diff options
Diffstat (limited to 'pkg/tcpip')
101 files changed, 11561 insertions, 1564 deletions
diff --git a/pkg/tcpip/checker/checker.go b/pkg/tcpip/checker/checker.go index c1745ba6a..ee264b726 100644 --- a/pkg/tcpip/checker/checker.go +++ b/pkg/tcpip/checker/checker.go @@ -320,6 +320,22 @@ func DstPort(port uint16) TransportChecker { } } +// NoChecksum creates a checker that checks if the checksum is zero. +func NoChecksum(noChecksum bool) TransportChecker { + return func(t *testing.T, h header.Transport) { + t.Helper() + + udp, ok := h.(header.UDP) + if !ok { + return + } + + if b := udp.Checksum() == 0; b != noChecksum { + t.Errorf("bad checksum state, got %t, want %t", b, noChecksum) + } + } +} + // SeqNum creates a checker that checks the sequence number. func SeqNum(seq uint32) TransportChecker { return func(t *testing.T, h header.Transport) { diff --git a/pkg/tcpip/header/BUILD b/pkg/tcpip/header/BUILD index 0cde694dc..d87797617 100644 --- a/pkg/tcpip/header/BUILD +++ b/pkg/tcpip/header/BUILD @@ -48,7 +48,7 @@ go_test( "//pkg/rand", "//pkg/tcpip", "//pkg/tcpip/buffer", - "@com_github_google_go-cmp//cmp:go_default_library", + "@com_github_google_go_cmp//cmp:go_default_library", ], ) @@ -64,6 +64,6 @@ go_test( deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", - "@com_github_google_go-cmp//cmp:go_default_library", + "@com_github_google_go_cmp//cmp:go_default_library", ], ) diff --git a/pkg/tcpip/header/arp.go b/pkg/tcpip/header/arp.go index 718a4720a..83189676e 100644 --- a/pkg/tcpip/header/arp.go +++ b/pkg/tcpip/header/arp.go @@ -14,14 +14,33 @@ package header -import "gvisor.dev/gvisor/pkg/tcpip" +import ( + "encoding/binary" + + "gvisor.dev/gvisor/pkg/tcpip" +) const ( // ARPProtocolNumber is the ARP network protocol number. ARPProtocolNumber tcpip.NetworkProtocolNumber = 0x0806 // ARPSize is the size of an IPv4-over-Ethernet ARP packet. - ARPSize = 2 + 2 + 1 + 1 + 2 + 2*6 + 2*4 + ARPSize = 28 +) + +// ARPHardwareType is the hardware type for LinkEndpoint in an ARP header. +type ARPHardwareType uint16 + +// Typical ARP HardwareType values. Some of the constants have to be specific +// values as they are egressed on the wire in the HTYPE field of an ARP header. +const ( + ARPHardwareNone ARPHardwareType = 0 + // ARPHardwareEther specifically is the HTYPE for Ethernet as specified + // in the IANA list here: + // + // https://www.iana.org/assignments/arp-parameters/arp-parameters.xhtml#arp-parameters-2 + ARPHardwareEther ARPHardwareType = 1 + ARPHardwareLoopback ARPHardwareType = 2 ) // ARPOp is an ARP opcode. @@ -36,54 +55,64 @@ const ( // ARP is an ARP packet stored in a byte array as described in RFC 826. type ARP []byte -func (a ARP) hardwareAddressSpace() uint16 { return uint16(a[0])<<8 | uint16(a[1]) } -func (a ARP) protocolAddressSpace() uint16 { return uint16(a[2])<<8 | uint16(a[3]) } -func (a ARP) hardwareAddressSize() int { return int(a[4]) } -func (a ARP) protocolAddressSize() int { return int(a[5]) } +const ( + hTypeOffset = 0 + protocolOffset = 2 + haAddressSizeOffset = 4 + protoAddressSizeOffset = 5 + opCodeOffset = 6 + senderHAAddressOffset = 8 + senderProtocolAddressOffset = senderHAAddressOffset + EthernetAddressSize + targetHAAddressOffset = senderProtocolAddressOffset + IPv4AddressSize + targetProtocolAddressOffset = targetHAAddressOffset + EthernetAddressSize +) + +func (a ARP) hardwareAddressType() ARPHardwareType { + return ARPHardwareType(binary.BigEndian.Uint16(a[hTypeOffset:])) +} + +func (a ARP) protocolAddressSpace() uint16 { return binary.BigEndian.Uint16(a[protocolOffset:]) } +func (a ARP) hardwareAddressSize() int { return int(a[haAddressSizeOffset]) } +func (a ARP) protocolAddressSize() int { return int(a[protoAddressSizeOffset]) } // Op is the ARP opcode. -func (a ARP) Op() ARPOp { return ARPOp(a[6])<<8 | ARPOp(a[7]) } +func (a ARP) Op() ARPOp { return ARPOp(binary.BigEndian.Uint16(a[opCodeOffset:])) } // SetOp sets the ARP opcode. func (a ARP) SetOp(op ARPOp) { - a[6] = uint8(op >> 8) - a[7] = uint8(op) + binary.BigEndian.PutUint16(a[opCodeOffset:], uint16(op)) } // SetIPv4OverEthernet configures the ARP packet for IPv4-over-Ethernet. func (a ARP) SetIPv4OverEthernet() { - a[0], a[1] = 0, 1 // htypeEthernet - a[2], a[3] = 0x08, 0x00 // IPv4ProtocolNumber - a[4] = 6 // macSize - a[5] = uint8(IPv4AddressSize) + binary.BigEndian.PutUint16(a[hTypeOffset:], uint16(ARPHardwareEther)) + binary.BigEndian.PutUint16(a[protocolOffset:], uint16(IPv4ProtocolNumber)) + a[haAddressSizeOffset] = EthernetAddressSize + a[protoAddressSizeOffset] = uint8(IPv4AddressSize) } // HardwareAddressSender is the link address of the sender. // It is a view on to the ARP packet so it can be used to set the value. func (a ARP) HardwareAddressSender() []byte { - const s = 8 - return a[s : s+6] + return a[senderHAAddressOffset : senderHAAddressOffset+EthernetAddressSize] } // ProtocolAddressSender is the protocol address of the sender. // It is a view on to the ARP packet so it can be used to set the value. func (a ARP) ProtocolAddressSender() []byte { - const s = 8 + 6 - return a[s : s+4] + return a[senderProtocolAddressOffset : senderProtocolAddressOffset+IPv4AddressSize] } // HardwareAddressTarget is the link address of the target. // It is a view on to the ARP packet so it can be used to set the value. func (a ARP) HardwareAddressTarget() []byte { - const s = 8 + 6 + 4 - return a[s : s+6] + return a[targetHAAddressOffset : targetHAAddressOffset+EthernetAddressSize] } // ProtocolAddressTarget is the protocol address of the target. // It is a view on to the ARP packet so it can be used to set the value. func (a ARP) ProtocolAddressTarget() []byte { - const s = 8 + 6 + 4 + 6 - return a[s : s+4] + return a[targetProtocolAddressOffset : targetProtocolAddressOffset+IPv4AddressSize] } // IsValid reports whether this is an ARP packet for IPv4 over Ethernet. @@ -91,10 +120,8 @@ func (a ARP) IsValid() bool { if len(a) < ARPSize { return false } - const htypeEthernet = 1 - const macSize = 6 - return a.hardwareAddressSpace() == htypeEthernet && + return a.hardwareAddressType() == ARPHardwareEther && a.protocolAddressSpace() == uint16(IPv4ProtocolNumber) && - a.hardwareAddressSize() == macSize && + a.hardwareAddressSize() == EthernetAddressSize && a.protocolAddressSize() == IPv4AddressSize } diff --git a/pkg/tcpip/header/eth.go b/pkg/tcpip/header/eth.go index b1e92d2d7..eaface8cb 100644 --- a/pkg/tcpip/header/eth.go +++ b/pkg/tcpip/header/eth.go @@ -53,6 +53,10 @@ const ( // (all bits set to 0). unspecifiedEthernetAddress = tcpip.LinkAddress("\x00\x00\x00\x00\x00\x00") + // EthernetBroadcastAddress is an ethernet address that addresses every node + // on a local link. + EthernetBroadcastAddress = tcpip.LinkAddress("\xff\xff\xff\xff\xff\xff") + // unicastMulticastFlagMask is the mask of the least significant bit in // the first octet (in network byte order) of an ethernet address that // determines whether the ethernet address is a unicast or multicast. If diff --git a/pkg/tcpip/header/icmpv4.go b/pkg/tcpip/header/icmpv4.go index 7908c5744..1a631b31a 100644 --- a/pkg/tcpip/header/icmpv4.go +++ b/pkg/tcpip/header/icmpv4.go @@ -72,6 +72,7 @@ const ( // Values for ICMP code as defined in RFC 792. const ( ICMPv4TTLExceeded = 0 + ICMPv4HostUnreachable = 1 ICMPv4PortUnreachable = 3 ICMPv4FragmentationNeeded = 4 ) diff --git a/pkg/tcpip/header/icmpv6.go b/pkg/tcpip/header/icmpv6.go index c7ee2de57..a13b4b809 100644 --- a/pkg/tcpip/header/icmpv6.go +++ b/pkg/tcpip/header/icmpv6.go @@ -110,9 +110,16 @@ const ( ICMPv6RedirectMsg ICMPv6Type = 137 ) -// Values for ICMP code as defined in RFC 4443. +// Values for ICMP destination unreachable code as defined in RFC 4443 section +// 3.1. const ( - ICMPv6PortUnreachable = 4 + ICMPv6NetworkUnreachable = 0 + ICMPv6Prohibited = 1 + ICMPv6BeyondScope = 2 + ICMPv6AddressUnreachable = 3 + ICMPv6PortUnreachable = 4 + ICMPv6Policy = 5 + ICMPv6RejectRoute = 6 ) // Type is the ICMP type field. diff --git a/pkg/tcpip/link/channel/BUILD b/pkg/tcpip/link/channel/BUILD index b8b93e78e..39ca774ef 100644 --- a/pkg/tcpip/link/channel/BUILD +++ b/pkg/tcpip/link/channel/BUILD @@ -10,6 +10,7 @@ go_library( "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/header", "//pkg/tcpip/stack", ], ) diff --git a/pkg/tcpip/link/channel/channel.go b/pkg/tcpip/link/channel/channel.go index 20b183da0..e12a5929b 100644 --- a/pkg/tcpip/link/channel/channel.go +++ b/pkg/tcpip/link/channel/channel.go @@ -23,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -296,3 +297,12 @@ func (e *Endpoint) AddNotify(notify Notification) *NotificationHandle { func (e *Endpoint) RemoveNotify(handle *NotificationHandle) { e.q.RemoveNotify(handle) } + +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. +func (*Endpoint) ARPHardwareType() header.ARPHardwareType { + return header.ARPHardwareNone +} + +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *Endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { +} 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..c18bb91fb 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" @@ -385,26 +386,35 @@ const ( _VIRTIO_NET_HDR_GSO_TCPV6 = 4 ) -// 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, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { if e.hdrSize > 0 { // Add ethernet header if needed. eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) pkt.LinkHeader = buffer.View(eth) ethHdr := &header.EthernetFields{ - DstAddr: r.RemoteLinkAddress, + DstAddr: remote, Type: protocol, } // Preserve the src address if it's set in the route. - if r.LocalLinkAddress != "" { - ethHdr.SrcAddr = r.LocalLinkAddress + if local != "" { + ethHdr.SrcAddr = local } else { ethHdr.SrcAddr = e.addr } eth.Encode(ethHdr) } +} + +// 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, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { + if e.hdrSize > 0 { + e.AddHeader(r.LocalLinkAddress, r.RemoteLinkAddress, protocol, pkt) + } + + var builder iovec.Builder fd := e.fds[pkt.Hash%uint32(len(e.fds))] if e.Capabilities()&stack.CapabilityHardwareGSO != 0 { @@ -430,47 +440,28 @@ 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 if e.hdrSize > 0 { - // Add ethernet header if needed. - ethHdrBuf = make([]byte, header.EthernetMinimumSize) - eth := header.Ethernet(ethHdrBuf) - ethHdr := &header.EthernetFields{ - DstAddr: pkt.EgressRoute.RemoteLinkAddress, - Type: pkt.NetworkProtocolNumber, - } - - // Preserve the src address if it's set in the route. - if pkt.EgressRoute.LocalLinkAddress != "" { - ethHdr.SrcAddr = pkt.EgressRoute.LocalLinkAddress - } else { - ethHdr.SrcAddr = e.addr - } - eth.Encode(ethHdr) - iovLen++ + e.AddHeader(pkt.EgressRoute.LocalLinkAddress, pkt.EgressRoute.RemoteLinkAddress, pkt.NetworkProtocolNumber, pkt) } - 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 +482,19 @@ 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(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 +591,14 @@ 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..7b995b85a 100644 --- a/pkg/tcpip/link/fdbased/endpoint_test.go +++ b/pkg/tcpip/link/fdbased/endpoint_test.go @@ -107,6 +107,10 @@ func (c *context) DeliverNetworkPacket(remote tcpip.LinkAddress, local tcpip.Lin c.ch <- packetInfo{remote, protocol, pkt} } +func (c *context) DeliverOutboundPacket(remote tcpip.LinkAddress, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + panic("unimplemented") +} + func TestNoEthernetProperties(t *testing.T) { c := newContext(t, &Options{MTU: mtu}) defer c.cleanup() @@ -500,3 +504,80 @@ 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 (d *fakeNetworkDispatcher) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + panic("unimplemented") +} + +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() diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go index 568c6874f..781cdd317 100644 --- a/pkg/tcpip/link/loopback/loopback.go +++ b/pkg/tcpip/link/loopback/loopback.go @@ -113,3 +113,11 @@ func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { return nil } + +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. +func (*endpoint) ARPHardwareType() header.ARPHardwareType { + return header.ARPHardwareLoopback +} + +func (e *endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { +} diff --git a/pkg/tcpip/link/muxed/BUILD b/pkg/tcpip/link/muxed/BUILD index 82b441b79..e7493e5c5 100644 --- a/pkg/tcpip/link/muxed/BUILD +++ b/pkg/tcpip/link/muxed/BUILD @@ -9,6 +9,7 @@ go_library( deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/header", "//pkg/tcpip/stack", ], ) diff --git a/pkg/tcpip/link/muxed/injectable.go b/pkg/tcpip/link/muxed/injectable.go index c69d6b7e9..56a611825 100644 --- a/pkg/tcpip/link/muxed/injectable.go +++ b/pkg/tcpip/link/muxed/injectable.go @@ -18,6 +18,7 @@ package muxed import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -129,6 +130,15 @@ func (m *InjectableEndpoint) Wait() { } } +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. +func (*InjectableEndpoint) ARPHardwareType() header.ARPHardwareType { + panic("unsupported operation") +} + +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (*InjectableEndpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { +} + // NewInjectableEndpoint creates a new multi-endpoint injectable endpoint. func NewInjectableEndpoint(routes map[tcpip.Address]stack.InjectableLinkEndpoint) *InjectableEndpoint { return &InjectableEndpoint{ diff --git a/pkg/tcpip/link/nested/BUILD b/pkg/tcpip/link/nested/BUILD new file mode 100644 index 000000000..2cdb23475 --- /dev/null +++ b/pkg/tcpip/link/nested/BUILD @@ -0,0 +1,32 @@ +load("//tools:defs.bzl", "go_library", "go_test") + +package(licenses = ["notice"]) + +go_library( + name = "nested", + srcs = [ + "nested.go", + ], + visibility = ["//visibility:public"], + deps = [ + "//pkg/sync", + "//pkg/tcpip", + "//pkg/tcpip/buffer", + "//pkg/tcpip/header", + "//pkg/tcpip/stack", + ], +) + +go_test( + name = "nested_test", + size = "small", + srcs = [ + "nested_test.go", + ], + deps = [ + "//pkg/tcpip", + "//pkg/tcpip/header", + "//pkg/tcpip/link/nested", + "//pkg/tcpip/stack", + ], +) diff --git a/pkg/tcpip/link/nested/nested.go b/pkg/tcpip/link/nested/nested.go new file mode 100644 index 000000000..d40de54df --- /dev/null +++ b/pkg/tcpip/link/nested/nested.go @@ -0,0 +1,152 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package nested provides helpers to implement the pattern of nested +// stack.LinkEndpoints. +package nested + +import ( + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/stack" +) + +// Endpoint is a wrapper around stack.LinkEndpoint and stack.NetworkDispatcher +// that can be used to implement nesting safely by providing lifecycle +// concurrency guards. +// +// See the tests in this package for example usage. +type Endpoint struct { + child stack.LinkEndpoint + embedder stack.NetworkDispatcher + + // mu protects