Define tcpip.Payloader in terms of io.Reader

Fixes #1509.

PiperOrigin-RevId: 353295589

pkg/sentry/kernel/task_block.go

@@ -69,7 +69,7 @@ func (t *Task) BlockWithTimeout(C chan struct{}, haveTimeout bool, timeout time.
 // syserror.ErrInterrupted if t is interrupted.
 //
 // Preconditions: The caller must be running on the task goroutine.
-func (t *Task) BlockWithDeadline(C chan struct{}, haveDeadline bool, deadline ktime.Time) error {
+func (t *Task) BlockWithDeadline(C <-chan struct{}, haveDeadline bool, deadline ktime.Time) error {
 	if !haveDeadline {
 		return t.block(C, nil)
 	}

pkg/sentry/socket/netstack/BUILD

@@ -41,7 +41,6 @@ go_library(
         "//pkg/syserr",
         "//pkg/syserror",
         "//pkg/tcpip",
-        "//pkg/tcpip/buffer",
         "//pkg/tcpip/header",
         "//pkg/tcpip/network/ipv4",
         "//pkg/tcpip/network/ipv6",

pkg/sentry/socket/netstack/netstack.go

@@ -19,7 +19,7 @@
 // be used to expose certain endpoints to the sentry while leaving others out,
 // for example, TCP endpoints and Unix-domain endpoints.
 //
-// Lock ordering: netstack => mm: ioSequencePayload copies user memory inside
+// Lock ordering: netstack => mm: ioSequenceReadWriter copies user memory inside
 // tcpip.Endpoint.Write(). Netstack is allowed to (and does) hold locks during
 // this operation.
 package netstack
@@ -55,7 +55,6 @@ import (
 	"gvisor.dev/gvisor/pkg/syserr"
 	"gvisor.dev/gvisor/pkg/syserror"
 	"gvisor.dev/gvisor/pkg/tcpip"
-	"gvisor.dev/gvisor/pkg/tcpip/buffer"
 	"gvisor.dev/gvisor/pkg/tcpip/header"
 	"gvisor.dev/gvisor/pkg/tcpip/stack"
 	"gvisor.dev/gvisor/pkg/tcpip/transport/tcp"
@@ -440,45 +439,10 @@ func (s *SocketOperations) WriteTo(ctx context.Context, _ *fs.File, dst io.Write
 	return int64(res.Count), nil
 }
 
-// ioSequencePayload implements tcpip.Payload.
-//
-// t copies user memory bytes on demand based on the requested size.
-type ioSequencePayload struct {
-	ctx context.Context
-	src usermem.IOSequence
-}
-
-// FullPayload implements tcpip.Payloader.FullPayload
-func (i *ioSequencePayload) FullPayload() ([]byte, *tcpip.Error) {
-	return i.Payload(int(i.src.NumBytes()))
-}
-
-// Payload implements tcpip.Payloader.Payload.
-func (i *ioSequencePayload) Payload(size int) ([]byte, *tcpip.Error) {
-	if max := int(i.src.NumBytes()); size > max {
-		size = max
-	}
-	v := buffer.NewView(size)
-	if _, err := i.src.CopyIn(i.ctx, v); err != nil {
-		// EOF can be returned only if src is a file and this means it
-		// is in a splice syscall and the error has to be ignored.
-		if err == io.EOF {
-			return v, nil
-		}
-		return nil, tcpip.ErrBadAddress
-	}
-	return v, nil
-}
-
-// DropFirst drops the first n bytes from underlying src.
-func (i *ioSequencePayload) DropFirst(n int) {
-	i.src = i.src.DropFirst(int(n))
-}
-
 // Write implements fs.FileOperations.Write.
 func (s *SocketOperations) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, _ int64) (int64, error) {
-	f := &ioSequencePayload{ctx: ctx, src: src}
-	n, err := s.Endpoint.Write(f, tcpip.WriteOptions{})
+	r := src.Reader(ctx)
+	n, err := s.Endpoint.Write(r, tcpip.WriteOptions{})
 	if err == tcpip.ErrWouldBlock {
 		return 0, syserror.ErrWouldBlock
 	}
@@ -486,69 +450,40 @@ func (s *SocketOperations) Write(ctx context.Context, _ *fs.File, src usermem.IO
 		return 0, syserr.TranslateNetstackError(err).ToError()
 	}
 
