			var addr tcpip.FullAddress
-			if gotPayload, _, err := rep.Read(&addr); test.expectRx {
+			var buf bytes.Buffer
+			opts := tcpip.ReadOptions{NeedRemoteAddr: true}
+			if res, err := rep.Read(&buf, len(data), opts); test.expectRx {
 				if err != nil {
-					t.Fatalf("reep.Read(_): %s", err)
+					t.Fatalf("rep.Read(_, %d, %#v): %s", len(data), opts, err)
 				}
-				if diff := cmp.Diff(buffer.View(data), gotPayload); diff != "" {
+				if diff := cmp.Diff(tcpip.ReadResult{
+					Count: buf.Len(),
+					Total: buf.Len(),
+					RemoteAddr: tcpip.FullAddress{
+						Addr: test.addAddress.AddressWithPrefix.Address,
+					},
+				}, res,
+					checker.IgnoreCmpPath("ControlMessages", "RemoteAddr.NIC", "RemoteAddr.Port"),
+				); diff != "" {
+					t.Errorf("rep.Read: unexpected result (-want +got):\n%s", diff)
+				}
+				if diff := cmp.Diff(data, buf.Bytes()); diff != "" {
 					t.Errorf("got UDP payload mismatch (-want +got):\n%s", diff)
 				}
-				if addr.Addr != test.addAddress.AddressWithPrefix.Address {
-					t.Errorf("got addr.Addr = %s, want = %s", addr.Addr, test.addAddress.AddressWithPrefix.Address)
-				}
-			} else {
-				if err != tcpip.ErrWouldBlock {
-					t.Fatalf("got rep.Read(nil) = (%x, _, %s), want = (_, _, %s)", gotPayload, err, tcpip.ErrWouldBlock)
-				}
+			} else if err != tcpip.ErrWouldBlock {
+				t.Fatalf("got rep.Read = (%v, %s) [with data %x], want = (_, %s)", res, err, buf.Bytes(), tcpip.ErrWouldBlock)
 			}
 		})
 	}

pkg/tcpip/tests/integration/multicast_broadcast_test.go

@@ -15,12 +15,14 @@
 package integration_test
 
 import (
+	"bytes"
 	"net"
 	"testing"
 
 	"github.com/google/go-cmp/cmp"
 	"gvisor.dev/gvisor/pkg/tcpip"
 	"gvisor.dev/gvisor/pkg/tcpip/buffer"
+	"gvisor.dev/gvisor/pkg/tcpip/checker"
 	"gvisor.dev/gvisor/pkg/tcpip/header"
 	"gvisor.dev/gvisor/pkg/tcpip/link/channel"
 	"gvisor.dev/gvisor/pkg/tcpip/link/loopback"
@@ -462,17 +464,23 @@ func TestIncomingMulticastAndBroadcast(t *testing.T) {
 			}
 
 			test.rxUDP(e, test.remoteAddr, test.dstAddr, data)
-			if gotPayload, _, err := ep.Read(nil); test.expectRx {
+			var buf bytes.Buffer
+			var opts tcpip.ReadOptions
+			if res, err := ep.Read(&buf, len(data), opts); test.expectRx {
 				if err != nil {
-					t.Fatalf("Read(nil): %s", err)
+					t.Fatalf("ep.Read(_, %d, %#v): %s", len(data), opts, err)
 				}
-				if diff := cmp.Diff(buffer.View(data), gotPayload); diff != "" {
+				if diff := cmp.Diff(tcpip.ReadResult{
+					Count: buf.Len(),
+					Total: buf.Len(),
+				}, res, checker.IgnoreCmpPath("ControlMessages")); diff != "" {
+					t.Errorf("ep.Read: unexpected result (-want +got):\n%s", diff)
+				}
+				if diff := cmp.Diff(data, buf.Bytes()); diff != "" {
 					t.Errorf("got UDP payload mismatch (-want +got):\n%s", diff)
 				}
-			} else {
-				if err != tcpip.ErrWouldBlock {
-					t.Fatalf("got Read(nil) = (%x, _, %s), want = (_, _, %s)", gotPayload, err, tcpip.ErrWouldBlock)
-				}
+			} else if err != tcpip.ErrWouldBlock {
+				t.Fatalf("got Read = (%v, %s) [with data %x], want = (_, %s)", res, err, buf.Bytes(), tcpip.ErrWouldBlock)
 			}
 		})
 	}
@@ -589,9 +597,19 @@ func TestReuseAddrAndBroadcast(t *testing.T) {
 					// Wait for the endpoint to become readable.
 					<-rep.ch
 