dispatcher. + mu sync.RWMutex + dispatcher stack.NetworkDispatcher +} + +var _ stack.GSOEndpoint = (*Endpoint)(nil) +var _ stack.LinkEndpoint = (*Endpoint)(nil) +var _ stack.NetworkDispatcher = (*Endpoint)(nil) + +// Init initializes a nested.Endpoint that uses embedder as the dispatcher for +// child on Attach. +// +// See the tests in this package for example usage. +func (e *Endpoint) Init(child stack.LinkEndpoint, embedder stack.NetworkDispatcher) { + e.child = child + e.embedder = embedder +} + +// DeliverNetworkPacket implements stack.NetworkDispatcher. +func (e *Endpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + e.mu.RLock() + d := e.dispatcher + e.mu.RUnlock() + if d != nil { + d.DeliverNetworkPacket(remote, local, protocol, pkt) + } +} + +// DeliverOutboundPacket implements stack.NetworkDispatcher.DeliverOutboundPacket. +func (e *Endpoint) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + e.mu.RLock() + d := e.dispatcher + e.mu.RUnlock() + if d != nil { + d.DeliverOutboundPacket(remote, local, protocol, pkt) + } +} + +// Attach implements stack.LinkEndpoint. +func (e *Endpoint) Attach(dispatcher stack.NetworkDispatcher) { + e.mu.Lock() + e.dispatcher = dispatcher + e.mu.Unlock() + // If we're attaching to a valid dispatcher, pass embedder as the dispatcher + // to our child, otherwise detach the child by giving it a nil dispatcher. + var pass stack.NetworkDispatcher + if dispatcher != nil { + pass = e.embedder + } + e.child.Attach(pass) +} + +// IsAttached implements stack.LinkEndpoint. +func (e *Endpoint) IsAttached() bool { + e.mu.RLock() + isAttached := e.dispatcher != nil + e.mu.RUnlock() + return isAttached +} + +// MTU implements stack.LinkEndpoint. +func (e *Endpoint) MTU() uint32 { + return e.child.MTU() +} + +// Capabilities implements stack.LinkEndpoint. +func (e *Endpoint) Capabilities() stack.LinkEndpointCapabilities { + return e.child.Capabilities() +} + +// MaxHeaderLength implements stack.LinkEndpoint. +func (e *Endpoint) MaxHeaderLength() uint16 { + return e.child.MaxHeaderLength() +} + +// LinkAddress implements stack.LinkEndpoint. +func (e *Endpoint) LinkAddress() tcpip.LinkAddress { + return e.child.LinkAddress() +} + +// WritePacket implements stack.LinkEndpoint. +func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { + return e.child.WritePacket(r, gso, protocol, pkt) +} + +// WritePackets implements stack.LinkEndpoint. +func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + return e.child.WritePackets(r, gso, pkts, protocol) +} + +// WriteRawPacket implements stack.LinkEndpoint. +func (e *Endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { + return e.child.WriteRawPacket(vv) +} + +// Wait implements stack.LinkEndpoint. +func (e *Endpoint) Wait() { + e.child.Wait() +} + +// GSOMaxSize implements stack.GSOEndpoint. +func (e *Endpoint) GSOMaxSize() uint32 { + if e, ok := e.child.(stack.GSOEndpoint); ok { + return e.GSOMaxSize() + } + return 0 +} + +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType +func (e *Endpoint) ARPHardwareType() header.ARPHardwareType { + return e.child.ARPHardwareType() +} + +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *Endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + e.child.AddHeader(local, remote, protocol, pkt) +} diff --git a/pkg/tcpip/link/nested/nested_test.go b/pkg/tcpip/link/nested/nested_test.go new file mode 100644 index 000000000..7d9249c1c --- /dev/null +++ b/pkg/tcpip/link/nested/nested_test.go @@ -0,0 +1,109 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package nested_test + +import ( + "testing" + + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/link/nested" + "gvisor.dev/gvisor/pkg/tcpip/stack" +) + +type parentEndpoint struct { + nested.Endpoint +} + +var _ stack.LinkEndpoint = (*parentEndpoint)(nil) +var _ stack.NetworkDispatcher = (*parentEndpoint)(nil) + +type childEndpoint struct { + stack.LinkEndpoint + dispatcher stack.NetworkDispatcher +} + +var _ stack.LinkEndpoint = (*childEndpoint)(nil) + +func (c *childEndpoint) Attach(dispatcher stack.NetworkDispatcher) { + c.dispatcher = dispatcher +} + +func (c *childEndpoint) IsAttached() bool { + return c.dispatcher != nil +} + +type counterDispatcher struct { + count int +} + +var _ stack.NetworkDispatcher = (*counterDispatcher)(nil) + +func (d *counterDispatcher) DeliverNetworkPacket(tcpip.LinkAddress, tcpip.LinkAddress, tcpip.NetworkProtocolNumber, *stack.PacketBuffer) { + d.count++ +} + +func (d *counterDispatcher) DeliverOutboundPacket(tcpip.LinkAddress, tcpip.LinkAddress, tcpip.NetworkProtocolNumber, *stack.PacketBuffer) { + panic("unimplemented") +} + +func TestNestedLinkEndpoint(t *testing.T) { + const emptyAddress = tcpip.LinkAddress("") + + var ( + childEP childEndpoint + nestedEP parentEndpoint + disp counterDispatcher + ) + nestedEP.Endpoint.Init(&childEP, &nestedEP) + + if childEP.IsAttached() { + t.Error("On init, childEP.IsAttached() = true, want = false") + } + if nestedEP.IsAttached() { + t.Error("On init, nestedEP.IsAttached() = true, want = false") + } + + nestedEP.Attach(&disp) + if disp.count != 0 { + t.Fatalf("After attach, got disp.count = %d, want = 0", disp.count) + } + if !childEP.IsAttached() { + t.Error("After attach, childEP.IsAttached() = false, want = true") + } + if !nestedEP.IsAttached() { + t.Error("After attach, nestedEP.IsAttached() = false, want = true") + } + + nestedEP.DeliverNetworkPacket(emptyAddress, emptyAddress, header.IPv4ProtocolNumber, &stack.PacketBuffer{}) + if disp.count != 1 { + t.Errorf("After first packet with dispatcher attached, got disp.count = %d, want = 1", disp.count) + } + + nestedEP.Attach(nil) + if childEP.IsAttached() { + t.Error("After detach, childEP.IsAttached() = true, want = false") + } + if nestedEP.IsAttached() { + t.Error("After detach, nestedEP.IsAttached() = true, want = false") + } + + disp.count = 0 + nestedEP.DeliverNetworkPacket(emptyAddress, emptyAddress, header.IPv4ProtocolNumber, &stack.PacketBuffer{}) + if disp.count != 0 { + t.Errorf("After second packet with dispatcher detached, got disp.count = %d, want = 0", disp.count) + } + +} diff --git a/pkg/tcpip/link/packetsocket/BUILD b/pkg/tcpip/link/packetsocket/BUILD new file mode 100644 index 000000000..6fff160ce --- /dev/null +++ b/pkg/tcpip/link/packetsocket/BUILD @@ -0,0 +1,14 @@ +load("//tools:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "packetsocket", + srcs = ["endpoint.go"], + visibility = ["//visibility:public"], + deps = [ + "//pkg/tcpip", + "//pkg/tcpip/link/nested", + "//pkg/tcpip/stack", + ], +) diff --git a/pkg/tcpip/link/packetsocket/endpoint.go b/pkg/tcpip/link/packetsocket/endpoint.go new file mode 100644 index 000000000..3922c2a04 --- /dev/null +++ b/pkg/tcpip/link/packetsocket/endpoint.go @@ -0,0 +1,50 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package packetsocket provides a link layer endpoint that provides the ability +// to loop outbound packets to any AF_PACKET sockets that may be interested in +// the outgoing packet. +package packetsocket + +import ( + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/link/nested" + "gvisor.dev/gvisor/pkg/tcpip/stack" +) + +type endpoint struct { + nested.Endpoint +} + +// New creates a new packetsocket LinkEndpoint. +func New(lower stack.LinkEndpoint) stack.LinkEndpoint { + e := &endpoint{} + e.Endpoint.Init(lower, e) + return e +} + +// WritePacket implements stack.LinkEndpoint.WritePacket. +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { + e.Endpoint.DeliverOutboundPacket(r.RemoteLinkAddress, r.LocalLinkAddress, protocol, pkt) + return e.Endpoint.WritePacket(r, gso, protocol, pkt) +} + +// WritePackets implements stack.LinkEndpoint.WritePackets. +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, proto tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() { + e.Endpoint.DeliverOutboundPacket(pkt.EgressRoute.RemoteLinkAddress, pkt.EgressRoute.LocalLinkAddress, pkt.NetworkProtocolNumber, pkt) + } + + return e.Endpoint.WritePackets(r, gso, pkts, proto) +} diff --git a/pkg/tcpip/link/qdisc/fifo/BUILD b/pkg/tcpip/link/qdisc/fifo/BUILD index 054c213bc..1d0079bd6 100644 --- a/pkg/tcpip/link/qdisc/fifo/BUILD +++ b/pkg/tcpip/link/qdisc/fifo/BUILD @@ -14,6 +14,7 @@ go_library( "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/header", "//pkg/tcpip/stack", ], ) diff --git a/pkg/tcpip/link/qdisc/fifo/endpoint.go b/pkg/tcpip/link/qdisc/fifo/endpoint.go index b5dfb7850..467083239 100644 --- a/pkg/tcpip/link/qdisc/fifo/endpoint.go +++ b/pkg/tcpip/link/qdisc/fifo/endpoint.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -106,6 +107,11 @@ func (e *endpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protoco e.dispatcher.DeliverNetworkPacket(remote, local, protocol, pkt) } +// DeliverOutboundPacket implements stack.NetworkDispatcher.DeliverOutboundPacket. +func (e *endpoint) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + e.dispatcher.DeliverOutboundPacket(remote, local, protocol, pkt) +} + // Attach implements stack.LinkEndpoint.Attach. func (e *endpoint) Attach(dispatcher stack.NetworkDispatcher) { e.dispatcher = dispatcher @@ -193,6 +199,8 @@ func (e *endpoint) WritePackets(_ *stack.Route, _ *stack.GSO, pkts stack.PacketB // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { + // TODO(gvisor.dev/issue/3267/): Queue these packets as well once + // WriteRawPacket takes PacketBuffer instead of VectorisedView. return e.lower.WriteRawPacket(vv) } @@ -207,3 +215,13 @@ func (e *endpoint) Wait() { e.wg.Wait() } + +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType +func (e *endpoint) ARPHardwareType() header.ARPHardwareType { + return e.lower.ARPHardwareType() +} + +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + e.lower.AddHeader(local, remote, protocol, pkt) +} diff --git a/pkg/tcpip/link/rawfile/rawfile_unsafe.go b/pkg/tcpip/link/rawfile/rawfile_unsafe.go index 44e25d475..f4c32c2da 100644 --- a/pkg/tcpip/link/rawfile/rawfile_unsafe.go +++ b/pkg/tcpip/link/rawfile/rawfile_unsafe.go @@ -66,39 +66,14 @@ func NonBlockingWrite(fd int, buf []byte) *tcpip.Error { return nil } -// NonBlockingWrite3 writes up to three byte slices to a file descriptor in a -// single syscall. It fails if partial data is written. -func NonBlockingWrite3(fd int, b1, b2, b3 []byte) *tcpip.Error { - // If the is no second buffer, issue a regular write. - if len(b2) == 0 { - return NonBlockingWrite(fd, b1) - } - - // We have two buffers. Build the iovec that represents them and issue - // a writev syscall. - iovec := [3]syscall.Iovec{ - { - Base: &b1[0], - Len: uint64(len(b1)), - }, - { - Base: &b2[0], - Len: uint64(len(b2)), - }, - } - iovecLen := uintptr(2) - - if len(b3) > 0 { - iovecLen++ - iovec[2].Base = &b3[0] - iovec[2].Len = uint64(len(b3)) - } - +// NonBlockingWriteIovec writes iovec to a file descriptor in a single syscall. +// It fails if partial data is written. +func NonBlockingWriteIovec(fd int, iovec []syscall.Iovec) *tcpip.Error { + iovecLen := uintptr(len(iovec)) _, _, e := syscall.RawSyscall(syscall.SYS_WRITEV, uintptr(fd), uintptr(unsafe.Pointer(&iovec[0])), iovecLen) if e != 0 { return TranslateErrno(e) } - return nil } diff --git a/pkg/tcpip/link/sharedmem/sharedmem.go b/pkg/tcpip/link/sharedmem/sharedmem.go index 0374a2441..507c76b76 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem.go +++ b/pkg/tcpip/link/sharedmem/sharedmem.go @@ -183,22 +183,29 @@ func (e *endpoint) LinkAddress() tcpip.LinkAddress { return e.addr } -// 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, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { - // Add the ethernet header here. +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + // Add ethernet header if needed. eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) pkt.LinkHeader = buffer.View(eth) ethHdr := &header.EthernetFields{ - DstAddr: r.RemoteLinkAddress, + DstAddr: remote, Type: protocol, } - if r.LocalLinkAddress != "" { - ethHdr.SrcAddr = r.LocalLinkAddress + + // Preserve the src address if it's set in the route. + if local != "" { + ethHdr.SrcAddr = local } else { ethHdr.SrcAddr = e.addr } eth.Encode(ethHdr) +} + +// 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, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { + e.AddHeader(r.LocalLinkAddress, r.RemoteLinkAddress, protocol, pkt) v := pkt.Data.ToView() // Transmit the packet. @@ -287,3 +294,8 @@ func (e *endpoint) dispatchLoop(d stack.NetworkDispatcher) { e.completed.Done() } + +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType +func (*endpoint) ARPHardwareType() header.ARPHardwareType { + return header.ARPHardwareEther +} diff --git a/pkg/tcpip/link/sharedmem/sharedmem_test.go b/pkg/tcpip/link/sharedmem/sharedmem_test.go index 28a2e88ba..8f3cd9449 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem_test.go +++ b/pkg/tcpip/link/sharedmem/sharedmem_test.go @@ -143,6 +143,10 @@ func (c *testContext) DeliverNetworkPacket(remoteLinkAddr, localLinkAddr tcpip.L c.packetCh <- struct{}{} } +func (c *testContext) DeliverOutboundPacket(remoteLinkAddr, localLinkAddr tcpip.LinkAddress, proto tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + panic("unimplemented") +} + func (c *testContext) cleanup() { c.ep.Close() closeFDs(&c.txCfg) diff --git a/pkg/tcpip/link/sniffer/BUILD b/pkg/tcpip/link/sniffer/BUILD index 230a8d53a..7cbc305e7 100644 --- a/pkg/tcpip/link/sniffer/BUILD +++ b/pkg/tcpip/link/sniffer/BUILD @@ -14,6 +14,7 @@ go_library( "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/header", + "//pkg/tcpip/link/nested", "//pkg/tcpip/stack", ], ) diff --git a/pkg/tcpip/link/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go index f2e47b6a7..509076643 100644 --- a/pkg/tcpip/link/sniffer/sniffer.go +++ b/pkg/tcpip/link/sniffer/sniffer.go @@ -31,6 +31,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/link/nested" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -48,18 +49,21 @@ var LogPackets uint32 = 1 var LogPacketsToPCAP uint32 = 1 type endpoint struct { - dispatcher stack.NetworkDispatcher - lower stack.LinkEndpoint + nested.Endpoint writer io.Writer maxPCAPLen uint32 } +var _ stack.GSOEndpoint = (*endpoint)(nil) +var _ stack.LinkEndpoint = (*endpoint)(nil) +var _ stack.NetworkDispatcher = (*endpoint)(nil) + // New creates a new sniffer link-layer endpoint. It wraps around another // endpoint and logs packets and they traverse the endpoint. func New(lower stack.LinkEndpoint) stack.LinkEndpoint { - return &endpoint{ - lower: lower, - } + sniffer := &endpoint{} + sniffer.Endpoint.Init(lower, sniffer) + return sniffer } func zoneOffset() (int32, error) { @@ -103,11 +107,12 @@ func NewWithWriter(lower stack.LinkEndpoint, writer io.Writer, snapLen uint32) ( if err := writePCAPHeader(writer, snapLen); err != nil { return nil, err } - return &endpoint{ - lower: lower, + sniffer := &endpoint{ writer: writer, maxPCAPLen: snapLen, - }, nil + } + sniffer.Endpoint.Init(lower, sniffer) + return sniffer, nil } // DeliverNetworkPacket implements the stack.NetworkDispatcher interface. It is @@ -115,50 +120,12 @@ func NewWithWriter(lower stack.LinkEndpoint, writer io.Writer, snapLen uint32) ( // logs the packet before forwarding to the actual dispatcher. func (e *endpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { e.dumpPacket("recv", nil, protocol, pkt) - e.dispatcher.DeliverNetworkPacket(remote, local, protocol, pkt) -} - -// Attach implements the stack.LinkEndpoint interface. It saves the dispatcher -// and registers with the lower endpoint as its dispatcher so that "e" is called -// for inbound packets. -func (e *endpoint) Attach(dispatcher stack.NetworkDispatcher) { - e.dispatcher = dispatcher - e.lower.Attach(e) -} - -// IsAttached implements stack.LinkEndpoint.IsAttached. -func (e *endpoint) IsAttached() bool { - return e.dispatcher != nil -} - -// MTU implements stack.LinkEndpoint.MTU. It just forwards the request to the -// lower endpoint. -func (e *endpoint) MTU() uint32 { - return e.lower.MTU() + e.Endpoint.DeliverNetworkPacket(remote, local, protocol, pkt) } -// Capabilities implements stack.LinkEndpoint.Capabilities. It just forwards the -// request to the lower endpoint. -func (e *endpoint) Capabilities() stack.LinkEndpointCapabilities { - return e.lower.Capabilities() -} - -// MaxHeaderLength implements the stack.LinkEndpoint interface. It just forwards -// the request to the lower endpoint. -func (e *endpoint) MaxHeaderLength() uint16 { - return e.lower.MaxHeaderLength() -} - -func (e *endpoint) LinkAddress() tcpip.LinkAddress { - return e.lower.LinkAddress() -} - -// GSOMaxSize returns the maximum GSO packet size. -func (e *endpoint) GSOMaxSize() uint32 { - if gso, ok := e.lower.