-	if int64(n) < src.NumBytes() {
-		return int64(n), syserror.ErrWouldBlock
+	if n < src.NumBytes() {
+		return n, syserror.ErrWouldBlock
 	}
 
-	return int64(n), nil
+	return n, nil
 }
 
-// readerPayload implements tcpip.Payloader.
-//
-// It allocates a view and reads from a reader on-demand, based on available
-// capacity in the endpoint.
-type readerPayload struct {
-	ctx   context.Context
-	r     io.Reader
-	count int64
-	err   error
+var _ tcpip.Payloader = (*limitedPayloader)(nil)
+
+type limitedPayloader struct {
+	io.LimitedReader
 }
 
-// FullPayload implements tcpip.Payloader.FullPayload.
-func (r *readerPayload) FullPayload() ([]byte, *tcpip.Error) {
-	return r.Payload(int(r.count))
-}
-
-// Payload implements tcpip.Payloader.Payload.
-func (r *readerPayload) Payload(size int) ([]byte, *tcpip.Error) {
-	if size > int(r.count) {
-		size = int(r.count)
-	}
-	v := buffer.NewView(size)
-	n, err := r.r.Read(v)
-	if n > 0 {
-		// We ignore the error here. It may re-occur on subsequent
-		// reads, but for now we can enqueue some amount of data.
-		r.count -= int64(n)
-		return v[:n], nil
-	}
-	if err == syserror.ErrWouldBlock {
-		return nil, tcpip.ErrWouldBlock
-	} else if err != nil {
-		r.err = err // Save for propation.
-		return nil, tcpip.ErrBadAddress
-	}
-
-	// There is no data and no error. Return an error, which will propagate
-	// r.err, which will be nil. This is the desired result: (0, nil).
-	return nil, tcpip.ErrBadAddress
+func (l limitedPayloader) Len() int {
+	return int(l.N)
 }
 
 // ReadFrom implements fs.FileOperations.ReadFrom.
 func (s *SocketOperations) ReadFrom(ctx context.Context, _ *fs.File, r io.Reader, count int64) (int64, error) {
-	f := &readerPayload{ctx: ctx, r: r, count: count}
-	n, err := s.Endpoint.Write(f, tcpip.WriteOptions{
+	f := limitedPayloader{
+		LimitedReader: io.LimitedReader{
+			R: r,
+			N: count,
+		},
+	}
+	n, err := s.Endpoint.Write(&f, tcpip.WriteOptions{
 		// Reads may be destructive but should be very fast,
 		// so we can't release the lock while copying data.
 		Atomic: true,
 	})
-	if err == tcpip.ErrWouldBlock {
-		return n, syserror.ErrWouldBlock
-	} else if err != nil {
-		return int64(n), f.err // Propagate error.
+	if err == tcpip.ErrBadBuffer {
+		err = nil
 	}
-
-	return int64(n), nil
+	return n, syserr.TranslateNetstackError(err).ToError()
 }
 
 // Readiness returns a mask of ready events for socket s.
@@ -2836,45 +2771,46 @@ func (s *socketOpsCommon) SendMsg(t *kernel.Task, src usermem.IOSequence, to []b
 		EndOfRecord: flags&linux.MSG_EOR != 0,
 	}
 
