diff options
Diffstat (limited to 'pkg/tcpip')
31 files changed, 700 insertions, 499 deletions
diff --git a/pkg/tcpip/buffer/view.go b/pkg/tcpip/buffer/view.go index 9a3c5d6c3..ea0c5413d 100644 --- a/pkg/tcpip/buffer/view.go +++ b/pkg/tcpip/buffer/view.go @@ -65,6 +65,16 @@ func (v View) ToVectorisedView() VectorisedView { return NewVectorisedView(len(v), []View{v}) } +// IsEmpty returns whether v is of length zero. +func (v View) IsEmpty() bool { + return len(v) == 0 +} + +// Size returns the length of v. +func (v View) Size() int { + return len(v) +} + // VectorisedView is a vectorised version of View using non contiguous memory. // It supports all the convenience methods supported by View. // 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, diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index 31a242482..1ad788a17 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -99,7 +99,7 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBu } func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { - h := header.ARP(pkt.NetworkHeader) + h := header.ARP(pkt.NetworkHeader().View()) if !h.IsValid() { return } @@ -110,17 +110,17 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { if e.linkAddrCache.CheckLocalAddress(e.nicID, header.IPv4ProtocolNumber, localAddr) == 0 { return // we have no useful answer, ignore the request } - hdr := buffer.NewPrependable(int(e.linkEP.MaxHeaderLength()) + header.ARPSize) - packet := header.ARP(hdr.Prepend(header.ARPSize)) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(e.linkEP.MaxHeaderLength()) + header.ARPSize, + }) + packet := header.ARP(pkt.NetworkHeader().Push(header.ARPSize)) packet.SetIPv4OverEthernet() packet.SetOp(header.ARPReply) copy(packet.HardwareAddressSender(), r.LocalLinkAddress[:]) copy(packet.ProtocolAddressSender(), h.ProtocolAddressTarget()) copy(packet.HardwareAddressTarget(), h.HardwareAddressSender()) copy(packet.ProtocolAddressTarget(), h.ProtocolAddressSender()) - e.linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, &stack.PacketBuffer{ - Header: hdr, - }) + _ = e.linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, pkt) fallthrough // also fill the cache from requests case header.ARPReply: addr := tcpip.Address(h.ProtocolAddressSender()) @@ -168,17 +168,17 @@ func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, remoteLinkAdd r.RemoteLinkAddress = header.EthernetBroadcastAddress } - hdr := buffer.NewPrependable(int(linkEP.MaxHeaderLength()) + header.ARPSize) - h := header.ARP(hdr.Prepend(header.ARPSize)) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(linkEP.MaxHeaderLength()) + header.ARPSize, + }) + h := header.ARP(pkt.NetworkHeader().Push(header.ARPSize)) h.SetIPv4OverEthernet() h.SetOp(header.ARPRequest) copy(h.HardwareAddressSender(), linkEP.LinkAddress()) copy(h.ProtocolAddressSender(), localAddr) copy(h.ProtocolAddressTarget(), addr) - return linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, &stack.PacketBuffer{ - Header: hdr, - }) + return linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, pkt) } // ResolveStaticAddress implements stack.LinkAddressResolver.ResolveStaticAddress. @@ -210,12 +210,10 @@ func (*protocol) Wait() {} // Parse implements stack.NetworkProtocol.Parse. func (*protocol) Parse(pkt *stack.PacketBuffer) (proto tcpip.TransportProtocolNumber, hasTransportHdr bool, ok bool) { - hdr, ok := pkt.Data.PullUp(header.ARPSize) + _, ok = pkt.NetworkHeader().Consume(header.ARPSize) if !ok { return 0, false, false } - pkt.NetworkHeader = hdr - pkt.Data.TrimFront(header.ARPSize) return 0, false, true } diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go index 94803a359..067d770f3 100644 --- a/pkg/tcpip/network/ipv4/icmp.go +++ b/pkg/tcpip/network/ipv4/icmp.go @@ -89,12 +89,14 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer) { return } + // Make a copy of data before pkt gets sent to raw socket. + // DeliverTransportPacket will take ownership of pkt. + replyData := pkt.Data.Clone(nil) + replyData.TrimFront(header.ICMPv4MinimumSize) + // It's possible that a raw socket expects to receive this. h.SetChecksum(wantChecksum) - e.dispatcher.DeliverTransportPacket(r, header.ICMPv4ProtocolNumber, &stack.PacketBuffer{ - Data: pkt.Data.Clone(nil), - NetworkHeader: append(buffer.View(nil), pkt.NetworkHeader...), - }) + e.dispatcher.DeliverTransportPacket(r, header.ICMPv4ProtocolNumber, pkt) remoteLinkAddr := r.RemoteLinkAddress @@ -116,24 +118,26 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer) { // Use the remote link address from the incoming packet. r.ResolveWith(remoteLinkAddr) - vv := pkt.Data.Clone(nil) - vv.TrimFront(header.ICMPv4MinimumSize) - hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv4MinimumSize) - pkt := header.ICMPv4(hdr.Prepend(header.ICMPv4MinimumSize)) - copy(pkt, h) - pkt.SetType(header.ICMPv4EchoReply) - pkt.SetChecksum(0) - pkt.SetChecksum(^header.Checksum(pkt, header.ChecksumVV(vv, 0))) + // Prepare a reply packet. + icmpHdr := make(header.ICMPv4, header.ICMPv4MinimumSize) + copy(icmpHdr, h) + icmpHdr.SetType(header.ICMPv4EchoReply) + icmpHdr.SetChecksum(0) + icmpHdr.SetChecksum(^header.Checksum(icmpHdr, header.ChecksumVV(replyData, 0))) + dataVV := buffer.View(icmpHdr).ToVectorisedView() + dataVV.Append(replyData) + replyPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(r.MaxHeaderLength()), + Data: dataVV, + }) + + // Send out the reply packet. sent := stats.ICMP.V4PacketsSent if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{ Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS, - }, &stack.PacketBuffer{ - Header: hdr, - Data: vv, - TransportHeader: buffer.View(pkt), - }); err != nil { + }, replyPkt); err != nil { sent.Dropped.Increment() return } diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index 9ff27a363..3cd48ceb3 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -21,7 +21,6 @@ package ipv4 import ( - "fmt" "sync/atomic" "gvisor.dev/gvisor/pkg/tcpip" @@ -127,14 +126,12 @@ func (e *endpoint) NetworkProtocolNumber() tcpip.NetworkProtocolNumber { } // writePacketFragments calls e.linkEP.WritePacket with each packet fragment to -// write. It assumes that the IP header is entirely in pkt.Header but does not -// assume that only the IP header is in pkt.Header. It assumes that the input -// packet's stated length matches the length of the header+payload. mtu -// includes the IP header and options. This does not support the DontFragment -// IP flag. +// write. It assumes that the IP header is already present in pkt.NetworkHeader. +// pkt.TransportHeader may be set. mtu includes the IP header and options. This +// does not support the DontFragment IP flag. func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, mtu int, pkt *stack.PacketBuffer) *tcpip.Error { // This packet is too big, it needs to be fragmented. - ip := header.IPv4(pkt.Header.View()) + ip := header.IPv4(pkt.NetworkHeader().View()) flags := ip.Flags() // Update mtu to take into account the header, which will exist in all @@ -148,88 +145,84 @@ func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, mtu int, outerMTU := innerMTU + int(ip.HeaderLength()) offset := ip.FragmentOffset() - originalAvailableLength := pkt.Header.AvailableLength() + + // Keep the length reserved for link-layer, we need to create fragments with + // the same reserved length. + reservedForLink := pkt.AvailableHeaderBytes() + + // Destroy the packet, pull all payloads out for fragmentation. + transHeader, data := pkt.TransportHeader().View(), pkt.Data + + // Where possible, the first fragment that is sent has the same + // number of bytes reserved for header as the input packet. The link-layer + // endpoint may depend on this for looking at, eg, L4 headers. + transFitsFirst := len(transHeader) <= innerMTU + for i := 0; i < n; i++ { - // Where possible, the first fragment that is sent has the same - // pkt.Header.UsedLength() as the input packet. The link-layer - // endpoint may depend on this for looking at, eg, L4 headers. - h := ip - if i > 0 { - pkt.Header = buffer.NewPrependable(int(ip.HeaderLength()) + originalAvailableLength) - h = header.IPv4(pkt.Header.Prepend(int(ip.HeaderLength()))) - copy(h, ip[:ip.HeaderLength()]) + reserve := reservedForLink + int(ip.HeaderLength()) + if i == 0 && transFitsFirst { + // Reserve for transport header if it's going to be put in the first + // fragment. + reserve += len(transHeader) + } + fragPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: reserve, + }) + fragPkt.NetworkProtocolNumber = header.IPv4ProtocolNumber + + // Copy data for the fragment. + avail := innerMTU + + if n := len(transHeader); n > 0 { + if n > avail { + n = avail + } + if i == 0 && transFitsFirst { + copy(fragPkt.TransportHeader().Push(n), transHeader) + } else { + fragPkt.Data.AppendView(transHeader[:n:n]) + } + transHeader = transHeader[n:] + avail -= n } + + if avail > 0 { + n := data.Size() + if n > avail { + n = avail + } + data.ReadToVV(&fragPkt.Data, n) + avail -= n + } + + copied := uint16(innerMTU - avail) + + // Set lengths in header and calculate checksum. + h := header.IPv4(fragPkt.NetworkHeader().Push(len(ip))) + copy(h, ip) if i != n-1 { h.SetTotalLength(uint16(outerMTU)) h.SetFlagsFragmentOffset(flags|header.IPv4FlagMoreFragments, offset) } else { - h.SetTotalLength(uint16(h.HeaderLength()) + uint16(pkt.Data.Size())) + h.SetTotalLength(uint16(h.HeaderLength()) + copied) h.SetFlagsFragmentOffset(flags, offset) } h.SetChecksum(0) h.SetChecksum(^h.CalculateChecksum()) - offset += uint16(innerMTU) - if i > 0 { - newPayload := pkt.Data.Clone(nil) - newPayload.CapLength(innerMTU) - if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, &stack.PacketBuffer{ - Header: pkt.Header, - Data: newPayload, - NetworkHeader: buffer.View(h), - NetworkProtocolNumber: header.IPv4ProtocolNumber, - }); err != nil { - return err - } - r.Stats().IP.PacketsSent.Increment() - pkt.Data.TrimFront(newPayload.Size()) - continue - } - // Special handling for the first fragment because it comes - // from the header. - if outerMTU >= pkt.Header.UsedLength() { - // This fragment can fit all of pkt.Header and possibly - // some of pkt.Data, too. - newPayload := pkt.Data.Clone(nil) - newPayloadLength := outerMTU - pkt.Header.UsedLength() - newPayload.CapLength(newPayloadLength) - if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, &stack.PacketBuffer{ - Header: pkt.Header, - Data: newPayload, - NetworkHeader: buffer.View(h), - NetworkProtocolNumber: header.IPv4ProtocolNumber, - }); err != nil { - return err - } - r.Stats().IP.PacketsSent.Increment() - pkt.Data.TrimFront(newPayloadLength) - } else { - // The fragment is too small to fit all of pkt.Header. - startOfHdr := pkt.Header - startOfHdr.TrimBack(pkt.Header.UsedLength() - outerMTU) - emptyVV := buffer.NewVectorisedView(0, []buffer.View{}) - if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, &stack.PacketBuffer{ - Header: startOfHdr, - Data: emptyVV, - NetworkHeader: buffer.View(h), - NetworkProtocolNumber: header.IPv4ProtocolNumber, - }); err != nil { - return err - } - r.Stats().IP.PacketsSent.Increment() - // Add the unused bytes of pkt.Header into the pkt.Data - // that remains to be sent. - restOfHdr := pkt.Header.View()[outerMTU:] - tmp := buffer.NewVectorisedView(len(restOfHdr), []buffer.View{buffer.NewViewFromBytes(restOfHdr)}) - tmp.Append(pkt.Data) - pkt.Data = tmp + offset += copied + + // Send out the fragment. + if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, fragPkt); err != nil { + return err } + r.Stats().IP.PacketsSent.Increment() } return nil } -func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadSize int, params stack.NetworkHeaderParams) header.IPv4 { - ip := header.IPv4(hdr.Prepend(header.IPv4MinimumSize)) - length := uint16(hdr.UsedLength() + payloadSize) +func (e *endpoint) addIPHeader(r *stack.Route, pkt *stack.PacketBuffer, params stack.NetworkHeaderParams) { + ip := header.IPv4(pkt.NetworkHeader().Push(header.IPv4MinimumSize)) + length := uint16(pkt.Size()) // RFC 6864 section 4.3 mandates uniqueness of ID values for non-atomic // datagrams. Since the DF bit is never being set here, all datagrams // are non-atomic and need an ID. @@ -245,14 +238,12 @@ func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadS DstAddr: r.RemoteAddress, }) ip.SetChecksum(^ip.CalculateChecksum()) - return ip + pkt.NetworkProtocolNumber = header.IPv4ProtocolNumber } // WritePacket writes a packet to the given destination address and protocol. func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt *stack.PacketBuffer) *tcpip.Error { - ip := e.addIPHeader(r, &pkt.Header, pkt.Data.Size(), params) - pkt.NetworkHeader = buffer.View(ip) - pkt.NetworkProtocolNumber = header.IPv4ProtocolNumber + e.addIPHeader(r, pkt, params) // iptables filtering. All packets that reach here are locally // generated. @@ -269,7 +260,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw // only NATted packets, but removing this check short circuits broadcasts // before they are sent out to other hosts. if pkt.NatDone { - netHeader := header.IPv4(pkt.NetworkHeader) + netHeader := header.IPv4(pkt.NetworkHeader().View()) ep, err := e.stack.FindNetworkEndpoint(header.IPv4ProtocolNumber, netHeader.DestinationAddress()) if err == nil { route := r.ReverseRoute(netHeader.SourceAddress(), netHeader.DestinationAddress()) @@ -286,7 +277,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw if r.Loop&stack.PacketOut == 0 { return nil } - if pkt.Header.UsedLength()+pkt.Data.Size() > int(e.linkEP.MTU()) && (gso == nil || gso.Type == stack.GSONone) { + if pkt.Size() > int(e.linkEP.MTU()) && (gso == nil || gso.Type == stack.GSONone) { return e.writePacketFragments(r, gso, int(e.linkEP.MTU()), pkt) } if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, pkt); err != nil { @@ -306,9 +297,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe } for pkt := pkts.Front(); pkt != nil; { - ip := e.addIPHeader(r, &pkt.Header, pkt.Data.Size(), params) - pkt.NetworkHeader = buffer.View(ip) - pkt.NetworkProtocolNumber = header.IPv4ProtocolNumber + e.addIPHeader(r, pkt, params) pkt = pkt.Next() } @@ -333,7 +322,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe continue } if _, ok := natPkts[pkt]; ok { - netHeader := header.IPv4(pkt.NetworkHeader) + netHeader := header.IPv4(pkt.NetworkHeader().View()) if ep, err := e.stack.FindNetworkEndpoint(header.IPv4ProtocolNumber, netHeader.DestinationAddress()); err == nil { src := netHeader.SourceAddress() dst := netHeader.DestinationAddress() @@ -402,17 +391,14 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBu r.Stats().IP.PacketsSent.Increment() - ip = ip[:ip.HeaderLength()] - pkt.Header = buffer.NewPrependableFromView(buffer.View(ip)) - pkt.Data.TrimFront(int(ip.HeaderLength())) return e.linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, pkt) } // HandlePacket is called by the link layer when new ipv4 packets arrive for // this endpoint. func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { - h := header.IPv4(pkt.NetworkHeader) - if !h.IsValid(pkt.Data.Size() + len(pkt.NetworkHeader) + len(pkt.TransportHeader)) { + h := header.IPv4(pkt.NetworkHeader().View()) + if !h.IsValid(pkt.Data.Size() + pkt.NetworkHeader().View().Size() + pkt.TransportHeader().View().Size()) { r.Stats().IP.MalformedPacketsReceived.Increment() return } @@ -426,7 +412,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { } if h.More() || h.FragmentOffset() != 0 { - if pkt.Data.Size()+len(pkt.TransportHeader) == 0 { + if pkt.Data.Size()+pkt.TransportHeader().View().Size() == 0 { // Drop the packet as it's marked as a fragment but has // no payload. r.Stats().IP.MalformedPacketsReceived.Increment() @@ -470,7 +456,6 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { } p := h.TransportProtocol() if p == header.ICMPv4ProtocolNumber { - pkt.NetworkHeader.CapLength(int(h.HeaderLength())) e.handleICMP(r, pkt) return } @@ -560,14 +545,19 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) (proto tcpip.TransportProtocolNu } ipHdr := header.IPv4(hdr) - // If there are options, pull those into hdr as well. - if headerLen := int(ipHdr.HeaderLength()); headerLen > header.IPv4MinimumSize && headerLen <= pkt.Data.Size() { - hdr, ok = pkt.Data.PullUp(headerLen) - if !ok { - panic(fmt.Sprintf("There are only %d bytes in pkt.Data, but there should be at least %d", pkt.Data.Size(), headerLen)) - } - ipHdr = header.IPv4(hdr) + // Header may have options, determine the true header length. + headerLen := int(ipHdr.HeaderLength()) + if headerLen < header.IPv4MinimumSize { + // TODO(gvisor.dev/issue/2404): Per RFC 791, IHL needs to be at least 5 in + // order for the packet to be valid. Figure out if we want to reject this + // case. + headerLen = header.IPv4MinimumSize + } + hdr, ok = pkt.NetworkHeader().Consume(headerLen) + if !ok { + return 0, false, false } + ipHdr = header.IPv4(hdr) // If this is a fragment, don't bother parsing the transport header. parseTransportHeader := true @@ -576,8 +566,6 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) (proto tcpip.TransportProtocolNu } pkt.NetworkProtocolNumber = header.IPv4ProtocolNumber - pkt.NetworkHeader = hdr - pkt.Data.TrimFront(len(hdr)) pkt.Data.CapLength(int(ipHdr.TotalLength()) - len(hdr)) return ipHdr.TransportProtocol(), parseTransportHeader, true } diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index ded91d83a..39ae19295 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -83,7 +83,7 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme return } h := header.ICMPv6(v) - iph := header.IPv6(pkt.NetworkHeader) + iph := header.IPv6(pkt.NetworkHeader().View()) // Validate ICMPv6 checksum before processing the packet. // @@ -276,8 +276,10 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme optsSerializer := header.NDPOptionsSerializer{ header.NDPTargetLinkLayerAddressOption(r.LocalLinkAddress), } - hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv6NeighborAdvertMinimumSize + int(optsSerializer.Length())) - packet := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborAdvertSize)) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(r.MaxHeaderLength()) + header.ICMPv6NeighborAdvertMinimumSize + int(optsSerializer.Length()), + }) + packet := header.ICMPv6(pkt.TransportHeader().Push(header.ICMPv6NeighborAdvertSize)) packet.SetType(header.ICMPv6NeighborAdvert) na := header.NDPNeighborAdvert(packet.NDPPayload()) na.SetSolicitedFlag(solicited) @@ -293,9 +295,7 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme // // The IP Hop Limit field has a value of 255, i.e., the packet // could not possibly have been forwarded by a router. - if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: stack.DefaultTOS}, &stack.PacketBuffer{ - Header: hdr, - }); err != nil { + if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: stack.DefaultTOS}, pkt); err != nil { sent.Dropped.Increment() return } @@ -384,7 +384,7 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme case header.ICMPv6EchoRequest: received.EchoRequest.Increment() - icmpHdr, ok := pkt.Data.PullUp(header.ICMPv6EchoMinimumSize) + icmpHdr, ok := pkt.TransportHeader().Consume(header.ICMPv6EchoMinimumSize) if !ok { received.Invalid.Increment() return @@ -409,16 +409,15 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme // Use the link address from the source of the original packet. r.ResolveWith(remoteLinkAddr) - pkt.Data.TrimFront(header.ICMPv6EchoMinimumSize) - hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv6EchoMinimumSize) - packet := header.ICMPv6(hdr.Prepend(header.ICMPv6EchoMinimumSize)) + replyPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(r.MaxHeaderLength()) + header.ICMPv6EchoMinimumSize, + Data: pkt.Data, + }) + packet := header.ICMPv6(replyPkt.TransportHeader().Push(header.ICMPv6EchoMinimumSize)) copy(packet, icmpHdr) packet.SetType(header.ICMPv6EchoReply) packet.SetChecksum(header.ICMPv6Checksum(packet, r.LocalAddress, r.RemoteAddress, pkt.Data)) - if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, &stack.PacketBuffer{ - Header: hdr, - Data: pkt.Data, - }); err != nil { + if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, replyPkt); err != nil { sent.Dropped.Increment() return } @@ -539,17 +538,19 @@ func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, remoteLinkAdd r.RemoteLinkAddress = header.EthernetAddressFromMulticastIPv6Address(snaddr) } - hdr := buffer.NewPrependable(int(linkEP.MaxHeaderLength()) + header.IPv6MinimumSize + header.ICMPv6NeighborAdvertSize) - pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborAdvertSize)) - pkt.SetType(header.ICMPv6NeighborSolicit) - copy(pkt[icmpV6OptOffset-len(addr):], addr) - pkt[icmpV6OptOffset] = ndpOptSrcLinkAddr - pkt[icmpV6LengthOffset] = 1 - copy(pkt[icmpV6LengthOffset+1:], linkEP.LinkAddress()) - pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) - - length := uint16(hdr.UsedLength()) - ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(linkEP.MaxHeaderLength()) + header.IPv6MinimumSize + header.ICMPv6NeighborAdvertSize, + }) + icmpHdr := header.ICMPv6(pkt.TransportHeader().Push(header.ICMPv6NeighborAdvertSize)) + icmpHdr.SetType(header.ICMPv6NeighborSolicit) + copy(icmpHdr[icmpV6OptOffset-len(addr):], addr) + icmpHdr[icmpV6OptOffset] = ndpOptSrcLinkAddr + icmpHdr[icmpV6LengthOffset] = 1 + copy(icmpHdr[icmpV6LengthOffset+1:], linkEP.LinkAddress()) + icmpHdr.SetChecksum(header.ICMPv6Checksum(icmpHdr, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) + + length := uint16(pkt.Size()) + ip := header.IPv6(pkt.NetworkHeader().Push(header.IPv6MinimumSize)) ip.Encode(&header.IPv6Fields{ PayloadLength: length, NextHeader: uint8(header.ICMPv6ProtocolNumber), @@ -559,9 +560,7 @@ func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, remoteLinkAdd }) // TODO(stijlist): count this in ICMP stats. - return linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, &stack.PacketBuffer{ - Header: hdr, - }) + return linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, pkt) } // ResolveStaticAddress implements stack.LinkAddressResolver. diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index d7d7fc611..0ade655b2 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -99,9 +99,9 @@ func (e *endpoint) GSOMaxSize() uint32 { return 0 } -func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadSize int, params stack.NetworkHeaderParams) header.IPv6 { - length := uint16(hdr.UsedLength() + payloadSize) - ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) +func (e *endpoint) addIPHeader(r *stack.Route, pkt *stack.PacketBuffer, params stack.NetworkHeaderParams) { + length := uint16(pkt.Size()) + ip := header.IPv6(pkt.NetworkHeader().Push(header.IPv6MinimumSize)) ip.Encode(&header.IPv6Fields{ PayloadLength: length, NextHeader: uint8(params.Protocol), @@ -110,26 +110,20 @@ func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadS SrcAddr: r.LocalAddress, DstAddr: r.RemoteAddress, }) - return ip + pkt.NetworkProtocolNumber = header.IPv6ProtocolNumber } // WritePacket writes a packet to the given destination address and protocol. func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt *stack.PacketBuffer) *tcpip.Error { - ip := e.addIPHeader(r, &pkt.Header, pkt.Data.Size(), params) - pkt.NetworkHeader = buffer.View(ip) - pkt.NetworkProtocolNumber = header.IPv6ProtocolNumber + e.addIPHeader(r, pkt, params) if r.Loop&stack.PacketLoop != 0 { - // The inbound path expects the network header to still be in - // the PacketBuffer's Data field. - views := make([]buffer.View, 1, 1+len(pkt.Data.Views())) - views[0] = pkt.Header.View() - views = append(views, pkt.Data.Views()...) loopedR := r.MakeLoopedRoute() - e.HandlePacket(&loopedR, &stack.PacketBuffer{ - Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), - }) + e.HandlePacket(&loopedR, stack.NewPacketBuffer(stack.PacketBufferOptions{ + // The inbound path expects an unparsed packet. + Data: buffer.NewVectorisedView(pkt.Size(), pkt.Views()), + })) loopedR.Release() } @@ -151,9 +145,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe } for pb := pkts.Front(); pb != nil; pb = pb.Next() { - ip := e.addIPHeader(r, &pb.Header, pb.Data.Size(), params) - pb.NetworkHeader = buffer.View(ip) - pb.NetworkProtocolNumber = header.IPv6ProtocolNumber + e.addIPHeader(r, pb, params) } n, err := e.linkEP.WritePackets(r, gso, pkts, ProtocolNumber) @@ -171,8 +163,8 @@ func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBuff // HandlePacket is called by the link layer when new ipv6 packets arrive for // this endpoint. func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { - h := header.IPv6(pkt.NetworkHeader) - if !h.IsValid(pkt.Data.Size() + len(pkt.NetworkHeader) + len(pkt.TransportHeader)) { + h := header.IPv6(pkt.NetworkHeader().View()) + if !h.IsValid(pkt.Data.Size() + pkt.NetworkHeader().View().Size() + pkt.TransportHeader().View().Size()) { r.Stats().IP.MalformedPacketsReceived.Increment() return } @@ -181,8 +173,8 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { // - Any IPv6 header bytes after the first 40 (i.e. extensions). // - The transport header, if present. // - Any other payload data. - vv := pkt.NetworkHeader[header.IPv6MinimumSize:].ToVectorisedView() - vv.AppendView(pkt.TransportHeader) + vv := pkt.NetworkHeader().View()[header.IPv6MinimumSize:].ToVectorisedView() + vv.AppendView(pkt.TransportHeader().View()) vv.Append(pkt.Data) it := header.MakeIPv6PayloadIterator(header.IPv6ExtensionHeaderIdentifier(h.NextHeader()), vv) hasFragmentHeader := false @@ -410,7 +402,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { // // For reassembled fragments, pkt.TransportHeader is unset, so this is a // no-op and pkt.Data begins with the transport header. - extHdr.Buf.TrimFront(len(pkt.TransportHeader)) + extHdr.Buf.TrimFront(pkt.TransportHeader().View().Size()) pkt.Data = extHdr.Buf if p := tcpip.TransportProtocolNumber(extHdr.Identifier); p == header.ICMPv6ProtocolNumber { @@ -581,17 +573,14 @@ traverseExtensions: } } - // Put the IPv6 header with extensions in pkt.NetworkHeader. - hdr, ok = pkt.Data.PullUp(header.IPv6MinimumSize + extensionsSize) + // Put the IPv6 header with extensions in pkt.NetworkHeader(). + hdr, ok = pkt.NetworkHeader().Consume(header.IPv6MinimumSize + extensionsSize) if !ok { panic(fmt.Sprintf("pkt.Data should have at least %d bytes, but only has %d.", header.IPv6MinimumSize+extensionsSize, pkt.Data.Size())) } ipHdr = header.IPv6(hdr) - - pkt.NetworkProtocolNumber = header.IPv6ProtocolNumber - pkt.NetworkHeader = hdr - pkt.Data.TrimFront(len(hdr)) pkt.Data.CapLength(int(ipHdr.PayloadLength())) + pkt.NetworkProtocolNumber = header.IPv6ProtocolNumber return nextHdr, foundNext, true } diff --git a/pkg/tcpip/stack/conntrack.go