Merge release-20200810.0-23-g47515f475 (automated)

9 files changed, 89 insertions, 92 deletions

diff --git a/pkg/tcpip/link/channel/channel.go b/pkg/tcpip/link/channel/channel.go
index e12a5929b..c95aef63c 100644
--- a/pkg/tcpip/link/channel/channel.go
+++ b/pkg/tcpip/link/channel/channel.go
@@ -274,7 +274,9 @@ func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe
 // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket.
 func (e *Endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error {
 	p := PacketInfo{
-		Pkt:   &stack.PacketBuffer{Data: vv},
+		Pkt: stack.NewPacketBuffer(stack.PacketBufferOptions{
+			Data: vv,
+		}),
 		Proto: 0,
 		GSO:   nil,
 	}
diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go
index c18bb91fb..975309fc8 100644
--- a/pkg/tcpip/link/fdbased/endpoint.go
+++ b/pkg/tcpip/link/fdbased/endpoint.go
@@ -390,8 +390,7 @@ const (
 func (e *endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) {
 	if e.hdrSize > 0 {
 		// Add ethernet header if needed.
-		eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize))
-		pkt.LinkHeader = buffer.View(eth)
+		eth := header.Ethernet(pkt.LinkHeader().Push(header.EthernetMinimumSize))
 		ethHdr := &header.EthernetFields{
 			DstAddr: remote,
 			Type:    protocol,
@@ -420,7 +419,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne
 	if e.Capabilities()&stack.CapabilityHardwareGSO != 0 {
 		vnetHdr := virtioNetHdr{}
 		if gso != nil {
-			vnetHdr.hdrLen = uint16(pkt.Header.UsedLength())
+			vnetHdr.hdrLen = uint16(pkt.HeaderSize())
 			if gso.NeedsCsum {
 				vnetHdr.flags = _VIRTIO_NET_HDR_F_NEEDS_CSUM
 				vnetHdr.csumStart = header.EthernetMinimumSize + gso.L3HdrLen
@@ -443,11 +442,9 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne
 		builder.Add(vnetHdrBuf)
 	}
 
-	builder.Add(pkt.Header.View())
-	for _, v := range pkt.Data.Views() {
+	for _, v := range pkt.Views() {
 		builder.Add(v)
 	}
-
 	return rawfile.NonBlockingWriteIovec(fd, builder.Build())
 }
 
@@ -463,7 +460,7 @@ func (e *endpoint) sendBatch(batchFD int, batch []*stack.PacketBuffer) (int, *tc
 		if e.Capabilities()&stack.CapabilityHardwareGSO != 0 {
 			vnetHdr := virtioNetHdr{}
 			if pkt.GSOOptions != nil {
-				vnetHdr.hdrLen = uint16(pkt.Header.UsedLength())
+				vnetHdr.hdrLen = uint16(pkt.HeaderSize())
 				if pkt.GSOOptions.NeedsCsum {
 					vnetHdr.flags = _VIRTIO_NET_HDR_F_NEEDS_CSUM
 					vnetHdr.csumStart = header.EthernetMinimumSize + pkt.GSOOptions.L3HdrLen
@@ -486,8 +483,7 @@ func (e *endpoint) sendBatch(batchFD int, batch []*stack.PacketBuffer) (int, *tc
 
 		var builder iovec.Builder
 		builder.Add(vnetHdrBuf)
-		builder.Add(pkt.Header.View())
-		for _, v := range pkt.Data.Views() {
+		for _, v := range pkt.Views() {
 			builder.Add(v)
 		}
 		iovecs := builder.Build()
diff --git a/pkg/tcpip/link/fdbased/mmap.go b/pkg/tcpip/link/fdbased/mmap.go
index 2dfd29aa9..c475dda20 100644
--- a/pkg/tcpip/link/fdbased/mmap.go
+++ b/pkg/tcpip/link/fdbased/mmap.go
@@ -18,6 +18,7 @@ package fdbased
 
 import (
 	"encoding/binary"
+	"fmt"
 	"syscall"
 
 	"golang.org/x/sys/unix"
@@ -170,10 +171,9 @@ func (d *packetMMapDispatcher) dispatch() (bool, *tcpip.Error) {
 	var (
 		p             tcpip.NetworkProtocolNumber
 		remote, local tcpip.LinkAddress
-		eth           header.Ethernet
 	)
 	if d.e.hdrSize > 0 {
-		eth = header.Ethernet(pkt)
+		eth := header.Ethernet(pkt)
 		p = eth.Type()
 		remote = eth.SourceAddress()
 		local = eth.DestinationAddress()
@@ -190,10 +190,14 @@ func (d *packetMMapDispatcher) dispatch() (bool, *tcpip.Error) {
 		}
 	}
 
