diff options
Diffstat (limited to 'pkg/tcpip')
68 files changed, 975 insertions, 499 deletions
diff --git a/pkg/tcpip/buffer/prependable.go b/pkg/tcpip/buffer/prependable.go index 48a2a2713..ba21f4eca 100644 --- a/pkg/tcpip/buffer/prependable.go +++ b/pkg/tcpip/buffer/prependable.go @@ -77,3 +77,9 @@ func (p *Prependable) Prepend(size int) []byte { p.usedIdx -= size return p.View()[:size:size] } + +// DeepCopy copies p and the bytes backing it. +func (p Prependable) DeepCopy() Prependable { + p.buf = append(View(nil), p.buf...) + return p +} diff --git a/pkg/tcpip/hash/jenkins/BUILD b/pkg/tcpip/hash/jenkins/BUILD index 0c5c20cea..e648efa71 100644 --- a/pkg/tcpip/hash/jenkins/BUILD +++ b/pkg/tcpip/hash/jenkins/BUILD @@ -7,9 +7,7 @@ go_library( name = "jenkins", srcs = ["jenkins.go"], importpath = "gvisor.dev/gvisor/pkg/tcpip/hash/jenkins", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], ) go_test( diff --git a/pkg/tcpip/link/channel/BUILD b/pkg/tcpip/link/channel/BUILD index 97a794986..7dbc05754 100644 --- a/pkg/tcpip/link/channel/BUILD +++ b/pkg/tcpip/link/channel/BUILD @@ -6,7 +6,7 @@ go_library( name = "channel", srcs = ["channel.go"], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/channel", - visibility = ["//:sandbox"], + visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/link/channel/channel.go b/pkg/tcpip/link/channel/channel.go index 22eefb564..70188551f 100644 --- a/pkg/tcpip/link/channel/channel.go +++ b/pkg/tcpip/link/channel/channel.go @@ -25,10 +25,9 @@ import ( // PacketInfo holds all the information about an outbound packet. type PacketInfo struct { - Header buffer.View - Payload buffer.View - Proto tcpip.NetworkProtocolNumber - GSO *stack.GSO + Pkt tcpip.PacketBuffer + Proto tcpip.NetworkProtocolNumber + GSO *stack.GSO } // Endpoint is link layer endpoint that stores outbound packets in a channel @@ -118,12 +117,11 @@ func (e *Endpoint) LinkAddress() tcpip.LinkAddress { } // WritePacket stores outbound packets into the channel. -func (e *Endpoint) WritePacket(_ *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (e *Endpoint) WritePacket(_ *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { p := PacketInfo{ - Header: hdr.View(), - Proto: protocol, - Payload: payload.ToView(), - GSO: gso, + Pkt: pkt, + Proto: protocol, + GSO: gso, } select { @@ -135,19 +133,20 @@ func (e *Endpoint) WritePacket(_ *stack.Route, gso *stack.GSO, hdr buffer.Prepen } // WritePackets stores outbound packets into the channel. -func (e *Endpoint) WritePackets(_ *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { - payloadView := payload.ToView() +func (e *Endpoint) WritePackets(_ *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + payloadView := pkts[0].Data.ToView() n := 0 packetLoop: - for i := range hdrs { - hdr := &hdrs[i].Hdr - off := hdrs[i].Off - size := hdrs[i].Size + for _, pkt := range pkts { + off := pkt.DataOffset + size := pkt.DataSize p := PacketInfo{ - Header: hdr.View(), - Proto: protocol, - Payload: buffer.NewViewFromBytes(payloadView[off : off+size]), - GSO: gso, + Pkt: tcpip.PacketBuffer{ + Header: pkt.Header, + Data: buffer.NewViewFromBytes(payloadView[off : off+size]).ToVectorisedView(), + }, + Proto: protocol, + GSO: gso, } select { @@ -162,12 +161,11 @@ packetLoop: } // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. -func (e *Endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { +func (e *Endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { p := PacketInfo{ - Header: packet.ToView(), - Proto: 0, - Payload: buffer.View{}, - GSO: nil, + Pkt: tcpip.PacketBuffer{Data: vv}, + Proto: 0, + GSO: nil, } select { diff --git a/pkg/tcpip/link/fdbased/BUILD b/pkg/tcpip/link/fdbased/BUILD index 8fa9e3984..897c94821 100644 --- a/pkg/tcpip/link/fdbased/BUILD +++ b/pkg/tcpip/link/fdbased/BUILD @@ -14,9 +14,7 @@ go_library( "packet_dispatchers.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/fdbased", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go index edef7db26..fa8a703d9 100644 --- a/pkg/tcpip/link/fdbased/endpoint.go +++ b/pkg/tcpip/link/fdbased/endpoint.go @@ -386,10 +386,11 @@ const ( // WritePacket writes outbound packets to the file descriptor. If it is not // currently writable, the packet is dropped. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { if e.hdrSize > 0 { // Add ethernet header if needed. - eth := header.Ethernet(hdr.Prepend(header.EthernetMinimumSize)) + eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) + pkt.LinkHeader = buffer.View(eth) ethHdr := &header.EthernetFields{ DstAddr: r.RemoteLinkAddress, Type: protocol, @@ -408,13 +409,13 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen vnetHdr := virtioNetHdr{} vnetHdrBuf := vnetHdrToByteSlice(&vnetHdr) if gso != nil { - vnetHdr.hdrLen = uint16(hdr.UsedLength()) + vnetHdr.hdrLen = uint16(pkt.Header.UsedLength()) if gso.NeedsCsum { vnetHdr.flags = _VIRTIO_NET_HDR_F_NEEDS_CSUM vnetHdr.csumStart = header.EthernetMinimumSize + gso.L3HdrLen vnetHdr.csumOffset = gso.CsumOffset } - if gso.Type != stack.GSONone && uint16(payload.Size()) > gso.MSS { + if gso.Type != stack.GSONone && uint16(pkt.Data.Size()) > gso.MSS { switch gso.Type { case stack.GSOTCPv4: vnetHdr.gsoType = _VIRTIO_NET_HDR_GSO_TCPV4 @@ -427,19 +428,19 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen } } - return rawfile.NonBlockingWrite3(e.fds[0], vnetHdrBuf, hdr.View(), payload.ToView()) + return rawfile.NonBlockingWrite3(e.fds[0], vnetHdrBuf, pkt.Header.View(), pkt.Data.ToView()) } - if payload.Size() == 0 { - return rawfile.NonBlockingWrite(e.fds[0], hdr.View()) + if pkt.Data.Size() == 0 { + return rawfile.NonBlockingWrite(e.fds[0], pkt.Header.View()) } - return rawfile.NonBlockingWrite3(e.fds[0], hdr.View(), payload.ToView(), nil) + return rawfile.NonBlockingWrite3(e.fds[0], pkt.Header.View(), pkt.Data.ToView(), nil) } // WritePackets writes outbound packets to the file descriptor. If it is not // currently writable, the packet is dropped. -func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { var ethHdrBuf []byte // hdr + data iovLen := 2 @@ -462,9 +463,9 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.Pac iovLen++ } - n := len(hdrs) + n := len(pkts) - views := payload.Views() + views := pkts[0].Data.Views() /* * Each bondary in views can add one more iovec. * @@ -482,14 +483,20 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.Pac viewOff := 0 off := 0 nextOff := 0 - for i := range hdrs { + for i := range pkts { + // TODO(b/134618279): Different packets may have different data + // in the future. We should handle this. + if !viewsEqual(pkts[i].Data.Views(), views) { + panic("All packets in pkts should have the same Data.") + } + prevIovecIdx := iovecIdx mmsgHdr := &mmsgHdrs[i] mmsgHdr.Msg.Iov = &iovec[iovecIdx] - packetSize := hdrs[i].Size - hdr := &hdrs[i].Hdr + packetSize := pkts[i].DataSize + hdr := &pkts[i].Header - off = hdrs[i].Off + off = pkts[i].DataOffset if off != nextOff { // We stop in a different point last time. size := packetSize @@ -554,9 +561,14 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.Pac return packets, nil } +// viewsEqual tests whether v1 and v2 refer to the same backing bytes. +func viewsEqual(vs1, vs2 []buffer.View) bool { + return len(vs1) == len(vs2) && (len(vs1) == 0 || &vs1[0] == &vs2[0]) +} + // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. -func (e *endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { - return rawfile.NonBlockingWrite(e.fds[0], packet.ToView()) +func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { + return rawfile.NonBlockingWrite(e.fds[0], vv.ToView()) } // InjectOutobund implements stack.InjectableEndpoint.InjectOutbound. diff --git a/pkg/tcpip/link/fdbased/endpoint_test.go b/pkg/tcpip/link/fdbased/endpoint_test.go index 7e08e033b..2066987eb 100644 --- a/pkg/tcpip/link/fdbased/endpoint_test.go +++ b/pkg/tcpip/link/fdbased/endpoint_test.go @@ -168,7 +168,10 @@ func testWritePacket(t *testing.T, plen int, eth bool, gsoMaxSize uint32) { L3HdrLen: header.IPv4MaximumHeaderSize, } } - if err := c.ep.WritePacket(r, gso, hdr, payload.ToVectorisedView(), proto); err != nil { + if err := c.ep.WritePacket(r, gso, proto, tcpip.PacketBuffer{ + Header: hdr, + Data: payload.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed: %v", err) } @@ -258,7 +261,10 @@ func TestPreserveSrcAddress(t *testing.T) { // WritePacket panics given a prependable with anything less than // the minimum size of the ethernet header. hdr := buffer.NewPrependable(header.EthernetMinimumSize) - if err := c.ep.WritePacket(r, nil /* gso */, hdr, buffer.VectorisedView{}, proto); err != nil { + if err := c.ep.WritePacket(r, nil /* gso */, proto, tcpip.PacketBuffer{ + Header: hdr, + Data: buffer.VectorisedView{}, + }); err != nil { t.Fatalf("WritePacket failed: %v", err) } diff --git a/pkg/tcpip/link/loopback/BUILD b/pkg/tcpip/link/loopback/BUILD index 23e4d1418..f35fcdff4 100644 --- a/pkg/tcpip/link/loopback/BUILD +++ b/pkg/tcpip/link/loopback/BUILD @@ -6,7 +6,7 @@ go_library( name = "loopback", srcs = ["loopback.go"], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/loopback", - visibility = ["//:sandbox"], + visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go index bc5d8a2f3..499cc608f 100644 --- a/pkg/tcpip/link/loopback/loopback.go +++ b/pkg/tcpip/link/loopback/loopback.go @@ -76,38 +76,38 @@ 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, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { - views := make([]buffer.View, 1, 1+len(payload.Views())) - views[0] = hdr.View() - views = append(views, payload.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. +func (e *endpoint) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { + views := make([]buffer.View, 1, 1+len(pkt.Data.Views())) + views[0] = pkt.Header.View() + views = append(views, pkt.Data.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(e, "" /* remote */, "" /* local */, protocol, tcpip.PacketBuffer{ - Data: buffer.NewVectorisedView(len(views[0])+payload.Size(), views), + Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), }) return nil } // WritePackets implements stack.LinkEndpoint.WritePackets. -func (e *endpoint) WritePackets(_ *stack.Route, _ *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, []tcpip.PacketBuffer, tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { panic("not implemented") } // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. -func (e *endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { +func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { // Reject the packet if it's shorter than an ethernet header. - if packet.Size() < header.EthernetMinimumSize { + if vv.Size() < header.EthernetMinimumSize { return tcpip.ErrBadAddress } - // There should be an ethernet header at the beginning of packet. - linkHeader := header.Ethernet(packet.First()[:header.EthernetMinimumSize]) - packet.TrimFront(len(linkHeader)) + // There should be an ethernet header at the beginning of vv. + linkHeader := header.Ethernet(vv.First()[:header.EthernetMinimumSize]) + vv.TrimFront(len(linkHeader)) e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, linkHeader.Type(), tcpip.PacketBuffer{ - Data: packet, + Data: vv, LinkHeader: buffer.View(linkHeader), }) diff --git a/pkg/tcpip/link/muxed/BUILD b/pkg/tcpip/link/muxed/BUILD index 1bab380b0..1ac7948b6 100644 --- a/pkg/tcpip/link/muxed/BUILD +++ b/pkg/tcpip/link/muxed/BUILD @@ -7,9 +7,7 @@ go_library( name = "muxed", srcs = ["injectable.go"], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/muxed", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/link/muxed/injectable.go b/pkg/tcpip/link/muxed/injectable.go index 9a8e8ebfe..445b22c17 100644 --- a/pkg/tcpip/link/muxed/injectable.go +++ b/pkg/tcpip/link/muxed/injectable.go @@ -87,26 +87,26 @@ func (m *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, // WritePackets writes outbound packets to the appropriate // LinkInjectableEndpoint based on the RemoteAddress. HandleLocal only works if // r.RemoteAddress has a route registered in this endpoint. -func (m *InjectableEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (m *InjectableEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { endpoint, ok := m.routes[r.RemoteAddress] if !ok { return 0, tcpip.ErrNoRoute } - return endpoint.WritePackets(r, gso, hdrs, payload, protocol) + return endpoint.WritePackets(r, gso, pkts, protocol) } // WritePacket writes outbound packets to the appropriate LinkInjectableEndpoint // based on the RemoteAddress. HandleLocal only works if r.RemoteAddress has a // route registered in this endpoint. -func (m *InjectableEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (m *InjectableEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { if endpoint, ok := m.routes[r.RemoteAddress]; ok { - return endpoint.WritePacket(r, gso, hdr, payload, protocol) + return endpoint.WritePacket(r, gso, protocol, pkt) } return tcpip.ErrNoRoute } // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. -func (m *InjectableEndpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { +func (m *InjectableEndpoint) WriteRawPacket(buffer.VectorisedView) *tcpip.Error { // WriteRawPacket doesn't get a route or network address, so there's // nowhere to write this. return tcpip.ErrNoRoute diff --git a/pkg/tcpip/link/muxed/injectable_test.go b/pkg/tcpip/link/muxed/injectable_test.go index 9cd300af8..63b249837 100644 --- a/pkg/tcpip/link/muxed/injectable_test.go +++ b/pkg/tcpip/link/muxed/injectable_test.go @@ -50,8 +50,10 @@ func TestInjectableEndpointDispatch(t *testing.T) { hdr.Prepend(1)[0] = 0xFA packetRoute := stack.Route{RemoteAddress: dstIP} - endpoint.WritePacket(&packetRoute, nil /* gso */, hdr, - buffer.NewViewFromBytes([]byte{0xFB}).ToVectorisedView(), ipv4.ProtocolNumber) + endpoint.WritePacket(&packetRoute, nil /* gso */, ipv4.ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buffer.NewViewFromBytes([]byte{0xFB}).ToVectorisedView(), + }) buf := make([]byte, 6500) bytesRead, err := sock.Read(buf) @@ -68,8 +70,10 @@ func TestInjectableEndpointDispatchHdrOnly(t *testing.T) { hdr := buffer.NewPrependable(1) hdr.Prepend(1)[0] = 0xFA packetRoute := stack.Route{RemoteAddress: dstIP} - endpoint.WritePacket(&packetRoute, nil /* gso */, hdr, - buffer.NewView(0).ToVectorisedView(), ipv4.ProtocolNumber) + endpoint.WritePacket(&packetRoute, nil /* gso */, ipv4.ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buffer.NewView(0).ToVectorisedView(), + }) buf := make([]byte, 6500) bytesRead, err := sock.Read(buf) if err != nil { diff --git a/pkg/tcpip/link/rawfile/BUILD b/pkg/tcpip/link/rawfile/BUILD index 05c7b8024..d8211e93d 100644 --- a/pkg/tcpip/link/rawfile/BUILD +++ b/pkg/tcpip/link/rawfile/BUILD @@ -13,9 +13,7 @@ go_library( "rawfile_unsafe.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/rawfile", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", "@org_golang_x_sys//unix:go_default_library", diff --git a/pkg/tcpip/link/sharedmem/BUILD b/pkg/tcpip/link/sharedmem/BUILD index 0a5ea3dc4..a4f9cdd69 100644 --- a/pkg/tcpip/link/sharedmem/BUILD +++ b/pkg/tcpip/link/sharedmem/BUILD @@ -12,9 +12,7 @@ go_library( "tx.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/sharedmem", - visibility = [ - "//:sandbox", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/log", "//pkg/tcpip", diff --git a/pkg/tcpip/link/sharedmem/pipe/BUILD b/pkg/tcpip/link/sharedmem/pipe/BUILD index 330ed5e94..6b5bc542c 100644 --- a/pkg/tcpip/link/sharedmem/pipe/BUILD +++ b/pkg/tcpip/link/sharedmem/pipe/BUILD @@ -12,7 +12,7 @@ go_library( "tx.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/sharedmem/pipe", - visibility = ["//:sandbox"], + visibility = ["//visibility:public"], ) go_test( diff --git a/pkg/tcpip/link/sharedmem/queue/BUILD b/pkg/tcpip/link/sharedmem/queue/BUILD index de1ce043d..8c9234d54 100644 --- a/pkg/tcpip/link/sharedmem/queue/BUILD +++ b/pkg/tcpip/link/sharedmem/queue/BUILD @@ -10,7 +10,7 @@ go_library( "tx.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/sharedmem/queue", - visibility = ["//:sandbox"], + visibility = ["//visibility:public"], deps = [ "//pkg/log", "//pkg/tcpip/link/sharedmem/pipe", diff --git a/pkg/tcpip/link/sharedmem/sharedmem.go b/pkg/tcpip/link/sharedmem/sharedmem.go index 2bace5298..080f9d667 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem.go +++ b/pkg/tcpip/link/sharedmem/sharedmem.go @@ -185,9 +185,10 @@ func (e *endpoint) LinkAddress() tcpip.LinkAddress { // WritePacket writes outbound packets to the file descriptor. If it is not // currently writable, the packet is dropped. -func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { // Add the ethernet header here. - eth := header.Ethernet(hdr.Prepend(header.EthernetMinimumSize)) + eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) + pkt.LinkHeader = buffer.View(eth) ethHdr := &header.EthernetFields{ DstAddr: r.RemoteLinkAddress, Type: protocol, @@ -199,10 +200,10 @@ func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, hdr buffer.Prependa } eth.Encode(ethHdr) - v := payload.ToView() + v := pkt.Data.ToView() // Transmit the packet. e.mu.Lock() - ok := e.tx.transmit(hdr.View(), v) + ok := e.tx.transmit(pkt.Header.View(), v) e.mu.Unlock() if !ok { @@ -213,13 +214,13 @@ func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, hdr buffer.Prependa } // WritePackets implements stack.LinkEndpoint.WritePackets. -func (e *endpoint) WritePackets(r *stack.Route, _ *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(r *stack.Route, _ *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { panic("not implemented") } // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. -func (e *endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { - v := packet.ToView() +func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { + v := vv.ToView() // Transmit the packet. e.mu.Lock() ok := e.tx.transmit(v, buffer.View{}) diff --git a/pkg/tcpip/link/sharedmem/sharedmem_test.go b/pkg/tcpip/link/sharedmem/sharedmem_test.go index 199406886..89603c48f 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem_test.go +++ b/pkg/tcpip/link/sharedmem/sharedmem_test.go @@ -273,7 +273,10 @@ func TestSimpleSend(t *testing.T) { randomFill(buf) proto := tcpip.NetworkProtocolNumber(rand.Intn(0x10000)) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), proto); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, proto, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed: %v", err) } @@ -342,7 +345,9 @@ func TestPreserveSrcAddressInSend(t *testing.T) { hdr := buffer.NewPrependable(header.EthernetMinimumSize) proto := tcpip.NetworkProtocolNumber(rand.Intn(0x10000)) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buffer.VectorisedView{}, proto); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, proto, tcpip.PacketBuffer{ + Header: hdr, + }); err != nil { t.Fatalf("WritePacket failed: %v", err) } @@ -396,7 +401,10 @@ func TestFillTxQueue(t *testing.T) { for i := queuePipeSize / 40; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } @@ -411,7 +419,10 @@ func TestFillTxQueue(t *testing.T) { // Next attempt to write must fail. hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != want { + if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != want { t.Fatalf("WritePacket return unexpected result: got %v, want %v", err, want) } } @@ -436,7 +447,10 @@ func TestFillTxQueueAfterBadCompletion(t *testing.T) { // Send two packets so that the id slice has at least two slots. for i := 2; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } } @@ -456,7 +470,10 @@ func TestFillTxQueueAfterBadCompletion(t *testing.T) { ids := make(map[uint64]struct{}) for i := queuePipeSize / 40; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } @@ -471,7 +488,10 @@ func TestFillTxQueueAfterBadCompletion(t *testing.T) { // Next attempt to write must fail. hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != want { + if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != want { t.Fatalf("WritePacket return unexpected result: got %v, want %v", err, want) } } @@ -494,7 +514,10 @@ func TestFillTxMemory(t *testing.T) { ids := make(map[uint64]struct{}) for i := queueDataSize / bufferSize; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } @@ -510,7 +533,10 @@ func TestFillTxMemory(t *testing.T) { // Next attempt to write must fail. hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber) + err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }) if want := tcpip.ErrWouldBlock; err != want { t.Fatalf("WritePacket return unexpected result: got %v, want %v", err, want) } @@ -535,7 +561,10 @@ func TestFillTxMemoryWithMultiBuffer(t *testing.T) { // until there is only one buffer left. for i := queueDataSize/bufferSize - 1; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } @@ -548,7 +577,10 @@ func TestFillTxMemoryWithMultiBuffer(t *testing.T) { { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) uu := buffer.NewView(bufferSize).ToVectorisedView() - if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, hdr, uu, header.IPv4ProtocolNumber); err != want { + if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: uu, + }); err != want { t.Fatalf("WritePacket return unexpected result: got %v, want %v", err, want) } } @@ -556,7 +588,10 @@ func TestFillTxMemoryWithMultiBuffer(t *testing.T) { // Attempt to write the one-buffer packet again. It must succeed. { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } } diff --git a/pkg/tcpip/link/sniffer/BUILD b/pkg/tcpip/link/sniffer/BUILD index 1756114e6..d6ae0368a 100644 --- a/pkg/tcpip/link/sniffer/BUILD +++ b/pkg/tcpip/link/sniffer/BUILD @@ -9,9 +9,7 @@ go_library( "sniffer.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/sniffer", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/log", "//pkg/tcpip", diff --git a/pkg/tcpip/link/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go index d71a03cd2..3392b7edd 100644 --- a/pkg/tcpip/link/sniffer/sniffer.go +++ b/pkg/tcpip/link/sniffer/sniffer.go @@ -49,6 +49,13 @@ var LogPackets uint32 = 1 // LogPacketsToFile must be accessed atomically. var LogPacketsToFile uint32 = 1 +var transportProtocolMinSizes map[tcpip.TransportProtocolNumber]int = map[tcpip.TransportProtocolNumber]int{ + header.ICMPv4ProtocolNumber: header.IPv4MinimumSize, + header.ICMPv6ProtocolNumber: header.IPv6MinimumSize, + header.UDPProtocolNumber: header.UDPMinimumSize, + header.TCPProtocolNumber: header.TCPMinimumSize, +} + type endpoint struct { dispatcher stack.NetworkDispatcher lower stack.LinkEndpoint @@ -193,19 +200,19 @@ func (e *endpoint) GSOMaxSize() uint32 { return 0 } -func (e *endpoint) dumpPacket(gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) { +func (e *endpoint) dumpPacket(gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { if atomic.LoadUint32(&LogPackets) == 1 && e.file == nil { - logPacket("send", protocol, hdr.View(), gso) + logPacket("send", protocol, pkt.Header.View(), gso) } if e.file != nil && atomic.LoadUint32(&LogPacketsToFile) == 1 { - hdrBuf := hdr.View() - length := len(hdrBuf) + payload.Size() + hdrBuf := pkt.Header.View() + length := len(hdrBuf) + pkt.Data.Size() if length > int(e.maxPCAPLen) { length = int(e.maxPCAPLen) } buf := bytes.NewBuffer(make([]byte, 0, pcapPacketHeaderLen+length)) - if err := binary.Write(buf, binary.BigEndian, newPCAPPacketHeader(uint32(length), uint32(len(hdrBuf)+payload.Size()))); err != nil { + if err := binary.Write(buf, binary.BigEndian, newPCAPPacketHeader(uint32(length), uint32(len(hdrBuf)+pkt.Data.Size()))); err != nil { panic(err) } if len(hdrBuf) > length { @@ -215,7 +222,7 @@ func (e *endpoint) dumpPacket(gso *stack.GSO, hdr buffer.Prependable, payload bu panic(err) } length -= len(hdrBuf) - logVectorisedView(payload, length, buf) + logVectorisedView(pkt.Data, length, buf) if _, err := e.file.Write(buf.Bytes()); err != nil { panic(err) } @@ -225,43 +232,46 @@ func (e *endpoint) dumpPacket(gso *stack.GSO, hdr buffer.Prependable, payload bu // WritePacket implements the stack.LinkEndpoint interface. It is called by // higher-level protocols to write packets; it just logs the packet and // forwards the request to the lower endpoint. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { - e.dumpPacket(gso, hdr, payload, protocol) - return e.lower.WritePacket(r, gso, hdr, payload, protocol) +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { + e.dumpPacket(gso, protocol, pkt) + return e.lower.WritePacket(r, gso, protocol, pkt) } // WritePackets implements the stack.LinkEndpoint interface. It is called by // higher-level protocols to write packets; it just logs the packet and // forwards the request to the lower endpoint. -func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { - view := payload.ToView() - for _, d := range hdrs { - e.dumpPacket(gso, d.Hdr, buffer.NewVectorisedView(d.Size, []buffer.View{view[d.Off:][:d.Size]}), protocol) +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + view := pkts[0].Data.ToView() + for _, pkt := range pkts { + e.dumpPacket(gso, protocol, tcpip.PacketBuffer{ + Header: pkt.Header, + Data: view[pkt.DataOffset:][:pkt.DataSize].ToVectorisedView(), + }) } - return e.lower.WritePackets(r, gso, hdrs, payload, protocol) + return e.lower.WritePackets(r, gso, pkts, protocol) } // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. -func (e *endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { +func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { if atomic.LoadUint32(&LogPackets) == 1 && e.file == nil { logPacket("send", 0, buffer.View("[raw packet, no header available]"), nil /* gso */) } if e.file != nil && atomic.LoadUint32(&LogPacketsToFile) == 1 { - length := packet.Size() + length := vv.Size() if length > int(e.maxPCAPLen) { length = int(e.maxPCAPLen) } buf := bytes.NewBuffer(make([]byte, 0, pcapPacketHeaderLen+length)) - if err := binary.Write(buf, binary.BigEndian, newPCAPPacketHeader(uint32(length), uint32(packet.Size()))); err != nil { + if err := binary.Write(buf, binary.BigEndian, newPCAPPacketHeader(uint32(length), uint32(vv.Size()))); err != nil { panic(err) } - logVectorisedView(packet, length, buf) + logVectorisedView(vv, length, buf) if _, err := e.file.Write(buf.Bytes()); err != nil { panic(err) } } - return e.lower.WriteRawPacket(packet) + return e.lower.WriteRawPacket(vv) } func logVectorisedView(vv buffer.VectorisedView, length int, buf *bytes.Buffer) { @@ -330,6 +340,13 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie return } + // We aren't guaranteed to have a transport header - it's possible for + // writes via raw endpoints to contain only network headers. + if minSize, ok := transportProtocolMinSizes[tcpip.TransportProtocolNumber(transProto)]; ok && len(b) < minSize { + log.Infof("%s %v -> %v transport protocol: %d, but no transport header found (possible raw packet)", prefix, src, dst, transProto) + return + } + // Figure out the transport layer info. transName := "unknown" srcPort := uint16(0) diff --git a/pkg/tcpip/link/tun/BUILD b/pkg/tcpip/link/tun/BUILD index 92dce8fac..a71a493fc 100644 --- a/pkg/tcpip/link/tun/BUILD +++ b/pkg/tcpip/link/tun/BUILD @@ -6,7 +6,5 @@ go_library( name = "tun", srcs = ["tun_unsafe.go"], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/tun", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], ) diff --git a/pkg/tcpip/link/waitable/BUILD b/pkg/tcpip/link/waitable/BUILD index 0746dc8ec..134837943 100644 --- a/pkg/tcpip/link/waitable/BUILD +++ b/pkg/tcpip/link/waitable/BUILD @@ -9,9 +9,7 @@ go_library( "waitable.