(stack.GSOEndpoint); ok { - return gso.GSOMaxSize() - } - return 0 +// DeliverOutboundPacket implements stack.NetworkDispatcher.DeliverOutboundPacket. +func (e *endpoint) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + e.Endpoint.DeliverOutboundPacket(remote, local, protocol, pkt) } func (e *endpoint) dumpPacket(prefix string, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { @@ -203,7 +170,7 @@ func (e *endpoint) dumpPacket(prefix string, gso *stack.GSO, protocol tcpip.Netw // forwards the request to the lower endpoint. func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { e.dumpPacket("send", gso, protocol, pkt) - return e.lower.WritePacket(r, gso, protocol, pkt) + return e.Endpoint.WritePacket(r, gso, protocol, pkt) } // WritePackets implements the stack.LinkEndpoint interface. It is called by @@ -213,7 +180,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() { e.dumpPacket("send", gso, protocol, pkt) } - return e.lower.WritePackets(r, gso, pkts, protocol) + return e.Endpoint.WritePackets(r, gso, pkts, protocol) } // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. @@ -221,12 +188,9 @@ func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { e.dumpPacket("send", nil, 0, &stack.PacketBuffer{ Data: vv, }) - return e.lower.WriteRawPacket(vv) + return e.Endpoint.WriteRawPacket(vv) } -// Wait implements stack.LinkEndpoint.Wait. -func (e *endpoint) Wait() { e.lower.Wait() } - func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer, gso *stack.GSO) { // Figure out the network layer info. var transProto uint8 diff --git a/pkg/tcpip/link/tun/device.go b/pkg/tcpip/link/tun/device.go index 6bc9033d0..04ae58e59 100644 --- a/pkg/tcpip/link/tun/device.go +++ b/pkg/tcpip/link/tun/device.go @@ -139,6 +139,7 @@ func attachOrCreateNIC(s *stack.Stack, name, prefix string, linkCaps stack.LinkE stack: s, nicID: id, name: name, + isTap: prefix == "tap", } endpoint.Endpoint.LinkEPCapabilities = linkCaps if endpoint.name == "" { @@ -271,21 +272,9 @@ func (d *Device) encodePkt(info *channel.PacketInfo) (buffer.View, bool) { if d.hasFlags(linux.IFF_TAP) { // Add ethernet header if not provided. if info.Pkt.LinkHeader == nil { - hdr := &header.EthernetFields{ - SrcAddr: info.Route.LocalLinkAddress, - DstAddr: info.Route.RemoteLinkAddress, - Type: info.Proto, - } - if hdr.SrcAddr == "" { - hdr.SrcAddr = d.endpoint.LinkAddress() - } - - eth := make(header.Ethernet, header.EthernetMinimumSize) - eth.Encode(hdr) - vv.AppendView(buffer.View(eth)) - } else { - vv.AppendView(info.Pkt.LinkHeader) + d.endpoint.AddHeader(info.Route.LocalLinkAddress, info.Route.RemoteLinkAddress, info.Proto, info.Pkt) } + vv.AppendView(info.Pkt.LinkHeader) } // Append upper headers. @@ -348,6 +337,7 @@ type tunEndpoint struct { stack *stack.Stack nicID tcpip.NICID name string + isTap bool } // DecRef decrements refcount of e, removes NIC if refcount goes to 0. @@ -356,3 +346,38 @@ func (e *tunEndpoint) DecRef() { e.stack.RemoveNIC(e.nicID) }) } + +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. +func (e *tunEndpoint) ARPHardwareType() header.ARPHardwareType { + if e.isTap { + return header.ARPHardwareEther + } + return header.ARPHardwareNone +} + +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *tunEndpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + if !e.isTap { + return + } + eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) + pkt.LinkHeader = buffer.View(eth) + hdr := &header.EthernetFields{ + SrcAddr: local, + DstAddr: remote, + Type: protocol, + } + if hdr.SrcAddr == "" { + hdr.SrcAddr = e.LinkAddress() + } + + eth.Encode(hdr) +} + +// MaxHeaderLength returns the maximum size of the link layer header. +func (e *tunEndpoint) MaxHeaderLength() uint16 { + if e.isTap { + return header.EthernetMinimumSize + } + return 0 +} diff --git a/pkg/tcpip/link/waitable/BUILD b/pkg/tcpip/link/waitable/BUILD index 0956d2c65..ee84c3d96 100644 --- a/pkg/tcpip/link/waitable/BUILD +++ b/pkg/tcpip/link/waitable/BUILD @@ -12,6 +12,7 @@ go_library( "//pkg/gate", "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/header", "//pkg/tcpip/stack", ], ) @@ -25,6 +26,7 @@ go_test( deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/header", "//pkg/tcpip/stack", ], ) diff --git a/pkg/tcpip/link/waitable/waitable.go b/pkg/tcpip/link/waitable/waitable.go index 949b3f2b2..b152a0f26 100644 --- a/pkg/tcpip/link/waitable/waitable.go +++ b/pkg/tcpip/link/waitable/waitable.go @@ -25,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/gate" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -59,6 +60,15 @@ func (e *Endpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protoco e.dispatchGate.Leave() } +// DeliverOutboundPacket implements stack.NetworkDispatcher.DeliverOutboundPacket. +func (e *Endpoint) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + if !e.dispatchGate.Enter() { + return + } + e.dispatcher.DeliverOutboundPacket(remote, local, protocol, pkt) + e.dispatchGate.Leave() +} + // Attach implements stack.LinkEndpoint.Attach. It saves the dispatcher and // registers with the lower endpoint as its dispatcher so that "e" is called // for inbound packets. @@ -147,3 +157,13 @@ func (e *Endpoint) WaitDispatch() { // Wait implements stack.LinkEndpoint.Wait. func (e *Endpoint) Wait() {} + +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. +func (e *Endpoint) ARPHardwareType() header.ARPHardwareType { + return e.lower.ARPHardwareType() +} + +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *Endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + e.lower.AddHeader(local, remote, protocol, pkt) +} diff --git a/pkg/tcpip/link/waitable/waitable_test.go b/pkg/tcpip/link/waitable/waitable_test.go index 63bf40562..c448a888f 100644 --- a/pkg/tcpip/link/waitable/waitable_test.go +++ b/pkg/tcpip/link/waitable/waitable_test.go @@ -19,6 +19,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -39,6 +40,10 @@ func (e *countedEndpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, e.dispatchCount++ } +func (e *countedEndpoint) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + panic("unimplemented") +} + func (e *countedEndpoint) Attach(dispatcher stack.NetworkDispatcher) { e.attachCount++ e.dispatcher = dispatcher @@ -81,9 +86,19 @@ func (e *countedEndpoint) WriteRawPacket(buffer.VectorisedView) *tcpip.Error { return nil } +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. +func (*countedEndpoint) ARPHardwareType() header.ARPHardwareType { + panic("unimplemented") +} + // Wait implements stack.LinkEndpoint.Wait. func (*countedEndpoint) Wait() {} +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *countedEndpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + panic("unimplemented") +} + func TestWaitWrite(t *testing.T) { ep := &countedEndpoint{} wep := New(ep) diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index 7f27a840d..31a242482 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -160,9 +160,12 @@ func (*protocol) LinkAddressProtocol() tcpip.NetworkProtocolNumber { } // LinkAddressRequest implements stack.LinkAddressResolver.LinkAddressRequest. -func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, linkEP stack.LinkEndpoint) *tcpip.Error { +func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, remoteLinkAddr tcpip.LinkAddress, linkEP stack.LinkEndpoint) *tcpip.Error { r := &stack.Route{ - RemoteLinkAddress: broadcastMAC, + RemoteLinkAddress: remoteLinkAddr, + } + if len(r.RemoteLinkAddress) == 0 { + r.RemoteLinkAddress = header.EthernetBroadcastAddress } hdr := buffer.NewPrependable(int(linkEP.MaxHeaderLength()) + header.ARPSize) @@ -181,7 +184,7 @@ func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, linkEP stack. // ResolveStaticAddress implements stack.LinkAddressResolver.ResolveStaticAddress. func (*protocol) ResolveStaticAddress(addr tcpip.Address) (tcpip.LinkAddress, bool) { if addr == header.IPv4Broadcast { - return broadcastMAC, true + return header.EthernetBroadcastAddress, true } if header.IsV4MulticastAddress(addr) { return header.EthernetAddressFromMulticastIPv4Address(addr), true @@ -216,8 +219,6 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) (proto tcpip.TransportProtocolNu return 0, false, true } -var broadcastMAC = tcpip.LinkAddress([]byte{0xff, 0xff, 0xff, 0xff, 0xff, 0xff}) - // NewProtocol returns an ARP network protocol. func NewProtocol() stack.NetworkProtocol { return &protocol{} diff --git a/pkg/tcpip/network/arp/arp_test.go b/pkg/tcpip/network/arp/arp_test.go index 66e67429c..a35a64a0f 100644 --- a/pkg/tcpip/network/arp/arp_test.go +++ b/pkg/tcpip/network/arp/arp_test.go @@ -32,10 +32,14 @@ import ( ) const ( - stackLinkAddr = tcpip.LinkAddress("\x0a\x0a\x0b\x0b\x0c\x0c") - stackAddr1 = tcpip.Address("\x0a\x00\x00\x01") - stackAddr2 = tcpip.Address("\x0a\x00\x00\x02") - stackAddrBad = tcpip.Address("\x0a\x00\x00\x03") + stackLinkAddr1 = tcpip.LinkAddress("\x0a\x0a\x0b\x0b\x0c\x0c") + stackLinkAddr2 = tcpip.LinkAddress("\x0b\x0b\x0c\x0c\x0d\x0d") + stackAddr1 = tcpip.Address("\x0a\x00\x00\x01") + stackAddr2 = tcpip.Address("\x0a\x00\x00\x02") + stackAddrBad = tcpip.Address("\x0a\x00\x00\x03") + + defaultChannelSize = 1 + defaultMTU = 65536 ) type testContext struct { @@ -50,8 +54,7 @@ func newTestContext(t *testing.T) *testContext { TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol4()}, }) - const defaultMTU = 65536 - ep := channel.New(256, defaultMTU, stackLinkAddr) + ep := channel.New(defaultChannelSize, defaultMTU, stackLinkAddr1) wep := stack.LinkEndpoint(ep) if testing.Verbose() { @@ -119,7 +122,7 @@ func TestDirectRequest(t *testing.T) { if !rep.IsValid() { t.Fatalf("invalid ARP response pi.Pkt.Header.UsedLength()=%d", pi.Pkt.Header.UsedLength()) } - if got, want := tcpip.LinkAddress(rep.HardwareAddressSender()), stackLinkAddr; got != want { + if got, want := tcpip.LinkAddress(rep.HardwareAddressSender()), stackLinkAddr1; got != want { t.Errorf("got HardwareAddressSender = %s, want = %s", got, want) } if got, want := tcpip.Address(rep.ProtocolAddressSender()), tcpip.Address(h.ProtocolAddressTarget()); got != want { @@ -144,3 +147,44 @@ func TestDirectRequest(t *testing.T) { t.Errorf("stackAddrBad: unexpected packet sent, Proto=%v", pkt.Proto) } } + +func TestLinkAddressRequest(t *testing.T) { + tests := []struct { + name string + remoteLinkAddr tcpip.LinkAddress + expectLinkAddr tcpip.LinkAddress + }{ + { + name: "Unicast", + remoteLinkAddr: stackLinkAddr2, + expectLinkAddr: stackLinkAddr2, + }, + { + name: "Multicast", + remoteLinkAddr: "", + expectLinkAddr: header.EthernetBroadcastAddress, + }, + } + + for _, test := range tests { + p := arp.NewProtocol() + linkRes, ok := p.(stack.LinkAddressResolver) + if !ok { + t.Fatal("expected ARP protocol to implement stack.LinkAddressResolver") + } + + linkEP := channel.New(defaultChannelSize, defaultMTU, stackLinkAddr1) + if err := linkRes.LinkAddressRequest(stackAddr1, stackAddr2, test.remoteLinkAddr, linkEP); err != nil { + t.Errorf("got p.LinkAddressRequest(%s, %s, %s, _) = %s", stackAddr1, stackAddr2, test.remoteLinkAddr, err) + } + + pkt, ok := linkEP.Read() + if !ok { + t.Fatal("expected to send a link address request") + } + + if got, want := pkt.Route.RemoteLinkAddress, test.expectLinkAddr; got != want { + t.Errorf("got pkt.Route.RemoteLinkAddress = %s, want = %s", got, want) + } + } +} diff --git a/pkg/tcpip/network/fragmentation/fragmentation.go b/pkg/tcpip/network/fragmentation/fragmentation.go index 2982450f8..ffbadb6e2 100644 --- a/pkg/tcpip/network/fragmentation/fragmentation.go +++ b/pkg/tcpip/network/fragmentation/fragmentation.go @@ -17,6 +17,7 @@ package fragmentation import ( + "errors" "fmt" "log" "time" @@ -25,20 +26,31 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/buffer" ) -// DefaultReassembleTimeout is based on the linux stack: net.ipv4.ipfrag_time. -const DefaultReassembleTimeout = 30 * time.Second +const ( + // DefaultReassembleTimeout is based on the linux stack: net.ipv4.ipfrag_time. + DefaultReassembleTimeout = 30 * time.Second -// HighFragThreshold is the threshold at which we start trimming old -// fragmented packets. Linux uses a default value of 4 MB. See -// net.ipv4.ipfrag_high_thresh for more information. -const HighFragThreshold = 4 << 20 // 4MB + // HighFragThreshold is the threshold at which we start trimming old + // fragmented packets. Linux uses a default value of 4 MB. See + // net.ipv4.ipfrag_high_thresh for more information. + HighFragThreshold = 4 << 20 // 4MB -// LowFragThreshold is the threshold we reach to when we start dropping -// older fragmented packets. It's important that we keep enough room for newer -// packets to be re-assembled. Hence, this needs to be lower than -// HighFragThreshold enough. Linux uses a default value of 3 MB. See -// net.ipv4.ipfrag_low_thresh for more information. -const LowFragThreshold = 3 << 20 // 3MB + // LowFragThreshold is the threshold we reach to when we start dropping + // older fragmented packets. It's important that we keep enough room for newer + // packets to be re-assembled. Hence, this needs to be lower than + // HighFragThreshold