-	v := &ioSequencePayload{t, src}
-	n, err := s.Endpoint.Write(v, opts)
-	dontWait := flags&linux.MSG_DONTWAIT != 0
-	if err == nil && (n >= v.src.NumBytes() || dontWait) {
-		// Complete write.
-		return int(n), nil
-	}
-	if err != nil && (err != tcpip.ErrWouldBlock || dontWait) {
-		return int(n), syserr.TranslateNetstackError(err)
-	}
-
-	// We'll have to block. Register for notification and keep trying to
-	// send all the data.
-	e, ch := waiter.NewChannelEntry(nil)
-	s.EventRegister(&e, waiter.EventOut)
-	defer s.EventUnregister(&e)
-
-	v.DropFirst(int(n))
-	total := n
+	r := src.Reader(t)
+	var (
+		total int64
+		entry waiter.Entry
+		ch    <-chan struct{}
+	)
 	for {
-		n, err = s.Endpoint.Write(v, opts)
-		v.DropFirst(int(n))
+		n, err := s.Endpoint.Write(r, opts)
 		total += n
-
-		if err != nil && err != tcpip.ErrWouldBlock && total == 0 {
-			return 0, syserr.TranslateNetstackError(err)
+		if flags&linux.MSG_DONTWAIT != 0 {
+			return int(total), syserr.TranslateNetstackError(err)
 		}
-
-		if err == nil && v.src.NumBytes() == 0 || err != nil && err != tcpip.ErrWouldBlock {
-			return int(total), nil
-		}
-
-		if err := t.BlockWithDeadline(ch, haveDeadline, deadline); err != nil {
-			if err == syserror.ETIMEDOUT {
-				return int(total), syserr.ErrTryAgain
+		switch err {
+		case nil:
+			if total == src.NumBytes() {
+				break
 			}
-			// handleIOError will consume errors from t.Block if needed.
-			return int(total), syserr.FromError(err)
+			fallthrough
+		case tcpip.ErrWouldBlock:
+			if ch == nil {
+				// We'll have to block. Register for notification and keep trying to
+				// send all the data.
+				entry, ch = waiter.NewChannelEntry(nil)
+				s.EventRegister(&entry, waiter.EventOut)
+				defer s.EventUnregister(&entry)
+			} else {
+				// Don't wait immediately after registration in case more data
+				// became available between when we last checked and when we setup
+				// the notification.
+				if err := t.BlockWithDeadline(ch, haveDeadline, deadline); err != nil {
+					if err == syserror.ETIMEDOUT {
+						return int(total), syserr.ErrTryAgain
+					}
+					// handleIOError will consume errors from t.Block if needed.
+					return int(total), syserr.FromError(err)
+				}
+			}
+			continue
 		}
+		return int(total), syserr.TranslateNetstackError(err)
 	}
 }
 

pkg/sentry/socket/netstack/netstack_vfs2.go

@@ -128,8 +128,8 @@ func (s *SocketVFS2) Write(ctx context.Context, src usermem.IOSequence, opts vfs
 		return 0, syserror.EOPNOTSUPP
 	}
 
-	f := &ioSequencePayload{ctx: ctx, src: src}
-	n, err := s.Endpoint.Write(f, tcpip.WriteOptions{})
+	r := src.Reader(ctx)
+	n, err := s.Endpoint.Write(r, tcpip.WriteOptions{})
 	if err == tcpip.ErrWouldBlock {
 		return 0, syserror.ErrWouldBlock
 	}
@@ -137,11 +137,11 @@ func (s *SocketVFS2) Write(ctx context.Context, src usermem.IOSequence, opts vfs
 		return 0, syserr.TranslateNetstackError(err).ToError()
 	}
 
-	if int64(n) < src.NumBytes() {
-		return int64(n), syserror.ErrWouldBlock
+	if n < src.NumBytes() {
+		return n, syserror.ErrWouldBlock
 	}
 
-	return int64(n), nil
+	return n, nil
 }
 
 // Accept implements the linux syscall accept(2) for sockets backed by

pkg/tcpip/adapters/gonet/gonet.go

@@ -16,6 +16,7 @@
 package gonet
 
 import (
+	"bytes"
 	"context"
 	"errors"
 	"io"
@@ -354,8 +355,6 @@ func (c *TCPConn) Write(b []byte) (int, error) {
 	default:
 	}
 
