diff options
Diffstat (limited to 'pkg/tcpip/link/fdbased')
-rw-r--r-- | pkg/tcpip/link/fdbased/BUILD | 1 | ||||
-rw-r--r-- | pkg/tcpip/link/fdbased/endpoint.go | 77 | ||||
-rw-r--r-- | pkg/tcpip/link/fdbased/endpoint_test.go | 73 | ||||
-rw-r--r-- | pkg/tcpip/link/fdbased/packet_dispatchers.go | 2 |
4 files changed, 104 insertions, 49 deletions
diff --git a/pkg/tcpip/link/fdbased/BUILD b/pkg/tcpip/link/fdbased/BUILD index aa6db9aea..507b44abc 100644 --- a/pkg/tcpip/link/fdbased/BUILD +++ b/pkg/tcpip/link/fdbased/BUILD @@ -15,6 +15,7 @@ go_library( visibility = ["//visibility:public"], deps = [ "//pkg/binary", + "//pkg/iovec", "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go index f34082e1a..6aa1badc7 100644 --- a/pkg/tcpip/link/fdbased/endpoint.go +++ b/pkg/tcpip/link/fdbased/endpoint.go @@ -45,6 +45,7 @@ import ( "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/iovec" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" @@ -406,6 +407,8 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne eth.Encode(ethHdr) } + var builder iovec.Builder + fd := e.fds[pkt.Hash%uint32(len(e.fds))] if e.Capabilities()&stack.CapabilityHardwareGSO != 0 { vnetHdr := virtioNetHdr{} @@ -430,29 +433,25 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne } vnetHdrBuf := binary.Marshal(make([]byte, 0, virtioNetHdrSize), binary.LittleEndian, vnetHdr) - return rawfile.NonBlockingWrite3(fd, vnetHdrBuf, pkt.Header.View(), pkt.Data.ToView()) + builder.Add(vnetHdrBuf) } - if pkt.Data.Size() == 0 { - return rawfile.NonBlockingWrite(fd, pkt.Header.View()) - } - if pkt.Header.UsedLength() == 0 { - return rawfile.NonBlockingWrite(fd, pkt.Data.ToView()) + builder.Add(pkt.Header.View()) + for _, v := range pkt.Data.Views() { + builder.Add(v) } - return rawfile.NonBlockingWrite3(fd, pkt.Header.View(), pkt.Data.ToView(), nil) + return rawfile.NonBlockingWriteIovec(fd, builder.Build()) } func (e *endpoint) sendBatch(batchFD int, batch []*stack.PacketBuffer) (int, *tcpip.Error) { // Send a batch of packets through batchFD. mmsgHdrs := make([]rawfile.MMsgHdr, 0, len(batch)) for _, pkt := range batch { - var ethHdrBuf []byte - iovLen := 0 + var eth header.Ethernet if e.hdrSize > 0 { // Add ethernet header if needed. - ethHdrBuf = make([]byte, header.EthernetMinimumSize) - eth := header.Ethernet(ethHdrBuf) + eth = make(header.Ethernet, header.EthernetMinimumSize) ethHdr := &header.EthernetFields{ DstAddr: pkt.EgressRoute.RemoteLinkAddress, Type: pkt.NetworkProtocolNumber, @@ -465,12 +464,11 @@ func (e *endpoint) sendBatch(batchFD int, batch []*stack.PacketBuffer) (int, *tc ethHdr.SrcAddr = e.addr } eth.Encode(ethHdr) - iovLen++ } - vnetHdr := virtioNetHdr{} var vnetHdrBuf []byte if e.Capabilities()&stack.CapabilityHardwareGSO != 0 { + vnetHdr := virtioNetHdr{} if pkt.GSOOptions != nil { vnetHdr.hdrLen = uint16(pkt.Header.UsedLength()) if pkt.GSOOptions.NeedsCsum { @@ -491,45 +489,20 @@ func (e *endpoint) sendBatch(batchFD int, batch []*stack.PacketBuffer) (int, *tc } } vnetHdrBuf = binary.Marshal(make([]byte, 0, virtioNetHdrSize), binary.LittleEndian, vnetHdr) - iovLen++ } - iovecs := make([]syscall.Iovec, iovLen+1+len(pkt.Data.Views())) + var builder iovec.Builder + builder.Add(vnetHdrBuf) + builder.Add(eth) + builder.Add(pkt.Header.View()) + for _, v := range pkt.Data.Views() { + builder.Add(v) + } + iovecs := builder.Build() + var mmsgHdr rawfile.MMsgHdr mmsgHdr.Msg.Iov = &iovecs[0] - iovecIdx := 0 - if vnetHdrBuf != nil { - v := &iovecs[iovecIdx] - v.Base = &vnetHdrBuf[0] - v.Len = uint64(len(vnetHdrBuf)) - iovecIdx++ - } - if ethHdrBuf != nil { - v := &iovecs[iovecIdx] - v.Base = ðHdrBuf[0] - v.Len = uint64(len(ethHdrBuf)) - iovecIdx++ - } - pktSize := uint64(0) - // Encode L3 Header - v := &iovecs[iovecIdx] - hdr := &pkt.Header - hdrView := hdr.View() - v.Base = &hdrView[0] - v.Len = uint64(len(hdrView)) - pktSize += v.Len - iovecIdx++ - - // Now encode the Transport Payload. - pktViews := pkt.Data.Views() - for i := range pktViews { - vec := &iovecs[iovecIdx] - iovecIdx++ - vec.Base = &pktViews[i][0] - vec.Len = uint64(len(pktViews[i])) - pktSize += vec.Len - } - mmsgHdr.Msg.Iovlen = uint64(iovecIdx) + mmsgHdr.Msg.Iovlen = uint64(len(iovecs)) mmsgHdrs = append(mmsgHdrs, mmsgHdr) } @@ -626,6 +599,14 @@ func (e *endpoint) GSOMaxSize() uint32 { return e.gsoMaxSize } +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. +func (e *endpoint) ARPHardwareType() header.ARPHardwareType { + if e.hdrSize > 0 { + return header.ARPHardwareEther + } + return header.ARPHardwareNone +} + // InjectableEndpoint is an injectable fd-based endpoint. The endpoint writes // to the FD, but does not read from it. All reads come from injected packets. type InjectableEndpoint struct { diff --git a/pkg/tcpip/link/fdbased/endpoint_test.go b/pkg/tcpip/link/fdbased/endpoint_test.go index eaee7e5d7..4bad930c7 100644 --- a/pkg/tcpip/link/fdbased/endpoint_test.go +++ b/pkg/tcpip/link/fdbased/endpoint_test.go @@ -500,3 +500,76 @@ func TestRecvMMsgDispatcherCapLength(t *testing.T) { } } + +// fakeNetworkDispatcher delivers packets to pkts. +type fakeNetworkDispatcher struct { + pkts []*stack.PacketBuffer +} + +func (d *fakeNetworkDispatcher) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + d.pkts = append(d.pkts, pkt) +} + +func TestDispatchPacketFormat(t *testing.T) { + for _, test := range []struct { + name string + newDispatcher func(fd int, e *endpoint) (linkDispatcher, error) + }{ + { + name: "readVDispatcher", + newDispatcher: newReadVDispatcher, + }, + { + name: "recvMMsgDispatcher", + newDispatcher: newRecvMMsgDispatcher, + }, + } { + t.Run(test.name, func(t *testing.T) { + // Create a socket pair to send/recv. + fds, err := syscall.Socketpair(syscall.AF_UNIX, syscall.SOCK_DGRAM, 0) + if err != nil { + t.Fatal(err) + } + defer syscall.Close(fds[0]) + defer syscall.Close(fds[1]) + + data := []byte{ + // Ethernet header. + 1, 2, 3, 4, 5, 60, + 1, 2, 3, 4, 5, 61, + 8, 0, + // Mock network header. + 40, 41, 42, 43, + } + err = syscall.Sendmsg(fds[1], data, nil, nil, 0) + if err != nil { + t.Fatal(err) + } + + // Create and run dispatcher once. + sink := &fakeNetworkDispatcher{} + d, err := test.newDispatcher(fds[0], &endpoint{ + hdrSize: header.EthernetMinimumSize, + dispatcher: sink, + }) + if err != nil { + t.Fatal(err) + } + if ok, err := d.dispatch(); !ok || err != nil { + t.Fatalf("d.dispatch() = %v, %v", ok, err) + } + + // Verify packet. + if got, want := len(sink.pkts), 1; got != want { + t.Fatalf("len(sink.pkts) = %d, want %d", got, want) + } + pkt := sink.pkts[0] + if got, want := len(pkt.LinkHeader), header.EthernetMinimumSize; got != want { + t.Errorf("len(pkt.LinkHeader) = %d, want %d", got, want) + } + if got, want := pkt.Data.Size(), 4; got != want { + t.Errorf("pkt.Data.Size() = %d, want %d", got, want) + } + }) + } +} diff --git a/pkg/tcpip/link/fdbased/packet_dispatchers.go b/pkg/tcpip/link/fdbased/packet_dispatchers.go index f04738cfb..d8f2504b3 100644 --- a/pkg/tcpip/link/fdbased/packet_dispatchers.go +++ b/pkg/tcpip/link/fdbased/packet_dispatchers.go @@ -278,7 +278,7 @@ func (d *recvMMsgDispatcher) dispatch() (bool, *tcpip.Error) { eth header.Ethernet ) if d.e.hdrSize > 0 { - eth = header.Ethernet(d.views[k][0]) + eth = header.Ethernet(d.views[k][0][:header.EthernetMinimumSize]) p = eth.Type() remote = eth.SourceAddress() local = eth.DestinationAddress() |