enough. Linux uses a default value of 3 MB. See + // net.ipv4.ipfrag_low_thresh for more information. + LowFragThreshold = 3 << 20 // 3MB + + // minBlockSize is the minimum block size for fragments. + minBlockSize = 1 +) + +var ( + // ErrInvalidArgs indicates to the caller that that an invalid argument was + // provided. + ErrInvalidArgs = errors.New("invalid args") +) // Fragmentation is the main structure that other modules // of the stack should use to implement IP Fragmentation. @@ -50,10 +62,13 @@ type Fragmentation struct { rList reassemblerList size int timeout time.Duration + blockSize uint16 } // NewFragmentation creates a new Fragmentation. // +// blockSize specifies the fragment block size, in bytes. +// // highMemoryLimit specifies the limit on the memory consumed // by the fragments stored by Fragmentation (overhead of internal data-structures // is not accounted). Fragments are dropped when the limit is reached. @@ -64,7 +79,7 @@ type Fragmentation struct { // reassemblingTimeout specifies the maximum time allowed to reassemble a packet. // Fragments are lazily evicted only when a new a packet with an // already existing fragmentation-id arrives after the timeout. -func NewFragmentation(highMemoryLimit, lowMemoryLimit int, reassemblingTimeout time.Duration) *Fragmentation { +func NewFragmentation(blockSize uint16, highMemoryLimit, lowMemoryLimit int, reassemblingTimeout time.Duration) *Fragmentation { if lowMemoryLimit >= highMemoryLimit { lowMemoryLimit = highMemoryLimit } @@ -73,17 +88,46 @@ func NewFragmentation(highMemoryLimit, lowMemoryLimit int, reassemblingTimeout t lowMemoryLimit = 0 } + if blockSize < minBlockSize { + blockSize = minBlockSize + } + return &Fragmentation{ reassemblers: make(map[uint32]*reassembler), highLimit: highMemoryLimit, lowLimit: lowMemoryLimit, timeout: reassemblingTimeout, + blockSize: blockSize, } } // Process processes an incoming fragment belonging to an ID and returns a // complete packet when all the packets belonging to that ID have been received. +// +// [first, last] is the range of the fragment bytes. +// +// first must be a multiple of the block size f is configured with. The size +// of the fragment data must be a multiple of the block size, unless there are +// no fragments following this fragment (more set to false). func (f *Fragmentation) Process(id uint32, first, last uint16, more bool, vv buffer.VectorisedView) (buffer.VectorisedView, bool, error) { + if first > last { + return buffer.VectorisedView{}, false, fmt.Errorf("first=%d is greater than last=%d: %w", first, last, ErrInvalidArgs) + } + + if first%f.blockSize != 0 { + return buffer.VectorisedView{}, false, fmt.Errorf("first=%d is not a multiple of block size=%d: %w", first, f.blockSize, ErrInvalidArgs) + } + + fragmentSize := last - first + 1 + if more && fragmentSize%f.blockSize != 0 { + return buffer.VectorisedView{}, false, fmt.Errorf("fragment size=%d bytes is not a multiple of block size=%d on non-final fragment: %w", fragmentSize, f.blockSize, ErrInvalidArgs) + } + + if l := vv.Size(); l < int(fragmentSize) { + return buffer.VectorisedView{}, false, fmt.Errorf("got fragment size=%d bytes less than the expected fragment size=%d bytes (first=%d last=%d): %w", l, fragmentSize, first, last, ErrInvalidArgs) + } + vv.CapLength(int(fragmentSize)) + f.mu.Lock() r, ok := f.reassemblers[id] if ok && r.tooOld(f.timeout) { diff --git a/pkg/tcpip/network/fragmentation/fragmentation_test.go b/pkg/tcpip/network/fragmentation/fragmentation_test.go index 72c0f53be..ebc3232e5 100644 --- a/pkg/tcpip/network/fragmentation/fragmentation_test.go +++ b/pkg/tcpip/network/fragmentation/fragmentation_test.go @@ -15,6 +15,7 @@ package fragmentation import ( + "errors" "reflect" "testing" "time" @@ -81,7 +82,7 @@ var processTestCases = []struct { func TestFragmentationProcess(t *testing.T) { for _, c := range processTestCases { t.Run(c.comment, func(t *testing.T) { - f := NewFragmentation(1024, 512, DefaultReassembleTimeout) + f := NewFragmentation(minBlockSize, 1024, 512, DefaultReassembleTimeout) for i, in := range c.in { vv, done, err := f.Process(in.id, in.first, in.last, in.more, in.vv) if err != nil { @@ -110,7 +111,7 @@ func TestFragmentationProcess(t *testing.T) { func TestReassemblingTimeout(t *testing.T) { timeout := time.Millisecond - f := NewFragmentation(1024, 512, timeout) + f := NewFragmentation(minBlockSize, 1024, 512, timeout) // Send first fragment with id = 0, first = 0, last = 0, and more = true. f.Process(0, 0, 0, true, vv(1, "0")) // Sleep more than the timeout. @@ -127,7 +128,7 @@ func TestReassemblingTimeout(t *testing.T) { } func TestMemoryLimits(t *testing.T) { - f := NewFragmentation(3, 1, DefaultReassembleTimeout) + f := NewFragmentation(minBlockSize, 3, 1, DefaultReassembleTimeout) // Send first fragment with id = 0. f.Process(0, 0, 0, true, vv(1, "0")) // Send first fragment with id = 1. @@ -151,7 +152,7 @@ func TestMemoryLimits(t *testing.T) { } func TestMemoryLimitsIgnoresDuplicates(t *testing.T) { - f := NewFragmentation(1, 0, DefaultReassembleTimeout) + f := NewFragmentation(minBlockSize, 1, 0, DefaultReassembleTimeout) // Send first fragment with id = 0. f.Process(0, 0, 0, true, vv(1, "0")) // Send the same packet again. @@ -163,3 +164,99 @@ func TestMemoryLimitsIgnoresDuplicates(t *testing.T) { t.Errorf("Wrong size, duplicates are not handled correctly: got=%d, want=%d.", got, want) } } + +func TestErrors(t *testing.T) { + const fragID = 5 + + tests := []struct { + name string + blockSize uint16 + first uint16 + last uint16 + more bool + data string + err error + }{ + { + name: "exact block size without more", + blockSize: 2, + first: 2, + last: 3, + more: false, + data: "01", + }, + { + name: "exact block size with more", + blockSize: 2, + first: 2, + last: 3, + more: true, + data: "01", + }, + { + name: "exact block size with more and extra data", + blockSize: 2, + first: 2, + last: 3, + more: true, + data: "012", + }, + { + name: "exact block size with more and too little data", + blockSize: 2, + first: 2, + last: 3, + more: true, + data: "0", + err: ErrInvalidArgs, + }, + { + name: "not exact block size with more", + blockSize: 2, + first: 2, + last: 2, + more: true, + data: "0", + err: ErrInvalidArgs, + }, + { + name: "not exact block size without more", + blockSize: 2, + first: 2, + last: 2, + more: false, + data: "0", + }, + { + name: "first not a multiple of block size", + blockSize: 2, + first: 3, + last: 4, + more: true, + data: "01", + err: ErrInvalidArgs, + }, + { + name: "first more than last", + blockSize: 2, + first: 4, + last: 3, + more: true, + data: "01", + err: ErrInvalidArgs, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + f := NewFragmentation(test.blockSize, HighFragThreshold, LowFragThreshold, DefaultReassembleTimeout) + _, done, err := f.Process(fragID, test.first, test.last, test.more, vv(len(test.data), test.data)) + if !errors.Is(err, test.err) { + t.Errorf("got Proceess(%d, %d, %d, %t, %q) = (_, _, %v), want = (_, _, %v)", fragID, test.first, test.last, test.more, test.data, err, test.err) + } + if done { + t.Errorf("got Proceess(%d, %d, %d, %t, %q) = (_, true, _), want = (_, false, _)", fragID, test.first, test.last, test.more, test.data) + } + }) + } +} diff --git a/pkg/tcpip/network/ip_test.go b/pkg/tcpip/network/ip_test.go index 7c8fb3e0a..615bae648 100644 --- a/pkg/tcpip/network/ip_test.go +++ b/pkg/tcpip/network/ip_test.go @@ -172,14 +172,24 @@ func (t *testObject) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.Ne } // WritePackets implements stack.LinkEndpoint.WritePackets. -func (t *testObject) WritePackets(_ *stack.Route, _ *stack.GSO, pkt stack.PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (*testObject) WritePackets(_ *stack.Route, _ *stack.GSO, pkt stack.PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { panic("not implemented") } -func (t *testObject) WriteRawPacket(_ buffer.VectorisedView) *tcpip.Error { +func (*testObject) WriteRawPacket(_ buffer.VectorisedView) *tcpip.Error { return tcpip.ErrNotSupported } +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. +func (*testObject) ARPHardwareType() header.ARPHardwareType { + panic("not implemented") +} + +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (*testObject) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + panic("not implemented") +} + func buildIPv4Route(local, remote tcpip.Address) (stack.Route, *tcpip.Error) { s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, diff --git a/pkg/tcpip/network/ipv4/BUILD b/pkg/tcpip/network/ipv4/BUILD index 78420d6e6..d142b4ffa 100644 --- a/pkg/tcpip/network/ipv4/BUILD +++ b/pkg/tcpip/network/ipv4/BUILD @@ -34,6 +34,6 @@ go_test( "//pkg/tcpip/transport/tcp", "//pkg/tcpip/transport/udp", "//pkg/waiter", - "@com_github_google_go-cmp//cmp:go_default_library", + "@com_github_google_go_cmp//cmp:go_default_library", ], ) diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go index 1b67aa066..83e71cb8c 100644 --- a/pkg/tcpip/network/ipv4/icmp.go +++ b/pkg/tcpip/network/ipv4/icmp.go @@ -129,6 +129,9 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer) { pkt.Data.TrimFront(header.ICMPv4MinimumSize) switch h.Code() { + case header.ICMPv4HostUnreachable: + e.handleControl(stack.ControlNoRoute, 0, pkt) + case header.ICMPv4PortUnreachable: e.handleControl(stack.ControlPortUnreachable, 0, pkt) diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index 7e9f16c90..0b5a35cce 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -45,6 +45,10 @@ const ( // buckets is the number of identifier buckets. buckets = 2048 + + // The size of a fragment block, in bytes, as per RFC 791 section 3.1, + // page 14. + fragmentblockSize = 8 ) type endpoint struct { @@ -66,7 +70,7 @@ func (p *protocol) NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWi prefixLen: addrWithPrefix.PrefixLen, linkEP: linkEP, dispatcher: dispatcher, - fragmentation: fragmentation.NewFragmentation(fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, fragmentation.DefaultReassembleTimeout), + fragmentation: fragmentation.NewFragmentation(fragmentblockSize, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, fragmentation.DefaultReassembleTimeout), protocol: p, stack: st, } @@ -225,12 +229,10 @@ func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, mtu int, func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadSize int, params stack.NetworkHeaderParams) header.IPv4 { ip := header.IPv4(hdr.Prepend(header.IPv4MinimumSize)) length := uint16(hdr.UsedLength() + payloadSize) - id := uint32(0) - if length > 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, params.Protocol, e.protocol.hashIV)%buckets], 1) - } + // RFC 6864 section 4.3 mandates uniqueness of ID values for non-atomic + // datagrams. Since the DF bit is never being set here, all datagrams + // are non-atomic and need an ID. + id := atomic.AddUint32(&e.protocol.ids[hashRoute(r, params.Protocol, e.protocol.hashIV)%buckets], 1) ip.Encode(&header.IPv4Fields{ IHL: header.IPv4MinimumSize, TotalLength: length, @@ -376,13 +378,12 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBu // Set the packet ID when zero. if ip.ID() == 0 { - id := uint32(0) - 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) + // RFC 6864 section 4.3 mandates uniqueness of ID values for + // non-atomic datagrams, so assign an ID to all such datagrams + // according to the definition given in RFC 6864 section 4. + if ip.Flags()&header.IPv4FlagDontFragment == 0 || ip.Flags()&header.IPv4FlagMoreFragments != 0 || ip.FragmentOffset() > 0 { + ip.SetID(uint16(atomic.AddUint32(&e.protocol.ids[hashRoute(r, 0 /* protocol */, e.protocol.hashIV)%buckets], 1))) } - ip.SetID(uint16(id)) } // Always set the checksum. diff --git a/pkg/tcpip/network/ipv4/ipv4_test.go b/pkg/tcpip/network/ipv4/ipv4_test.go index 11e579c4b..4f82c45e2 100644 --- a/pkg/tcpip/network/ipv4/ipv4_test.go +++ b/pkg/tcpip/network/ipv4/ipv4_test.go @@ -519,6 +519,11 @@ func TestReceiveFragments(t *testing.T) { // UDP header plus a payload of 0..256 in increments of 2. ipv4Payload2 := udpGen(128, 2) udpPayload2 := ipv4Payload2[header.UDPMinimumSize:] + // UDP header plus a payload of 0..256 in increments of 3. + // Used to test cases where the fragment blocks are not a multiple of + // the fragment block size of 8 (RFC 791 section 3.1 page 14). + ipv4Payload3 := udpGen(127, 3) + udpPayload3 := ipv4Payload3[header.UDPMinimumSize:] type fragmentData struct { id uint16 @@ -545,6 +550,18 @@ func TestReceiveFragments(t *testing.T) { expectedPayloads: [][]byte{udpPayload1}, }, { + name: "No fragmentation with size not a multiple of fragment block size", + fragments: []fragmentData{ + { + id: 1, + flags: 0, + fragmentOffset: 0, + payload: ipv4Payload3, + }, + }, + expectedPayloads: [][]byte{udpPayload3}, + }, + { name: "More fragments without payload", fragments: []fragmentData{ { @@ -587,6 +604,42 @@ func TestReceiveFragments(t *testing.T) { expectedPayloads: [][]byte{udpPayload1}, }, { + name: "Two fragments with last fragment size not a multiple of fragment block size", + fragments: []fragmentData{ + { + id: 1, + flags: header.IPv4FlagMoreFragments, + fragmentOffset: 0, + payload: ipv4Payload3[:64], + }, + { + id: 1, + flags: 0, + fragmentOffset: 64, + payload: ipv4Payload3[64:], + }, + }, + expectedPayloads: [][]byte{udpPayload3}, + }, + { + name: "Two fragments with first fragment size not a multiple of fragment block size", + fragments: []fragmentData{ + { + id: 1, + flags: header.IPv4FlagMoreFragments, + fragmentOffset: 0, + payload: ipv4Payload3[:63], + }, + { + id: 1, + flags: 0, + fragmentOffset: 63, + payload: ipv4Payload3[63:], + }, + }, + expectedPayloads: nil, + }, + { name: "Second fragment has MoreFlags set", fragments: []fragmentData{ { diff --git a/pkg/tcpip/network/ipv6/BUILD b/pkg/tcpip/network/ipv6/BUILD index 3f71fc520..feada63dc 100644 --- a/pkg/tcpip/network/ipv6/BUILD +++ b/pkg/tcpip/network/ipv6/BUILD @@ -39,6 +39,6 @@ go_test( "//pkg/tcpip/transport/icmp", "//pkg/tcpip/transport/udp", "//pkg/waiter", - "@com_github_google_go-cmp//cmp:go_default_library", + "@com_github_google_go_cmp//cmp:go_default_library", ], ) diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index 2ff7eedf4..24600d877 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -128,6 +128,8 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme } pkt.Data.TrimFront(header.ICMPv6DstUnreachableMinimumSize) switch header.ICMPv6(hdr).Code() { + case header.ICMPv6NetworkUnreachable: + e.handleControl(stack.ControlNetworkUnreachable, 0, pkt) case header.ICMPv6PortUnreachable: e.handleControl(stack.ControlPortUnreachable, 0, pkt) } @@ -494,8 +496,6 @@ const ( icmpV6LengthOffset = 25 ) -var broadcastMAC = tcpip.LinkAddress([]byte{0xff, 0xff, 0xff, 0xff, 0xff, 0xff}) - var _ stack.LinkAddressResolver = (*protocol)(nil) // LinkAddressProtocol implements