-					if gotPayload, _, err := rep.ep.Read(nil); err != nil {
-						t.Errorf("(eps[%d] write) eps[%d].Read(nil): %s", i, j, err)
-					} else if diff := cmp.Diff(buffer.View(data), gotPayload); diff != "" {
+					var buf bytes.Buffer
+					result, err := rep.ep.Read(&buf, len(data), tcpip.ReadOptions{})
+					if err != nil {
+						t.Errorf("(eps[%d] write) eps[%d].Read: %s", i, j, err)
+						continue
+					}
+					if diff := cmp.Diff(tcpip.ReadResult{
+						Count: buf.Len(),
+						Total: buf.Len(),
+					}, result, checker.IgnoreCmpPath("ControlMessages")); diff != "" {
+						t.Errorf("(eps[%d] write) eps[%d].Read: unexpected result (-want +got):\n%s", i, j, diff)
+					}
+					if diff := cmp.Diff([]byte(data), buf.Bytes()); diff != "" {
 						t.Errorf("(eps[%d] write) got UDP payload from eps[%d] mismatch (-want +got):\n%s", i, j, diff)
 					}
 				}
@@ -719,10 +737,20 @@ func TestUDPAddRemoveMembershipSocketOption(t *testing.T) {
 						t.Fatalf("ep.SetSockOpt(&%#v): %s", addOpt, err)
 					}
 					test.rxUDP(e, test.remoteAddr, test.multicastAddr, data)
-					if gotPayload, _, err := ep.Read(nil); err != nil {
-						t.Fatalf("ep.Read(nil): %s", err)
-					} else if diff := cmp.Diff(buffer.View(data), gotPayload); diff != "" {
-						t.Errorf("got UDP payload mismatch (-want +got):\n%s", diff)
+					var buf bytes.Buffer
+					result, err := ep.Read(&buf, len(data), tcpip.ReadOptions{})
+					if err != nil {
+						t.Fatalf("ep.Read: %s", err)
+					} else {
+						if diff := cmp.Diff(tcpip.ReadResult{
+							Count: buf.Len(),
+							Total: buf.Len(),
+						}, result, checker.IgnoreCmpPath("ControlMessages")); diff != "" {
+							t.Errorf("ep.Read: unexpected result (-want +got):\n%s", diff)
+						}
+						if diff := cmp.Diff(data, buf.Bytes()); diff != "" {
+							t.Errorf("got UDP payload mismatch (-want +got):\n%s", diff)
+						}
 					}
 
 					// We should not receive UDP packets to the group once we leave
@@ -731,8 +759,8 @@ func TestUDPAddRemoveMembershipSocketOption(t *testing.T) {
 					if err := ep.SetSockOpt(&removeOpt); err != nil {
 						t.Fatalf("ep.SetSockOpt(&%#v): %s", removeOpt, err)
 					}
-					if gotPayload, _, err := ep.Read(nil); err != tcpip.ErrWouldBlock {
-						t.Fatalf("got ep.Read(nil) = (%x, _, %s), want = (nil, _, %s)", gotPayload, err, tcpip.ErrWouldBlock)
+					if _, err := ep.Read(&buf, 1, tcpip.ReadOptions{}); err != tcpip.ErrWouldBlock {
+						t.Fatalf("got ep.Read = (_, %s), want = (_, %s)", err, tcpip.ErrWouldBlock)
 					}
 				})
 			}

pkg/tcpip/tests/integration/route_test.go

@@ -15,11 +15,14 @@
 package integration_test
 
 import (
+	"bytes"
+	"math"
 	"testing"
 
 	"github.com/google/go-cmp/cmp"
 	"gvisor.dev/gvisor/pkg/tcpip"
 	"gvisor.dev/gvisor/pkg/tcpip/buffer"
+	"gvisor.dev/gvisor/pkg/tcpip/checker"
 	"gvisor.dev/gvisor/pkg/tcpip/header"
 	"gvisor.dev/gvisor/pkg/tcpip/link/channel"
 	"gvisor.dev/gvisor/pkg/tcpip/link/loopback"
@@ -203,17 +206,26 @@ func TestLocalPing(t *testing.T) {
 			// Wait for the endpoint to become readable.
 			<-ch
 