-	pkt = pkt[d.e.hdrSize:]
-	d.e.dispatcher.DeliverNetworkPacket(remote, local, p, &stack.PacketBuffer{
-		Data:       buffer.View(pkt).ToVectorisedView(),
-		LinkHeader: buffer.View(eth),
+	pbuf := stack.NewPacketBuffer(stack.PacketBufferOptions{
+		Data: buffer.View(pkt).ToVectorisedView(),
 	})
+	if d.e.hdrSize > 0 {
+		if _, ok := pbuf.LinkHeader().Consume(d.e.hdrSize); !ok {
+			panic(fmt.Sprintf("LinkHeader().Consume(%d) must succeed", d.e.hdrSize))
+		}
+	}
+	d.e.dispatcher.DeliverNetworkPacket(remote, local, p, pbuf)
 	return true, nil
 }
diff --git a/pkg/tcpip/link/fdbased/packet_dispatchers.go b/pkg/tcpip/link/fdbased/packet_dispatchers.go
index d8f2504b3..8c3ca86d6 100644
--- a/pkg/tcpip/link/fdbased/packet_dispatchers.go
+++ b/pkg/tcpip/link/fdbased/packet_dispatchers.go
@@ -103,7 +103,7 @@ func (d *readVDispatcher) dispatch() (bool, *tcpip.Error) {
 	d.allocateViews(BufConfig)
 
 	n, err := rawfile.BlockingReadv(d.fd, d.iovecs)
-	if err != nil {
+	if n == 0 || err != nil {
 		return false, err
 	}
 	if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 {
@@ -111,17 +111,22 @@ func (d *readVDispatcher) dispatch() (bool, *tcpip.Error) {
 		// isn't used and it isn't in a view.
 		n -= virtioNetHdrSize
 	}
-	if n <= d.e.hdrSize {
-		return false, nil
-	}
+
+	used := d.capViews(n, BufConfig)
+	pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{
+		Data: buffer.NewVectorisedView(n, append([]buffer.View(nil), d.views[:used]...)),
+	})
 
 	var (
 		p             tcpip.NetworkProtocolNumber
 		remote, local tcpip.LinkAddress
-		eth           header.Ethernet
 	)
 	if d.e.hdrSize > 0 {
-		eth = header.Ethernet(d.views[0][:header.EthernetMinimumSize])
+		hdr, ok := pkt.LinkHeader().Consume(d.e.hdrSize)
+		if !ok {
+			return false, nil
+		}
+		eth := header.Ethernet(hdr)
 		p = eth.Type()
 		remote = eth.SourceAddress()
 		local = eth.DestinationAddress()
@@ -138,13 +143,6 @@ func (d *readVDispatcher) dispatch() (bool, *tcpip.Error) {
 		}
 	}
 
-	used := d.capViews(n, BufConfig)
-	pkt := &stack.PacketBuffer{
-		Data:       buffer.NewVectorisedView(n, append([]buffer.View(nil), d.views[:used]...)),
-		LinkHeader: buffer.View(eth),
-	}
-	pkt.Data.TrimFront(d.e.hdrSize)
-
 	d.e.dispatcher.DeliverNetworkPacket(remote, local, p, pkt)
 
 	// Prepare e.views for another packet: release used views.
@@ -268,17 +266,22 @@ func (d *recvMMsgDispatcher) dispatch() (bool, *tcpip.Error) {
 		if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 {
 			n -= virtioNetHdrSize
 		}
-		if n <= d.e.hdrSize {
-			return false, nil
-		}
+
+		used := d.capViews(k, int(n), BufConfig)
+		pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{
+			Data: buffer.NewVectorisedView(int(n), append([]buffer.View(nil), d.views[k][:used]...)),
+		})
 
 		var (
 			p             tcpip.NetworkProtocolNumber
 			remote, local tcpip.LinkAddress
-			eth           header.Ethernet
 		)
 		if d.e.hdrSize > 0 {
-			eth = header.Ethernet(d.views[k][0][:header.EthernetMinimumSize])
+			hdr, ok := pkt.LinkHeader().Consume(d.e.hdrSize)
+			if !ok {
+				return false, nil
+			}
+			eth := header.Ethernet(hdr)
 			p = eth.Type()
 			remote = eth.SourceAddress()
 			local = eth.DestinationAddress()
@@ -295,12 +298,6 @@ func (d *recvMMsgDispatcher) dispatch() (bool, *tcpip.Error) {
 			}
 		}
 