stack.LinkAddressResolver. @@ -504,7 +504,7 @@ func (*protocol) LinkAddressProtocol() tcpip.NetworkProtocolNumber { } // LinkAddressRequest implements stack.LinkAddressResolver. -func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, linkEP stack.LinkEndpoint) *tcpip.Error { +func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, remoteLinkAddr tcpip.LinkAddress, linkEP stack.LinkEndpoint) *tcpip.Error { snaddr := header.SolicitedNodeAddr(addr) // TODO(b/148672031): Use stack.FindRoute instead of manually creating the @@ -513,8 +513,12 @@ func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, linkEP stack. r := &stack.Route{ LocalAddress: localAddr, RemoteAddress: snaddr, - RemoteLinkAddress: header.EthernetAddressFromMulticastIPv6Address(snaddr), + RemoteLinkAddress: remoteLinkAddr, + } + if len(r.RemoteLinkAddress) == 0 { + r.RemoteLinkAddress = header.EthernetAddressFromMulticastIPv6Address(snaddr) } + hdr := buffer.NewPrependable(int(linkEP.MaxHeaderLength()) + header.IPv6MinimumSize + header.ICMPv6NeighborAdvertSize) pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborAdvertSize)) pkt.SetType(header.ICMPv6NeighborSolicit) diff --git a/pkg/tcpip/network/ipv6/icmp_test.go b/pkg/tcpip/network/ipv6/icmp_test.go index 52a01b44e..f86aaed1d 100644 --- a/pkg/tcpip/network/ipv6/icmp_test.go +++ b/pkg/tcpip/network/ipv6/icmp_test.go @@ -34,6 +34,9 @@ const ( linkAddr0 = tcpip.LinkAddress("\x02\x02\x03\x04\x05\x06") linkAddr1 = tcpip.LinkAddress("\x0a\x0b\x0c\x0d\x0e\x0e") linkAddr2 = tcpip.LinkAddress("\x0a\x0b\x0c\x0d\x0e\x0f") + + defaultChannelSize = 1 + defaultMTU = 65536 ) var ( @@ -257,8 +260,7 @@ func newTestContext(t *testing.T) *testContext { }), } - const defaultMTU = 65536 - c.linkEP0 = channel.New(256, defaultMTU, linkAddr0) + c.linkEP0 = channel.New(defaultChannelSize, defaultMTU, linkAddr0) wrappedEP0 := stack.LinkEndpoint(endpointWithResolutionCapability{LinkEndpoint: c.linkEP0}) if testing.Verbose() { @@ -271,7 +273,7 @@ func newTestContext(t *testing.T) *testContext { t.Fatalf("AddAddress lladdr0: %v", err) } - c.linkEP1 = channel.New(256, defaultMTU, linkAddr1) + c.linkEP1 = channel.New(defaultChannelSize, defaultMTU, linkAddr1) wrappedEP1 := stack.LinkEndpoint(endpointWithResolutionCapability{LinkEndpoint: c.linkEP1}) if err := c.s1.CreateNIC(1, wrappedEP1); err != nil { t.Fatalf("CreateNIC failed: %v", err) @@ -951,3 +953,47 @@ func TestICMPChecksumValidationWithPayloadMultipleViews(t *testing.T) { }) } } + +func TestLinkAddressRequest(t *testing.T) { + snaddr := header.SolicitedNodeAddr(lladdr0) + mcaddr := header.EthernetAddressFromMulticastIPv6Address(snaddr) + + tests := []struct { + name string + remoteLinkAddr tcpip.LinkAddress + expectLinkAddr tcpip.LinkAddress + }{ + { + name: "Unicast", + remoteLinkAddr: linkAddr1, + expectLinkAddr: linkAddr1, + }, + { + name: "Multicast", + remoteLinkAddr: "", + expectLinkAddr: mcaddr, + }, + } + + for _, test := range tests { + p := NewProtocol() + linkRes, ok := p.(stack.LinkAddressResolver) + if !ok { + t.Fatalf("expected IPv6 protocol to implement stack.LinkAddressResolver") + } + + linkEP := channel.New(defaultChannelSize, defaultMTU, linkAddr0) + if err := linkRes.LinkAddressRequest(lladdr0, lladdr1, test.remoteLinkAddr, linkEP); err != nil { + t.Errorf("got p.LinkAddressRequest(%s, %s, %s, _) = %s", lladdr0, lladdr1, test.remoteLinkAddr, err) + } + + pkt, ok := linkEP.Read() + if !ok { + t.Fatal("expected to send a link address request") + } + + if got, want := pkt.Route.RemoteLinkAddress, test.expectLinkAddr; got != want { + t.Errorf("got pkt.Route.RemoteLinkAddress = %s, want = %s", got, want) + } + } +} diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index 95fbcf2d1..5483ae4ee 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -467,7 +467,7 @@ func (p *protocol) NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWi linkEP: linkEP, linkAddrCache: linkAddrCache, dispatcher: dispatcher, - fragmentation: fragmentation.NewFragmentation(fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, fragmentation.DefaultReassembleTimeout), + fragmentation: fragmentation.NewFragmentation(header.IPv6FragmentExtHdrFragmentOffsetBytesPerUnit, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, fragmentation.DefaultReassembleTimeout), protocol: p, }, nil } diff --git a/pkg/tcpip/network/ipv6/ipv6_test.go b/pkg/tcpip/network/ipv6/ipv6_test.go index 213ff64f2..84bac14ff 100644 --- a/pkg/tcpip/network/ipv6/ipv6_test.go +++ b/pkg/tcpip/network/ipv6/ipv6_test.go @@ -678,13 +678,18 @@ type fragmentData struct { } func TestReceiveIPv6Fragments(t *testing.T) { - const nicID = 1 - const udpPayload1Length = 256 - const udpPayload2Length = 128 - const fragmentExtHdrLen = 8 - // Note, not all routing extension headers will be 8 bytes but this test - // uses 8 byte routing extension headers for most sub tests. - const routingExtHdrLen = 8 + const ( + nicID = 1 + udpPayload1Length = 256 + udpPayload2Length = 128 + // Used to test cases where the fragment blocks are not a multiple of + // the fragment block size of 8 (RFC 8200 section 4.5). + udpPayload3Length = 127 + fragmentExtHdrLen = 8 + // Note, not all routing extension headers will be 8 bytes but this test + // uses 8 byte routing extension headers for most sub tests. + routingExtHdrLen = 8 + ) udpGen := func(payload []byte, multiplier uint8) buffer.View { payloadLen := len(payload) @@ -716,6 +721,10 @@ func TestReceiveIPv6Fragments(t *testing.T) { udpPayload2 := udpPayload2Buf[:] ipv6Payload2 := udpGen(udpPayload2, 2) + var udpPayload3Buf [udpPayload3Length]byte + udpPayload3 := udpPayload3Buf[:] + ipv6Payload3 := udpGen(udpPayload3, 3) + tests := []struct { name string expectedPayload []byte @@ -751,6 +760,24 @@ func TestReceiveIPv6Fragments(t *testing.T) { expectedPayloads: [][]byte{udpPayload1}, }, { + name: "Atomic fragment with size not a multiple of fragment block size", + fragments: []fragmentData{ + { + nextHdr: fragmentExtHdrID, + data: buffer.NewVectorisedView( + fragmentExtHdrLen+len(ipv6Payload3), + []buffer.View{ + // Fragment extension header. + buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 0, 0, 0, 0, 0}), + + ipv6Payload3, + }, + ), + }, + }, + expectedPayloads: [][]byte{udpPayload3}, + }, + { name: "Two fragments", fragments: []fragmentData{ { @@ -785,6 +812,74 @@ func TestReceiveIPv6Fragments(t *testing.T) { expectedPayloads: [][]byte{udpPayload1}, }, { + name: "Two fragments with last fragment size not a multiple of fragment block size", + fragments: []fragmentData{ + { + nextHdr: fragmentExtHdrID, + data: buffer.NewVectorisedView( + fragmentExtHdrLen+64, + []buffer.View{ + // Fragment extension header. + // + // Fragment offset = 0, More = true, ID = 1 + buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 1, 0, 0, 0, 1}), + + ipv6Payload3[:64], + }, + ), + }, + { + nextHdr: fragmentExtHdrID, + data: buffer.NewVectorisedView( + fragmentExtHdrLen+len(ipv6Payload3)-64, + []buffer.View{ + // Fragment extension header. + // + // Fragment offset = 8, More = false, ID = 1 + buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 64, 0, 0, 0, 1}), + + ipv6Payload3[64:], + }, + ), + }, + }, + expectedPayloads: [][]byte{udpPayload3}, + }, + { + name: "Two fragments with first fragment size not a multiple of fragment block size", + fragments: []fragmentData{ + { + nextHdr: fragmentExtHdrID, + data: buffer.NewVectorisedView( + fragmentExtHdrLen+63, + []buffer.View{ + // Fragment extension header. + // + // Fragment offset = 0, More = true, ID = 1 + buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 1, 0, 0, 0, 1}), + + ipv6Payload3[:63], + }, + ), + }, + { + nextHdr: fragmentExtHdrID, + data: buffer.NewVectorisedView( + fragmentExtHdrLen+len(ipv6Payload3)-63, + []buffer.View{ + // Fragment extension header. + // + // Fragment offset = 8, More = false, ID = 1 + buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 64, 0, 0, 0, 1}), + + ipv6Payload3[63:], + }, + ), + }, + }, + expectedPayloads: nil, + }, + { name: "Two fragments with different IDs", fragments: []fragmentData{ { diff --git a/pkg/tcpip/ports/ports.go b/pkg/tcpip/ports/ports.go index edc29ad27..f6d592eb5 100644 --- a/pkg/tcpip/ports/ports.go +++ b/pkg/tcpip/ports/ports.go @@ -52,6 +52,9 @@ type Flags struct { // // LoadBalanced takes precidence over MostRecent. LoadBalanced bool + + // TupleOnly represents TCP SO_REUSEADDR. + TupleOnly bool } // Bits converts the Flags to their bitset form. @@ -63,6 +66,9 @@ func (f Flags) Bits() BitFlags { if f.LoadBalanced { rf |= LoadBalancedFlag } + if f.TupleOnly { + rf |= TupleOnlyFlag + } return rf } @@ -98,6 +104,9 @@ const ( // LoadBalancedFlag represents Flags.LoadBalanced. LoadBalancedFlag + // TupleOnlyFlag represents Flags.TupleOnly. + TupleOnlyFlag + // nextFlag is the value that the next added flag will have. // // It is used to calculate FlagMask below. It is also the number of @@ -106,6 +115,10 @@ const ( // FlagMask is a bit mask for BitFlags. FlagMask = nextFlag - 1 + + // MultiBindFlagMask contains the flags that allow binding the same + // tuple multiple times. + MultiBindFlagMask = MostRecentFlag | LoadBalancedFlag ) // ToFlags converts the bitset into a Flags struct. @@ -113,6 +126,7 @@ func (f BitFlags) ToFlags() Flags { return Flags{ MostRecent: f&MostRecentFlag != 0, LoadBalanced: f&LoadBalancedFlag != 0, + TupleOnly: f&TupleOnlyFlag != 0, } } @@ -175,9 +189,54 @@ func (c FlagCounter) IntersectionRefs() BitFlags { return intersection } +type destination struct { + addr tcpip.Address + port uint16 +} + +func makeDestination(a tcpip.FullAddress) destination { + return destination{ + a.Addr, + a.Port, + } +} + +// portNode is never empty. When it has no elements, it is removed from the +// map that references it. +type portNode map[destination]FlagCounter + +// intersectionRefs calculates the intersection of flag bit values which affect +// the specified destination. +// +// If no destinations are present, all flag values are returned as there are no +// entries to limit possible flag values of a new entry. +// +// In addition to the intersection, the number of intersecting refs is +// returned. +func (p portNode) intersectionRefs(dst destination) (BitFlags, int) { + intersection := FlagMask + var count int + + for d, f := range p { + if d == dst { + intersection &= f.IntersectionRefs() + count++ + continue + } + // Wildcard destinations affect all destinations for TupleOnly. + if d.addr == anyIPAddress || dst.addr == anyIPAddress { + // Only bitwise and the TupleOnlyFlag. + intersection &= ((^TupleOnlyFlag) | f.IntersectionRefs()) + count++ + } + } + + return intersection, count +} + // deviceNode is never empty. When it has no elements, it is removed from the // map that references it. -type deviceNode map[tcpip.NICID]FlagCounter +type deviceNode map[tcpip.NICID]portNode // isAvailable checks whether binding is possible by device. If not binding to a // device, check against all FlagCounters. If binding to a specific device, check @@ -186,17 +245,15 @@ type deviceNode map[tcpip.NICID]FlagCounter // If either of the port reuse flags is enabled on any of the nodes, all nodes // sharing a port must share at least one reuse flag. This matches Linux's // behavior. -func (d deviceNode) isAvailable(flags Flags, bindToDevice tcpip.NICID) bool { +func (d deviceNode) isAvailable(flags Flags, bindToDevice tcpip.NICID, dst destination) bool { flagBits := flags.Bits() if bindToDevice == 0 { - // Trying to binding all devices. - if flagBits == 0 { - // Can't bind because the (addr,port) is already bound. - return false - } intersection := FlagMask for _, p := range d { - i := p.IntersectionRefs() + i, c := p.intersectionRefs(dst) + if c == 0 { + continue + } intersection &= i if intersection&flagBits == 0 { // Can't bind because the (addr,port) was @@ -210,16 +267,17 @@ func (d deviceNode) isAvailable(flags Flags, bindToDevice tcpip.NICID) bool { intersection := FlagMask if p, ok := d[0]; ok { - intersection = p.IntersectionRefs() - if intersection&flagBits == 0 { + var c int + intersection, c = p.intersectionRefs(dst) + if c > 0 && intersection&flagBits == 0 { return false } } if p, ok := d[bindToDevice]; ok { - i := p.IntersectionRefs() + i, c := p.intersectionRefs(dst) intersection &= i - if intersection&flagBits == 0 { + if c > 0 && intersection&flagBits == 0 { return false } } @@ -233,12 +291,12 @@ type bindAddresses map[tcpip.Address]deviceNode // isAvailable checks whether an IP address is available to bind to. If the // address is the "any" address, check all other addresses. Otherwise, just // check against the "any" address and the provided address. -func (b bindAddresses) isAvailable(addr tcpip.Address, flags Flags, bindToDevice tcpip.NICID) bool { +func (b bindAddresses) isAvailable(addr tcpip.Address, flags Flags, bindToDevice tcpip.NICID, dst destination) bool { if addr == anyIPAddress { // If binding to the "any" address then check that there are no conflicts // with all addresses. for _, d := range b { - if !d.isAvailable(flags, bindToDevice) { + if !d.isAvailable(flags, bindToDevice, dst) { return false } } @@ -247,14 +305,14 @@ func (b bindAddresses) isAvailable(addr tcpip.Address, flags Flags, bindToDevice // Check that there is no conflict with the "any" address. if d, ok := b[anyIPAddress]; ok { - if !d.isAvailable(flags, bindToDevice) { + if !d.isAvailable(flags, bindToDevice, dst) { return false } } // Check that this is no conflict with the provided address. if d, ok := b[addr]; ok { - if !d.isAvailable(flags, bindToDevice) { + if !d.isAvailable(flags, bindToDevice, dst) { return false } } @@ -320,17 +378,17 @@ func (s *PortManager) pickEphemeralPort(offset, count uint32, testPort func(p ui } // IsPortAvailable tests if the given port is available on all given protocols. -func (s *PortManager) IsPortAvailable(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, flags Flags, bindToDevice tcpip.NICID) bool { +func (s *PortManager) IsPortAvailable(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, flags Flags, bindToDevice tcpip.NICID, dest tcpip.FullAddress) bool { s.mu.Lock() defer s.mu.Unlock() - return s.isPortAvailableLocked(networks, transport, addr, port, flags, bindToDevice) + return s.isPortAvailableLocked(networks, transport, addr, port, flags, bindToDevice, makeDestination(dest)) } -func (s *PortManager) isPortAvailableLocked(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, flags Flags, bindToDevice tcpip.NICID) bool { +func (s *PortManager) isPortAvailableLocked(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, flags Flags, bindToDevice tcpip.NICID, dst destination) bool { for _, network := range networks { desc := portDescriptor{network, transport, port} if addrs, ok := s.allocatedPorts[desc]; ok { - if !addrs.isAvailable(addr, flags, bindToDevice) { + if !addrs.isAvailable(addr, flags, bindToDevice, dst) { return