b/pkg/tcpip/stack/conntrack.go index 470c265aa..7dd344b4f 100644 --- a/pkg/tcpip/stack/conntrack.go +++ b/pkg/tcpip/stack/conntrack.go @@ -199,12 +199,12 @@ type bucket struct { func packetToTupleID(pkt *PacketBuffer) (tupleID, *tcpip.Error) { // TODO(gvisor.dev/issue/170): Need to support for other // protocols as well. - netHeader := header.IPv4(pkt.NetworkHeader) - if netHeader == nil || netHeader.TransportProtocol() != header.TCPProtocolNumber { + netHeader := header.IPv4(pkt.NetworkHeader().View()) + if len(netHeader) < header.IPv4MinimumSize || netHeader.TransportProtocol() != header.TCPProtocolNumber { return tupleID{}, tcpip.ErrUnknownProtocol } - tcpHeader := header.TCP(pkt.TransportHeader) - if tcpHeader == nil { + tcpHeader := header.TCP(pkt.TransportHeader().View()) + if len(tcpHeader) < header.TCPMinimumSize { return tupleID{}, tcpip.ErrUnknownProtocol } @@ -344,8 +344,8 @@ func handlePacketPrerouting(pkt *PacketBuffer, conn *conn, dir direction) { return } - netHeader := header.IPv4(pkt.NetworkHeader) - tcpHeader := header.TCP(pkt.TransportHeader) + netHeader := header.IPv4(pkt.NetworkHeader().View()) + tcpHeader := header.TCP(pkt.TransportHeader().View()) // For prerouting redirection, packets going in the original direction // have their destinations modified and replies have their sources @@ -377,8 +377,8 @@ func handlePacketOutput(pkt *PacketBuffer, conn *conn, gso *GSO, r *Route, dir d return } - netHeader := header.IPv4(pkt.NetworkHeader) - tcpHeader := header.TCP(pkt.TransportHeader) + netHeader := header.IPv4(pkt.NetworkHeader().View()) + tcpHeader := header.TCP(pkt.TransportHeader().View()) // For output redirection, packets going in the original direction // have their destinations modified and replies have their sources @@ -396,8 +396,7 @@ func handlePacketOutput(pkt *PacketBuffer, conn *conn, gso *GSO, r *Route, dir d // Calculate the TCP checksum and set it. tcpHeader.SetChecksum(0) - hdr := &pkt.Header - length := uint16(pkt.Data.Size()+hdr.UsedLength()) - uint16(netHeader.HeaderLength()) + length := uint16(pkt.Size()) - uint16(netHeader.HeaderLength()) xsum := r.PseudoHeaderChecksum(header.TCPProtocolNumber, length) if gso != nil && gso.NeedsCsum { tcpHeader.SetChecksum(xsum) @@ -423,7 +422,7 @@ func (ct *ConnTrack) handlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r *Rou } // TODO(gvisor.dev/issue/170): Support other transport protocols. - if pkt.NetworkHeader == nil || header.IPv4(pkt.NetworkHeader).TransportProtocol() != header.TCPProtocolNumber { + if nh := pkt.NetworkHeader().View(); nh.IsEmpty() || header.IPv4(nh).TransportProtocol() != header.TCPProtocolNumber { return false } @@ -433,8 +432,8 @@ func (ct *ConnTrack) handlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r *Rou return true } - tcpHeader := header.TCP(pkt.TransportHeader) - if tcpHeader == nil { + tcpHeader := header.TCP(pkt.TransportHeader().View()) + if len(tcpHeader) < header.TCPMinimumSize { return false } @@ -455,7 +454,7 @@ func (ct *ConnTrack) handlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r *Rou // Mark the connection as having been used recently so it isn't reaped. conn.lastUsed = time.Now() // Update connection state. - conn.updateLocked(header.TCP(pkt.TransportHeader), hook) + conn.updateLocked(header.TCP(pkt.TransportHeader().View()), hook) return false } @@ -474,7 +473,7 @@ func (ct *ConnTrack) maybeInsertNoop(pkt *PacketBuffer, hook Hook) { } // We only track TCP connections. - if pkt.NetworkHeader == nil || header.IPv4(pkt.NetworkHeader).TransportProtocol() != header.TCPProtocolNumber { + if nh := pkt.NetworkHeader().View(); nh.IsEmpty() || header.IPv4(nh).TransportProtocol() != header.TCPProtocolNumber { return } @@ -486,7 +485,7 @@ func (ct *ConnTrack) maybeInsertNoop(pkt *PacketBuffer, hook Hook) { return } conn := newConn(tid, tid.reply(), manipNone, hook) - conn.updateLocked(header.TCP(pkt.TransportHeader), hook) + conn.updateLocked(header.TCP(pkt.TransportHeader().View()), hook) ct.insertConn(conn) } diff --git a/pkg/tcpip/stack/headertype_string.go b/pkg/tcpip/stack/headertype_string.go new file mode 100644 index 000000000..5efddfaaf --- /dev/null +++ b/pkg/tcpip/stack/headertype_string.go @@ -0,0 +1,39 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at // +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Code generated by "stringer -type headerType ."; DO NOT EDIT. + +package stack + +import "strconv" + +func _() { + // An "invalid array index" compiler error signifies that the constant values have changed. + // Re-run the stringer command to generate them again. + var x [1]struct{} + _ = x[linkHeader-0] + _ = x[networkHeader-1] + _ = x[transportHeader-2] + _ = x[numHeaderType-3] +} + +const _headerType_name = "linkHeadernetworkHeadertransportHeadernumHeaderType" + +var _headerType_index = [...]uint8{0, 10, 23, 38, 51} + +func (i headerType) String() string { + if i < 0 || i >= headerType(len(_headerType_index)-1) { + return "headerType(" + strconv.FormatInt(int64(i), 10) + ")" + } + return _headerType_name[_headerType_index[i]:_headerType_index[i+1]] +} diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go index 110ba073d..c37da814f 100644 --- a/pkg/tcpip/stack/iptables.go +++ b/pkg/tcpip/stack/iptables.go @@ -394,7 +394,7 @@ func (it *IPTables) checkRule(hook Hook, pkt *PacketBuffer, table Table, ruleIdx rule := table.Rules[ruleIdx] // Check whether the packet matches the IP header filter. - if !rule.Filter.match(header.IPv4(pkt.NetworkHeader), hook, nicName) { + if !rule.Filter.match(header.IPv4(pkt.NetworkHeader().View()), hook, nicName) { // Continue on to the next rule. return RuleJump, ruleIdx + 1 } diff --git a/pkg/tcpip/stack/iptables_targets.go b/pkg/tcpip/stack/iptables_targets.go index dc88033c7..5f1b2af64 100644 --- a/pkg/tcpip/stack/iptables_targets.go +++ b/pkg/tcpip/stack/iptables_targets.go @@ -99,7 +99,7 @@ func (rt RedirectTarget) Action(pkt *PacketBuffer, ct *ConnTrack, hook Hook, gso } // Drop the packet if network and transport header are not set. - if pkt.NetworkHeader == nil || pkt.TransportHeader == nil { + if pkt.NetworkHeader().View().IsEmpty() || pkt.TransportHeader().View().IsEmpty() { return RuleDrop, 0 } @@ -118,17 +118,16 @@ func (rt RedirectTarget) Action(pkt *PacketBuffer, ct *ConnTrack, hook Hook, gso // TODO(gvisor.dev/issue/170): Check Flags in RedirectTarget if // we need to change dest address (for OUTPUT chain) or ports. - netHeader := header.IPv4(pkt.NetworkHeader) + netHeader := header.IPv4(pkt.NetworkHeader().View()) switch protocol := netHeader.TransportProtocol(); protocol { case header.UDPProtocolNumber: - udpHeader := header.UDP(pkt.TransportHeader) + udpHeader := header.UDP(pkt.TransportHeader().View()) udpHeader.SetDestinationPort(rt.MinPort) // Calculate UDP checksum and set it. if hook == Output { udpHeader.SetChecksum(0) - hdr := &pkt.Header - length := uint16(pkt.Data.Size()+hdr.UsedLength()) - uint16(netHeader.HeaderLength()) + length := uint16(pkt.Size()) - uint16(netHeader.HeaderLength()) // Only calculate the checksum if offloading isn't supported. if r.Capabilities()&CapabilityTXChecksumOffload == 0 { diff --git a/pkg/tcpip/stack/ndp.go b/pkg/tcpip/stack/ndp.go index 5174e639c..93567806b 100644 --- a/pkg/tcpip/stack/ndp.go +++ b/pkg/tcpip/stack/ndp.go @@ -746,12 +746,16 @@ func (ndp *ndpState) sendDADPacket(addr tcpip.Address, ref *referencedNetworkEnd panic(fmt.Sprintf("ndp: route resolution not immediate for route to send NDP NS for DAD (%s -> %s on NIC(%d))", header.IPv6Any, snmc, ndp.nic.ID())) } - hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv6NeighborSolicitMinimumSize) - pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborSolicitMinimumSize)) - pkt.SetType(header.ICMPv6NeighborSolicit) - ns := header.NDPNeighborSolicit(pkt.NDPPayload()) + icmpData := header.ICMPv6(buffer.NewView(header.ICMPv6NeighborSolicitMinimumSize)) + icmpData.SetType(header.ICMPv6NeighborSolicit) + ns := header.NDPNeighborSolicit(icmpData.NDPPayload()) ns.SetTargetAddress(addr) - pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) + icmpData.SetChecksum(header.ICMPv6Checksum(icmpData, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) + + pkt := NewPacketBuffer(PacketBufferOptions{ + ReserveHeaderBytes: int(r.MaxHeaderLength()), + Data: buffer.View(icmpData).ToVectorisedView(), + }) sent := r.Stats().ICMP.V6PacketsSent if err := r.WritePacket(nil, @@ -759,7 +763,7 @@ func (ndp *ndpState) sendDADPacket(addr tcpip.Address, ref *referencedNetworkEnd Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: DefaultTOS, - }, &PacketBuffer{Header: hdr}, + }, pkt, ); err != nil { sent.Dropped.Increment() return err @@ -1897,12 +1901,16 @@ func (ndp *ndpState) startSolicitingRouters() { } } payloadSize := header.ICMPv6HeaderSize + header.NDPRSMinimumSize + int(optsSerializer.Length()) - hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + payloadSize) - pkt := header.ICMPv6(hdr.Prepend(payloadSize)) - pkt.SetType(header.ICMPv6RouterSolicit) - rs := header.NDPRouterSolicit(pkt.NDPPayload()) + icmpData := header.ICMPv6(buffer.NewView(payloadSize)) + icmpData.SetType(header.ICMPv6RouterSolicit) + rs := header.NDPRouterSolicit(icmpData.NDPPayload()) rs.Options().Serialize(optsSerializer) - pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) + icmpData.SetChecksum(header.ICMPv6Checksum(icmpData, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) + + pkt := NewPacketBuffer(PacketBufferOptions{ + ReserveHeaderBytes: int(r.MaxHeaderLength()), + Data: buffer.View(icmpData).ToVectorisedView(), + }) sent := r.Stats().ICMP.V6PacketsSent if err := r.WritePacket(nil, @@ -1910,7 +1918,7 @@ func (ndp *ndpState) startSolicitingRouters() { Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: DefaultTOS, - }, &PacketBuffer{Header: hdr}, + }, pkt, ); err != nil { sent.Dropped.Increment() log.Printf("startSolicitingRouters: error writing NDP router solicit message on NIC(%d); err = %s", ndp.nic.ID(), err) diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index eaaf756cd..2315ea5b9 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -1299,7 +1299,7 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp } } - src, dst := netProto.ParseAddresses(pkt.NetworkHeader) + src, dst := netProto.ParseAddresses(pkt.NetworkHeader().View()) if n.stack.handleLocal && !n.isLoopback() && n.getRef(protocol, src) != nil { // The source address is one of our own, so we never should have gotten a @@ -1401,24 +1401,19 @@ func (n *NIC) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tc func (n *NIC) forwardPacket(r *Route, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { // TODO(b/143425874) Decrease the TTL field in forwarded packets. - // TODO(b/151227689): Avoid copying the packet when forwarding. We can do this - // by having lower layers explicity write each header instead of just - // pkt.Header. - // pkt may have set its NetworkHeader and TransportHeader. If we're - // forwarding, we'll have to copy them into pkt.Header. - pkt.Header = buffer.NewPrependable(int(n.linkEP.MaxHeaderLength()) + len(pkt.NetworkHeader) + len(pkt.TransportHeader)) - if n := copy(pkt.Header.Prepend(len(pkt.TransportHeader)), pkt.TransportHeader); n != len(pkt.TransportHeader) { - panic(fmt.Sprintf("copied %d bytes, expected %d", n, len(pkt.TransportHeader))) - } - if n := copy(pkt.Header.Prepend(len(pkt.NetworkHeader)), pkt.NetworkHeader); n != len(pkt.NetworkHeader) { - panic(fmt.Sprintf("copied %d bytes, expected %d", n, len(pkt.NetworkHeader))) - } + // pkt may have set its header and may not have enough headroom for link-layer + // header for the other link to prepend. Here we create a new packet to + // forward. + fwdPkt := NewPacketBuffer(PacketBufferOptions{ + ReserveHeaderBytes: int(n.linkEP.MaxHeaderLength()), + Data: buffer.NewVectorisedView(pkt.Size(), pkt.Views()), + }) - // WritePacket takes ownership of pkt, calculate numBytes first. - numBytes := pkt.Header.UsedLength() + pkt.Data.Size() + // WritePacket takes ownership of fwdPkt, calculate numBytes first. + numBytes := fwdPkt.Size() - if err := n.linkEP.WritePacket(r, nil /* gso */, protocol, pkt); err != nil { + if err := n.linkEP.WritePacket(r, nil /* gso */, protocol, fwdPkt); err != nil { r.Stats().IP.OutgoingPacketErrors.Increment() return } @@ -1443,34 +1438,31 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN // validly formed. n.stack.demux.deliverRawPacket(r, protocol, pkt) - // TransportHeader is nil only when pkt is an ICMP packet or was reassembled + // TransportHeader is empty only when pkt is an ICMP packet or was reassembled // from fragments. - if pkt.TransportHeader == nil { + if pkt.TransportHeader().View().IsEmpty() { // TODO(gvisor.dev/issue/170): ICMP packets don't have their TransportHeader // fields set yet, parse it here. See icmp/protocol.go:protocol.Parse for a // full explanation. if protocol == header.ICMPv4ProtocolNumber || protocol == header.ICMPv6ProtocolNumber { // ICMP packets may be longer, but until icmp.Parse is implemented, here // we parse it using the minimum size. - transHeader, ok := pkt.Data.PullUp(transProto.MinimumPacketSize()) - if !ok { + if _, ok := pkt.TransportHeader().Consume(transProto.MinimumPacketSize()); !ok { n.stack.stats.MalformedRcvdPackets.Increment() return } - pkt.TransportHeader = transHeader - pkt.Data.TrimFront(len(pkt.TransportHeader)) } else { // This is either a bad packet or was re-assembled from fragments. transProto.Parse(pkt) } } - if len(pkt.TransportHeader) < transProto.MinimumPacketSize() { + if pkt.TransportHeader().View().Size() < transProto.MinimumPacketSize() { n.stack.stats.MalformedRcvdPackets.Increment() return } - srcPort, dstPort, err := transProto.ParsePorts(pkt.TransportHeader) + srcPort, dstPort, err := transProto.ParsePorts(pkt.TransportHeader().View()) if err != nil { n.stack.stats.MalformedRcvdPackets.Increment() return diff --git a/pkg/tcpip/stack/packet_buffer.go b/pkg/tcpip/stack/packet_buffer.go index 9e871f968..17b8beebb 100644 --- a/pkg/tcpip/stack/packet_buffer.go +++ b/pkg/tcpip/stack/packet_buffer.go @@ -14,16 +14,43 @@ package stack import ( + "fmt" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" ) +type headerType int + +const ( + linkHeader headerType = iota + networkHeader + transportHeader + numHeaderType +) + +// PacketBufferOptions specifies options for PacketBuffer creation. +type PacketBufferOptions struct { + // ReserveHeaderBytes is the number of bytes to reserve for headers. Total + // number of bytes pushed onto the headers must not exceed this value. + ReserveHeaderBytes int + + // Data is the initial unparsed data for the new packet. If set, it will be + // owned by the new packet. + Data buffer.VectorisedView +} + // A PacketBuffer contains all the data of a network packet. // // As a PacketBuffer traverses up the stack, it may be necessary to pass it to -// multiple endpoints. Clone() should be called in such cases so that -// modifications to the Data field do not affect other copies. +// multiple endpoints. +// +// The whole packet is expected to be a series of bytes in the following order: +// LinkHeader, NetworkHeader, TransportHeader, and Data. Any of them can be +// empty. Use of PacketBuffer in any other order is unsupported. +// +// PacketBuffer must be created with NewPacketBuffer. type PacketBuffer struct { _ sync.NoCopy @@ -31,36 +58,27 @@ type PacketBuffer struct { // PacketBuffers. PacketBufferEntry - // Data holds the payload of the packet. For inbound packets, it also - // holds the headers, which are consumed as the packet moves up the - // stack. Headers are guaranteed not to be split across views. + // Data holds the payload of the packet. + // + // For inbound packets, Data is initially the whole packet. Then gets moved to + // headers via PacketHeader.Consume, when the packet is being parsed. // - // The bytes backing Data are immutable, but Data itself may be trimmed - // or otherwise modified. + // For outbound packets, Data is the innermost layer, defined by the protocol. + // Headers are pushed in front of it via PacketHeader.Push. + // + // The bytes backing Data are immutable, a.k.a. users shouldn't write to its + // backing storage. Data buffer.VectorisedView - // Header holds the headers of outbound packets. As a packet is passed - // down the stack, each layer adds to Header. Note that forwarded - // packets don't populate Headers on their way out -- their headers and - // payload are never parsed out and remain in Data. - // - // TODO(gvisor.dev/issue/170): Forwarded packets don't currently - // populate Header, but should. This will be doable once early parsing - // (https://github.com/google/gvisor/pull/1995) is supported. - Header buffer.Prependable + // headers stores metadata about each header. + headers [numHeaderType]headerInfo - // These fields are used by both inbound and outbound packets. They - // typically overlap with the Data and Header fields. + // header is the internal storage for outbound packets. Headers will be pushed + // (prepended) on this storage as the packet is being constructed. // - // The bytes backing these views are immutable. Each field may be nil - // if either it has not been set yet or no such header exists (e.g. - // packets sent via loopback may not have a link header). - // - // These fields may be Views into other slices (either Data or Header). - // SR dosen't support this, so deep copies are necessary in some cases. - LinkHeader buffer.View - NetworkHeader buffer.View - TransportHeader buffer.View + // TODO(gvisor.dev/issue/2404): Switch to an implementation that header and + // data are held in the same underlying buffer storage. + header buffer.Prependable // NetworkProtocol is only valid when NetworkHeader is set. // TODO(gvisor.dev/issue/3574): Remove the separately passed