-	v := buffer.NewViewFromBytes(b)
-
 	// We must handle two soft failure conditions simultaneously:
 	//  1. Write may write nothing and return tcpip.ErrWouldBlock.
 	//     If this happens, we need to register for notifications if we have
@@ -368,22 +367,23 @@ func (c *TCPConn) Write(b []byte) (int, error) {
 	// There is no guarantee that all of the condition #1s will occur before
 	// all of the condition #2s or visa-versa.
 	var (
-		err      *tcpip.Error
-		nbytes   int
-		reg      bool
-		notifyCh chan struct{}
+		r      bytes.Reader
+		nbytes int
+		entry  waiter.Entry
+		ch     <-chan struct{}
 	)
-	for nbytes < len(b) && (err == tcpip.ErrWouldBlock || err == nil) {
-		if err == tcpip.ErrWouldBlock {
-			if !reg {
-				// Only register once.
-				reg = true
+	for nbytes != len(b) {
+		r.Reset(b[nbytes:])
+		n, err := c.ep.Write(&r, tcpip.WriteOptions{})
+		nbytes += int(n)
+		switch err {
+		case nil:
+		case tcpip.ErrWouldBlock:
+			if ch == nil {
+				entry, ch = waiter.NewChannelEntry(nil)
 
-				// Create wait queue entry that notifies a channel.
-				var waitEntry waiter.Entry
-				waitEntry, notifyCh = waiter.NewChannelEntry(nil)
-				c.wq.EventRegister(&waitEntry, waiter.EventOut)
-				defer c.wq.EventUnregister(&waitEntry)
+				c.wq.EventRegister(&entry, waiter.EventOut)
+				defer c.wq.EventUnregister(&entry)
 			} else {
 				// Don't wait immediately after registration in case more data
 				// became available between when we last checked and when we setup
@@ -391,22 +391,15 @@ func (c *TCPConn) Write(b []byte) (int, error) {
 				select {
 				case <-deadline:
 					return nbytes, c.newOpError("write", &timeoutError{})
-				case <-notifyCh:
+				case <-ch:
+					continue
 				}
 			}
+		default:
+			return nbytes, c.newOpError("write", errors.New(err.String()))
 		}
-
-		var n int64
-		n, err = c.ep.Write(tcpip.SlicePayload(v), tcpip.WriteOptions{})
-		nbytes += int(n)
-		v.TrimFront(int(n))
 	}
-
-	if err == nil {
-		return nbytes, nil
-	}
-
-	return nbytes, c.newOpError("write", errors.New(err.String()))
+	return nbytes, nil
 }
 
 // Close implements net.Conn.Close.
@@ -644,16 +637,18 @@ func (c *UDPConn) WriteTo(b []byte, addr net.Addr) (int, error) {
 	}
 
 	// If we're being called by Write, there is no addr
-	wopts := tcpip.WriteOptions{}
+	writeOptions := tcpip.WriteOptions{}
 	if addr != nil {
 		ua := addr.(*net.UDPAddr)
-		wopts.To = &tcpip.FullAddress{Addr: tcpip.Address(ua.IP), Port: uint16(ua.Port)}
+		writeOptions.To = &tcpip.FullAddress{
+			Addr: tcpip.Address(ua.IP),
+			Port: uint16(ua.Port),
+		}
 	}
 
-	v := buffer.NewView(len(b))
-	copy(v, b)
-
-	n, err := c.ep.Write(tcpip.SlicePayload(v), wopts)
+	var r bytes.Reader
+	r.Reset(b)
+	n, err := c.ep.Write(&r, writeOptions)
 	if err == tcpip.ErrWouldBlock {
 		// Create wait queue entry that notifies a channel.
 		waitEntry, notifyCh := waiter.NewChannelEntry(nil)
@@ -666,7 +661,7 @@ func (c *UDPConn) WriteTo(b []byte, addr net.Addr) (int, error) {
 			case <-notifyCh:
 			}
 