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/waitable", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/gate", "//pkg/tcpip", diff --git a/pkg/tcpip/link/waitable/waitable.go b/pkg/tcpip/link/waitable/waitable.go index b440970e0..a8de38979 100644 --- a/pkg/tcpip/link/waitable/waitable.go +++ b/pkg/tcpip/link/waitable/waitable.go @@ -99,12 +99,12 @@ func (e *Endpoint) LinkAddress() tcpip.LinkAddress { // WritePacket implements stack.LinkEndpoint.WritePacket. It is called by // higher-level protocols to write packets. It only forwards packets to the // lower endpoint if Wait or WaitWrite haven't been called. -func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { if !e.writeGate.Enter() { return nil } - err := e.lower.WritePacket(r, gso, hdr, payload, protocol) + err := e.lower.WritePacket(r, gso, protocol, pkt) e.writeGate.Leave() return err } @@ -112,23 +112,23 @@ func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen // WritePackets implements stack.LinkEndpoint.WritePackets. It is called by // higher-level protocols to write packets. It only forwards packets to the // lower endpoint if Wait or WaitWrite haven't been called. -func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { if !e.writeGate.Enter() { - return len(hdrs), nil + return len(pkts), nil } - n, err := e.lower.WritePackets(r, gso, hdrs, payload, protocol) + n, err := e.lower.WritePackets(r, gso, pkts, protocol) e.writeGate.Leave() return n, err } // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. -func (e *Endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { +func (e *Endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { if !e.writeGate.Enter() { return nil } - err := e.lower.WriteRawPacket(packet) + err := e.lower.WriteRawPacket(vv) e.writeGate.Leave() return err } diff --git a/pkg/tcpip/link/waitable/waitable_test.go b/pkg/tcpip/link/waitable/waitable_test.go index df2e70e54..31b11a27a 100644 --- a/pkg/tcpip/link/waitable/waitable_test.go +++ b/pkg/tcpip/link/waitable/waitable_test.go @@ -65,18 +65,18 @@ func (e *countedEndpoint) LinkAddress() tcpip.LinkAddress { return e.linkAddr } -func (e *countedEndpoint) WritePacket(r *stack.Route, _ *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (e *countedEndpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { e.writeCount++ return nil } // WritePackets implements stack.LinkEndpoint.WritePackets. -func (e *countedEndpoint) WritePackets(r *stack.Route, _ *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { - e.writeCount += len(hdrs) - return len(hdrs), nil +func (e *countedEndpoint) WritePackets(r *stack.Route, _ *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + e.writeCount += len(pkts) + return len(pkts), nil } -func (e *countedEndpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { +func (e *countedEndpoint) WriteRawPacket(buffer.VectorisedView) *tcpip.Error { e.writeCount++ return nil } @@ -89,21 +89,21 @@ func TestWaitWrite(t *testing.T) { wep := New(ep) // Write and check that it goes through. - wep.WritePacket(nil, nil /* gso */, buffer.Prependable{}, buffer.VectorisedView{}, 0) + wep.WritePacket(nil, nil /* gso */, 0, tcpip.PacketBuffer{}) if want := 1; ep.writeCount != want { t.Fatalf("Unexpected writeCount: got=%v, want=%v", ep.writeCount, want) } // Wait on dispatches, then try to write. It must go through. wep.WaitDispatch() - wep.WritePacket(nil, nil /* gso */, buffer.Prependable{}, buffer.VectorisedView{}, 0) + wep.WritePacket(nil, nil /* gso */, 0, tcpip.PacketBuffer{}) if want := 2; ep.writeCount != want { t.Fatalf("Unexpected writeCount: got=%v, want=%v", ep.writeCount, want) } // Wait on writes, then try to write. It must not go through. wep.WaitWrite() - wep.WritePacket(nil, nil /* gso */, buffer.Prependable{}, buffer.VectorisedView{}, 0) + wep.WritePacket(nil, nil /* gso */, 0, tcpip.PacketBuffer{}) if want := 2; ep.writeCount != want { t.Fatalf("Unexpected writeCount: got=%v, want=%v", ep.writeCount, want) } diff --git a/pkg/tcpip/network/arp/BUILD b/pkg/tcpip/network/arp/BUILD index df0d3a8c0..e7617229b 100644 --- a/pkg/tcpip/network/arp/BUILD +++ b/pkg/tcpip/network/arp/BUILD @@ -7,9 +7,7 @@ go_library( name = "arp", srcs = ["arp.go"], importpath = "gvisor.dev/gvisor/pkg/tcpip/network/arp", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index 0ee509ebe..da8482509 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -79,16 +79,16 @@ func (e *endpoint) MaxHeaderLength() uint16 { func (e *endpoint) Close() {} -func (e *endpoint) WritePacket(*stack.Route, *stack.GSO, buffer.Prependable, buffer.VectorisedView, stack.NetworkHeaderParams, stack.PacketLooping) *tcpip.Error { +func (e *endpoint) WritePacket(*stack.Route, *stack.GSO, stack.NetworkHeaderParams, stack.PacketLooping, tcpip.PacketBuffer) *tcpip.Error { return tcpip.ErrNotSupported } // WritePackets implements stack.NetworkEndpoint.WritePackets. -func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, []stack.PacketDescriptor, buffer.VectorisedView, stack.NetworkHeaderParams, stack.PacketLooping) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, []tcpip.PacketBuffer, stack.NetworkHeaderParams, stack.PacketLooping) (int, *tcpip.Error) { return 0, tcpip.ErrNotSupported } -func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error { +func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, loop stack.PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error { return tcpip.ErrNotSupported } @@ -106,14 +106,16 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt tcpip.PacketBuffer) { return // we have no useful answer, ignore the request } hdr := buffer.NewPrependable(int(e.linkEP.MaxHeaderLength()) + header.ARPSize) - pkt := header.ARP(hdr.Prepend(header.ARPSize)) - pkt.SetIPv4OverEthernet() - pkt.SetOp(header.ARPReply) - copy(pkt.HardwareAddressSender(), r.LocalLinkAddress[:]) - copy(pkt.ProtocolAddressSender(), h.ProtocolAddressTarget()) - copy(pkt.HardwareAddressTarget(), h.HardwareAddressSender()) - copy(pkt.ProtocolAddressTarget(), h.ProtocolAddressSender()) - e.linkEP.WritePacket(r, nil /* gso */, hdr, buffer.VectorisedView{}, ProtocolNumber) + packet := header.ARP(hdr.Prepend(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, tcpip.PacketBuffer{ + Header: hdr, + }) fallthrough // also fill the cache from requests case header.ARPReply: addr := tcpip.Address(h.ProtocolAddressSender()) @@ -165,7 +167,9 @@ func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, linkEP stack. copy(h.ProtocolAddressSender(), localAddr) copy(h.ProtocolAddressTarget(), addr) - return linkEP.WritePacket(r, nil /* gso */, hdr, buffer.VectorisedView{}, ProtocolNumber) + return linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + }) } // ResolveStaticAddress implements stack.LinkAddressResolver. diff --git a/pkg/tcpip/network/arp/arp_test.go b/pkg/tcpip/network/arp/arp_test.go index 47098bfdc..8e6048a21 100644 --- a/pkg/tcpip/network/arp/arp_test.go +++ b/pkg/tcpip/network/arp/arp_test.go @@ -110,13 +110,13 @@ func TestDirectRequest(t *testing.T) { for i, address := range []tcpip.Address{stackAddr1, stackAddr2} { t.Run(strconv.Itoa(i), func(t *testing.T) { inject(address) - pkt := <-c.linkEP.C - if pkt.Proto != arp.ProtocolNumber { - t.Fatalf("expected ARP response, got network protocol number %d", pkt.Proto) + pi := <-c.linkEP.C + if pi.Proto != arp.ProtocolNumber { + t.Fatalf("expected ARP response, got network protocol number %d", pi.Proto) } - rep := header.ARP(pkt.Header) + rep := header.ARP(pi.Pkt.Header.View()) if !rep.IsValid() { - t.Fatalf("invalid ARP response len(pkt.Header)=%d", len(pkt.Header)) + t.Fatalf("invalid ARP response pi.Pkt.Header.UsedLength()=%d", pi.Pkt.Header.UsedLength()) } if got, want := tcpip.LinkAddress(rep.HardwareAddressSender()), stackLinkAddr; got != want { t.Errorf("got HardwareAddressSender = %s, want = %s", got, want) diff --git a/pkg/tcpip/network/fragmentation/BUILD b/pkg/tcpip/network/fragmentation/BUILD index 2cad0a0b6..acf1e022c 100644 --- a/pkg/tcpip/network/fragmentation/BUILD +++ b/pkg/tcpip/network/fragmentation/BUILD @@ -25,7 +25,7 @@ go_library( "reassembler_list.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/network/fragmentation", - visibility = ["//:sandbox"], + visibility = ["//visibility:public"], deps = [ "//pkg/log", "//pkg/tcpip", @@ -44,11 +44,3 @@ go_test( embed = [":fragmentation"], deps = ["//pkg/tcpip/buffer"], ) - -filegroup( - name = "autogen", - srcs = [ - "reassembler_list.go", - ], - visibility = ["//:sandbox"], -) diff --git a/pkg/tcpip/network/ip_test.go b/pkg/tcpip/network/ip_test.go index fe499d47e..4144a7837 100644 --- a/pkg/tcpip/network/ip_test.go +++ b/pkg/tcpip/network/ip_test.go @@ -150,29 +150,29 @@ func (*testObject) Wait() {} // WritePacket is called by network endpoints after producing a packet and // writing it to the link endpoint. This is used by the test object to verify // that the produced packet is as expected. -func (t *testObject) WritePacket(_ *stack.Route, _ *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (t *testObject) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { var prot tcpip.TransportProtocolNumber var srcAddr tcpip.Address var dstAddr tcpip.Address if t.v4 { - h := header.IPv4(hdr.View()) + h := header.IPv4(pkt.Header.View()) prot = tcpip.TransportProtocolNumber(h.Protocol()) srcAddr = h.SourceAddress() dstAddr = h.DestinationAddress() } else { - h := header.IPv6(hdr.View()) + h := header.IPv6(pkt.Header.View()) prot = tcpip.TransportProtocolNumber(h.NextHeader()) srcAddr = h.SourceAddress() dstAddr = h.DestinationAddress() } - t.checkValues(prot, payload, srcAddr, dstAddr) + t.checkValues(prot, pkt.Data, srcAddr, dstAddr) return nil } // WritePackets implements stack.LinkEndpoint.WritePackets. -func (t *testObject) WritePackets(_ *stack.Route, _ *stack.GSO, hdr []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (t *testObject) WritePackets(_ *stack.Route, _ *stack.GSO, pkt []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { panic("not implemented") } @@ -239,7 +239,10 @@ func TestIPv4Send(t *testing.T) { if err != nil { t.Fatalf("could not find route: %v", err) } - if err := ep.WritePacket(&r, nil /* gso */, hdr, payload.ToVectorisedView(), stack.NetworkHeaderParams{Protocol: 123, TTL: 123, TOS: stack.DefaultTOS}, stack.PacketOut); err != nil { + if err := ep.WritePacket(&r, nil /* gso */, stack.NetworkHeaderParams{Protocol: 123, TTL: 123, TOS: stack.DefaultTOS}, stack.PacketOut, tcpip.PacketBuffer{ + Header: hdr, + Data: payload.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed: %v", err) } } @@ -477,7 +480,10 @@ func TestIPv6Send(t *testing.T) { if err != nil { t.Fatalf("could not find route: %v", err) } - if err := ep.WritePacket(&r, nil /* gso */, hdr, payload.ToVectorisedView(), stack.NetworkHeaderParams{Protocol: 123, TTL: 123, TOS: stack.DefaultTOS}, stack.PacketOut); err != nil { + if err := ep.WritePacket(&r, nil /* gso */, stack.NetworkHeaderParams{Protocol: 123, TTL: 123, TOS: stack.DefaultTOS}, stack.PacketOut, tcpip.PacketBuffer{ + Header: hdr, + Data: payload.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed: %v", err) } } diff --git a/pkg/tcpip/network/ipv4/BUILD b/pkg/tcpip/network/ipv4/BUILD index 58e537aad..aeddfcdd4 100644 --- a/pkg/tcpip/network/ipv4/BUILD +++ b/pkg/tcpip/network/ipv4/BUILD @@ -10,9 +10,7 @@ go_library( "ipv4.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/network/ipv4", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go index ce771631c..32bf39e43 100644 --- a/pkg/tcpip/network/ipv4/icmp.go +++ b/pkg/tcpip/network/ipv4/icmp.go @@ -99,7 +99,11 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt tcpip.PacketBuffer) { pkt.SetChecksum(0) pkt.SetChecksum(^header.Checksum(pkt, header.ChecksumVV(vv, 0))) sent := stats.ICMP.V4PacketsSent - if err := r.WritePacket(nil /* gso */, hdr, vv, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}); err != nil { + if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + Data: vv, + TransportHeader: buffer.View(pkt), + }); err != nil { sent.Dropped.Increment() return } diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index ac16c8add..e645cf62c 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -117,13 +117,14 @@ func (e *endpoint) GSOMaxSize() uint32 { } // writePacketFragments calls e.linkEP.WritePacket with each packet fragment to -// write. It assumes that the IP header is entirely in hdr but does not assume -// that only the IP header is in hdr. It assumes that the input packet's stated -// length matches the length of the hdr+payload. 