protocol @@ -89,20 +107,137 @@ type PacketBuffer struct { PktType tcpip.PacketType } -// Clone makes a copy of pk. It clones the Data field, which creates a new -// VectorisedView but does not deep copy the underlying bytes. -// -// Clone also does not deep copy any of its other fields. +// NewPacketBuffer creates a new PacketBuffer with opts. +func NewPacketBuffer(opts PacketBufferOptions) *PacketBuffer { + pk := &PacketBuffer{ + Data: opts.Data, + } + if opts.ReserveHeaderBytes != 0 { + pk.header = buffer.NewPrependable(opts.ReserveHeaderBytes) + } + return pk +} + +// ReservedHeaderBytes returns the number of bytes initially reserved for +// headers. +func (pk *PacketBuffer) ReservedHeaderBytes() int { + return pk.header.UsedLength() + pk.header.AvailableLength() +} + +// AvailableHeaderBytes returns the number of bytes currently available for +// headers. This is relevant to PacketHeader.Push method only. +func (pk *PacketBuffer) AvailableHeaderBytes() int { + return pk.header.AvailableLength() +} + +// LinkHeader returns the handle to link-layer header. +func (pk *PacketBuffer) LinkHeader() PacketHeader { + return PacketHeader{ + pk: pk, + typ: linkHeader, + } +} + +// NetworkHeader returns the handle to network-layer header. +func (pk *PacketBuffer) NetworkHeader() PacketHeader { + return PacketHeader{ + pk: pk, + typ: networkHeader, + } +} + +// TransportHeader returns the handle to transport-layer header. +func (pk *PacketBuffer) TransportHeader() PacketHeader { + return PacketHeader{ + pk: pk, + typ: transportHeader, + } +} + +// HeaderSize returns the total size of all headers in bytes. +func (pk *PacketBuffer) HeaderSize() int { + // Note for inbound packets (Consume called), headers are not stored in + // pk.header. Thus, calculation of size of each header is needed. + var size int + for i := range pk.headers { + size += len(pk.headers[i].buf) + } + return size +} + +// Size returns the size of packet in bytes. +func (pk *PacketBuffer) Size() int { + return pk.HeaderSize() + pk.Data.Size() +} + +// Views returns the underlying storage of the whole packet. +func (pk *PacketBuffer) Views() []buffer.View { + // Optimization for outbound packets that headers are in pk.header. + useHeader := true + for i := range pk.headers { + if !canUseHeader(&pk.headers[i]) { + useHeader = false + break + } + } + + dataViews := pk.Data.Views() + + var vs []buffer.View + if useHeader { + vs = make([]buffer.View, 0, 1+len(dataViews)) + vs = append(vs, pk.header.View()) + } else { + vs = make([]buffer.View, 0, len(pk.headers)+len(dataViews)) + for i := range pk.headers { + if v := pk.headers[i].buf; len(v) > 0 { + vs = append(vs, v) + } + } + } + return append(vs, dataViews...) +} + +func canUseHeader(h *headerInfo) bool { + // h.offset will be negative if the header was pushed in to prependable + // portion, or doesn't matter when it's empty. + return len(h.buf) == 0 || h.offset < 0 +} + +func (pk *PacketBuffer) push(typ headerType, size int) buffer.View { + h := &pk.headers[typ] + if h.buf != nil { + panic(fmt.Sprintf("push must not be called twice: type %s", typ)) + } + h.buf = buffer.View(pk.header.Prepend(size)) + h.offset = -pk.header.UsedLength() + return h.buf +} + +func (pk *PacketBuffer) consume(typ headerType, size int) (v buffer.View, consumed bool) { + h := &pk.headers[typ] + if h.buf != nil { + panic(fmt.Sprintf("consume must not be called twice: type %s", typ)) + } + v, ok := pk.Data.PullUp(size) + if !ok { + return + } + pk.Data.TrimFront(size) + h.buf = v + return h.buf, true +} + +// Clone makes a shallow copy of pk. // -// FIXME(b/153685824): Data gets copied but not other header references. +// Clone should be called in such cases so that no modifications is done to +// underlying packet payload. func (pk *PacketBuffer) Clone() *PacketBuffer { - return &PacketBuffer{ + newPk := &PacketBuffer{ PacketBufferEntry: pk.PacketBufferEntry, Data: pk.Data.Clone(nil), - Header: pk.Header, - LinkHeader: pk.LinkHeader, - NetworkHeader: pk.NetworkHeader, - TransportHeader: pk.TransportHeader, + headers: pk.headers, + header: pk.header, Hash: pk.Hash, Owner: pk.Owner, EgressRoute: pk.EgressRoute, @@ -110,4 +245,55 @@ func (pk *PacketBuffer) Clone() *PacketBuffer { NetworkProtocolNumber: pk.NetworkProtocolNumber, NatDone: pk.NatDone, } + return newPk +} + +// headerInfo stores metadata about a header in a packet. +type headerInfo struct { + // buf is the memorized slice for both prepended and consumed header. + // When header is prepended, buf serves as memorized value, which is a slice + // of pk.header. When header is consumed, buf is the slice pulled out from + // pk.Data, which is the only place to hold this header. + buf buffer.View + + // offset will be a negative number denoting the offset where this header is + // from the end of pk.header, if it is prepended. Otherwise, zero. + offset int +} + +// PacketHeader is a handle object to a header in the underlying packet. +type PacketHeader struct { + pk *PacketBuffer + typ headerType +} + +// View returns the underlying storage of h. +func (h PacketHeader) View() buffer.View { + return h.pk.headers[h.typ].buf +} + +// Push pushes size bytes in the front of its residing packet, and returns the +// backing storage. Callers may only call one of Push or Consume once on each +// header in the lifetime of the underlying packet. +func (h PacketHeader) Push(size int) buffer.View { + return h.pk.push(h.typ, size) +} + +// Consume moves the first size bytes of the unparsed data portion in the packet +// to h, and returns the backing storage. In the case of data is shorter than +// size, consumed will be false, and the state of h will not be affected. +// Callers may only call one of Push or Consume once on each header in the +// lifetime of the underlying packet. +func (h PacketHeader) Consume(size int) (v buffer.View, consumed bool) { + return h.pk.consume(h.typ, size) +} + +// PayloadSince returns packet payload starting from and including a particular +// header. This method isn't optimized and should be used in test only. +func PayloadSince(h PacketHeader) buffer.View { + var v buffer.View + for _, hinfo := range h.pk.headers[h.typ:] { + v = append(v, hinfo.buf...) + } + return append(v, h.pk.Data.ToView()...) } diff --git a/pkg/tcpip/stack/route.go b/pkg/tcpip/stack/route.go index 9ce0a2c22..e267bebb0 100644 --- a/pkg/tcpip/stack/route.go +++ b/pkg/tcpip/stack/route.go @@ -173,7 +173,7 @@ func (r *Route) WritePacket(gso *GSO, params NetworkHeaderParams, pkt *PacketBuf } // WritePacket takes ownership of pkt, calculate numBytes first. - numBytes := pkt.Header.UsedLength() + pkt.Data.Size() + numBytes := pkt.Size() err := r.ref.ep.WritePacket(r, gso, params, pkt) if err != nil { @@ -203,8 +203,7 @@ func (r *Route) WritePackets(gso *GSO, pkts PacketBufferList, params NetworkHead writtenBytes := 0 for i, pb := 0, pkts.Front(); i < n && pb != nil; i, pb = i+1, pb.Next() { - writtenBytes += pb.Header.UsedLength() - writtenBytes += pb.Data.Size() + writtenBytes += pb.Size() } r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(writtenBytes)) diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index 4612be4e7..bd6f49eb8 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -430,9 +430,12 @@ func send4(r *stack.Route, ident uint16, data buffer.View, ttl uint8, owner tcpi return tcpip.ErrInvalidEndpointState } - hdr := buffer.NewPrependable(header.ICMPv4MinimumSize + int(r.MaxHeaderLength())) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: header.ICMPv4MinimumSize + int(r.MaxHeaderLength()), + }) + pkt.Owner = owner - icmpv4 := header.ICMPv4(hdr.Prepend(header.ICMPv4MinimumSize)) + icmpv4 := header.ICMPv4(pkt.TransportHeader().Push(header.ICMPv4MinimumSize)) copy(icmpv4, data) // Set the ident to the user-specified port. Sequence number should // already be set by the user. @@ -447,15 +450,12 @@ func send4(r *stack.Route, ident uint16, data buffer.View, ttl uint8, owner tcpi icmpv4.SetChecksum(0) icmpv4.SetChecksum(^header.Checksum(icmpv4, header.Checksum(data, 0))) + pkt.Data = data.ToVectorisedView() + if ttl == 0 { ttl = r.DefaultTTL() } - return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}, &stack.PacketBuffer{ - Header: hdr, - Data: data.ToVectorisedView(), - TransportHeader: buffer.View(icmpv4), - Owner: owner, - }) + return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}, pkt) } func send6(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Error { @@ -463,9 +463,11 @@ func send6(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Err return tcpip.ErrInvalidEndpointState } - hdr := buffer.NewPrependable(header.ICMPv6MinimumSize + int(r.MaxHeaderLength())) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: header.ICMPv6MinimumSize + int(r.MaxHeaderLength()), + }) - icmpv6 := header.ICMPv6(hdr.Prepend(header.ICMPv6MinimumSize)) + icmpv6 := header.ICMPv6(pkt.TransportHeader().Push(header.ICMPv6MinimumSize)) copy(icmpv6, data) // Set the ident. Sequence number is provided by the user. icmpv6.SetIdent(ident) @@ -477,15 +479,12 @@ func send6(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Err dataVV := data.ToVectorisedView() icmpv6.SetChecksum(header.ICMPv6Checksum(icmpv6, r.LocalAddress, r.RemoteAddress, dataVV)) + pkt.Data = dataVV if ttl == 0 { ttl = r.DefaultTTL() } - return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}, &stack.PacketBuffer{ - Header: hdr, - Data: dataVV, - TransportHeader: buffer.View(icmpv6), - }) + return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}, pkt) } // checkV4MappedLocked determines the effective network protocol and converts @@ -748,14 +747,18 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk // Only accept echo replies. switch e.NetProto { case header.IPv4ProtocolNumber: - h := header.ICMPv4(pkt.TransportHeader) + h := header.ICMPv4(pkt.TransportHeader().View()) + // TODO(b/129292233): Determine if len(h) check is still needed after early + // parsing. if len(h) < header.ICMPv4MinimumSize || h.Type() != header.ICMPv4EchoReply { e.stack.Stats().DroppedPackets.