-			n, err = c.ep.Write(tcpip.SlicePayload(v), wopts)
+			n, err = c.ep.Write(&r, writeOptions)
 			if err != tcpip.ErrWouldBlock {
 				break
 			}

pkg/tcpip/network/ipv6/icmp_test.go

@@ -15,6 +15,7 @@
 package ipv6
 
 import (
+	"bytes"
 	"context"
 	"net"
 	"reflect"
@@ -638,7 +639,6 @@ func TestLinkResolution(t *testing.T) {
 	pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6EchoMinimumSize))
 	pkt.SetType(header.ICMPv6EchoRequest)
 	pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{}))
-	payload := tcpip.SlicePayload(hdr.View())
 
 	// We can't send our payload directly over the route because that
 	// doesn't provoke NDP discovery.
@@ -648,8 +648,12 @@ func TestLinkResolution(t *testing.T) {
 		t.Fatalf("NewEndpoint(_) = (_, %s), want = (_, nil)", err)
 	}
 
-	if _, err := ep.Write(payload, tcpip.WriteOptions{To: &tcpip.FullAddress{NIC: nicID, Addr: lladdr1}}); err != nil {
-		t.Fatalf("ep.Write(_): %s", err)
+	{
+		var r bytes.Reader
+		r.Reset(hdr.View())
+		if _, err := ep.Write(&r, tcpip.WriteOptions{To: &tcpip.FullAddress{NIC: nicID, Addr: lladdr1}}); err != nil {
+			t.Fatalf("ep.Write(_): %s", err)
+		}
 	}
 	for _, args := range []routeArgs{
 		{src: c.linkEP0, dst: c.linkEP1, typ: header.ICMPv6NeighborSolicit, remoteLinkAddr: header.EthernetAddressFromMulticastIPv6Address(header.SolicitedNodeAddr(lladdr1))},

pkg/tcpip/sample/tun_tcp_connect/BUILD

@@ -8,7 +8,6 @@ go_binary(
     visibility = ["//:sandbox"],
     deps = [
         "//pkg/tcpip",
-        "//pkg/tcpip/buffer",
         "//pkg/tcpip/header",
         "//pkg/tcpip/link/fdbased",
         "//pkg/tcpip/link/rawfile",

pkg/tcpip/sample/tun_tcp_connect/main.go

@@ -41,7 +41,7 @@
 package main
 
 import (
-	"bufio"
+	"bytes"
 	"fmt"
 	"log"
 	"math/rand"
@@ -51,7 +51,6 @@ import (
 	"time"
 
 	"gvisor.dev/gvisor/pkg/tcpip"
-	"gvisor.dev/gvisor/pkg/tcpip/buffer"
 	"gvisor.dev/gvisor/pkg/tcpip/header"
 	"gvisor.dev/gvisor/pkg/tcpip/link/fdbased"
 	"gvisor.dev/gvisor/pkg/tcpip/link/rawfile"
@@ -71,24 +70,21 @@ func writer(ch chan struct{}, ep tcpip.Endpoint) {
 		close(ch)
 	}()
 
-	r := bufio.NewReader(os.Stdin)
-	for {
-		v := buffer.NewView(1024)
-		n, err := r.Read(v)
-		if err != nil {
-			return
-		}
-
-		v.CapLength(n)
-		for len(v) > 0 {
-			n, err := ep.Write(tcpip.SlicePayload(v), tcpip.WriteOptions{})
-			if err != nil {
-				fmt.Println("Write failed:", err)
-				return
+	var b bytes.Buffer
+	if err := func() error {
+		for {
+			if _, err := b.ReadFrom(os.Stdin); err != nil {
+				return fmt.Errorf("b.ReadFrom failed: %w", err)
 			}
 