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, hdr buffer.Prependable, payload buffer.VectorisedView, mtu int) *tcpip.Error { +// 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. +func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, mtu int, pkt tcpip.PacketBuffer) *tcpip.Error { // This packet is too big, it needs to be fragmented. - ip := header.IPv4(hdr.View()) + ip := header.IPv4(pkt.Header.View()) flags := ip.Flags() // Update mtu to take into account the header, which will exist in all @@ -137,62 +138,77 @@ func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, hdr buff outerMTU := innerMTU + int(ip.HeaderLength()) offset := ip.FragmentOffset() - originalAvailableLength := hdr.AvailableLength() + originalAvailableLength := pkt.Header.AvailableLength() for i := 0; i < n; i++ { // Where possible, the first fragment that is sent has the same - // hdr.UsedLength() as the input packet. The link-layer endpoint may depends - // on this for looking at, eg, L4 headers. + // 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 { - hdr = buffer.NewPrependable(int(ip.HeaderLength()) + originalAvailableLength) - h = header.IPv4(hdr.Prepend(int(ip.HeaderLength()))) + pkt.Header = buffer.NewPrependable(int(ip.HeaderLength()) + originalAvailableLength) + h = header.IPv4(pkt.Header.Prepend(int(ip.HeaderLength()))) copy(h, ip[:ip.HeaderLength()]) } if i != n-1 { h.SetTotalLength(uint16(outerMTU)) h.SetFlagsFragmentOffset(flags|header.IPv4FlagMoreFragments, offset) } else { - h.SetTotalLength(uint16(h.HeaderLength()) + uint16(payload.Size())) + h.SetTotalLength(uint16(h.HeaderLength()) + uint16(pkt.Data.Size())) h.SetFlagsFragmentOffset(flags, offset) } h.SetChecksum(0) h.SetChecksum(^h.CalculateChecksum()) offset += uint16(innerMTU) if i > 0 { - newPayload := payload.Clone([]buffer.View{}) + newPayload := pkt.Data.Clone(nil) newPayload.CapLength(innerMTU) - if err := e.linkEP.WritePacket(r, gso, hdr, newPayload, ProtocolNumber); err != nil { + if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, tcpip.PacketBuffer{ + Header: pkt.Header, + Data: newPayload, + NetworkHeader: buffer.View(h), + }); err != nil { return err } r.Stats().IP.PacketsSent.Increment() - payload.TrimFront(newPayload.Size()) + pkt.Data.TrimFront(newPayload.Size()) continue } - // Special handling for the first fragment because it comes from the hdr. - if outerMTU >= hdr.UsedLength() { - // This fragment can fit all of hdr and possibly some of payload, too. - newPayload := payload.Clone([]buffer.View{}) - newPayloadLength := outerMTU - hdr.UsedLength() + // 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, hdr, newPayload, ProtocolNumber); err != nil { + if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, tcpip.PacketBuffer{ + Header: pkt.Header, + Data: newPayload, + NetworkHeader: buffer.View(h), + }); err != nil { return err } r.Stats().IP.PacketsSent.Increment() - payload.TrimFront(newPayloadLength) + pkt.Data.TrimFront(newPayloadLength) } else { - // The fragment is too small to fit all of hdr. - startOfHdr := hdr - startOfHdr.TrimBack(hdr.UsedLength() - outerMTU) + // 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, startOfHdr, emptyVV, ProtocolNumber); err != nil { + if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, tcpip.PacketBuffer{ + Header: startOfHdr, + Data: emptyVV, + NetworkHeader: buffer.View(h), + }); err != nil { return err } r.Stats().IP.PacketsSent.Increment() - // Add the unused bytes of hdr into the payload that remains to be sent. - restOfHdr := hdr.View()[outerMTU:] + // 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(payload) - payload = tmp + tmp.Append(pkt.Data) + pkt.Data = tmp } } return nil @@ -222,18 +238,20 @@ func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadS } // WritePacket writes a packet to the given destination address and protocol. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) *tcpip.Error { - ip := e.addIPHeader(r, &hdr, payload.Size(), params) +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, loop stack.PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error { + ip := e.addIPHeader(r, &pkt.Header, pkt.Data.Size(), params) + pkt.NetworkHeader = buffer.View(ip) if loop&stack.PacketLoop != 0 { - views := make([]buffer.View, 1, 1+len(payload.Views())) - views[0] = hdr.View() - views = append(views, payload.Views()...) + // 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, tcpip.PacketBuffer{ - Data: buffer.NewVectorisedView(len(views[0])+payload.Size(), views), - NetworkHeader: buffer.View(ip), + Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), }) loopedR.Release() @@ -241,10 +259,10 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen if loop&stack.PacketOut == 0 { return nil } - if hdr.UsedLength()+payload.Size() > int(e.linkEP.MTU()) && (gso == nil || gso.Type == stack.GSONone) { - return e.writePacketFragments(r, gso, hdr, payload, int(e.linkEP.MTU())) + if pkt.Header.UsedLength()+pkt.Data.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, hdr, payload, ProtocolNumber); err != nil { + if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, pkt); err != nil { return err } r.Stats().IP.PacketsSent.Increment() @@ -252,34 +270,35 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen } // WritePackets implements stack.NetworkEndpoint.WritePackets. -func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) { if loop&stack.PacketLoop != 0 { panic("multiple packets in local loop") } if loop&stack.PacketOut == 0 { - return len(hdrs), nil + return len(pkts), nil } - for i := range hdrs { - e.addIPHeader(r, &hdrs[i].Hdr, hdrs[i].Size, params) + for i := range pkts { + ip := e.addIPHeader(r, &pkts[i].Header, pkts[i].DataSize, params) + pkts[i].NetworkHeader = buffer.View(ip) } - n, err := e.linkEP.WritePackets(r, gso, hdrs, payload, ProtocolNumber) + n, err := e.linkEP.WritePackets(r, gso, pkts, ProtocolNumber) r.Stats().IP.PacketsSent.IncrementBy(uint64(n)) return n, err } // WriteHeaderIncludedPacket writes a packet already containing a network // header through the given route. -func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error { +func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, loop stack.PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error { // The packet already has an IP header, but there are a few required // checks. - ip := header.IPv4(payload.First()) - if !ip.IsValid(payload.Size()) { + ip := header.IPv4(pkt.Data.First()) + if !ip.IsValid(pkt.Data.Size()) { return tcpip.ErrInvalidOptionValue } // Always set the total length. - ip.SetTotalLength(uint16(payload.Size())) + ip.SetTotalLength(uint16(pkt.Data.Size())) // Set the source address when zero. if ip.SourceAddress() == tcpip.Address(([]byte{0, 0, 0, 0})) { @@ -293,7 +312,7 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.Vect // Set the packet ID when zero. if ip.ID() == 0 { id := uint32(0) - if payload.Size() > header.IPv4MaximumHeaderSize+8 { + if pkt.Data.Size() > header.IPv4MaximumHeaderSize+8 { // Packets of 68 bytes or less are required by RFC 791 to not be // fragmented, so we only assign ids to larger packets. id = atomic.AddUint32(&e.protocol.ids[hashRoute(r, 0 /* protocol */, e.protocol.hashIV)%buckets], 1) @@ -306,18 +325,18 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.Vect ip.SetChecksum(^ip.CalculateChecksum()) if loop&stack.PacketLoop != 0 { - e.HandlePacket(r, tcpip.PacketBuffer{ - Data: payload, - NetworkHeader: buffer.View(ip), - }) + e.HandlePacket(r, pkt.Clone()) } if loop&stack.PacketOut == 0 { return nil } - hdr := buffer.NewPrependableFromView(payload.ToView()) r.Stats().IP.PacketsSent.Increment() - return e.linkEP.WritePacket(r, nil /* gso */, hdr, buffer.VectorisedView{}, ProtocolNumber) + + 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 diff --git a/pkg/tcpip/network/ipv4/ipv4_test.go b/pkg/tcpip/network/ipv4/ipv4_test.go index 01dfb5f20..e900f1b45 100644 --- a/pkg/tcpip/network/ipv4/ipv4_test.go +++ b/pkg/tcpip/network/ipv4/ipv4_test.go @@ -113,12 +113,12 @@ func makeHdrAndPayload(hdrLength int, extraLength int, viewSizes []int) (buffer. // comparePayloads compared the contents of all the packets against the contents // of the source packet. -func compareFragments(t *testing.T, packets []packetInfo, sourcePacketInfo packetInfo, mtu uint32) { +func compareFragments(t *testing.T, packets []tcpip.PacketBuffer, sourcePacketInfo tcpip.PacketBuffer, mtu uint32) { t.Helper() // Make a complete array of the sourcePacketInfo packet. source := header.IPv4(packets[0].Header.View()[:header.IPv4MinimumSize]) source = append(source, sourcePacketInfo.Header.View()...) - source = append(source, sourcePacketInfo.Payload.ToView()...) + source = append(source, sourcePacketInfo.Data.ToView()...) // Make a copy of the IP header, which will be modified in some fields to make // an expected header. @@ -132,7 +132,7 @@ func compareFragments(t *testing.T, packets []packetInfo, sourcePacketInfo packe for i, packet := range packets { // Confirm that the packet is valid. allBytes := packet.Header.View().ToVectorisedView() - allBytes.Append(packet.Payload) + allBytes.Append(packet.Data) ip := header.IPv4(allBytes.ToView()) if !ip.IsValid(len(ip)) { t.Errorf("IP packet is invalid:\n%s", hex.Dump(ip)) @@ -173,7 +173,7 @@ func compareFragments(t *testing.T, packets []packetInfo, sourcePacketInfo packe type errorChannel struct { *channel.Endpoint - Ch chan packetInfo + Ch chan tcpip.PacketBuffer packetCollectorErrors []*tcpip.Error } @@ -183,17 +183,11 @@ type errorChannel struct { func newErrorChannel(size int, mtu uint32, linkAddr tcpip.LinkAddress, packetCollectorErrors []*tcpip.Error) *errorChannel { return &errorChannel{ Endpoint: channel.New(size, mtu, linkAddr), - Ch: make(chan packetInfo, size), + Ch: make(chan tcpip.PacketBuffer, size), packetCollectorErrors: packetCollectorErrors, } } -// packetInfo holds all the information about an outbound packet. -type packetInfo struct { - Header buffer.Prependable - Payload buffer.VectorisedView -} - // Drain removes all outbound packets from the channel and counts them. func (e *errorChannel) Drain() int { c := 0 @@ -208,14 +202,9 @@ func (e *errorChannel) Drain() int { } // WritePacket stores outbound packets into the channel. -func (e *errorChannel) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { - p := packetInfo{ - Header: hdr, - Payload: payload, - } - +func (e *errorChannel) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { select { - case e.Ch <- p: + case e.Ch <- pkt: default: } @@ -292,18 +281,21 @@ func TestFragmentation(t *testing.T) { for _, ft := range fragTests { t.Run(ft.description, func(t *testing.T) { hdr, payload := makeHdrAndPayload(ft.hdrLength, ft.extraLength, ft.payloadViewsSizes) - source := packetInfo{ + source := tcpip.PacketBuffer{ Header: hdr, // Save the source payload because WritePacket will modify it. - Payload: payload.Clone([]buffer.View{}), + Data: payload.Clone(nil), } c := buildContext(t, nil, ft.mtu) - err := c.Route.WritePacket(ft.gso, hdr, payload, stack.NetworkHeaderParams{Protocol: tcp.ProtocolNumber, TTL: 42, TOS: stack.DefaultTOS}) + err := c.Route.WritePacket(ft.gso, stack.NetworkHeaderParams{Protocol: tcp.ProtocolNumber, TTL: 42, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + Data: payload, + }) if err != nil { t.Errorf("err got %v, want %v", err, nil) } - var results []packetInfo + var results []tcpip.PacketBuffer L: for { select { @@ -345,7 +337,10 @@ func TestFragmentationErrors(t *testing.T) { t.Run(ft.description, func(t *testing.T) { hdr, payload := makeHdrAndPayload(ft.hdrLength, header.IPv4MinimumSize, ft.payloadViewsSizes) c := buildContext(t, ft.packetCollectorErrors, ft.mtu) - err := c.Route.WritePacket(&stack.GSO{}, hdr, payload, stack.NetworkHeaderParams{Protocol: tcp.ProtocolNumber, TTL: 42, TOS: stack.DefaultTOS}) + err := c.Route.WritePacket(&stack.GSO{}, stack.NetworkHeaderParams{Protocol: tcp.ProtocolNumber, TTL: 42, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + Data: payload, + }) for i := 0; i < len(ft.packetCollectorErrors)-1; i++ { if got, want := ft.packetCollectorErrors[i], (*tcpip.Error)(nil); got != want { t.Errorf("ft.packetCollectorErrors[%d] got %v, want %v", i, got, want) diff --git a/pkg/tcpip/network/ipv6/BUILD b/pkg/tcpip/network/ipv6/BUILD index f06622a8b..e4e273460 100644 --- a/pkg/tcpip/network/ipv6/BUILD +++ b/pkg/tcpip/network/ipv6/BUILD @@ -10,9 +10,7 @@ go_library( "ipv6.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/network/ipv6", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index 6629951c6..1c3410618 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -226,7 +226,9 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt tcpip.P // // 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 */, hdr, buffer.VectorisedView{}, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: stack.DefaultTOS}); err != nil { + if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + }); err != nil { sent.Dropped.Increment() return } @@ -291,7 +293,10 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt tcpip.P copy(packet, h) packet.SetType(header.ICMPv6EchoReply) packet.SetChecksum(header.ICMPv6Checksum(packet, r.LocalAddress, r.RemoteAddress, pkt.Data)) - if err := r.WritePacket(nil /* gso */, hdr, pkt.Data, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}); err != nil { + if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + Data: pkt.Data, + }); err != nil { sent.Dropped.Increment() return } @@ -417,7 +422,9 @@ func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, linkEP stack. }) // TODO(stijlist): count this in ICMP stats. - return linkEP.WritePacket(r, nil /* gso */, hdr, buffer.VectorisedView{}, ProtocolNumber) + return linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + }) } // ResolveStaticAddress implements stack.LinkAddressResolver. diff --git a/pkg/tcpip/network/ipv6/icmp_test.go b/pkg/tcpip/network/ipv6/icmp_test.go index 6037a1ef8..335f634d5 100644 --- a/pkg/tcpip/network/ipv6/icmp_test.go +++ b/pkg/tcpip/network/ipv6/icmp_test.go @@ -55,7 +55,7 @@ func (*stubLinkEndpoint) LinkAddress() tcpip.LinkAddress { return "" } -func (*stubLinkEndpoint) WritePacket(*stack.Route, *stack.GSO, buffer.Prependable, buffer.VectorisedView, tcpip.NetworkProtocolNumber) *tcpip.Error { +func (*stubLinkEndpoint) WritePacket(*stack.Route, *stack.GSO, tcpip.NetworkProtocolNumber, tcpip.PacketBuffer) *tcpip.Error { return nil } @@ -276,22 +276,22 @@ type routeArgs struct { func routeICMPv6Packet(t *testing.T, args routeArgs, fn func(*testing.T, header.ICMPv6)) { t.Helper() - pkt := <-args.src.C + pi := <-args.src.C { - views := []buffer.View{pkt.Header, pkt.Payload} - size := len(pkt.Header) + len(pkt.Payload) + views := []buffer.View{pi.Pkt.Header.View(), pi.Pkt.Data.ToView()} + size := pi.Pkt.Header.UsedLength() + pi.Pkt.Data.Size() vv := buffer.NewVectorisedView(size, views) - args.dst.InjectLinkAddr(pkt.Proto, args.dst.LinkAddress(), tcpip.PacketBuffer{ + args.dst.InjectLinkAddr(pi.Proto, args.dst.LinkAddress(), tcpip.PacketBuffer{ Data: vv, }) } - if pkt.Proto != ProtocolNumber { - t.Errorf("unexpected protocol number %d", pkt.Proto) + if pi.Proto != ProtocolNumber { + t.Errorf("unexpected protocol number %d", pi.Proto) return } - ipv6 := header.IPv6(pkt.Header) + ipv6 := header.IPv6(pi.Pkt.Header.View()) transProto := tcpip.TransportProtocolNumber(ipv6.NextHeader()) if transProto != header.ICMPv6ProtocolNumber { t.Errorf("unexpected transport protocol number %d", transProto) diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index 4cee848a1..dd31f0fb7 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -112,18 +112,20 @@ func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadS } // WritePacket writes a packet to the given destination address and protocol. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) *tcpip.Error { - ip := e.addIPHeader(r, &hdr, payload.Size(), params) +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, loop stack.PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error { + ip := e.addIPHeader(r, &pkt.Header, pkt.Data.Size(), params) + pkt.NetworkHeader = buffer.View(ip) if loop&stack.PacketLoop != 0 { - views := make([]buffer.View, 1, 1+len(payload.Views())) - views[0] = hdr.View() - views = append(views, payload.Views()...) + // 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, tcpip.PacketBuffer{ - Data: buffer.NewVectorisedView(len(views[0])+payload.Size(), views), - NetworkHeader: buffer.View(ip), + Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), }) loopedR.Release() @@ -133,32 +135,33 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen } r.Stats().IP.PacketsSent.Increment() - return e.linkEP.WritePacket(r, gso, hdr, payload, ProtocolNumber) + return e.linkEP.WritePacket(r, gso, ProtocolNumber, pkt) } // WritePackets implements stack.LinkEndpoint.WritePackets. -func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) { if loop&stack.PacketLoop != 0 { panic("not implemented") } if loop&stack.PacketOut == 0 { - return len(hdrs), nil + return len(pkts), nil } - for i := range hdrs { - hdr := &hdrs[i].Hdr - size := hdrs[i].Size - e.addIPHeader(r, hdr, size, params) + for i := range pkts { + hdr := &pkts[i].Header + size := pkts[i].DataSize + ip := e.addIPHeader(r, hdr, size, params) + pkts[i].NetworkHeader = buffer.View(ip) } - n, err := e.linkEP.WritePackets(r, gso, hdrs, payload, ProtocolNumber) + n, err := e.linkEP.WritePackets(r, gso, pkts, ProtocolNumber) r.Stats().IP.PacketsSent.IncrementBy(uint64(n)) return n, err } // WriteHeaderIncludedPacker implements stack.NetworkEndpoint. It is not yet // supported by IPv6. -func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error { +func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, loop stack.PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error { // TODO(b/119580726): Support IPv6 header-included packets. return tcpip.ErrNotSupported } diff --git a/pkg/tcpip/packet_buffer.go b/pkg/tcpip/packet_buffer.go index 10b04239d..ab24372e7 100644 --- a/pkg/tcpip/packet_buffer.go +++ b/pkg/tcpip/packet_buffer.go @@ -31,12 +31,27 @@ type PacketBuffer struct { // or otherwise modified. Data buffer.VectorisedView + // DataOffset is used for GSO output. It is the offset into the Data + // field where the payload of this packet starts. + DataOffset int + + // DataSize is used for GSO output. It is the size of this packet's + // payload. + DataSize int + + // Header holds the headers of outbound packets. As a packet is passed + // down the stack, each layer adds to Header. + Header buffer.Prependable + + // These fields are used by both inbound and outbound packets. They + // typically overlap with the Data and Header fields. + // // 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 Views. SR dosen't support this, - // so deep copies are necessary in some cases. + // 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 @@ -44,11 +59,9 @@ type PacketBuffer struct { // 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. func (pk PacketBuffer) Clone() PacketBuffer { - return PacketBuffer{ - Data: pk.Data.Clone(nil), - LinkHeader: pk.LinkHeader, - NetworkHeader: pk.NetworkHeader, - TransportHeader: pk.TransportHeader, - } + pk.Data = pk.Data.Clone(nil) + return pk } diff --git a/pkg/tcpip/packet_buffer_state.go b/pkg/tcpip/packet_buffer_state.go index 04c4cf136..ad3cc24fa 100644 --- a/pkg/tcpip/packet_buffer_state.go +++ b/pkg/tcpip/packet_buffer_state.go @@ -20,6 +20,7 @@ import "gvisor.dev/gvisor/pkg/tcpip/buffer" func (pk *PacketBuffer) beforeSave() { // Non-Data fields may be slices of the Data field. This causes // problems for SR, so during save we make each header independent. + pk.Header = pk.Header.DeepCopy() pk.LinkHeader = append(buffer.View(nil), pk.LinkHeader...) pk.NetworkHeader = append(buffer.View(nil), pk.NetworkHeader...) pk.TransportHeader = append(buffer.View(nil), pk.TransportHeader...) diff --git a/pkg/tcpip/ports/BUILD b/pkg/tcpip/ports/BUILD index 11efb4e44..4839f0a65 100644 --- a/pkg/tcpip/ports/BUILD +++ b/pkg/tcpip/ports/BUILD @@ -7,7 +7,7 @@ go_library( name = "ports", srcs = ["ports.go"], importpath = "gvisor.dev/gvisor/pkg/tcpip/ports", - visibility = ["//:sandbox"], + visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", ], diff --git a/pkg/tcpip/seqnum/BUILD b/pkg/tcpip/seqnum/BUILD index 29b7d761c..b31ddba2f 100644 --- a/pkg/tcpip/seqnum/BUILD +++ b/pkg/tcpip/seqnum/BUILD @@ -6,7 +6,5 @@ go_library( name = "seqnum", srcs = ["seqnum.go"], importpath = "gvisor.dev/gvisor/pkg/tcpip/seqnum", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], ) diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD index 460db3cf8..69077669a 100644 --- a/pkg/tcpip/stack/BUILD +++ b/pkg/tcpip/stack/BUILD @@ -31,9 +31,7 @@ go_library( "transport_demuxer.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/stack", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/ilist", "//pkg/rand", @@ -87,11 +85,3 @@ go_test( "//pkg/tcpip", ], ) - -filegroup( - name = "autogen", - srcs = [ - "linkaddrentry_list.go", - ], - visibility = ["//:sandbox"], -) diff --git a/pkg/tcpip/stack/ndp.go b/pkg/tcpip/stack/ndp.go index 8357dca77..cfdd0496e 100644 --- a/pkg/tcpip/stack/ndp.go +++ b/pkg/tcpip/stack/ndp.go @@ -421,7 +421,9 @@ func (ndp *ndpState) doDuplicateAddressDetection(addr tcpip.Address, remaining u pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) sent := r.Stats().ICMP.V6PacketsSent - if err := r.WritePacket(nil, hdr, buffer.VectorisedView{}, NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: DefaultTOS}); err != nil { + if err := r.WritePacket(nil, NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + }); err != nil { sent.Dropped.Increment() return false, err } diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go index 494244368..5b901f947 100644 --- a/pkg/tcpip/stack/ndp_test.go +++ b/pkg/tcpip/stack/ndp_test.go @@ -344,7 +344,7 @@ func TestDADResolve(t *testing.T) { } // Check NDP packet. - checker.IPv6(t, p.Header.ToVectorisedView().First(), + checker.IPv6(t, p.Pkt.Header.View().ToVectorisedView().First(), checker.TTL(header.NDPHopLimit), checker.NDPNS( checker.NDPNSTargetAddress(addr1))) diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 9ed9e1e7c..3f8d7312c 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -812,15 +812,15 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link } else { // n doesn't have a destination endpoint. // Send the packet out of n. - hdr := buffer.NewPrependableFromView(pkt.Data.First()) + pkt.Header = buffer.NewPrependableFromView(pkt.Data.First()) pkt.Data.RemoveFirst() // TODO(b/128629022): use route.WritePacket. - if err := n.linkEP.WritePacket(&r, nil /* gso */, hdr, pkt.Data, protocol); err != nil { + if err := n.linkEP.WritePacket(&r, nil /* gso */, protocol, pkt); err != nil { r.Stats().IP.OutgoingPacketErrors.Increment() } else { n.stats.Tx.Packets.Increment() - n.stats.Tx.Bytes.IncrementBy(uint64(hdr.UsedLength() + pkt.Data.Size())) + n.stats.Tx.Bytes.IncrementBy(uint64(pkt.Header.UsedLength() + pkt.Data.Size())) } } return diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go index c0026f5a3..61fd46d66 100644 --- a/pkg/tcpip/stack/registration.go +++ b/pkg/tcpip/stack/registration.go @@ -232,16 +232,17 @@ type NetworkEndpoint interface { MaxHeaderLength() uint16 // WritePacket writes a packet to the given destination address and - // protocol. - WritePacket(r *Route, gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params NetworkHeaderParams, loop PacketLooping) *tcpip.Error + // protocol. It sets pkt.NetworkHeader. pkt.TransportHeader must have + // already been set. + WritePacket(r *Route, gso *GSO, params NetworkHeaderParams, loop PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error // WritePackets writes packets to the given destination address and - // protocol. - WritePackets(r *Route, gso *GSO, hdrs []PacketDescriptor, payload buffer.VectorisedView, params NetworkHeaderParams, loop PacketLooping) (int, *tcpip.Error) + // protocol. pkts must not be zero length. + WritePackets(r *Route, gso *GSO, pkts []tcpip.PacketBuffer, params NetworkHeaderParams, loop PacketLooping) (int, *tcpip.Error) // WriteHeaderIncludedPacket writes a packet that includes a network // header to the given destination address. - WriteHeaderIncludedPacket(r *Route, payload buffer.VectorisedView, loop PacketLooping) *tcpip.Error + WriteHeaderIncludedPacket(r *Route, loop PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error // ID returns the network protocol endpoint ID. ID() *NetworkEndpointID @@ -361,25 +362,27 @@ type LinkEndpoint interface { // link endpoint. LinkAddress() tcpip.LinkAddress - // WritePacket writes a packet with the given protocol through the given - // route. + // WritePacket writes a packet with the given protocol through the + // given route. It sets pkt.LinkHeader if a link layer header exists. + // pkt.NetworkHeader and pkt.TransportHeader must have already been + // set. // // To participate in transparent bridging, a LinkEndpoint implementation // should call eth.Encode with header.EthernetFields.SrcAddr set to // r.LocalLinkAddress if it is provided. - WritePacket(r *Route, gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error + WritePacket(r *Route, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error // WritePackets writes packets with the given protocol through the - // given route. + // given route. pkts must not be zero length. // // Right now, WritePackets is used only when the software segmentation // offload is enabled. If it will be used for something else, it may // require to change syscall filters. - WritePackets(r *Route, gso *GSO, hdrs []PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) + WritePackets(r *Route, gso *GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) // WriteRawPacket writes a packet directly to the link. The packet // should already have an ethernet header. - WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error + WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error // Attach attaches the data link layer endpoint to the network-layer // dispatcher of the stack. diff --git a/pkg/tcpip/stack/route.go b/pkg/tcpip/stack/route.go index 1a0a51b57..34307ae07 100644 --- a/pkg/tcpip/stack/route.go +++ b/pkg/tcpip/stack/route.go @@ -17,7 +17,6 @@ package stack import ( "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" ) @@ -154,54 +153,36 @@ func (r *Route) IsResolutionRequired() bool { } // WritePacket writes the packet through the given route. -func (r *Route) WritePacket(gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params NetworkHeaderParams) *tcpip.Error { +func (r *Route) WritePacket(gso *GSO, params NetworkHeaderParams, pkt tcpip.PacketBuffer) *tcpip.Error { if !r.ref.isValidForOutgoing() { return tcpip.ErrInvalidEndpointState } - err := r.ref.ep.WritePacket(r, gso, hdr, payload, params, r.Loop) + err := r.ref.ep.WritePacket(r, gso, params, r.Loop, pkt) if err != nil { r.Stats().IP.OutgoingPacketErrors.Increment() } else { r.ref.nic.stats.Tx.Packets.Increment() - r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(hdr.UsedLength() + payload.Size())) + r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(pkt.Header.UsedLength() + pkt.Data.Size())) } return err } -// PacketDescriptor is a packet descriptor which contains a packet header and -// offset and size of packet data in a payload view. -type PacketDescriptor struct { - Hdr buffer.Prependable - Off int - Size int -} - -// NewPacketDescriptors allocates a set of packet descriptors. -func NewPacketDescriptors(n int, hdrSize int) []PacketDescriptor { - buf := make([]byte, n*hdrSize) - hdrs := make([]PacketDescriptor, n) - for i := range hdrs { - hdrs[i].Hdr = buffer.NewEmptyPrependableFromView(buf[i*hdrSize:][:hdrSize]) - } - return hdrs -} - // WritePackets writes the set of packets through the given route. -func (r *Route) WritePackets(gso *GSO, hdrs []PacketDescriptor, payload buffer.VectorisedView, params NetworkHeaderParams) (int, *tcpip.Error) { +func (r *Route) WritePackets(gso *GSO, pkts []tcpip.PacketBuffer, params NetworkHeaderParams) (int, *tcpip.Error) { if !r.ref.isValidForOutgoing() { return 0, tcpip.ErrInvalidEndpointState } - n, err := r.ref.ep.WritePackets(r, gso, hdrs, payload, params, r.Loop) + n, err := r.ref.ep.WritePackets(r, gso, pkts, params, r.Loop) if err != nil { - r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(len(hdrs) - n)) + r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(len(pkts) - n)) } r.ref.nic.stats.Tx.Packets.IncrementBy(uint64(n)) payloadSize := 0 for i := 0; i < n; i++ { - r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(hdrs[i].Hdr.UsedLength())) - payloadSize += hdrs[i].Size + r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(pkts[i].Header.UsedLength())) + payloadSize += pkts[i].DataSize } r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(payloadSize)) return n, err @@ -209,17 +190,17 @@ func (r *Route) WritePackets(gso *GSO, hdrs []PacketDescriptor, payload buffer.V // WriteHeaderIncludedPacket writes a packet already containing a network // header through the given route. -func (r *Route) WriteHeaderIncludedPacket(payload buffer.VectorisedView) *tcpip.Error { +func (r *Route) WriteHeaderIncludedPacket(pkt tcpip.PacketBuffer) *tcpip.Error { if !r.ref.isValidForOutgoing() { return tcpip.ErrInvalidEndpointState } - if err := r.ref.ep.WriteHeaderIncludedPacket(r, payload, r.Loop); err != nil { + if err := r.ref.ep.WriteHeaderIncludedPacket(r, r.Loop, pkt); err != nil { r.Stats().IP.OutgoingPacketErrors.Increment() return err } r.ref.nic.stats.Tx.Packets.Increment() - r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(payload.Size())) + r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(pkt.Data.Size())) return nil } diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 2f8d8e822..0e88643a4 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -1366,10 +1366,10 @@ func (s *Stack) WritePacket(nicID tcpip.NICID, dst tcpip.LinkAddress, netProto t } fakeHeader := make(header.Ethernet, header.EthernetMinimumSize) fakeHeader.Encode(ðFields) - ethHeader := buffer.View(fakeHeader).ToVectorisedView() - ethHeader.Append(payload) + vv := buffer.View(fakeHeader).ToVectorisedView() + vv.Append(payload) - if err := nic.linkEP.WriteRawPacket(ethHeader); err != nil { + if err := nic.linkEP.WriteRawPacket(vv); err != nil { return err } diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index bf1d6974c..8fc034ca1 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -122,39 +122,38 @@ func (f *fakeNetworkEndpoint) Capabilities() stack.LinkEndpointCapabilities { return f.ep.Capabilities() } -func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) *tcpip.Error { +func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, loop stack.PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error { // Increment the sent packet count in the protocol descriptor. f.proto.sendPacketCount[int(r.RemoteAddress[0])%len(f.proto.sendPacketCount)]++ // Add the protocol's header to the packet and send it to the link // endpoint. - b := hdr.Prepend(fakeNetHeaderLen) + b := pkt.Header.Prepend(fakeNetHeaderLen) b[0] = r.RemoteAddress[0] b[1] = f.id.LocalAddress[0] b[2] = byte(params.Protocol) if loop&stack.PacketLoop != 0 { - views := make([]buffer.View, 1, 1+len(payload.Views())) - views[0] = hdr.View() - views = append(views, payload.Views()...) - vv := buffer.NewVectorisedView(len(views[0])+payload.Size(), views) + views := make([]buffer.View, 1, 1+len(pkt.Data.Views())) + views[0] = pkt.Header.View() + views = append(views, pkt.Data.Views()...) f.HandlePacket(r, tcpip.PacketBuffer{ - Data: vv, + Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), }) } if loop&stack.PacketOut == 0 { return nil } - return f.ep.WritePacket(r, gso, hdr, payload, fakeNetNumber) + return f.ep.WritePacket(r, gso, fakeNetNumber, pkt) } // WritePackets implements stack.LinkEndpoint.WritePackets. -func (f *fakeNetworkEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) { +func (f *fakeNetworkEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) { panic("not implemented") } -func (*fakeNetworkEndpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error { +func (*fakeNetworkEndpoint) WriteHeaderIncludedPacket(r *stack.Route, loop stack.PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error { return tcpip.ErrNotSupported } @@ -330,7 +329,10 @@ func sendTo(s *stack.Stack, addr tcpip.Address, payload buffer.View) *tcpip.Erro func send(r stack.Route, payload buffer.View) *tcpip.Error { hdr := buffer.NewPrependable(int(r.MaxHeaderLength())) - return r.WritePacket(nil /* gso */, hdr, payload.ToVectorisedView(), stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}) + return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + Data: payload.ToVectorisedView(), + }) } func testSendTo(t *testing.T, s *stack.Stack, addr tcpip.Address, ep *channel.Endpoint, payload buffer.View) { diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index 2cacea99a..748ce4ea5 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -83,7 +83,10 @@ func (f *fakeTransportEndpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions if err != nil { return 0, nil, err } - if err := f.route.WritePacket(nil /* gso */, hdr, buffer.View(v).ToVectorisedView(), stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}); err != nil { + if err := f.route.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + Data: buffer.View(v).ToVectorisedView(), + }); err != nil { return 0, nil, err } @@ -617,10 +620,10 @@ func TestTransportForwarding(t *testing.T) { t.Fatal("Response packet not forwarded") } - if dst := p.Header[0]; dst != 3 { + if dst := p.Pkt.Header.View()[0]; dst != 3 { t.Errorf("Response packet has incorrect destination addresss: got = %d, want = 3", dst) } - if src := p.Header[1]; src != 1 { + if src := p.Pkt.Header.View()[1]; src != 1 { t.Errorf("Response packet has incorrect source addresss: got = %d, want = 3", src) } } diff --git a/pkg/tcpip/transport/icmp/BUILD b/pkg/tcpip/transport/icmp/BUILD index 9254c3dea..d8c5b5058 100644 --- a/pkg/tcpip/transport/icmp/BUILD +++ b/pkg/tcpip/transport/icmp/BUILD @@ -38,11 +38,3 @@ go_library( "//pkg/waiter", ], ) - -filegroup( - name = "autogen", - srcs = [ - "icmp_packet_list.go", - ], - visibility = ["//:sandbox"], -) diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index 70e008d36..9c40931b5 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -429,7 +429,11 @@ func send4(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Err if ttl == 0 { ttl = r.DefaultTTL() } - return r.WritePacket(nil /* gso */, hdr, data.ToVectorisedView(), stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}) + return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + Data: data.ToVectorisedView(), + TransportHeader: buffer.View(icmpv4), + }) } func send6(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Error { @@ -455,7 +459,11 @@ func send6(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Err if ttl == 0 { ttl = r.DefaultTTL() } - return r.WritePacket(nil /* gso */, hdr, dataVV, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}) + return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + Data: dataVV, + TransportHeader: buffer.View(icmpv6), + }) } func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress, allowMismatch bool) (tcpip.NetworkProtocolNumber, *tcpip.Error) { diff --git a/pkg/tcpip/transport/packet/BUILD b/pkg/tcpip/transport/packet/BUILD index 8ea2e6ee5..44b58ff6b 100644 --- a/pkg/tcpip/transport/packet/BUILD +++ b/pkg/tcpip/transport/packet/BUILD @@ -36,11 +36,3 @@ go_library( "//pkg/waiter", ], ) - -filegroup( - name = "autogen", - srcs = [ - "packet_list.go", - ], - visibility = ["//:sandbox"], -) diff --git a/pkg/tcpip/transport/raw/BUILD b/pkg/tcpip/transport/raw/BUILD index 4af49218c..00991ac8e 100644 --- a/pkg/tcpip/transport/raw/BUILD +++ b/pkg/tcpip/transport/raw/BUILD @@ -38,11 +38,3 @@ go_library( "//pkg/waiter", ], ) - -filegroup( - name = "autogen", - srcs = [ - "raw_packet_list.go", - ], - visibility = ["//:sandbox"], -) diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index 230a1537a..5aafe2615 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -338,13 +338,18 @@ func (e *endpoint) finishWrite(payloadBytes []byte, route *stack.Route) (int64, switch e.NetProto { case header.IPv4ProtocolNumber: if !e.associated { - if err := route.WriteHeaderIncludedPacket(buffer.View(payloadBytes).ToVectorisedView()); err != nil { + if err := route.WriteHeaderIncludedPacket(tcpip.PacketBuffer{ + Data: buffer.View(payloadBytes).ToVectorisedView(), + }); err != nil { return 0, nil, err } break } hdr := buffer.NewPrependable(len(payloadBytes) + int(route.MaxHeaderLength())) - if err := route.WritePacket(nil /* gso */, hdr, buffer.View(payloadBytes).ToVectorisedView(), stack.NetworkHeaderParams{Protocol: e.TransProto, TTL: route.DefaultTTL(), TOS: stack.DefaultTOS}); err != nil { + if err := route.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: e.TransProto, TTL: route.DefaultTTL(), TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + Data: buffer.View(payloadBytes).ToVectorisedView(), + }); err != nil { return 0, nil, err } diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD index 3f47b328d..dd1728f9c 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -61,14 +61,6 @@ go_library( ], ) -filegroup( - name = "autogen", - srcs = [ - "tcp_segment_list.go", - ], - visibility = ["//:sandbox"], -) - go_test( name = "tcp_test", size = "medium", diff --git a/pkg/tcpip/transport/tcp/accept.go b/pkg/tcpip/transport/tcp/accept.go index f24b51b91..f543a6105 100644 --- a/pkg/tcpip/transport/tcp/accept.go +++ b/pkg/tcpip/transport/tcp/accept.go @@ -243,7 +243,7 @@ func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, i n.initGSO() // Register new endpoint so that packets are routed to it. - if err := n.stack.RegisterTransportEndpoint(n.boundNICID, n.effectiveNetProtos, ProtocolNumber, n.ID, n, n.reusePort, n.bindToDevice); err != nil { + if err := n.stack.RegisterTransportEndpoint(n.boundNICID, n.effectiveNetProtos, ProtocolNumber, n.ID, n, n.reusePort, n.boundBindToDevice); err != nil { n.Close() return nil, err } @@ -419,8 +419,8 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { // TODO(b/143300739): Use the userMSS of the listening socket // for accepted sockets. - switch s.flags { - case header.TCPFlagSyn: + switch { + case s.flags == header.TCPFlagSyn: opts := parseSynSegmentOptions(s) if incSynRcvdCount() { // Only handle the syn if the following conditions hold @@ -464,7 +464,7 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { e.stack.Stats().TCP.ListenOverflowSynCookieSent.Increment() } - case header.TCPFlagAck: + case (s.flags & header.TCPFlagAck) != 0: if e.acceptQueueIsFull() { // Silently drop the ack as the application can't accept // the connection at this point. The ack will be @@ -478,6 +478,14 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { } if !synCookiesInUse() { + // When not using SYN cookies, as per RFC 793, section 3.9, page 64: + // Any acknowledgment is bad if it arrives on a connection still in + // the LISTEN state. An acceptable reset segment should be formed + // for any arriving ACK-bearing segment. The RST should be + // formatted as follows: + // + // <SEQ=SEG.ACK><CTL=RST> + // // Send a reset as this is an ACK for which there is no // half open connections and we are not using cookies // yet. diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index a114c06c1..4206db8b6 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -299,6 +299,15 @@ func (h *handshake) synRcvdState(s *segment) *tcpip.Error { return nil } + // RFC 793, Section 3.9, page 69, states that in the SYN-RCVD state, a + // sequence number outside of the window causes an ACK with the proper seq + // number and "After sending the acknowledgment, drop the unacceptable + // segment and return." + if !s.sequenceNumber.InWindow(h.ackNum, h.rcvWnd) { + h.ep.sendRaw(buffer.VectorisedView{}, header.TCPFlagAck, h.iss+1, h.ackNum, h.rcvWnd) + return nil + } + if s.flagIsSet(header.TCPFlagSyn) && s.sequenceNumber != h.ackNum-1 { // We received two SYN segments with different sequence // numbers, so we reset this and restart the whole @@ -631,13 +640,14 @@ func (e *endpoint) sendTCP(r *stack.Route, id stack.TransportEndpointID, data bu return nil } -func buildTCPHdr(r *stack.Route, id stack.TransportEndpointID, d *stack.PacketDescriptor, data buffer.VectorisedView, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) { +func buildTCPHdr(r *stack.Route, id stack.TransportEndpointID, pkt *tcpip.PacketBuffer, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) { optLen := len(opts) - hdr := &d.Hdr - packetSize := d.Size - off := d.Off + hdr := &pkt.Header + packetSize := pkt.DataSize + off := pkt.DataOffset // Initialize the header. tcp := header.TCP(hdr.Prepend(header.TCPMinimumSize + optLen)) + pkt.TransportHeader = buffer.View(tcp) tcp.Encode(&header.TCPFields{ SrcPort: id.LocalPort, DstPort: id.RemotePort, @@ -659,7 +669,7 @@ func buildTCPHdr(r *stack.Route, id stack.TransportEndpointID, d *stack.PacketDe // header and data and get the right sum of the TCP packet. tcp.SetChecksum(xsum) } else if r.Capabilities()&stack.CapabilityTXChecksumOffload == 0 { - xsum = header.ChecksumVVWithOffset(data, xsum, off, packetSize) + xsum = header.ChecksumVVWithOffset(pkt.Data, xsum, off, packetSize) tcp.SetChecksum(^tcp.CalculateChecksum(xsum)) } @@ -674,7 +684,13 @@ func sendTCPBatch(r *stack.Route, id stack.TransportEndpointID, data buffer.Vect mss := int(gso.MSS) n := (data.Size() + mss - 1) / mss - hdrs := stack.NewPacketDescriptors(n, header.TCPMinimumSize+int(r.MaxHeaderLength())+optLen) + // Allocate one big slice for all the headers. + hdrSize := header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen + buf := make([]byte, n*hdrSize) + pkts := make([]tcpip.PacketBuffer, n) + for i := range pkts { + pkts[i].Header = buffer.NewEmptyPrependableFromView(buf[i*hdrSize:][:hdrSize]) + } size := data.Size() off := 0 @@ -684,16 +700,17 @@ func sendTCPBatch(r *stack.Route, id stack.TransportEndpointID, data buffer.Vect packetSize = size } size -= packetSize - hdrs[i].Off = off - hdrs[i].Size = packetSize - buildTCPHdr(r, id, &hdrs[i], data, flags, seq, ack, rcvWnd, opts, gso) + pkts[i].DataOffset = off + pkts[i].DataSize = packetSize + pkts[i].Data = data + buildTCPHdr(r, id, &pkts[i], flags, seq, ack, rcvWnd, opts, gso) off += packetSize seq = seq.Add(seqnum.Size(packetSize)) } if ttl == 0 { ttl = r.DefaultTTL() } - sent, err := r.WritePackets(gso, hdrs, data, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}) + sent, err := r.WritePackets(gso, pkts, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}) if err != nil { r.Stats().TCP.SegmentSendErrors.IncrementBy(uint64(n - sent)) } @@ -713,17 +730,18 @@ func sendTCP(r *stack.Route, id stack.TransportEndpointID, data buffer.Vectorise return sendTCPBatch(r, id, data, ttl, tos, flags, seq, ack, rcvWnd, opts, gso) } - d := &stack.PacketDescriptor{ - Hdr: buffer.NewPrependable(header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen), - Off: 0, - Size: data.Size(), + pkt := tcpip.PacketBuffer{ + Header: buffer.NewPrependable(header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen), + DataOffset: 0, + DataSize: data.Size(), + Data: data, } - buildTCPHdr(r, id, d, data, flags, seq, ack, rcvWnd, opts, gso) + buildTCPHdr(r, id, &pkt, flags, seq, ack, rcvWnd, opts, gso) if ttl == 0 { ttl = r.DefaultTTL() } - if err := r.WritePacket(gso, d.Hdr, data, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}); err != nil { + if err := r.WritePacket(gso, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}, pkt); err != nil { r.Stats().TCP.SegmentSendErrors.Increment() return err } @@ -862,6 +880,33 @@ func (e *endpoint) completeWorkerLocked() { } } +// transitionToStateCloseLocked ensures that the endpoint is +// cleaned up from the transport demuxer, "before" moving to +// StateClose. This will ensure that no packet will be +// delivered to this endpoint from the demuxer when the endpoint +// is transitioned to StateClose. +func (e *endpoint) transitionToStateCloseLocked() { + if e.state == StateClose { + return + } + e.cleanupLocked() + e.state = StateClose +} + +// tryDeliverSegmentFromClosedEndpoint attempts to deliver the parsed +// segment to any other endpoint other than the current one. This is called +// only when the endpoint is in StateClose and we want to deliver the segment +// to any other listening endpoint. We reply with RST if we cannot find one. +func (e *endpoint) tryDeliverSegmentFromClosedEndpoint(s *segment) { + ep := e.stack.FindTransportEndpoint(e.NetProto, e.TransProto, e.ID, &s.route) + if ep == nil { + replyWithReset(s) + s.decRef() + return + } + ep.