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() return } case header.IPv6ProtocolNumber: - h := header.ICMPv6(pkt.TransportHeader) + h := header.ICMPv6(pkt.TransportHeader().View()) + // TODO(b/129292233): Determine if len(h) check is still needed after early + // parsing. if len(h) < header.ICMPv6MinimumSize || h.Type() != header.ICMPv6EchoReply { e.stack.Stats().DroppedPackets.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() @@ -791,7 +794,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk } // ICMP socket's data includes ICMP header. - packet.data = pkt.TransportHeader.ToVectorisedView() + packet.data = pkt.TransportHeader().View().ToVectorisedView() packet.data.Append(pkt.Data) e.rcvList.PushBack(packet) diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go index df478115d..1b03ad6bb 100644 --- a/pkg/tcpip/transport/packet/endpoint.go +++ b/pkg/tcpip/transport/packet/endpoint.go @@ -433,9 +433,9 @@ func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, // Push new packet into receive list and increment the buffer size. var packet packet // TODO(gvisor.dev/issue/173): Return network protocol. - if len(pkt.LinkHeader) > 0 { + if !pkt.LinkHeader().View().IsEmpty() { // Get info directly from the ethernet header. - hdr := header.Ethernet(pkt.LinkHeader) + hdr := header.Ethernet(pkt.LinkHeader().View()) packet.senderAddr = tcpip.FullAddress{ NIC: nicID, Addr: tcpip.Address(hdr.SourceAddress()), @@ -458,9 +458,14 @@ func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, case tcpip.PacketHost: packet.data = pkt.Data case tcpip.PacketOutgoing: - // Strip Link Header from the Header. - pkt.Header = buffer.NewPrependableFromView(pkt.Header.View()[len(pkt.LinkHeader):]) - combinedVV := pkt.Header.View().ToVectorisedView() + // Strip Link Header. + var combinedVV buffer.VectorisedView + if v := pkt.NetworkHeader().View(); !v.IsEmpty() { + combinedVV.AppendView(v) + } + if v := pkt.TransportHeader().View(); !v.IsEmpty() { + combinedVV.AppendView(v) + } combinedVV.Append(pkt.Data) packet.data = combinedVV default: @@ -471,9 +476,8 @@ func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, // Raw packets need their ethernet headers prepended before // queueing. var linkHeader buffer.View - var combinedVV buffer.VectorisedView if pkt.PktType != tcpip.PacketOutgoing { - if len(pkt.LinkHeader) == 0 { + if pkt.LinkHeader().View().IsEmpty() { // We weren't provided with an actual ethernet header, // so fake one. ethFields := header.EthernetFields{ @@ -485,19 +489,14 @@ func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, fakeHeader.Encode(ðFields) linkHeader = buffer.View(fakeHeader) } else { - linkHeader = append(buffer.View(nil), pkt.LinkHeader...) + linkHeader = append(buffer.View(nil), pkt.LinkHeader().View()...) } - combinedVV = linkHeader.ToVectorisedView() - } - if pkt.PktType == tcpip.PacketOutgoing { - // For outgoing packets the Link, Network and Transport - // headers are in the pkt.Header fields normally unless - // a Raw socket is in use in which case pkt.Header could - // be nil. - combinedVV.AppendView(pkt.Header.View()) + combinedVV := linkHeader.ToVectorisedView() + combinedVV.Append(pkt.Data) + packet.data = combinedVV + } else { + packet.data = buffer.NewVectorisedView(pkt.Size(), pkt.Views()) } - combinedVV.Append(pkt.Data) - packet.data = combinedVV } packet.timestampNS = ep.stack.Clock().NowNanoseconds() diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index f85a68554..edc2b5b61 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -352,18 +352,23 @@ func (e *endpoint) finishWrite(payloadBytes []byte, route *stack.Route) (int64, } if e.hdrIncluded { - if err := route.WriteHeaderIncludedPacket(&stack.PacketBuffer{ + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buffer.View(payloadBytes).ToVectorisedView(), - }); err != nil { + }) + if err := route.WriteHeaderIncludedPacket(pkt); err != nil { return 0, nil, err } } else { - hdr := buffer.NewPrependable(len(payloadBytes) + int(route.MaxHeaderLength())) - if err := route.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: e.TransProto, TTL: route.DefaultTTL(), TOS: stack.DefaultTOS}, &stack.PacketBuffer{ - Header: hdr, - Data: buffer.View(payloadBytes).ToVectorisedView(), - Owner: e.owner, - }); err != nil { + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(route.MaxHeaderLength()), + Data: buffer.View(payloadBytes).ToVectorisedView(), + }) + pkt.Owner = e.owner + if err := route.WritePacket(nil /* gso */, stack.NetworkHeaderParams{ + Protocol: e.TransProto, + TTL: route.DefaultTTL(), + TOS: stack.DefaultTOS, + }, pkt); err != nil { return 0, nil, err } } @@ -691,12 +696,13 @@ func (e *endpoint) HandlePacket(route *stack.Route, pkt *stack.PacketBuffer) { // slice. Save/restore doesn't support overlapping slices and will fail. var combinedVV buffer.VectorisedView if e.TransportEndpointInfo.NetProto == header.IPv4ProtocolNumber { - headers := make(buffer.View, 0, len(pkt.NetworkHeader)+len(pkt.TransportHeader)) - headers = append(headers, pkt.NetworkHeader...) - headers = append(headers, pkt.TransportHeader...) + network, transport := pkt.NetworkHeader().View(), pkt.TransportHeader().View() + headers := make(buffer.View, 0, len(network)+len(transport)) + headers = append(headers, network...) + headers = append(headers, transport...) combinedVV = headers.ToVectorisedView() } else { - combinedVV = append(buffer.View(nil), pkt.TransportHeader...).ToVectorisedView() + combinedVV = append(buffer.View(nil), pkt.TransportHeader().View()...).ToVectorisedView() } combinedVV.Append(pkt.Data) packet.data = combinedVV diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 46702906b..290172ac9 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -746,11 +746,7 @@ func (e *endpoint) sendTCP(r *stack.Route, tf tcpFields, data buffer.VectorisedV func buildTCPHdr(r *stack.Route, tf tcpFields, pkt *stack.PacketBuffer, gso *stack.GSO) { optLen := len(tf.opts) - hdr := &pkt.Header - packetSize := pkt.Data.Size() - // Initialize the header. - tcp := header.TCP(hdr.Prepend(header.TCPMinimumSize + optLen)) - pkt.TransportHeader = buffer.View(tcp) + tcp := header.TCP(pkt.TransportHeader().Push(header.TCPMinimumSize + optLen)) tcp.Encode(&header.TCPFields{ SrcPort: tf.id.LocalPort, DstPort: tf.id.RemotePort, @@ -762,8 +758,7 @@ func buildTCPHdr(r *stack.Route, tf tcpFields, pkt *stack.PacketBuffer, gso *sta }) copy(tcp[header.TCPMinimumSize:], tf.opts) - length := uint16(hdr.UsedLength() + packetSize) - xsum := r.PseudoHeaderChecksum(ProtocolNumber, length) + xsum := r.PseudoHeaderChecksum(ProtocolNumber, uint16(pkt.Size())) // Only calculate the checksum if offloading isn't supported. if gso != nil && gso.NeedsCsum { // This is called CHECKSUM_PARTIAL in the Linux kernel. We @@ -801,17 +796,18 @@ func sendTCPBatch(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso packetSize = size } size -= packetSize - var pkt stack.PacketBuffer - pkt.Header = buffer.NewPrependable(hdrSize) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: hdrSize, + }) pkt.Hash = tf.txHash pkt.Owner = owner pkt.EgressRoute = r pkt.GSOOptions = gso pkt.NetworkProtocolNumber = r.NetworkProtocolNumber() data.ReadToVV(&pkt.Data, packetSize) - buildTCPHdr(r, tf, &pkt, gso) + buildTCPHdr(r, tf, pkt, gso) tf.seq = tf.seq.Add(seqnum.Size(packetSize)) - pkts.PushBack(&pkt) + pkts.PushBack(pkt) } if tf.ttl == 0 { @@ -837,12 +833,12 @@ func sendTCP(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso *stac return sendTCPBatch(r, tf, data, gso, owner) } - pkt := &stack.PacketBuffer{ - Header: buffer.NewPrependable(header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen), - Data: data, - Hash: tf.txHash, - Owner: owner, - } + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen, + Data: data, + }) + pkt.Hash = tf.txHash + pkt.Owner = owner buildTCPHdr(r, tf, pkt, gso) if tf.ttl == 0 { diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index 49a673b42..c5afa2680 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -21,7 +21,6 @@ package tcp import ( - "fmt" "runtime" "strings" "time" @@ -547,22 +546,22 @@ func (p *protocol) SynRcvdCounter() *synRcvdCounter { // Parse implements stack.TransportProtocol.Parse. func (*protocol) Parse(pkt *stack.PacketBuffer) bool { - hdr, ok := pkt.Data.PullUp(header.TCPMinimumSize) + // TCP header is variable length, peek at it first. + hdrLen := header.TCPMinimumSize + hdr, ok := pkt.Data.PullUp(hdrLen) if !ok { return false } // If the header has options, pull those up as well. if offset := int(header.TCP(hdr).DataOffset()); offset > header.TCPMinimumSize && offset <= pkt.Data.Size() { - hdr, ok = pkt.Data.PullUp(offset) - if !ok { - panic(fmt.Sprintf("There should be at least %d bytes in pkt.Data.", offset)) - } + // TODO(gvisor.dev/issue/2404): Figure out whether to reject this kind of + // packets. + hdrLen = offset } - pkt.TransportHeader = hdr - pkt.Data.TrimFront(len(hdr)) - return true + _, ok = pkt.TransportHeader().Consume(hdrLen) + return ok } // NewProtocol returns a TCP transport protocol. diff --git a/pkg/tcpip/transport/tcp/segment.go b/pkg/tcpip/transport/tcp/segment.go index bb60dc29d..94307d31a 100644 --- a/pkg/tcpip/transport/tcp/segment.go +++ b/pkg/tcpip/transport/tcp/segment.go @@ -68,7 +68,7 @@ func newSegment(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketB route: r.Clone(), } s.data = pkt.Data.Clone(s.views[:]) - s.hdr = header.TCP(pkt.TransportHeader) + s.hdr = header.TCP(pkt.TransportHeader().View()) s.rcvdTime = time.Now() return s } diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 4a2b6c03a..73608783c 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -986,13 +986,16 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { // sendUDP sends a UDP segment via the provided network endpoint and under the // provided identity. func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort uint16, ttl uint8, useDefaultTTL bool, tos uint8, owner tcpip.PacketOwner, noChecksum bool) *tcpip.Error { - // Allocate a buffer for the UDP header. - hdr := buffer.NewPrependable(header.UDPMinimumSize + int(r.MaxHeaderLength())) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: header.UDPMinimumSize + int(r.MaxHeaderLength()), + Data: data, + }) + pkt.Owner = owner - // Initialize the header. - udp := header.UDP(hdr.Prepend(header.UDPMinimumSize)) + // Initialize the UDP header. + udp := header.UDP(pkt.TransportHeader().Push(header.UDPMinimumSize)) - length := uint16(hdr.UsedLength() + data.Size()) + length := uint16(pkt.Size()) udp.Encode(&header.UDPFields{ SrcPort: localPort, DstPort: remotePort, @@ -1019,12 +1022,7 @@ func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort u Protocol: ProtocolNumber, TTL: ttl, TOS: tos, - }, &stack.PacketBuffer{ - Header: hdr, - Data: data, - TransportHeader: buffer.View(udp), - Owner: owner, - }); err != nil { + }, pkt); err != nil { r.Stats().UDP.PacketSendErrors.Increment() return err } @@ -1372,7 +1370,7 @@ func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { // endpoint. func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketBuffer) { // Get the header then trim it from the view. - hdr := header.UDP(pkt.TransportHeader) + hdr := header.UDP(pkt.TransportHeader().View()) if int(hdr.Length()) > pkt.Data.Size()+header.UDPMinimumSize { // Malformed packet. e.stack.Stats().UDP.MalformedPacketsReceived.Increment() @@ -1443,9 +1441,9 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk // Save any useful information from the network header to the packet. switch r.NetProto { case header.IPv4ProtocolNumber: - packet.tos, _ = header.IPv4(pkt.NetworkHeader).TOS() + packet.tos, _ = header.IPv4(pkt.NetworkHeader().View()).TOS() case header.IPv6ProtocolNumber: - packet.tos, _ = header.IPv6(pkt.NetworkHeader).TOS() + packet.tos, _ = header.IPv6(pkt.NetworkHeader().View()).TOS() } // TODO(gvisor.dev/issue/3556): r.LocalAddress may be a multicast or broadcast diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go index 0e7464e3a..63d4bed7c 100644 --- a/pkg/tcpip/transport/udp/protocol.go +++ b/pkg/tcpip/transport/udp/protocol.go @@ -82,7 +82,7 @@ func (*protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) { // HandleUnknownDestinationPacket handles packets targeted at this protocol but // that don't match any existing endpoint. func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketBuffer) bool { - hdr := header.UDP(pkt.TransportHeader) + hdr := header.UDP(pkt.TransportHeader().View()) if int(hdr.Length()) > pkt.Data.Size()+header.UDPMinimumSize { // Malformed packet. r.Stack().Stats().UDP.MalformedPacketsReceived.Increment() @@ -130,7 +130,7 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans } headerLen := int(r.MaxHeaderLength()) + header.ICMPv4MinimumSize available := int(mtu) - headerLen - payloadLen := len(pkt.NetworkHeader) + len(pkt.TransportHeader) + pkt.Data.Size() + payloadLen := pkt.NetworkHeader().View().Size() + pkt.TransportHeader().View().Size() + pkt.Data.Size() if payloadLen > available { payloadLen = available } @@ -139,22 +139,21 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans // For example, a raw or packet socket may use what UDP // considers an unreachable destination. Thus we deep copy pkt // to prevent multiple ownership and SR errors. - newHeader := append(buffer.View(nil), pkt.NetworkHeader...) - newHeader = append(newHeader, pkt.TransportHeader...) + newHeader := append(buffer.View(nil), pkt.NetworkHeader().View()...) + newHeader = append(newHeader, pkt.TransportHeader().View()...) payload := newHeader.ToVectorisedView() payload.AppendView(pkt.Data.ToView()) payload.CapLength(payloadLen) - hdr := buffer.NewPrependable(headerLen) - pkt := header.ICMPv4(hdr.Prepend(header.ICMPv4MinimumSize)) - pkt.SetType(header.ICMPv4DstUnreachable) - pkt.SetCode(header.ICMPv4PortUnreachable) - pkt.SetChecksum(header.ICMPv4Checksum(pkt, payload)) - r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, &stack.PacketBuffer{ - Header: hdr, - TransportHeader: buffer.View(pkt), - Data: payload, + icmpPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: headerLen, + Data: payload, }) + icmpHdr := header.ICMPv4(icmpPkt.TransportHeader().Push(header.ICMPv4MinimumSize)) + icmpHdr.SetType(header.ICMPv4DstUnreachable) + icmpHdr.SetCode(header.ICMPv4PortUnreachable) + icmpHdr.SetChecksum(header.ICMPv4Checksum(icmpHdr, icmpPkt.Data)) + r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, icmpPkt) case header.IPv6AddressSize: if !r.Stack().AllowICMPMessage() { @@ -175,24 +174,24 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans } headerLen := int(r.MaxHeaderLength()) + header.ICMPv6DstUnreachableMinimumSize available := int(mtu) - headerLen - payloadLen := len(pkt.NetworkHeader) + len(pkt.TransportHeader) + pkt.Data.Size() + network, transport := pkt.NetworkHeader().View(), pkt.TransportHeader().View() + payloadLen := len(network) + len(transport) + pkt.Data.Size() if payloadLen > available { payloadLen = available } - payload := buffer.NewVectorisedView(len(pkt.NetworkHeader)+len(pkt.TransportHeader), []buffer.View{pkt.NetworkHeader, pkt.TransportHeader}) + payload := buffer.NewVectorisedView(len(network)+len(transport), []buffer.View{network, transport}) payload.Append(pkt.Data) payload.CapLength(payloadLen) - hdr := buffer.NewPrependable(headerLen) - pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6DstUnreachableMinimumSize)) - pkt.SetType(header.ICMPv6DstUnreachable) - pkt.SetCode(header.ICMPv6PortUnreachable) - pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, payload)) - r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, &stack.PacketBuffer{ - Header: hdr, - TransportHeader: buffer.View(pkt), - Data: payload, + icmpPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: headerLen, + Data: payload, }) + icmpHdr := header.ICMPv6(icmpPkt.TransportHeader().Push(header.ICMPv6DstUnreachableMinimumSize)) + icmpHdr.SetType(header.ICMPv6DstUnreachable) + icmpHdr.SetCode(header.ICMPv6PortUnreachable) + icmpHdr.SetChecksum(header.ICMPv6Checksum(icmpHdr, r.LocalAddress, r.RemoteAddress, icmpPkt.Data)) + r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, icmpPkt) } return true } @@ -215,14 +214,8 @@ func (*protocol) Wait() {} // Parse implements stack.TransportProtocol.Parse. func (*protocol) Parse(pkt *stack.PacketBuffer) bool { - h, ok := pkt.Data.PullUp(header.UDPMinimumSize) - if !ok { - // Packet is too small - return false - } - pkt.TransportHeader = h - pkt.Data.TrimFront(header.UDPMinimumSize) - return true + _, ok := pkt.TransportHeader().Consume(header.UDPMinimumSize) + return ok } // NewProtocol returns a UDP transport protocol. |