-			v.TrimFront(int(n))
+			for b.Len() != 0 {
+				if _, err := ep.Write(&b, tcpip.WriteOptions{Atomic: true}); err != nil {
+					return fmt.Errorf("ep.Write failed: %s", err)
+				}
+			}
 		}
+	}(); err != nil {
+		fmt.Println(err)
 	}
 }
 

pkg/tcpip/sample/tun_tcp_echo/main.go

@@ -20,6 +20,7 @@
 package main
 
 import (
+	"bytes"
 	"flag"
 	"io"
 	"log"
@@ -58,7 +59,9 @@ func (e *tcpipError) Error() string {
 }
 
 func (e *endpointWriter) Write(p []byte) (int, error) {
-	n, err := e.ep.Write(tcpip.SlicePayload(p), tcpip.WriteOptions{})
+	var r bytes.Reader
+	r.Reset(p)
+	n, err := e.ep.Write(&r, tcpip.WriteOptions{})
 	if err != nil {
 		return int(n), &tcpipError{
 			inner: err,

pkg/tcpip/stack/transport_test.go

@@ -15,6 +15,7 @@
 package stack_test
 
 import (
+	"bytes"
 	"io"
 	"testing"
 
@@ -95,10 +96,11 @@ func (f *fakeTransportEndpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions
 		return 0, tcpip.ErrNoRoute
 	}
 
-	v, err := p.FullPayload()
-	if err != nil {
-		return 0, err
+	v := make([]byte, p.Len())
+	if _, err := io.ReadFull(p, v); err != nil {
+		return 0, tcpip.ErrBadBuffer
 	}
+
 	pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{
 		ReserveHeaderBytes: int(f.route.MaxHeaderLength()) + fakeTransHeaderLen,
 		Data:               buffer.View(v).ToVectorisedView(),
@@ -520,8 +522,10 @@ func TestTransportSend(t *testing.T) {
 	}
 
 	// Create buffer that will hold the payload.
-	view := buffer.NewView(30)
-	if _, err := ep.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil {
+	b := make([]byte, 30)
+	var r bytes.Reader
+	r.Reset(b)
+	if _, err := ep.Write(&r, tcpip.WriteOptions{}); err != nil {
 		t.Fatalf("write failed: %v", err)
 	}
 

pkg/tcpip/tcpip.go

@@ -29,6 +29,7 @@
 package tcpip
 
 import (
+	"bytes"
 	"errors"
 	"fmt"
 	"io"
@@ -471,30 +472,15 @@ type FullAddress struct {
 // This interface allows the endpoint to request the amount of data it needs
 // based on internal buffers without exposing them.
 type Payloader interface {
-	// FullPayload returns all available bytes.
-	FullPayload() ([]byte, *Error)
+	io.Reader
 
-	// Payload returns a slice containing at most size bytes.
-	Payload(size int) ([]byte, *Error)
+	// Len returns the number of bytes of the unread portion of the
+	// Reader.
+	Len() int
 }
 
-// SlicePayload implements Payloader for slices.
-//
-// This is typically used for tests.
-type SlicePayload []byte
-
-// FullPayload implements Payloader.FullPayload.
-func (s SlicePayload) FullPayload() ([]byte, *Error) {
-	return s, nil
-}
-
-// Payload implements Payloader.Payload.
-func (s SlicePayload) Payload(size int) ([]byte, *Error) {
-	if size > len(s) {
-		size = len(s)
-	}
-	return s[:size], nil
-}
+var _ Payloader = (*bytes.Buffer)(nil)
+var _ Payloader = (*bytes.Reader)(nil)
 
 var _ io.Writer = (*SliceWriter)(nil)
 

pkg/tcpip/tests/integration/forward_test.go

@@ -436,9 +436,10 @@ func TestForwarding(t *testing.T) {
 					write := func(ep tcpip.Endpoint, data []byte) {
 						t.Helper()
 
-						dataPayload := tcpip.SlicePayload(data)
+						var r bytes.Reader
+						r.Reset(data)
 						var wOpts tcpip.WriteOptions
-						n, err := ep.Write(dataPayload, wOpts)
+						n, err := ep.Write(&r, wOpts)
 						if err != nil {
 							t.Fatalf("ep.Write(_, %#v): %s", wOpts, err)
 						}
@@ -486,7 +487,7 @@ func TestForwarding(t *testing.T) {
 
 					read(serverCH, serverEP, data, clientAddr)
 