(*endpoint).enqueueSegment(s) +} + func (e *endpoint) handleReset(s *segment) (ok bool, err *tcpip.Error) { if e.rcv.acceptable(s.sequenceNumber, 0) { // RFC 793, page 37 states that "in all states @@ -891,12 +936,8 @@ func (e *endpoint) handleReset(s *segment) (ok bool, err *tcpip.Error) { // general "connection reset" signal. Enter the CLOSED state, // delete the TCB, and return. case StateCloseWait: - e.state = StateClose + e.transitionToStateCloseLocked() e.HardError = tcpip.ErrAborted - // We need to set this explicitly here because otherwise - // the port registrations will not be released till the - // endpoint is actively closed by the application. - e.workerCleanup = true e.mu.Unlock() return false, nil default: @@ -912,6 +953,20 @@ func (e *endpoint) handleReset(s *segment) (ok bool, err *tcpip.Error) { func (e *endpoint) handleSegments() *tcpip.Error { checkRequeue := true for i := 0; i < maxSegmentsPerWake; i++ { + e.mu.RLock() + state := e.state + e.mu.RUnlock() + if state == StateClose { + // When we get into StateClose while processing from the queue, + // return immediately and let the protocolMainloop handle it. + // + // We can reach StateClose only while processing a previous segment + // or a notification from the protocolMainLoop (caller goroutine). + // This means that with this return, the segment dequeue below can + // never occur on a closed endpoint. + return nil + } + s := e.segmentQueue.dequeue() if s == nil { checkRequeue = false @@ -1157,7 +1212,7 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { // to the TCP_FIN_WAIT2 timeout was hit. Just // mark the socket as closed. e.mu.Lock() - e.state = StateClose + e.transitionToStateCloseLocked() e.mu.Unlock() return nil }, @@ -1229,7 +1284,9 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { return err } } - if e.state != StateError { + if e.state != StateClose && e.state != StateError { + // Only block the worker if the endpoint + // is not in closed state or error state. close(e.drainDone) <-e.undrain } @@ -1316,12 +1373,24 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { if e.state != StateError { e.stack.Stats().TCP.EstablishedResets.Increment() e.stack.Stats().TCP.CurrentEstablished.Decrement() - e.state = StateClose + e.transitionToStateCloseLocked() } // Lock released below. epilogue() + // epilogue removes the endpoint from the transport-demuxer and + // unlocks e.mu. Now that no new segments can get enqueued to this + // endpoint, try to re-match the segment to a different endpoint + // as the current endpoint is closed. + for { + s := e.segmentQueue.dequeue() + if s == nil { + break + } + e.tryDeliverSegmentFromClosedEndpoint(s) + } + // A new SYN was received during TIME_WAIT and we need to abort // the timewait and redirect the segment to the listener queue if reuseTW != nil { diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 04c92c04c..9d4a87e30 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -340,6 +340,9 @@ type endpoint struct { // TCP should never broadcast but Linux nevertheless supports enabling/ // disabling SO_BROADCAST, albeit as a NOOP. broadcast bool + // Values used to reserve a port or register a transport endpoint + // (which ever happens first). + boundBindToDevice tcpip.NICID // effectiveNetProtos contains the network protocols actually in use. In // most cases it will only contain "netProto", but in cases like IPv6 @@ -730,12 +733,13 @@ func (e *endpoint) Close() { // in Listen() when trying to register. if e.state == StateListen && e.isPortReserved { if e.isRegistered { - e.stack.StartTransportEndpointCleanup(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.bindToDevice) + e.stack.StartTransportEndpointCleanup(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.boundBindToDevice) e.isRegistered = false } - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.bindToDevice) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.boundBindToDevice) e.isPortReserved = false + e.boundBindToDevice = 0 } // Mark endpoint as closed. @@ -791,14 +795,15 @@ func (e *endpoint) cleanupLocked() { e.workerCleanup = false if e.isRegistered { - e.stack.StartTransportEndpointCleanup(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.bindToDevice) + e.stack.StartTransportEndpointCleanup(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.boundBindToDevice) e.isRegistered = false } if e.isPortReserved { - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.bindToDevice) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.boundBindToDevice) e.isPortReserved = false } + e.boundBindToDevice = 0 e.route.Release() e.stack.CompleteTransportEndpointCleanup(e) @@ -1741,7 +1746,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tc if e.ID.LocalPort != 0 { // The endpoint is bound to a port, attempt to register it. - err := e.stack.RegisterTransportEndpoint(nicID, netProtos, ProtocolNumber, e.ID, e, e.reusePort, e.bindToDevice) + err := e.stack.RegisterTransportEndpoint(nicID, netProtos, ProtocolNumber, e.ID, e, e.reusePort, e.boundBindToDevice) if err != nil { return err } @@ -1778,7 +1783,10 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tc id.LocalPort = p switch e.stack.RegisterTransportEndpoint(nicID, netProtos, ProtocolNumber, id, e, e.reusePort, e.bindToDevice) { case nil: + // Port picking successful. Save the details of + // the selected port. e.ID = id + e.boundBindToDevice = e.bindToDevice return true, nil case tcpip.ErrPortInUse: return false, nil @@ -1794,7 +1802,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tc // before Connect: in such a case we don't want to hold on to // reservations anymore. if e.isPortReserved { - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, origID.LocalAddress, origID.LocalPort, e.bindToDevice) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, origID.LocalAddress, origID.LocalPort, e.boundBindToDevice) e.isPortReserved = false } @@ -1950,7 +1958,7 @@ func (e *endpoint) listen(backlog int) *tcpip.Error { } // Register the endpoint. - if err := e.stack.RegisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.reusePort, e.bindToDevice); err != nil { + if err := e.stack.RegisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.reusePort, e.boundBindToDevice); err != nil { return err } @@ -2031,6 +2039,7 @@ func (e *endpoint) Bind(addr tcpip.FullAddress) (err *tcpip.Error) { return err } + e.boundBindToDevice = e.bindToDevice e.isPortReserved = true e.effectiveNetProtos = netProtos e.ID.LocalPort = port @@ -2044,8 +2053,9 @@ func (e *endpoint) Bind(addr tcpip.FullAddress) (err *tcpip.Error) { e.ID.LocalPort = 0 e.ID.LocalAddress = "" e.boundNICID = 0 + e.boundBindToDevice = 0 } - }(e.bindToDevice) + }(e.boundBindToDevice) // If an address is specified, we must ensure that it's one of our // local addresses. diff --git a/pkg/tcpip/transport/tcp/rcv.go b/pkg/tcpip/transport/tcp/rcv.go index 068b90fb6..857dc445f 100644 --- a/pkg/tcpip/transport/tcp/rcv.go +++ b/pkg/tcpip/transport/tcp/rcv.go @@ -218,7 +218,7 @@ func (r *receiver) consumeSegment(s *segment, segSeq seqnum.Value, segLen seqnum case StateClosing: r.ep.state = StateTimeWait case StateLastAck: - r.ep.state = StateClose + r.ep.transitionToStateCloseLocked() } r.ep.mu.Unlock() } diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index b443fe9dc..c4b45aa6f 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -454,6 +454,112 @@ func TestConnectResetAfterClose(t *testing.T) { } } +// TestClosingWithEnqueuedSegments tests handling of +// still enqueued segments when the endpoint transitions +// to StateClose. The in-flight segments would be re-enqueued +// to a any listening endpoint. +func TestClosingWithEnqueuedSegments(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) + ep := c.EP + c.EP = nil + + if got, want := tcp.EndpointState(ep.State()), tcp.StateEstablished; got != want { + t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) + } + + // Send a FIN for ESTABLISHED --> CLOSED-WAIT + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + Flags: header.TCPFlagFin | header.TCPFlagAck, + SeqNum: 790, + AckNum: c.IRS.Add(1), + RcvWnd: 30000, + }) + + // Get the ACK for the FIN we sent. + checker.IPv4(t, c.GetPacket(), + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS)+1), + checker.AckNum(791), + checker.TCPFlags(header.TCPFlagAck), + ), + ) + + if got, want := tcp.EndpointState(ep.State()), tcp.StateCloseWait; got != want { + t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) + } + + // Close the application endpoint for CLOSE_WAIT --> LAST_ACK + ep.Close() + + // Get the FIN + checker.IPv4(t, c.GetPacket(), + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS)+1), + checker.AckNum(791), + checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), + ), + ) + + if got, want := tcp.EndpointState(ep.State()), tcp.StateLastAck; got != want { + t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) + } + + // Pause the endpoint`s protocolMainLoop. + ep.(interface{ StopWork() }).StopWork() + + // Enqueue last ACK followed by an ACK matching the endpoint + // + // Send Last ACK for LAST_ACK --> CLOSED + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + Flags: header.TCPFlagAck, + SeqNum: 791, + AckNum: c.IRS.Add(2), + RcvWnd: 30000, + }) + + // Send a packet with ACK set, this would generate RST when + // not using SYN cookies as in this test. + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + Flags: header.TCPFlagAck | header.TCPFlagFin, + SeqNum: 792, + AckNum: c.IRS.Add(2), + RcvWnd: 30000, + }) + + // Unpause endpoint`s protocolMainLoop. + ep.(interface{ ResumeWork() }).ResumeWork() + + // Wait for the protocolMainLoop to resume and update state. + time.Sleep(1 * time.Millisecond) + + // Expect the endpoint to be closed. + if got, want := tcp.EndpointState(ep.State()), tcp.StateClose; got != want { + t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) + } + + // Check if the endpoint was moved to CLOSED and netstack a reset in + // response to the ACK packet that we sent after last-ACK. + checker.IPv4(t, c.GetPacket(), + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS)+2), + checker.AckNum(793), + checker.TCPFlags(header.TCPFlagAck|header.TCPFlagRst), + ), + ) +} + func TestSimpleReceive(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() @@ -686,6 +792,96 @@ func TestSendRstOnListenerRxSynAckV6(t *testing.T) { checker.SeqNum(200))) } +func TestSendRstOnListenerRxAckV4(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.Create(-1 /* epRcvBuf */) + + if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatal("Bind failed:", err) + } + + if err := c.EP.Listen(10 /* backlog */); err != nil { + t.Fatal("Listen failed:", err) + } + + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagFin | header.TCPFlagAck, + SeqNum: 100, + AckNum: 200, + }) + + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck), + checker.SeqNum(200))) +} + +func TestSendRstOnListenerRxAckV6(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.CreateV6Endpoint(true /* v6Only */) + + if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatal("Bind failed:", err) + } + + if err := c.EP.Listen(10 /* backlog */); err != nil { + t.Fatal("Listen failed:", err) + } + + c.SendV6Packet(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagFin | header.TCPFlagAck, + SeqNum: 100, + AckNum: 200, + }) + + checker.IPv6(t, c.GetV6Packet(), checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck), + checker.SeqNum(200))) +} + +// TestListenShutdown tests for the listening endpoint not processing +// any receive when it is on read shutdown. +func TestListenShutdown(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.Create(-1 /* epRcvBuf */) + + if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatal("Bind failed:", err) + } + + if err := c.EP.Listen(10 /* backlog */); err != nil { + t.Fatal("Listen failed:", err) + } + + if err := c.EP.Shutdown(tcpip.ShutdownRead); err != nil { + t.Fatal("Shutdown failed:", err) + } + + // Wait for the endpoint state to be propagated. + time.Sleep(10 * time.Millisecond) + + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: 100, + AckNum: 200, + }) + + c.CheckNoPacket("Packet received when listening socket