-		used := d.capViews(k, int(n), BufConfig)
-		pkt := &stack.PacketBuffer{
-			Data:       buffer.NewVectorisedView(int(n), append([]buffer.View(nil), d.views[k][:used]...)),
-			LinkHeader: buffer.View(eth),
-		}
-		pkt.Data.TrimFront(d.e.hdrSize)
 		d.e.dispatcher.DeliverNetworkPacket(remote, local, p, pkt)
 
 		// Prepare e.views for another packet: release used views.
diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go
index 781cdd317..38aa694e4 100644
--- a/pkg/tcpip/link/loopback/loopback.go
+++ b/pkg/tcpip/link/loopback/loopback.go
@@ -77,16 +77,16 @@ func (*endpoint) Wait() {}
 // WritePacket implements stack.LinkEndpoint.WritePacket. It delivers outbound
 // packets to the network-layer dispatcher.
 func (e *endpoint) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error {
-	views := make([]buffer.View, 1, 1+len(pkt.Data.Views()))
-	views[0] = pkt.Header.View()
-	views = append(views, pkt.Data.Views()...)
+	// Construct data as the unparsed portion for the loopback packet.
+	data := buffer.NewVectorisedView(pkt.Size(), pkt.Views())
 
 	// Because we're immediately turning around and writing the packet back
 	// to the rx path, we intentionally don't preserve the remote and local
 	// link addresses from the stack.Route we're passed.
-	e.dispatcher.DeliverNetworkPacket("" /* remote */, "" /* local */, protocol, &stack.PacketBuffer{
-		Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views),
+	newPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{
+		Data: data,
 	})
+	e.dispatcher.DeliverNetworkPacket("" /* remote */, "" /* local */, protocol, newPkt)
 
 	return nil
 }
@@ -98,18 +98,17 @@ func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, stack.PacketBufferList
 
 // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket.
 func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error {
+	pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{
+		Data: vv,
+	})
 	// There should be an ethernet header at the beginning of vv.
-	hdr, ok := vv.PullUp(header.EthernetMinimumSize)
+	hdr, ok := pkt.LinkHeader().Consume(header.EthernetMinimumSize)
 	if !ok {
 		// Reject the packet if it's shorter than an ethernet header.
 		return tcpip.ErrBadAddress
 	}
 	linkHeader := header.Ethernet(hdr)
-	vv.TrimFront(len(linkHeader))
-	e.dispatcher.DeliverNetworkPacket("" /* remote */, "" /* local */, linkHeader.Type(), &stack.PacketBuffer{
-		Data:       vv,
-		LinkHeader: buffer.View(linkHeader),
-	})
+	e.dispatcher.DeliverNetworkPacket("" /* remote */, "" /* local */, linkHeader.Type(), pkt)
 
 	return nil
 }
diff --git a/pkg/tcpip/link/sharedmem/sharedmem.go b/pkg/tcpip/link/sharedmem/sharedmem.go
index 507c76b76..7fb8a6c49 100644
--- a/pkg/tcpip/link/sharedmem/sharedmem.go
+++ b/pkg/tcpip/link/sharedmem/sharedmem.go
@@ -186,8 +186,7 @@ func (e *endpoint) LinkAddress() tcpip.LinkAddress {
 // AddHeader implements stack.LinkEndpoint.AddHeader.
 func (e *endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) {
 	// Add ethernet header if needed.
-	eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize))
-	pkt.LinkHeader = buffer.View(eth)
+	eth := header.Ethernet(pkt.LinkHeader().Push(header.EthernetMinimumSize))
 	ethHdr := &header.EthernetFields{
 		DstAddr: remote,
 		Type:    protocol,
@@ -207,10 +206,10 @@ func (e *endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.Net
 func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error {
 	e.AddHeader(r.LocalLinkAddress, r.RemoteLinkAddress, protocol, pkt)
 