-					data = tcpip.SlicePayload([]byte{5, 6, 7, 8, 9, 10, 11, 12})
+					data = []byte{5, 6, 7, 8, 9, 10, 11, 12}
 					write(serverEP, data)
 					read(epsAndAddrs.clientReadableCH, epsAndAddrs.clientEP, data, serverAddr)
 				})

pkg/tcpip/tests/integration/link_resolution_test.go

@@ -209,8 +209,10 @@ func TestPing(t *testing.T) {
 			defer ep.Close()
 
 			icmpBuf := test.icmpBuf(t)
+			var r bytes.Reader
+			r.Reset(icmpBuf)
 			wOpts := tcpip.WriteOptions{To: &tcpip.FullAddress{Addr: test.remoteAddr}}
-			if n, err := ep.Write(tcpip.SlicePayload(icmpBuf), wOpts); err != nil {
+			if n, err := ep.Write(&r, wOpts); err != nil {
 				t.Fatalf("ep.Write(_, _): %s", err)
 			} else if want := int64(len(icmpBuf)); n != want {
 				t.Fatalf("got ep.Write(_, _) = (%d, _), want = (%d, _)", n, want)
@@ -360,9 +362,11 @@ func TestTCPLinkResolutionFailure(t *testing.T) {
 			// Wait for an error due to link resolution failing, or the endpoint to be
 			// writable.
 			<-ch
+			var r bytes.Reader
+			r.Reset([]byte{0})
 			var wOpts tcpip.WriteOptions
-			if n, err := clientEP.Write(tcpip.SlicePayload(nil), wOpts); err != test.expectedWriteErr {
-				t.Errorf("got clientEP.Write(nil, %#v) = (%d, %s), want = (_, %s)", wOpts, n, err, test.expectedWriteErr)
+			if n, err := clientEP.Write(&r, wOpts); err != test.expectedWriteErr {
+				t.Errorf("got clientEP.Write(_, %#v) = (%d, %s), want = (_, %s)", wOpts, n, err, test.expectedWriteErr)
 			}
 
 			if test.expectedWriteErr == nil {

pkg/tcpip/tests/integration/loopback_test.go

@@ -232,7 +232,9 @@ func TestLoopbackAcceptAllInSubnetUDP(t *testing.T) {
 					Port: localPort,
 				},
 			}
-			n, err := sep.Write(tcpip.SlicePayload(data), wopts)
+			var r bytes.Reader
+			r.Reset(data)
+			n, err := sep.Write(&r, wopts)
 			if err != nil {
 				t.Fatalf("sep.Write(_, _): %s", err)
 			}

pkg/tcpip/tests/integration/multicast_broadcast_test.go

@@ -586,8 +586,10 @@ func TestReuseAddrAndBroadcast(t *testing.T) {
 						Port: localPort,
 					},
 				}
-				data := tcpip.SlicePayload([]byte{byte(i), 2, 3, 4})
-				if n, err := wep.ep.Write(data, writeOpts); err != nil {
+				data := []byte{byte(i), 2, 3, 4}
+				var r bytes.Reader
+				r.Reset(data)
+				if n, err := wep.ep.Write(&r, writeOpts); err != nil {
 					t.Fatalf("eps[%d].Write(_, _): %s", i, err)
 				} else if want := int64(len(data)); n != want {
 					t.Fatalf("got eps[%d].Write(_, _) = (%d, nil), want = (%d, nil)", i, n, want)

pkg/tcpip/tests/integration/route_test.go

@@ -194,9 +194,11 @@ func TestLocalPing(t *testing.T) {
 				return
 			}
 
-			payload := tcpip.SlicePayload(test.icmpBuf(t))
+			payload := test.icmpBuf(t)
+			var r bytes.Reader
+			r.Reset(payload)
 			var wOpts tcpip.WriteOptions
-			if n, err := ep.Write(payload, wOpts); err != nil {
+			if n, err := ep.Write(&r, wOpts); err != nil {
 				t.Fatalf("ep.Write(%#v, %#v): %s", payload, wOpts, err)
 			} else if n != int64(len(payload)) {
 				t.Fatalf("got ep.Write(%#v, %#v) = (%d, nil), want = (%d, nil)", payload, wOpts, n, len(payload))
@@ -329,12 +331,14 @@ func TestLocalUDP(t *testing.T) {
 						Port: 80,
 					}
 