was shutdown") +} + func TestTOSV4(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() @@ -4477,7 +4673,7 @@ func TestListenSynRcvdQueueFull(t *testing.T) { SrcPort: context.TestPort, DstPort: context.StackPort, Flags: header.TCPFlagSyn, - SeqNum: seqnum.Value(789), + SeqNum: irs, RcvWnd: 30000, }) @@ -4629,6 +4825,125 @@ func TestListenBacklogFullSynCookieInUse(t *testing.T) { } } +func TestSynRcvdBadSeqNumber(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + // Create TCP endpoint. + var err *tcpip.Error + c.EP, err = c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &c.WQ) + if err != nil { + t.Fatalf("NewEndpoint failed: %s", err) + } + + // Bind to wildcard. + if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatalf("Bind failed: %s", err) + } + + // Start listening. + if err := c.EP.Listen(10); err != nil { + t.Fatalf("Listen failed: %s", err) + } + + // Send a SYN to get a SYN-ACK. This should put the ep into SYN-RCVD state + irs := seqnum.Value(789) + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: irs, + RcvWnd: 30000, + }) + + // Receive the SYN-ACK reply. + b := c.GetPacket() + tcpHdr := header.TCP(header.IPv4(b).Payload()) + iss := seqnum.Value(tcpHdr.SequenceNumber()) + tcpCheckers := []checker.TransportChecker{ + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagAck | header.TCPFlagSyn), + checker.AckNum(uint32(irs) + 1), + } + checker.IPv4(t, b, checker.TCP(tcpCheckers...)) + + // Now send a packet with an out-of-window sequence number + largeSeqnum := irs + seqnum.Value(tcpHdr.WindowSize()) + 1 + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: largeSeqnum, + AckNum: iss + 1, + RcvWnd: 30000, + }) + + // Should receive an ACK with the expected SEQ number + b = c.GetPacket() + tcpCheckers = []checker.TransportChecker{ + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagAck), + checker.AckNum(uint32(irs) + 1), + checker.SeqNum(uint32(iss + 1)), + } + checker.IPv4(t, b, checker.TCP(tcpCheckers...)) + + // Now that the socket replied appropriately with the ACK, + // complete the connection to test that the large SEQ num + // did not change the state from SYN-RCVD. + + // Send ACK to move to ESTABLISHED state. + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: irs + 1, + AckNum: iss + 1, + RcvWnd: 30000, + }) + + newEP, _, err := c.EP.Accept() + + if err != nil && err != tcpip.ErrWouldBlock { + t.Fatalf("Accept failed: %s", err) + } + + if err == tcpip.ErrWouldBlock { + // Try to accept the connections in the backlog. + we, ch := waiter.NewChannelEntry(nil) + c.WQ.EventRegister(&we, waiter.EventIn) + defer c.WQ.EventUnregister(&we) + + // Wait for connection to be established. + select { + case <-ch: + newEP, _, err = c.EP.Accept() + if err != nil { + t.Fatalf("Accept failed: %s", err) + } + + case <-time.After(1 * time.Second): + t.Fatalf("Timed out waiting for accept") + } + } + + // Now verify that the TCP socket is usable and in a connected state. + data := "Don't panic" + _, _, err = newEP.Write(tcpip.SlicePayload(buffer.NewViewFromBytes([]byte(data))), tcpip.WriteOptions{}) + + if err != nil { + t.Fatalf("Write failed: %s", err) + } + + pkt := c.GetPacket() + tcpHdr = header.TCP(header.IPv4(pkt).Payload()) + if string(tcpHdr.Payload()) != data { + t.Fatalf("Unexpected data: got %s, want %s", string(tcpHdr.Payload()), data) + } +} + func TestPassiveConnectionAttemptIncrement(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() diff --git a/pkg/tcpip/transport/tcp/testing/context/BUILD b/pkg/tcpip/transport/tcp/testing/context/BUILD index 19b0d31c5..b33ec2087 100644 --- a/pkg/tcpip/transport/tcp/testing/context/BUILD +++ b/pkg/tcpip/transport/tcp/testing/context/BUILD @@ -8,7 +8,7 @@ go_library( srcs = ["context.go"], importpath = "gvisor.dev/gvisor/pkg/tcpip/transport/tcp/testing/context", visibility = [ - "//:sandbox", + "//visibility:public", ], deps = [ "//pkg/tcpip", diff --git a/pkg/tcpip/transport/tcp/testing/context/context.go b/pkg/tcpip/transport/tcp/testing/context/context.go index 0a733fa94..04fdaaed1 100644 --- a/pkg/tcpip/transport/tcp/testing/context/context.go +++ b/pkg/tcpip/transport/tcp/testing/context/context.go @@ -236,9 +236,9 @@ func (c *Context) GetPacket() []byte { if p.Proto != ipv4.ProtocolNumber { c.t.Fatalf("Bad network protocol: got %v, wanted %v", p.Proto, ipv4.ProtocolNumber) } - b := make([]byte, len(p.Header)+len(p.Payload)) - copy(b, p.Header) - copy(b[len(p.Header):], p.Payload) + + hdr := p.Pkt.Header.View() + b := append(hdr[:len(hdr):len(hdr)], p.Pkt.Data.ToView()...) if p.GSO != nil && p.GSO.L3HdrLen != header.IPv4MinimumSize { c.t.Errorf("L3HdrLen %v (expected %v)", p.GSO.L3HdrLen, header.IPv4MinimumSize) @@ -264,9 +264,9 @@ func (c *Context) GetPacketNonBlocking() []byte { if p.Proto != ipv4.ProtocolNumber { c.t.Fatalf("Bad network protocol: got %v, wanted %v", p.Proto, ipv4.ProtocolNumber) } - b := make([]byte, len(p.Header)+len(p.Payload)) - copy(b, p.Header) - copy(b[len(p.Header):], p.Payload) + + hdr := p.Pkt.Header.View() + b := append(hdr[:len(hdr):len(hdr)], p.Pkt.Data.ToView()...) checker.IPv4(c.t, b, checker.SrcAddr(StackAddr), checker.DstAddr(TestAddr)) return b @@ -488,9 +488,9 @@ func (c *Context) GetV6Packet() []byte { if p.Proto != ipv6.ProtocolNumber { c.t.Fatalf("Bad network protocol: got %v, wanted %v", p.Proto, ipv6.ProtocolNumber) } - b := make([]byte, len(p.Header)+len(p.Payload)) - copy(b, p.Header) - copy(b[len(p.Header):], p.Payload) + b := make([]byte, p.Pkt.Header.UsedLength()+p.Pkt.Data.Size()) + copy(b, p.Pkt.Header.View()) + copy(b[p.Pkt.Header.UsedLength():], p.Pkt.Data.ToView()) checker.IPv6(c.t, b, checker.SrcAddr(StackV6Addr), checker.DstAddr(TestV6Addr)) return b diff --git a/pkg/tcpip/transport/udp/BUILD b/pkg/tcpip/transport/udp/BUILD index c9460aa0d..8d4c3808f 100644 --- a/pkg/tcpip/transport/udp/BUILD +++ b/pkg/tcpip/transport/udp/BUILD @@ -59,11 +59,3 @@ go_test( "//pkg/waiter", ], ) - -filegroup( - name = "autogen", - srcs = [ - "udp_packet_list.go", - ], - visibility = ["//:sandbox"], -) diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index dda7af910..24cb88c13 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -104,6 +104,10 @@ type endpoint struct { bindToDevice tcpip.NICID broadcast bool + // Values used to reserve a port or register a transport endpoint. + // (which ever happens first). + boundBindToDevice tcpip.NICID + // sendTOS represents IPv4 TOS or IPv6 TrafficClass, // applied while sending packets. Defaults to 0 as on Linux. sendTOS uint8 @@ -175,8 +179,9 @@ func (e *endpoint) Close() { switch e.state { case StateBound, StateConnected: - e.stack.UnregisterTransportEndpoint(e.RegisterNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.bindToDevice) - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.bindToDevice) + e.stack.UnregisterTransportEndpoint(e.RegisterNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.boundBindToDevice) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.boundBindToDevice) + e.boundBindToDevice = 0 } for _, mem := range e.multicastMemberships { @@ -817,7 +822,11 @@ func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort u if useDefaultTTL { ttl = r.DefaultTTL() } - if err := r.WritePacket(nil /* gso */, hdr, data, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}); err != nil { + if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}, tcpip.PacketBuffer{ + Header: hdr, + Data: data, + TransportHeader: buffer.View(udp), + }); err != nil { r.Stats().UDP.PacketSendErrors.Increment() return err } @@ -867,7 +876,10 @@ func (e *endpoint) Disconnect() *tcpip.Error { if e.state != StateConnected { return nil } - id := stack.TransportEndpointID{} + var ( + id stack.TransportEndpointID + btd tcpip.NICID + ) // Exclude ephemerally bound endpoints. if e.BindNICID != 0 || e.ID.LocalAddress == "" { var err *tcpip.Error @@ -875,7 +887,7 @@ func (e *endpoint) Disconnect() *tcpip.Error { LocalPort: e.ID.LocalPort, LocalAddress: e.ID.LocalAddress, } - id, err = e.registerWithStack(e.RegisterNICID, e.effectiveNetProtos, id) + id, btd, err = e.registerWithStack(e.RegisterNICID, e.effectiveNetProtos, id) if err != nil { return err } @@ -883,13 +895,14 @@ func (e *endpoint) Disconnect() *tcpip.Error { } else { if e.ID.LocalPort != 0 { // Release the ephemeral port. - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.bindToDevice) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.boundBindToDevice) } e.state = StateInitial } - e.stack.UnregisterTransportEndpoint(e.RegisterNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.bindToDevice) + e.stack.UnregisterTransportEndpoint(e.RegisterNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.boundBindToDevice) e.ID = id + e.boundBindToDevice = btd e.route.Release() e.route = stack.Route{} e.dstPort = 0 @@ -958,17 +971,18 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { } } - id, err = e.registerWithStack(nicID, netProtos, id) + id, btd, err := e.registerWithStack(nicID, netProtos, id) if err != nil { return err } // Remove the old registration. if e.ID.LocalPort != 0 { - e.stack.UnregisterTransportEndpoint(e.RegisterNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.bindToDevice) + e.stack.UnregisterTransportEndpoint(e.RegisterNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.boundBindToDevice) } e.ID = id + e.boundBindToDevice = btd e.route = r.Clone() e.dstPort = addr.Port e.RegisterNICID = nicID @@ -1026,11 +1040,11 @@ func (*endpoint) Accept() (tcpip.Endpoint, *waiter.Queue, *tcpip.Error) { return nil, nil, tcpip.ErrNotSupported } -func (e *endpoint) registerWithStack(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, id stack.TransportEndpointID) (stack.TransportEndpointID, *tcpip.Error) { +func (e *endpoint) registerWithStack(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, id stack.TransportEndpointID) (stack.TransportEndpointID, tcpip.NICID, *tcpip.Error) { if e.ID.LocalPort == 0 { port, err := e.stack.ReservePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, e.reusePort, e.bindToDevice) if err != nil { - return id, err + return id, e.bindToDevice, err } id.LocalPort = port } @@ -1039,7 +1053,7 @@ func (e *endpoint) registerWithStack(nicID tcpip.NICID, netProtos []tcpip.Networ if err != nil { e.stack.ReleasePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, e.bindToDevice) } - return id, err + return id, e.bindToDevice, err } func (e *endpoint) bindLocked(addr tcpip.FullAddress) *tcpip.Error { @@ -1078,12 +1092,13 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) *tcpip.Error { LocalPort: addr.Port, LocalAddress: addr.Addr, } - id, err = e.registerWithStack(nicID, netProtos, id) + id, btd, err := e.registerWithStack(nicID, netProtos, id) if err != nil { return err } e.ID = id + e.boundBindToDevice = btd e.RegisterNICID = nicID e.effectiveNetProtos = netProtos diff --git a/pkg/tcpip/transport/udp/endpoint_state.go b/pkg/tcpip/transport/udp/endpoint_state.go index b227e353b..43fb047ed 100644 --- a/pkg/tcpip/transport/udp/endpoint_state.go +++ b/pkg/tcpip/transport/udp/endpoint_state.go @@ -109,7 +109,7 @@ func (e *endpoint) Resume(s *stack.Stack) { // pass it to the reservation machinery. id := e.ID e.ID.LocalPort = 0 - e.ID, err = e.registerWithStack(e.RegisterNICID, e.effectiveNetProtos, id) + e.ID, e.boundBindToDevice, err = e.registerWithStack(e.RegisterNICID, e.effectiveNetProtos, id) if err != nil { panic(err) } diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go index 43f11b700..259c3072a 100644 --- a/pkg/tcpip/transport/udp/protocol.go +++ b/pkg/tcpip/transport/udp/protocol.go @@ -135,7 +135,10 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans pkt.SetType(header.ICMPv4DstUnreachable) pkt.SetCode(header.ICMPv4PortUnreachable) pkt.SetChecksum(header.ICMPv4Checksum(pkt, payload)) - r.WritePacket(nil /* gso */, hdr, payload, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}) + r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + Data: payload, + }) case header.IPv6AddressSize: if !r.Stack().AllowICMPMessage() { @@ -169,7 +172,10 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans pkt.SetType(header.ICMPv6DstUnreachable) pkt.SetCode(header.ICMPv6PortUnreachable) pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, payload)) - r.WritePacket(nil /* gso */, hdr, payload, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}) + r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + Data: payload, + }) } return true } diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index 30ee9801b..7051a7a9c 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -356,9 +356,9 @@ func (c *testContext) getPacketAndVerify(flow testFlow, checkers ...checker.Netw if p.Proto != flow.netProto() { c.t.Fatalf("Bad network protocol: got %v, wanted %v", p.Proto, flow.netProto()) } - b := make([]byte, len(p.Header)+len(p.Payload)) - copy(b, p.Header) - copy(b[len(p.Header):], p.Payload) + + hdr := p.Pkt.Header.View() + b := append(hdr[:len(hdr):len(hdr)], p.Pkt.Data.ToView()...) h := flow.header4Tuple(outgoing) checkers := append( @@ -1453,8 +1453,8 @@ func TestV4UnknownDestination(t *testing.T) { select { case p := <-c.linkEP.C: var pkt []byte - pkt = append(pkt, p.Header...) - pkt = append(pkt, p.Payload...) + pkt = append(pkt, p.Pkt.Header.View()...) + pkt = append(pkt, p.Pkt.Data.ToView()...) if got, want := len(pkt), header.IPv4MinimumProcessableDatagramSize; got > want { t.Fatalf("got an ICMP packet of size: %d, want: sz <= %d", got, want) } @@ -1527,8 +1527,8 @@ func TestV6UnknownDestination(t *testing.T) { select { case p := <-c.linkEP.C: var pkt []byte - pkt = append(pkt, p.Header...) - pkt = append(pkt, p.Payload...) + pkt = append(pkt, p.Pkt.Header.View()...) + pkt = append(pkt, p.Pkt.Data.ToView()...) if got, want := len(pkt), header.IPv6MinimumMTU; got > want { t.Fatalf("got an ICMP packet of size: %d, want: sz <= %d", got, want) } |