false } } @@ -342,14 +400,16 @@ func (s *PortManager) isPortAvailableLocked(networks []tcpip.NetworkProtocolNumb // reserved by another endpoint. If port is zero, ReservePort will search for // an unreserved ephemeral port and reserve it, returning its value in the // "port" return value. -func (s *PortManager) ReservePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, flags Flags, bindToDevice tcpip.NICID) (reservedPort uint16, err *tcpip.Error) { +func (s *PortManager) ReservePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, flags Flags, bindToDevice tcpip.NICID, dest tcpip.FullAddress) (reservedPort uint16, err *tcpip.Error) { s.mu.Lock() defer s.mu.Unlock() + dst := makeDestination(dest) + // If a port is specified, just try to reserve it for all network // protocols. if port != 0 { - if !s.reserveSpecificPort(networks, transport, addr, port, flags, bindToDevice) { + if !s.reserveSpecificPort(networks, transport, addr, port, flags, bindToDevice, dst) { return 0, tcpip.ErrPortInUse } return port, nil @@ -357,15 +417,16 @@ func (s *PortManager) ReservePort(networks []tcpip.NetworkProtocolNumber, transp // A port wasn't specified, so try to find one. return s.PickEphemeralPort(func(p uint16) (bool, *tcpip.Error) { - return s.reserveSpecificPort(networks, transport, addr, p, flags, bindToDevice), nil + return s.reserveSpecificPort(networks, transport, addr, p, flags, bindToDevice, dst), nil }) } // reserveSpecificPort tries to reserve the given port on all given protocols. -func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, flags Flags, bindToDevice tcpip.NICID) bool { - if !s.isPortAvailableLocked(networks, transport, addr, port, flags, bindToDevice) { +func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, flags Flags, bindToDevice tcpip.NICID, dst destination) bool { + if !s.isPortAvailableLocked(networks, transport, addr, port, flags, bindToDevice, dst) { return false } + flagBits := flags.Bits() // Reserve port on all network protocols. @@ -381,9 +442,65 @@ func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber d = make(deviceNode) m[addr] = d } - n := d[bindToDevice] + p := d[bindToDevice] + if p == nil { + p = make(portNode) + } + n := p[dst] n.AddRef(flagBits) - d[bindToDevice] = n + p[dst] = n + d[bindToDevice] = p + } + + return true +} + +// ReserveTuple adds a port reservation for the tuple on all given protocol. +func (s *PortManager) ReserveTuple(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, flags Flags, bindToDevice tcpip.NICID, dest tcpip.FullAddress) bool { + flagBits := flags.Bits() + dst := makeDestination(dest) + + s.mu.Lock() + defer s.mu.Unlock() + + // It is easier to undo the entire reservation, so if we find that the + // tuple can't be fully added, finish and undo the whole thing. + undo := false + + // Reserve port on all network protocols. + for _, network := range networks { + desc := portDescriptor{network, transport, port} + m, ok := s.allocatedPorts[desc] + if !ok { + m = make(bindAddresses) + s.allocatedPorts[desc] = m + } + d, ok := m[addr] + if !ok { + d = make(deviceNode) + m[addr] = d + } + p := d[bindToDevice] + if p == nil { + p = make(portNode) + } + + n := p[dst] + if n.TotalRefs() != 0 && n.IntersectionRefs()&flagBits == 0 { + // Tuple already exists. + undo = true + } + n.AddRef(flagBits) + p[dst] = n + d[bindToDevice] = p + } + + if undo { + // releasePortLocked decrements the counts (rather than setting + // them to zero), so it will undo the incorrect incrementing + // above. + s.releasePortLocked(networks, transport, addr, port, flagBits, bindToDevice, dst) + return false } return true @@ -391,12 +508,14 @@ func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber // ReleasePort releases the reservation on a port/IP combination so that it can // be reserved by other endpoints. -func (s *PortManager) ReleasePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, flags Flags, bindToDevice tcpip.NICID) { +func (s *PortManager) ReleasePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, flags Flags, bindToDevice tcpip.NICID, dest tcpip.FullAddress) { s.mu.Lock() defer s.mu.Unlock() - flagBits := flags.Bits() + s.releasePortLocked(networks, transport, addr, port, flags.Bits(), bindToDevice, makeDestination(dest)) +} +func (s *PortManager) releasePortLocked(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, flags BitFlags, bindToDevice tcpip.NICID, dst destination) { for _, network := range networks { desc := portDescriptor{network, transport, port} if m, ok := s.allocatedPorts[desc]; ok { @@ -404,21 +523,32 @@ func (s *PortManager) ReleasePort(networks []tcpip.NetworkProtocolNumber, transp if !ok { continue } - n, ok := d[bindToDevice] + p, ok := d[bindToDevice] if !ok { continue } - n.refs[flagBits]-- - d[bindToDevice] = n - if n.TotalRefs() == 0 { - delete(d, bindToDevice) + n, ok := p[dst] + if !ok { + continue + } + n.DropRef(flags) + if n.TotalRefs() > 0 { + p[dst] = n + continue } - if len(d) == 0 { - delete(m, addr) + delete(p, dst) + if len(p) > 0 { + continue + } + delete(d, bindToDevice) + if len(d) > 0 { + continue } - if len(m) == 0 { - delete(s.allocatedPorts, desc) + delete(m, addr) + if len(m) > 0 { + continue } + delete(s.allocatedPorts, desc) } } } diff --git a/pkg/tcpip/ports/ports_test.go b/pkg/tcpip/ports/ports_test.go index d6969d050..58db5868c 100644 --- a/pkg/tcpip/ports/ports_test.go +++ b/pkg/tcpip/ports/ports_test.go @@ -36,6 +36,7 @@ type portReserveTestAction struct { flags Flags release bool device tcpip.NICID + dest tcpip.FullAddress } func TestPortReservation(t *testing.T) { @@ -272,6 +273,54 @@ func TestPortReservation(t *testing.T) { {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true, LoadBalanced: true}, want: nil}, {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true}, want: tcpip.ErrPortInUse}, }, + }, { + tname: "bind tuple with reuseaddr, and then wildcard with reuseaddr", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, flags: Flags{TupleOnly: true}, dest: tcpip.FullAddress{Addr: fakeIPAddress, Port: 24}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{TupleOnly: true}, dest: tcpip.FullAddress{}, want: nil}, + }, + }, { + tname: "bind tuple with reuseaddr, and then wildcard", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, flags: Flags{TupleOnly: true}, dest: tcpip.FullAddress{Addr: fakeIPAddress, Port: 24}, want: nil}, + {port: 24, ip: fakeIPAddress, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind wildcard with reuseaddr, and then tuple with reuseaddr", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, flags: Flags{TupleOnly: true}, dest: tcpip.FullAddress{}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{TupleOnly: true}, dest: tcpip.FullAddress{Addr: fakeIPAddress, Port: 24}, want: nil}, + }, + }, { + tname: "bind tuple with reuseaddr, and then wildcard", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{TupleOnly: true}, dest: tcpip.FullAddress{Addr: fakeIPAddress, Port: 24}, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind two tuples with reuseaddr", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, flags: Flags{TupleOnly: true}, dest: tcpip.FullAddress{Addr: fakeIPAddress, Port: 24}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{TupleOnly: true}, dest: tcpip.FullAddress{Addr: fakeIPAddress, Port: 25}, want: nil}, + }, + }, { + tname: "bind two tuples", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, dest: tcpip.FullAddress{Addr: fakeIPAddress, Port: 24}, want: nil}, + {port: 24, ip: fakeIPAddress, dest: tcpip.FullAddress{Addr: fakeIPAddress, Port: 25}, want: nil}, + }, + }, { + tname: "bind wildcard, and then tuple with reuseaddr", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, dest: tcpip.FullAddress{}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{TupleOnly: true}, dest: tcpip.FullAddress{Addr: fakeIPAddress, Port: 24}, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind wildcard twice with reuseaddr", + actions: []portReserveTestAction{ + {port: 24, ip: anyIPAddress, flags: Flags{TupleOnly: true}, want: nil}, + {port: 24, ip: anyIPAddress, flags: Flags{TupleOnly: true}, want: nil}, + }, }, } { t.Run(test.tname, func(t *testing.T) { @@ -280,19 +329,18 @@ func TestPortReservation(t *testing.T) { for _, test := range test.actions { if test.release { - pm.ReleasePort(net, fakeTransNumber, test.ip, test.port, test.flags, test.device) + pm.ReleasePort(net, fakeTransNumber, test.ip, test.port, test.flags, test.device, test.dest) continue } - gotPort, err := pm.ReservePort(net, fakeTransNumber, test.ip, test.port, test.flags, test.device) + gotPort, err := pm.ReservePort(net, fakeTransNumber, test.ip, test.port, test.flags, test.device, test.dest) if err != test.want { - t.Fatalf("ReservePort(.., .., %s, %d, %+v, %d) = %v, want %v", test.ip, test.port, test.flags, test.device, err, test.want) + t.Fatalf("ReservePort(.., .., %s, %d, %+v, %d, %v) = %v, want %v", test.ip, test.port, test.flags, test.device, test.dest, err, test.want) } if test.port == 0 && (gotPort == 0 || gotPort < FirstEphemeral) { - t.Fatalf("ReservePort(.., .., .., 0) = %d, want port number >= %d to be picked", gotPort, FirstEphemeral) + t.Fatalf("ReservePort(.., .., .., 0, ..) = %d, want port number >= %d to be picked", gotPort, FirstEphemeral) } } }) - } } diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD index 24f52b735..1c58bed2d 100644 --- a/pkg/tcpip/stack/BUILD +++ b/pkg/tcpip/stack/BUILD @@ -16,6 +16,18 @@ go_template_instance( ) go_template_instance( + name = "neighbor_entry_list", + out = "neighbor_entry_list.go", + package = "stack", + prefix = "neighborEntry", + template = "//pkg/ilist:generic_list", + types = { + "Element": "*neighborEntry", + "Linker": "*neighborEntry", + }, +) + +go_template_instance( name = "packet_buffer_list", out = "packet_buffer_list.go", package = "stack", @@ -27,6 +39,18 @@ go_template_instance( }, ) +go_template_instance( + name = "tuple_list", + out = "tuple_list.go", + package = "stack", + prefix = "tuple", + template = "//pkg/ilist:generic_list", + types = { + "Element": "*tuple", + "Linker": "*tuple", + }, +) + go_library( name = "stack", srcs = [ @@ -35,12 +59,18 @@ go_library( "forwarder.go", "icmp_rate_limit.go", "iptables.go", + "iptables_state.go", "iptables_targets.go", "iptables_types.go", "linkaddrcache.go", "linkaddrentry_list.go", "ndp.go", + "neighbor_cache.go", + "neighbor_entry.go", + "neighbor_entry_list.go", + "neighborstate_string.go", "nic.go", + "nud.go", "packet_buffer.go", "packet_buffer_list.go", "rand.go", @@ -48,7 +78,9 @@ go_library( "route.go", "stack.go", "stack_global_state.go", + "stack_options.go", "transport_demuxer.go", + "tuple_list.go", ], visibility = ["//visibility:public"], deps = [ @@ -74,10 +106,12 @@ go_test( size = "medium", srcs = [ "ndp_test.go", + "nud_test.go", "stack_test.go", "transport_demuxer_test.go", "transport_test.go", ], + shard_count = 20, deps = [ ":stack", "//pkg/rand", @@ -93,7 +127,8 @@ go_test( "//pkg/tcpip/transport/icmp", "//pkg/tcpip/transport/udp", "//pkg/waiter", - "@com_github_google_go-cmp//cmp:go_default_library", + "@com_github_google_go_cmp//cmp:go_default_library", + "@com_github_google_go_cmp//cmp/cmpopts:go_default_library", ], ) @@ -101,8 +136,11 @@ go_test( name = "stack_test", size = "small", srcs = [ + "fake_time_test.go", "forwarder_test.go", "linkaddrcache_test.go", + "neighbor_cache_test.go", + "neighbor_entry_test.go", "nic_test.go", ], library = ":stack", @@ -112,5 +150,8 @@ go_test( "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/header", + "@com_github_dpjacques_clockwork//:go_default_library", + "@com_github_google_go_cmp//cmp:go_default_library", + "@com_github_google_go_cmp//cmp/cmpopts:go_default_library", ], ) diff --git a/pkg/tcpip/stack/conntrack.go b/pkg/tcpip/stack/conntrack.go index 05bf62788..559a1c4dd 100644 --- a/pkg/tcpip/stack/conntrack.go +++ b/pkg/tcpip/stack/conntrack.go @@ -26,280 +26,321 @@ import ( ) // Connection tracking is used to track and manipulate packets for NAT rules. -// The connection is created for a packet if it does not exist. Every connection -// contains two tuples (original and reply). The tuples are manipulated if there -// is a matching NAT rule. The packet is modified by looking at the tuples in the -// Prerouting and Output hooks. +// The connection is created for a packet if it does not exist. Every +// connection contains two tuples (original and reply). The tuples are +// manipulated if there is a matching NAT rule. The packet is modified by +// looking at the tuples in the Prerouting and Output hooks. +// +// Currently, only TCP tracking is supported. + +// Our hash table has 16K buckets. +// TODO(gvisor.dev/issue/170): These should be tunable. +const numBuckets = 1 << 14 // Direction of the tuple. -type ctDirection int +type direction int const ( - dirOriginal ctDirection = iota + dirOriginal direction = iota dirReply ) -// Status of connection. -// TODO(gvisor.dev/issue/170): Add other states of connection. -type connStatus int - -const ( - connNew connStatus = iota - connEstablished -) - // Manipulation type for the connection. type manipType int const ( - manipDstPrerouting manipType = iota + manipNone manipType = iota + manipDstPrerouting manipDstOutput ) -// connTrackMutable is the manipulatable part of the tuple. -type connTrackMutable struct { - // addr is source address of the tuple. - addr tcpip.Address - - // port is source port of the tuple. - port uint16 - - // protocol is network layer protocol. - protocol tcpip.NetworkProtocolNumber -} - -// connTrackImmutable is the non-manipulatable part of the tuple. -type connTrackImmutable struct { - // addr is destination address of the tuple. - addr tcpip.Address +// tuple holds a connection's identifying and manipulating data in one +// direction. It is immutable. +// +// +stateify savable +type tuple struct { + // tupleEntry is used to build an intrusive list of tuples. + tupleEntry - // direction is direction (original or reply) of the tuple. - direction ctDirection + tupleID - // port is destination port of the tuple. - port uint16 + // conn is the connection tracking entry this tuple belongs to. + conn *conn - // protocol is transport layer protocol. - protocol tcpip.TransportProtocolNumber + // direction is the direction of the tuple. + direction direction } -// connTrackTuple represents the tuple which is created from the -// packet. -type connTrackTuple struct { - // dst is non-manipulatable part of the tuple. - dst connTrackImmutable - - // src is manipulatable part of the tuple. - src connTrackMutable +// tupleID uniquely identifies a connection in one direction. It currently +// contains enough information to distinguish between any TCP or UDP +// connection, and will need to be extended to support other protocols. +// +// +stateify savable +type tupleID struct { + srcAddr tcpip.Address + srcPort uint16 + dstAddr tcpip.Address + dstPort uint16 + transProto tcpip.TransportProtocolNumber + netProto tcpip.NetworkProtocolNumber } -// connTrackTupleHolder is the container of tuple and connection. -type ConnTrackTupleHolder struct { - // conn is pointer to the connection tracking entry. - conn *connTrack - - // tuple is original or reply tuple. - tuple connTrackTuple +// reply creates the reply tupleID. +func (ti tupleID) reply() tupleID { + return tupleID{ + srcAddr: ti.dstAddr, + srcPort: ti.dstPort, + dstAddr: ti.srcAddr, + dstPort: ti.srcPort, + transProto: ti.transProto, + netProto: ti.netProto, + } } -// connTrack is the connection. -type connTrack struct { - // originalTupleHolder contains tuple in original direction. - originalTupleHolder ConnTrackTupleHolder - - // replyTupleHolder contains tuple in reply direction. - replyTupleHolder ConnTrackTupleHolder - - // status indicates connection is new or established. - status connStatus +// conn is a tracked connection. +// +// +stateify savable +type conn struct { + // original is the tuple in original direction. It is immutable. + original tuple - // timeout indicates the time connection should be active. - timeout time.Duration + // reply is the tuple in reply direction. It is immutable. + reply tuple - // manip indicates if the packet should be manipulated. + // manip indicates if the packet should be manipulated. It is immutable. manip manipType - // tcb is TCB control block. It is used to keep track of states - // of tcp connection. - tcb tcpconntrack.TCB - // tcbHook indicates if the packet is inbound or outbound to - // update the state of tcb. + // update the state of tcb. It is immutable. tcbHook Hook + + // mu protects all mutable state. + mu sync.Mutex `state:"nosave"` + // tcb is TCB control block. It is used to keep track of states + // of tcp connection and is protected by mu. + tcb tcpconntrack.TCB + // lastUsed is the last time the connection saw a relevant packet, and + // is updated by each packet on the connection. It is protected by mu. + lastUsed time.Time `state:".