-	v := pkt.Data.ToView()
+	views := pkt.Views()
 	// Transmit the packet.
 	e.mu.Lock()
-	ok := e.tx.transmit(pkt.Header.View(), v)
+	ok := e.tx.transmit(views...)
 	e.mu.Unlock()
 
 	if !ok {
@@ -227,10 +226,10 @@ func (e *endpoint) WritePackets(r *stack.Route, _ *stack.GSO, pkts stack.PacketB
 
 // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket.
 func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error {
-	v := vv.ToView()
+	views := vv.Views()
 	// Transmit the packet.
 	e.mu.Lock()
-	ok := e.tx.transmit(v, buffer.View{})
+	ok := e.tx.transmit(views...)
 	e.mu.Unlock()
 
 	if !ok {
@@ -276,16 +275,18 @@ func (e *endpoint) dispatchLoop(d stack.NetworkDispatcher) {
 			rxb[i].Size = e.bufferSize
 		}
 
-		if n < header.EthernetMinimumSize {
+		pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{
+			Data: buffer.View(b).ToVectorisedView(),
+		})
+
+		hdr, ok := pkt.LinkHeader().Consume(header.EthernetMinimumSize)
+		if !ok {
 			continue
 		}
+		eth := header.Ethernet(hdr)
 
 		// Send packet up the stack.
-		eth := header.Ethernet(b[:header.EthernetMinimumSize])
-		d.DeliverNetworkPacket(eth.SourceAddress(), eth.DestinationAddress(), eth.Type(), &stack.PacketBuffer{
-			Data:       buffer.View(b[header.EthernetMinimumSize:]).ToVectorisedView(),
-			LinkHeader: buffer.View(eth),
-		})
+		d.DeliverNetworkPacket(eth.SourceAddress(), eth.DestinationAddress(), eth.Type(), pkt)
 	}
 
 	// Clean state.
diff --git a/pkg/tcpip/link/sharedmem/tx.go b/pkg/tcpip/link/sharedmem/tx.go
index 6b8d7859d..44f421c2d 100644
--- a/pkg/tcpip/link/sharedmem/tx.go
+++ b/pkg/tcpip/link/sharedmem/tx.go
@@ -18,6 +18,7 @@ import (
 	"math"
 	"syscall"
 
+	"gvisor.dev/gvisor/pkg/tcpip/buffer"
 	"gvisor.dev/gvisor/pkg/tcpip/link/sharedmem/queue"
 )
 
@@ -76,9 +77,9 @@ func (t *tx) cleanup() {
 	syscall.Munmap(t.data)
 }
 
-// transmit sends a packet made up of up to two buffers. Returns a boolean that
-// specifies whether the packet was successfully transmitted.
-func (t *tx) transmit(a, b []byte) bool {
+// transmit sends a packet made of bufs. Returns a boolean that specifies
+// whether the packet was successfully transmitted.
+func (t *tx) transmit(bufs ...buffer.View) bool {
 	// Pull completions from the tx queue and add their buffers back to the
 	// pool so that we can reuse them.
 	for {
@@ -93,7 +94,10 @@ func (t *tx) transmit(a, b []byte) bool {
 	}
 
 	bSize := t.bufs.entrySize
-	total := uint32(len(a) + len(b))
+	total := uint32(0)
+	for _, data := range bufs {
+		total += uint32(len(data))
+	}
 	bufCount := (total + bSize - 1) / bSize
 
 	// Allocate enough buffers to hold all the data.
@@ -115,7 +119,7 @@ func (t *tx) transmit(a, b []byte) bool {
 	// Copy data into allocated buffers.
 	nBuf := buf
 	var dBuf []byte
-	for _, data := range [][]byte{a, b} {
+	for _, data := range bufs {
 		for len(data) > 0 {
 			if len(dBuf) == 0 {
 				dBuf = t.data[nBuf.Offset:][:nBuf.Size]
diff --git a/pkg/tcpip/link/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go
index 509076643..4fb127978 100644
--- a/pkg/tcpip/link/sniffer/sniffer.go
+++ b/pkg/tcpip/link/sniffer/sniffer.go
@@ -134,7 +134,7 @@ func (e *endpoint) dumpPacket(prefix string, gso *stack.GSO, protocol tcpip.Netw
 		logPacket(prefix, protocol, pkt, gso)
 	}
 	if writer != nil && atomic.LoadUint32(&LogPacketsToPCAP) == 1 {
-		totalLength := pkt.Header.UsedLength() + pkt.Data.Size()
+		totalLength := pkt.Size()
 		length := totalLength
 		if max := int(e.maxPCAPLen); length > max {
 			length = max
@@ -155,12 +155,11 @@ func (e *endpoint) dumpPacket(prefix string, gso *stack.GSO, protocol tcpip.Netw
 				length -= n
 			}
 		}
-		write(pkt.Header.View())
-		for _, view := range pkt.Data.Views() {
+		for _, v := range pkt.Views() {
 			if length == 0 {
 				break
 			}
-			write(view)
+			write(v)
 		}
 	}
 }
@@ -185,9 +184,9 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe
 
 // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket.
 func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error {
-	e.dumpPacket("send", nil, 0, &stack.PacketBuffer{
+	e.dumpPacket("send", nil, 0, stack.NewPacketBuffer(stack.PacketBufferOptions{
 		Data: vv,
-	})
+	}))
 	return e.Endpoint.WriteRawPacket(vv)
 }
 
@@ -201,12 +200,8 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.P
 	var fragmentOffset uint16
 	var moreFragments bool
 
-	// Create a clone of pkt, including any headers if present. Avoid allocating
-	// backing memory for the clone.
-	views := [8]buffer.View{}
-	vv := buffer.NewVectorisedView(0, views[:0])
-	vv.AppendView(pkt.Header.View())
-	vv.Append(pkt.Data)
+	// Examine the packet using a new VV. Backing storage must not be written.
+	vv := buffer.NewVectorisedView(pkt.Size(), pkt.Views())
 
 	switch protocol {
 	case header.IPv4ProtocolNumber:
diff --git a/pkg/tcpip/link/tun/device.go b/pkg/tcpip/link/tun/device.go
index 22b0a12bd..3b1510a33 100644
--- a/pkg/tcpip/link/tun/device.go
+++ b/pkg/tcpip/link/tun/device.go
@@ -215,12 +215,11 @@ func (d *Device) Write(data []byte) (int64, error) {
 		remote = tcpip.LinkAddress(zeroMAC[:])
 	}
 
-	pkt := &stack.PacketBuffer{
-		Data: buffer.View(data).ToVectorisedView(),
-	}
-	if ethHdr != nil {
-		pkt.LinkHeader = buffer.View(ethHdr)
-	}
+	pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{
+		ReserveHeaderBytes: len(ethHdr),
+		Data:               buffer.View(data).ToVectorisedView(),
+	})
+	copy(pkt.LinkHeader().Push(len(ethHdr)), ethHdr)
 	endpoint.InjectLinkAddr(protocol, remote, pkt)
 	return dataLen, nil
 }
@@ -265,21 +264,22 @@ func (d *Device) encodePkt(info *channel.PacketInfo) (buffer.View, bool) {
 	// If the packet does not already have link layer header, and the route
 	// does not exist, we can't compute it. This is possibly a raw packet, tun
 	// device doesn't support this at the moment.
-	if info.Pkt.LinkHeader == nil && info.Route.RemoteLinkAddress == "" {
+	if info.Pkt.LinkHeader().View().IsEmpty() && info.Route.RemoteLinkAddress == "" {
 		return nil, false
 	}
 
 	// Ethernet header (TAP only).
 	if d.hasFlags(linux.IFF_TAP) {
 		// Add ethernet header if not provided.
-		if info.Pkt.LinkHeader == nil {
+		if info.Pkt.LinkHeader().View().IsEmpty() {
 			d.endpoint.AddHeader(info.Route.LocalLinkAddress, info.Route.RemoteLinkAddress, info.Proto, info.Pkt)
 		}
-		vv.AppendView(info.Pkt.LinkHeader)
+		vv.AppendView(info.Pkt.LinkHeader().View())
 	}
 
 	// Append upper headers.
-	vv.AppendView(buffer.View(info.Pkt.Header.View()[len(info.Pkt.LinkHeader):]))
+	vv.AppendView(info.Pkt.NetworkHeader().View())
+	vv.AppendView(info.Pkt.TransportHeader().View())
 	// Append data payload.
 	vv.Append(info.Pkt.Data)
 
@@ -361,8 +361,7 @@ func (e *tunEndpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.
 	if !e.isTap {
 		return
 	}
-	eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize))
-	pkt.LinkHeader = buffer.View(eth)
+	eth := header.Ethernet(pkt.LinkHeader().Push(header.EthernetMinimumSize))
 	hdr := &header.EthernetFields{
 		SrcAddr: local,
 		DstAddr: remote,