-			var addr tcpip.FullAddress
-			v, _, err := ep.Read(&addr)
+			var buf bytes.Buffer
+			opts := tcpip.ReadOptions{NeedRemoteAddr: true}
+			res, err := ep.Read(&buf, math.MaxUint16, opts)
 			if err != nil {
-				t.Fatalf("ep.Read(_): %s", err)
+				t.Fatalf("ep.Read(_, %d, %#v): %s", math.MaxUint16, opts, err)
 			}
-			if diff := cmp.Diff(v[icmpDataOffset:], buffer.View(payload[icmpDataOffset:])); diff != "" {
+			if diff := cmp.Diff(tcpip.ReadResult{
+				Count:      buf.Len(),
+				Total:      buf.Len(),
+				RemoteAddr: tcpip.FullAddress{Addr: test.localAddr},
+			}, res, checker.IgnoreCmpPath(
+				"ControlMessages",
+				"RemoteAddr.NIC",
+				"RemoteAddr.Port",
+			)); diff != "" {
+				t.Errorf("ep.Read: unexpected result (-want +got):\n%s", diff)
+			}
+			if diff := cmp.Diff(buf.Bytes()[icmpDataOffset:], []byte(payload[icmpDataOffset:])); diff != "" {
 				t.Errorf("received data mismatch (-want +got):\n%s", diff)
 			}
-			if addr.Addr != test.localAddr {
-				t.Errorf("got addr.Addr = %s, want = %s", addr.Addr, test.localAddr)
-			}
 
 			test.checkLinkEndpoint(t, e)
 		})
@@ -338,14 +350,27 @@ func TestLocalUDP(t *testing.T) {
 					<-serverCH
 
 					var clientAddr tcpip.FullAddress
-					if v, _, err := server.Read(&clientAddr); err != nil {
+					var readBuf bytes.Buffer
+					if read, err := server.Read(&readBuf, math.MaxUint16, tcpip.ReadOptions{NeedRemoteAddr: true}); err != nil {
 						t.Fatalf("server.Read(_): %s", err)
 					} else {
-						if diff := cmp.Diff(buffer.View(clientPayload), v); diff != "" {
-							t.Errorf("server read clientPayload mismatch (-want +got):\n%s", diff)
+						clientAddr = read.RemoteAddr
+
+						if diff := cmp.Diff(tcpip.ReadResult{
+							Count: readBuf.Len(),
+							Total: readBuf.Len(),
+							RemoteAddr: tcpip.FullAddress{
+								Addr: test.canBePrimaryAddr.AddressWithPrefix.Address,
+							},
+						}, read, checker.IgnoreCmpPath(
+							"ControlMessages",
+							"RemoteAddr.NIC",
+							"RemoteAddr.Port",
+						)); diff != "" {
+							t.Errorf("server.Read: unexpected result (-want +got):\n%s", diff)
 						}
-						if clientAddr.Addr != test.canBePrimaryAddr.AddressWithPrefix.Address {
-							t.Errorf("got clientAddr.Addr = %s, want = %s", clientAddr.Addr, test.canBePrimaryAddr.AddressWithPrefix.Address)
+						if diff := cmp.Diff(buffer.View(clientPayload), buffer.View(readBuf.Bytes())); diff != "" {
+							t.Errorf("server read clientPayload mismatch (-want +got):\n%s", diff)
 						}
 						if t.Failed() {
 							t.FailNow()
@@ -367,15 +392,23 @@ func TestLocalUDP(t *testing.T) {
 					// Wait for the client endpoint to become readable.
 					<-clientCH
 
-					var gotServerAddr tcpip.FullAddress
-					if v, _, err := client.Read(&gotServerAddr); err != nil {
+					readBuf.Reset()
+					if read, err := client.Read(&readBuf, math.MaxUint16, tcpip.ReadOptions{NeedRemoteAddr: true}); err != nil {
 						t.Fatalf("client.Read(_): %s", err)
 					} else {
-						if diff := cmp.Diff(buffer.View(serverPayload), v); diff != "" {
-							t.Errorf("client read serverPayload mismatch (-want +got):\n%s", diff)
+						if diff := cmp.Diff(tcpip.ReadResult{
+							Count:      readBuf.Len(),
+							Total:      readBuf.Len(),
+							RemoteAddr: tcpip.FullAddress{Addr: serverAddr.Addr},
+						}, read, checker.IgnoreCmpPath(
+							"ControlMessages",
+							"RemoteAddr.NIC",
+							"RemoteAddr.Port",
+						)); diff != "" {
+							t.Errorf("client.Read: unexpected result (-want +got):\n%s", diff)
 						}
-						if gotServerAddr.Addr != serverAddr.Addr {
-							t.Errorf("got gotServerAddr.Addr = %s, want = %s", gotServerAddr.Addr, serverAddr.Addr)
+						if diff := cmp.Diff(buffer.View(serverPayload), buffer.View(readBuf.Bytes())); diff != "" {
+							t.Errorf("client read serverPayload mismatch (-want +got):\n%s", diff)
 						}
 						if t.Failed() {
 							t.FailNow()