-					clientPayload := tcpip.SlicePayload([]byte{1, 2, 3, 4})
+					clientPayload := []byte{1, 2, 3, 4}
 					{
+						var r bytes.Reader
+						r.Reset(clientPayload)
 						wOpts := tcpip.WriteOptions{
 							To: &serverAddr,
 						}
-						if n, err := client.Write(clientPayload, wOpts); err != subTest.expectedWriteErr {
+						if n, err := client.Write(&r, wOpts); err != subTest.expectedWriteErr {
 							t.Fatalf("got client.Write(%#v, %#v) = (%d, %s), want = (_, %s)", clientPayload, wOpts, n, err, subTest.expectedWriteErr)
 						} else if subTest.expectedWriteErr != nil {
 							// Nothing else to test if we expected not to be able to send the
@@ -376,12 +380,14 @@ func TestLocalUDP(t *testing.T) {
 						}
 					}
 
-					serverPayload := tcpip.SlicePayload([]byte{1, 2, 3, 4})
+					serverPayload := []byte{1, 2, 3, 4}
 					{
+						var r bytes.Reader
+						r.Reset(serverPayload)
 						wOpts := tcpip.WriteOptions{
 							To: &clientAddr,
 						}
-						if n, err := server.Write(serverPayload, wOpts); err != nil {
+						if n, err := server.Write(&r, wOpts); err != nil {
 							t.Fatalf("server.Write(%#v, %#v): %s", serverPayload, wOpts, err)
 						} else if n != int64(len(serverPayload)) {
 							t.Fatalf("got server.Write(%#v, %#v) = (%d, nil), want = (%d, nil)", serverPayload, wOpts, n, len(serverPayload))

pkg/tcpip/transport/icmp/endpoint.go

@@ -313,11 +313,12 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, *tc
 		route = r
 	}
 
-	v, err := p.FullPayload()
-	if err != nil {
-		return 0, err
+	v := make([]byte, p.Len())
+	if _, err := io.ReadFull(p, v); err != nil {
+		return 0, tcpip.ErrBadBuffer
 	}
 
+	var err *tcpip.Error
 	switch e.NetProto {
 	case header.IPv4ProtocolNumber:
 		err = send4(route, e.ID.LocalPort, v, e.ttl, e.owner)

pkg/tcpip/transport/packet/endpoint.go

@@ -207,7 +207,7 @@ func (ep *endpoint) Read(dst io.Writer, opts tcpip.ReadOptions) (tcpip.ReadResul
 	return res, nil
 }
 
-func (*endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, *tcpip.Error) {
+func (*endpoint) Write(tcpip.Payloader, tcpip.WriteOptions) (int64, *tcpip.Error) {
 	// TODO(gvisor.dev/issue/173): Implement.
 	return 0, tcpip.ErrInvalidOptionValue
 }

pkg/tcpip/transport/raw/endpoint.go

@@ -280,9 +280,9 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, *tc
 		return 0, tcpip.ErrInvalidEndpointState
 	}
 
-	payloadBytes, err := p.FullPayload()
-	if err != nil {
-		return 0, err
+	payloadBytes := make([]byte, p.Len())
+	if _, err := io.ReadFull(p, payloadBytes); err != nil {
+		return 0, tcpip.ErrBadBuffer
 	}
 
 	// If this is an unassociated socket and callee provided a nonzero

pkg/tcpip/transport/tcp/BUILD

@@ -99,7 +99,6 @@ go_test(
         "//pkg/rand",
         "//pkg/sync",
         "//pkg/tcpip",
-        "//pkg/tcpip/buffer",
         "//pkg/tcpip/checker",
         "//pkg/tcpip/header",
         "//pkg/tcpip/link/loopback",