(unixTime)"` } -// ConnTrackTable contains a map of all existing connections created for -// NAT rules. -type ConnTrackTable struct { - // connMu protects connTrackTable. - connMu sync.RWMutex +// timedOut returns whether the connection timed out based on its state. +func (cn *conn) timedOut(now time.Time) bool { + const establishedTimeout = 5 * 24 * time.Hour + const defaultTimeout = 120 * time.Second + cn.mu.Lock() + defer cn.mu.Unlock() + if cn.tcb.State() == tcpconntrack.ResultAlive { + // Use the same default as Linux, which doesn't delete + // established connections for 5(!) days. + return now.Sub(cn.lastUsed) > establishedTimeout + } + // Use the same default as Linux, which lets connections in most states + // other than established remain for <= 120 seconds. + return now.Sub(cn.lastUsed) > defaultTimeout +} - // connTrackTable maintains a map of tuples needed for connection tracking - // for iptables NAT rules. The key for the map is an integer calculated - // using seed, source address, destination address, source port and - // destination port. - CtMap map[uint32]ConnTrackTupleHolder +// update the connection tracking state. +// +// Precondition: ct.mu must be held. +func (ct *conn) updateLocked(tcpHeader header.TCP, hook Hook) { + // Update the state of tcb. tcb assumes it's always initialized on the + // client. However, we only need to know whether the connection is + // established or not, so the client/server distinction isn't important. + // TODO(gvisor.dev/issue/170): Add support in tcpconntrack to handle + // other tcp states. + if ct.tcb.IsEmpty() { + ct.tcb.Init(tcpHeader) + } else if hook == ct.tcbHook { + ct.tcb.UpdateStateOutbound(tcpHeader) + } else { + ct.tcb.UpdateStateInbound(tcpHeader) + } +} +// ConnTrack tracks all connections created for NAT rules. Most users are +// expected to only call handlePacket, insertRedirectConn, and maybeInsertNoop. +// +// ConnTrack keeps all connections in a slice of buckets, each of which holds a +// linked list of tuples. This gives us some desirable properties: +// - Each bucket has its own lock, lessening lock contention. +// - The slice is large enough that lists stay short (<10 elements on average). +// Thus traversal is fast. +// - During linked list traversal we reap expired connections. This amortizes +// the cost of reaping them and makes reapUnused faster. +// +// Locks are ordered by their location in the buckets slice. That is, a +// goroutine that locks buckets[i] can only lock buckets[j] s.t. i < j. +// +// +stateify savable +type ConnTrack struct { // seed is a one-time random value initialized at stack startup - // and is used in calculation of hash key for connection tracking - // table. - Seed uint32 + // and is used in the calculation of hash keys for the list of buckets. + // It is immutable. + seed uint32 + + // mu protects the buckets slice, but not buckets' contents. Only take + // the write lock if you are modifying the slice or saving for S/R. + mu sync.RWMutex `state:"nosave"` + + // buckets is protected by mu. + buckets []bucket } -// packetToTuple converts packet to a tuple in original direction. -func packetToTuple(pkt *PacketBuffer, hook Hook) (connTrackTuple, *tcpip.Error) { - var tuple connTrackTuple +// +stateify savable +type bucket struct { + // mu protects tuples. + mu sync.Mutex `state:"nosave"` + tuples tupleList +} - netHeader := header.IPv4(pkt.NetworkHeader) +// packetToTupleID converts packet to a tuple ID. It fails when pkt lacks a valid +// TCP header. +func packetToTupleID(pkt *PacketBuffer) (tupleID, *tcpip.Error) { // TODO(gvisor.dev/issue/170): Need to support for other // protocols as well. + netHeader := header.IPv4(pkt.NetworkHeader) if netHeader == nil || netHeader.TransportProtocol() != header.TCPProtocolNumber { - return tuple, tcpip.ErrUnknownProtocol + return tupleID{}, tcpip.ErrUnknownProtocol } tcpHeader := header.TCP(pkt.TransportHeader) if tcpHeader == nil { - return tuple, tcpip.ErrUnknownProtocol + return tupleID{}, tcpip.ErrUnknownProtocol } - tuple.src.addr = netHeader.SourceAddress() - tuple.src.port = tcpHeader.SourcePort() - tuple.src.protocol = header.IPv4ProtocolNumber - - tuple.dst.addr = netHeader.DestinationAddress() - tuple.dst.port = tcpHeader.DestinationPort() - tuple.dst.protocol = netHeader.TransportProtocol() - - return tuple, nil + return tupleID{ + srcAddr: netHeader.SourceAddress(), + srcPort: tcpHeader.SourcePort(), + dstAddr: netHeader.DestinationAddress(), + dstPort: tcpHeader.DestinationPort(), + transProto: netHeader.TransportProtocol(), + netProto: header.IPv4ProtocolNumber, + }, nil } -// getReplyTuple creates reply tuple for the given tuple. -func getReplyTuple(tuple connTrackTuple) connTrackTuple { - var replyTuple connTrackTuple - replyTuple.src.addr = tuple.dst.addr - replyTuple.src.port = tuple.dst.port - replyTuple.src.protocol = tuple.src.protocol - replyTuple.dst.addr = tuple.src.addr - replyTuple.dst.port = tuple.src.port - replyTuple.dst.protocol = tuple.dst.protocol - replyTuple.dst.direction = dirReply - - return replyTuple +// newConn creates new connection. +func newConn(orig, reply tupleID, manip manipType, hook Hook) *conn { + conn := conn{ + manip: manip, + tcbHook: hook, + lastUsed: time.Now(), + } + conn.original = tuple{conn: &conn, tupleID: orig} + conn.reply = tuple{conn: &conn, tupleID: reply, direction: dirReply} + return &conn } -// makeNewConn creates new connection. -func makeNewConn(tuple, replyTuple connTrackTuple) connTrack { - var conn connTrack - conn.status = connNew - conn.originalTupleHolder.tuple = tuple - conn.originalTupleHolder.conn = &conn - conn.replyTupleHolder.tuple = replyTuple - conn.replyTupleHolder.conn = &conn +// connFor gets the conn for pkt if it exists, or returns nil +// if it does not. It returns an error when pkt does not contain a valid TCP +// header. +// TODO(gvisor.dev/issue/170): Only TCP packets are supported. Need to support +// other transport protocols. +func (ct *ConnTrack) connFor(pkt *PacketBuffer) (*conn, direction) { + tid, err := packetToTupleID(pkt) + if err != nil { + return nil, dirOriginal + } - return conn -} + bucket := ct.bucket(tid) + now := time.Now() + + ct.mu.RLock() + defer ct.mu.RUnlock() + ct.buckets[bucket].mu.Lock() + defer ct.buckets[bucket].mu.Unlock() + + // Iterate over the tuples in a bucket, cleaning up any unused + // connections we find. + for other := ct.buckets[bucket].tuples.Front(); other != nil; other = other.Next() { + // Clean up any timed-out connections we happen to find. + if ct.reapTupleLocked(other, bucket, now) { + // The tuple expired. + continue + } + if tid == other.tupleID { + return other.conn, other.direction + } + } -// getTupleHash returns hash of the tuple. The fields used for -// generating hash are seed (generated once for stack), source address, -// destination address, source port and destination ports. -func (ct *ConnTrackTable) getTupleHash(tuple connTrackTuple) uint32 { - h := jenkins.Sum32(ct.Seed) - h.Write([]byte(tuple.src.addr)) - h.Write([]byte(tuple.dst.addr)) - portBuf := make([]byte, 2) - binary.LittleEndian.PutUint16(portBuf, tuple.src.port) - h.Write([]byte(portBuf)) - binary.LittleEndian.PutUint16(portBuf, tuple.dst.port) - h.Write([]byte(portBuf)) - - return h.Sum32() + return nil, dirOriginal } -// connTrackForPacket returns connTrack for packet. -// TODO(gvisor.dev/issue/170): Only TCP packets are supported. Need to support other -// transport protocols. -func (ct *ConnTrackTable) connTrackForPacket(pkt *PacketBuffer, hook Hook, createConn bool) (*connTrack, ctDirection) { - var dir ctDirection - tuple, err := packetToTuple(pkt, hook) +func (ct *ConnTrack) insertRedirectConn(pkt *PacketBuffer, hook Hook, rt RedirectTarget) *conn { + tid, err := packetToTupleID(pkt) if err != nil { - return nil, dir - } - - ct.connMu.Lock() - defer ct.connMu.Unlock() - - connTrackTable := ct.CtMap - hash := ct.getTupleHash(tuple) - - var conn *connTrack - switch createConn { - case true: - // If connection does not exist for the hash, create a new - // connection. - replyTuple := getReplyTuple(tuple) - replyHash := ct.getTupleHash(replyTuple) - newConn := makeNewConn(tuple, replyTuple) - conn = &newConn - - // Add tupleHolders to the map. - // TODO(gvisor.dev/issue/170): Need to support collisions using linked list. - ct.CtMap[hash] = conn.originalTupleHolder - ct.CtMap[replyHash] = conn.replyTupleHolder - default: - tupleHolder, ok := connTrackTable[hash] - if !ok { - return nil, dir - } - - // If this is the reply of new connection, set the connection - // status as ESTABLISHED. - conn = tupleHolder.conn - if conn.status == connNew && tupleHolder.tuple.dst.direction == dirReply { - conn.status = connEstablished - } - if tupleHolder.conn == nil { - panic("tupleHolder has null connection tracking entry") - } + return nil + } + if hook != Prerouting && hook != Output { + return nil + } - dir = tupleHolder.tuple.dst.direction + // Create a new connection and change the port as per the iptables + // rule. This tuple will be used to manipulate the packet in + // handlePacket. + replyTID := tid.reply() + replyTID.srcAddr = rt.MinIP + replyTID.srcPort = rt.MinPort + var manip manipType + switch hook { + case Prerouting: + manip = manipDstPrerouting + case Output: + manip = manipDstOutput } - return conn, dir + conn := newConn(tid, replyTID, manip, hook) + ct.insertConn(conn) + return conn } -// SetNatInfo will manipulate the tuples according to iptables NAT rules. -func (ct *ConnTrackTable) SetNatInfo(pkt *PacketBuffer, rt RedirectTarget, hook Hook) { - // Get the connection. Connection is always created before this - // function is called. - conn, _ := ct.connTrackForPacket(pkt, hook, false) - if conn == nil { - panic("connection should be created to manipulate tuples.") +// insertConn inserts conn into the appropriate table bucket. +func (ct *ConnTrack) insertConn(conn *conn) { + // Lock the buckets in the correct order. + tupleBucket := ct.bucket(conn.original.tupleID) + replyBucket := ct.bucket(conn.reply.tupleID) + ct.mu.RLock() + defer ct.mu.RUnlock() + if tupleBucket < replyBucket { + ct.buckets[tupleBucket].mu.Lock() + ct.buckets[replyBucket].mu.Lock() + } else if tupleBucket > replyBucket { + ct.buckets[replyBucket].mu.Lock() + ct.buckets[tupleBucket].mu.Lock() + } else { + // Both tuples are in the same bucket. + ct.buckets[tupleBucket].mu.Lock() } - replyTuple := conn.replyTupleHolder.tuple - replyHash := ct.getTupleHash(replyTuple) - // TODO(gvisor.dev/issue/170): Support only redirect of ports. Need to - // support changing of address for Prerouting. - - // Change the port as per the iptables rule. This tuple will be used - // to manipulate the packet in HandlePacket. - conn.replyTupleHolder.tuple.src.addr = rt.MinIP - conn.replyTupleHolder.tuple.src.port = rt.MinPort - newHash := ct.getTupleHash(conn.replyTupleHolder.tuple) + // Now that we hold the locks, ensure the tuple hasn't been inserted by + // another thread. + alreadyInserted := false + for other := ct.buckets[tupleBucket].tuples.Front(); other != nil; other = other.Next() { + if other.tupleID == conn.original.tupleID { + alreadyInserted = true + break + } + } - // Add the changed tuple to the map. - ct.connMu.Lock() - defer ct.connMu.Unlock() - ct.CtMap[newHash] = conn.replyTupleHolder - if hook == Output { - conn.replyTupleHolder.conn.manip = manipDstOutput + if !alreadyInserted { + // Add the tuple to the map. + ct.buckets[tupleBucket].tuples.PushFront(&conn.original) + ct.buckets[replyBucket].tuples.PushFront(&conn.reply) } - // Delete the old tuple. - delete(ct.CtMap, replyHash) + // Unlocking can happen in any order. + ct.buckets[tupleBucket].mu.Unlock() + if tupleBucket != replyBucket { + ct.buckets[replyBucket].mu.Unlock() + } } // handlePacketPrerouting manipulates ports for packets in Prerouting hook. -// TODO(gvisor.dev/issue/170): Change address for Prerouting hook.. -func handlePacketPrerouting(pkt *PacketBuffer, conn *connTrack, dir ctDirection) { +// TODO(gvisor.dev/issue/170): Change address for Prerouting hook. +func handlePacketPrerouting(pkt *PacketBuffer, conn *conn, dir direction) { + // If this is a noop entry, don't do anything. + if conn.manip == manipNone { + return + } + netHeader := header.IPv4(pkt.NetworkHeader) tcpHeader := header.TCP(pkt.TransportHeader) @@ -308,21 +349,31 @@ func handlePacketPrerouting(pkt *PacketBuffer, conn *connTrack, dir ctDirection) // modified. switch dir { case dirOriginal: - port := conn.replyTupleHolder.tuple.src.port + port := conn.reply.srcPort tcpHeader.SetDestinationPort(port) - netHeader.SetDestinationAddress(conn.replyTupleHolder.tuple.src.addr) + netHeader.SetDestinationAddress(conn.reply.srcAddr) case dirReply: - port := conn.originalTupleHolder.tuple.dst.port + port := conn.original.dstPort tcpHeader.SetSourcePort(port) - netHeader.SetSourceAddress(conn.originalTupleHolder.tuple.dst.addr) + netHeader.SetSourceAddress(conn.original.dstAddr) } + // TODO(gvisor.dev/issue/170): TCP checksums aren't usually validated + // on inbound packets, so we don't recalculate them. However, we should + // support cases when they are validated, e.g. when we can't offload + // receive checksumming. + netHeader.SetChecksum(0) netHeader.SetChecksum(^netHeader.CalculateChecksum()) } // handlePacketOutput manipulates ports for packets in Output hook. -func handlePacketOutput(pkt *PacketBuffer, conn *connTrack, gso *GSO, r *Route, dir ctDirection) { +func handlePacketOutput(pkt *PacketBuffer, conn *conn, gso *GSO, r *Route, dir direction) { + // If this is a noop entry, don't do anything. + if conn.manip == manipNone { + return + } + netHeader := header.IPv4(pkt.NetworkHeader) tcpHeader := header.TCP(pkt.TransportHeader) @@ -331,13 +382,13 @@ func handlePacketOutput(pkt *PacketBuffer, conn *connTrack, gso *GSO, r *Route, // modified. For prerouting redirection, we only reach this point // when replying, so packet sources are modified. if conn.manip == manipDstOutput && dir == dirOriginal { - port := conn.replyTupleHolder.tuple.src.port + port := conn.reply.srcPort tcpHeader.SetDestinationPort(port) - netHeader.SetDestinationAddress(conn.replyTupleHolder.tuple.src.addr) + netHeader.SetDestinationAddress(conn.reply.srcAddr) } else { - port := conn.originalTupleHolder.tuple.dst.port + port := conn.original.dstPort tcpHeader.SetSourcePort(port) - netHeader.SetSourceAddress(conn.originalTupleHolder.tuple.dst.addr) + netHeader.SetSourceAddress(conn.original.dstAddr) } // Calculate the TCP checksum and set it. @@ -356,33 +407,32 @@ func handlePacketOutput(pkt *PacketBuffer, conn *connTrack, gso *GSO, r *Route, netHeader.SetChecksum(^netHeader.CalculateChecksum()) } -// HandlePacket will manipulate the port and address of the packet if the -// connection exists. -func (ct *ConnTrackTable) HandlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r *Route) { +// handlePacket will manipulate the port and address of the packet if the +// connection exists. Returns whether, after the packet traverses the tables, +// it should create a new entry in the table. +func (ct *ConnTrack) handlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r *Route) bool { if pkt.NatDone { - return + return false } if hook != Prerouting && hook != Output { - return + return false } - conn, dir := ct.connTrackForPacket(pkt, hook, false) - // Connection or Rule not found for the packet. - if conn == nil { - return + // TODO(gvisor.dev/issue/170): Support other transport protocols. + if pkt.NetworkHeader == nil || header.IPv4(pkt.NetworkHeader).TransportProtocol() != header.TCPProtocolNumber { + return false } - netHeader := header.IPv4(pkt.NetworkHeader) - // TODO(gvisor.dev/issue/170): Need to support for other transport - // protocols as well. - if netHeader == nil || netHeader.TransportProtocol() != header.TCPProtocolNumber { - return + conn, dir := ct.connFor(pkt) + // Connection or Rule not found for the packet. + if conn == nil { + return true } tcpHeader := header.TCP(pkt.TransportHeader) if tcpHeader == nil { - return + return false } switch hook { @@ -396,39 +446,161 @@ func (ct *ConnTrackTable) HandlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r // Update the state of tcb. // TODO(gvisor.dev/issue/170): Add support in tcpcontrack to handle // other tcp states. - var st tcpconntrack.Result - if conn.tcb.IsEmpty() { - conn.tcb.Init(tcpHeader) - conn.tcbHook = hook - } else { - switch hook { - case conn.tcbHook: - st = conn.tcb.UpdateStateOutbound(tcpHeader) - default: - st = conn.tcb.UpdateStateInbound(tcpHeader) - } + conn.mu.Lock() + defer conn.mu.Unlock() + + // Mark the connection as having been used recently so it isn't reaped. + conn.lastUsed = time.Now() + // Update connection state. + conn.updateLocked(header.TCP(pkt.TransportHeader), hook) + + return false +} + +// maybeInsertNoop tries to insert a no-op connection entry to keep connections +// from getting clobbered when replies arrive. It only inserts if there isn't +// already a connection for pkt. +// +// This should be called after traversing iptables rules only, to ensure that +// pkt.NatDone is set correctly. +func (ct *ConnTrack) maybeInsertNoop(pkt *PacketBuffer, hook Hook) { + // If there were a rule applying to this packet, it would be marked + // with NatDone. + if pkt.NatDone { + return } - // Delete conntrack if tcp connection is closed. - if st == tcpconntrack.ResultClosedByPeer || st == tcpconntrack.ResultClosedBySelf || st == tcpconntrack.ResultReset { - ct.deleteConnTrack(conn) + // We only track TCP connections. + if pkt.NetworkHeader == nil || header.IPv4(pkt.NetworkHeader).TransportProtocol() != header.TCPProtocolNumber { + return } -} -// deleteConnTrack deletes the connection. -func (ct *ConnTrackTable) deleteConnTrack(conn *connTrack) { - if conn == nil { + // This is the first packet we're seeing for the TCP connection. Insert + // the noop entry (an identity mapping) so that the response doesn't + // get NATed, breaking the connection. + tid, err := packetToTupleID(pkt) + if err != nil { return } + conn := newConn(tid, tid.reply(), manipNone, hook) + conn.updateLocked(header.TCP(pkt.TransportHeader), hook) + ct.insertConn(conn) +} - tuple := conn.originalTupleHolder.tuple - hash := ct.getTupleHash(tuple) - replyTuple := conn.replyTupleHolder.tuple - replyHash := ct.getTupleHash(replyTuple) +// bucket gets the conntrack bucket for a tupleID. +func (ct *ConnTrack) bucket(id tupleID) int { + h := jenkins.Sum32(ct.seed) + h.Write([]byte(id.srcAddr)) + h.Write([]byte(id.dstAddr)) + shortBuf := make([]byte, 2) + binary.LittleEndian.PutUint16(shortBuf, id.srcPort) + h.Write([]byte(shortBuf)) + binary.LittleEndian.PutUint16(shortBuf, id.dstPort) + h.Write([]byte(shortBuf)) + binary.LittleEndian.PutUint16(shortBuf, uint16(id.transProto)) + h.Write([]byte(shortBuf)) + binary.LittleEndian.PutUint16(shortBuf, uint16(id.netProto)) + h.Write([]byte(shortBuf)) + ct.mu.RLock() + defer ct.mu.RUnlock() + return int(h.Sum32()) % len(ct.buckets) +} - ct.connMu.Lock() - defer ct.connMu.Unlock() +// reapUnused deletes timed out entries from the conntrack map. The rules for +// reaping are: +// - Most reaping occurs in connFor, which is called on each packet. connFor +// cleans up the bucket the packet's connection maps to. Thus calls to +// reapUnused should be fast. +// - Each call to reapUnused traverses a fraction of the conntrack table. +// Specifically, it traverses len(ct.buckets)/fractionPerReaping. +// - After reaping, reapUnused decides when it should next run based on the +// ratio of expired connections to examined connections. If the ratio is +// greater than maxExpiredPct, it schedules the next run quickly. Otherwise it +// slightly increases the interval between runs. +// - maxFullTraversal caps the time it takes to traverse the entire table. +// +// reapUnused returns the next bucket that should be checked and the time after +// which it should be called again. +func (ct *ConnTrack) reapUnused(start int, prevInterval time.Duration) (int, time.Duration) { + // TODO(gvisor.dev/issue/170): This can be more finely controlled, as + // it is in Linux via sysctl. + const fractionPerReaping = 128 + const maxExpiredPct = 50 + const maxFullTraversal = 60 * time.Second + const minInterval = 10 * time.Millisecond + const maxInterval = maxFullTraversal / fractionPerReaping + + now := time.Now() + checked := 0 + expired := 0 + var idx int + ct.mu.RLock() + defer ct.mu.RUnlock() + for i := 0; i < len(ct.buckets)/fractionPerReaping; i++ { + idx = (i + start) % len(ct.buckets) + ct.buckets[idx].mu.Lock() + for tuple := ct.buckets[idx].tuples.Front(); tuple != nil; tuple = tuple.Next() { + checked++ + if ct.reapTupleLocked(tuple, idx, now) { + expired++ + } + } + ct.buckets[idx].mu.Unlock() + } + // We already checked buckets[idx]. + idx++ + + // If half or more of the connections are expired, the table has gotten + // stale. Reschedule quickly. + expiredPct := 0 + if checked != 0 { + expiredPct = expired * 100 / checked + } + if expiredPct > maxExpiredPct { + return idx, minInterval + } + if interval := prevInterval + minInterval; interval <= maxInterval { + // Increment the interval between runs. + return idx, interval + } + // We've hit the maximum interval. + return idx, maxInterval +} + +// reapTupleLocked tries to remove tuple and its reply from the table. It +// returns whether the tuple's connection has timed out. +// +// Preconditions: ct.mu is locked for reading and bucket is locked. +func (ct *ConnTrack) reapTupleLocked(tuple *tuple, bucket int, now time.Time) bool { + if !tuple.conn.timedOut(now) { + return false + } + + // To maintain lock order, we can only reap these tuples if the reply + // appears later in the table. + replyBucket := ct.bucket(tuple.reply()) + if bucket > replyBucket { + return true + } + + // Don't re-lock if both tuples are in the same bucket. + differentBuckets := bucket != replyBucket + if differentBuckets { + ct.buckets[replyBucket].mu.Lock() + } + + // We have the buckets locked and can remove both tuples. + if tuple.direction == dirOriginal { + ct.buckets[replyBucket].tuples.Remove(&tuple.conn.reply) + } else { + ct.buckets[replyBucket].tuples.Remove(&tuple.conn.original) + } + ct.buckets[bucket].tuples.Remove(tuple) + + // Don't re-unlock if both tuples are in the same bucket. + if differentBuckets { + ct.buckets[replyBucket].mu.Unlock() + } - delete(ct.CtMap, hash) - delete(ct.CtMap, replyHash) + return true } diff --git a/pkg/tcpip/stack/fake_time_test.go b/pkg/tcpip/stack/fake_time_test.go new file mode 100644 index 000000000..92c8cb534 --- /dev/null +++ b/pkg/tcpip/stack/fake_time_test.go @@ -0,0 +1,209 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package stack + +import ( + "container/heap" + "sync" + "time" + + "github.com/dpjacques/clockwork" + "gvisor.dev/gvisor/pkg/tcpip" +) + +type fakeClock struct { + clock clockwork.FakeClock + + // mu protects the fields below. + mu sync.RWMutex + + // times is min-heap of times. A heap is used for quick retrieval of the next + // upcoming time of scheduled work. + times *timeHeap + + // waitGroups stores one WaitGroup for all work scheduled to execute at the + // same time via AfterFunc. This allows parallel execution of all functions + // passed to AfterFunc scheduled for the same time. + waitGroups map[time.Time]*sync.WaitGroup +} + +func newFakeClock() *fakeClock { + return &fakeClock{ + clock: clockwork.NewFakeClock(), + times: &timeHeap{}, + waitGroups: make(map[time.Time]*sync.WaitGroup), + } +} + +var _ tcpip.Clock = (*fakeClock)(nil) + +// NowNanoseconds implements tcpip.Clock.NowNanoseconds. +func (fc *fakeClock) NowNanoseconds() int64 { + return fc.clock.Now().UnixNano() +} + +// NowMonotonic implements tcpip.Clock.NowMonotonic. +func (fc *fakeClock) NowMonotonic() int64 { + return fc.NowNanoseconds() +} + +// AfterFunc implements tcpip.Clock.AfterFunc. +func (fc *fakeClock) AfterFunc(d time.Duration, f func()) tcpip.Timer { + until := fc.clock.Now().Add(d) + wg := fc.addWait(until) + return &fakeTimer{ + clock: fc, + until: until, + timer: fc.clock.AfterFunc(d, func() { + defer wg.Done() + f() + }), + } +} + +// addWait adds an additional wait to the WaitGroup for parallel execution of +// all work scheduled for t. Returns a reference to the WaitGroup modified. +func (fc *fakeClock) addWait(t time.Time) *sync.WaitGroup { + fc.mu.RLock() + wg, ok := fc.waitGroups[t] + fc.mu.RUnlock() + + if ok { + wg.Add(1) + return wg + } + + fc.mu.Lock() + heap.Push(fc.times, t) + fc.mu.Unlock() + + wg = &sync.WaitGroup{} + wg.Add(1) + + fc.mu.Lock() + fc.waitGroups[t] = wg + fc.mu.Unlock() + + return wg +} + +// removeWait removes a wait from the WaitGroup for parallel execution of all +// work scheduled for t. +func (fc *fakeClock) removeWait(t time.Time) { + fc.mu.RLock() + defer fc.mu.RUnlock() + + wg := fc.waitGroups[t] + wg.Done() +} + +// advance executes all work that have been scheduled to execute within d from +// the current fake time. Blocks until all work has completed execution. +func (fc *fakeClock) advance(d time.Duration) { + // Block until all the work is done + until := fc.clock.Now().Add(d) + for { + fc.mu.Lock() + if fc.times.Len() == 0 { + fc.mu.Unlock() + return + } + + t := heap.Pop(fc.times).(time.Time) + if t.After(until) { + // No work to do + heap.Push(fc.times, t) + fc.mu.Unlock() + return + } + fc.mu.Unlock() + + diff := t.Sub(fc.clock.Now()) + fc.clock.Advance(diff) + + fc.mu.RLock() + wg := fc.waitGroups[t] + fc.mu.RUnlock() + + wg.Wait() + + fc.mu.Lock() + delete(fc.waitGroups, t) + fc.mu.Unlock() + } +} + +type fakeTimer struct { + clock *fakeClock + timer clockwork.Timer + + mu sync.RWMutex + until time.Time +} + +var _ tcpip.Timer = (*fakeTimer)(nil) + +// Reset implements tcpip.Timer.Reset. +func (ft *fakeTimer) Reset(d time.Duration) { + if !ft.timer.Reset(d) { + return + } + + ft.mu.Lock() + defer ft.mu.Unlock() + + ft.clock.removeWait(ft.until) + ft.until = ft.clock.clock.Now().Add(d) + ft.clock.addWait(ft.until) +} + +// Stop implements tcpip.Timer.Stop. +func (ft *fakeTimer) Stop() bool { + if !ft.timer.Stop() { + return false + } + + ft.mu.RLock() + defer ft.mu.RUnlock() + + ft.clock.removeWait(ft.until) + return true +} + +type timeHeap []time.Time + +var _ heap.Interface = (*timeHeap)(nil) + +func (h timeHeap) Len() int { + return len(h) +} + +func (h timeHeap) Less(i, j int) bool { + return h[i].Before(h[j]) +} + +func (h timeHeap) Swap(i, j int) { + h[i], h[j] = h[j], h[i] +} + +func (h *timeHeap) Push(x interface{}) { + *h = append(*h, x.(time.Time)) +} + +func (h *timeHeap) Pop() interface{} { + last := (*h)[len(*h)-1] + *h = (*h)[:len(*h)-1] + return last +} diff --git a/pkg/tcpip/stack/forwarder_test.go b/pkg/tcpip/stack/forwarder_test.go index a6546cef0..c962693f5 100644 --- a/pkg/tcpip/stack/forwarder_test.go +++ b/pkg/tcpip/stack/forwarder_test.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" ) const ( @@ -120,10 +121,12 @@ func (*fwdTestNetworkEndpoint) Close() {} type fwdTestNetworkProtocol struct { addrCache *linkAddrCache addrResolveDelay time.Duration - onLinkAddressResolved func(cache *linkAddrCache, addr tcpip.Address) + onLinkAddressResolved func(cache *linkAddrCache, addr tcpip.Address, _ tcpip.LinkAddress) onResolveStaticAddress func(tcpip.Address) (tcpip.LinkAddress, bool) } +var _ LinkAddressResolver = (*fwdTestNetworkProtocol)(nil) + func (f *fwdTestNetworkProtocol) Number() tcpip.NetworkProtocolNumber { return fwdTestNetNumber } @@ -173,10 +176,10 @@ func (f *fwdTestNetworkProtocol) Close() {} func (f *fwdTestNetworkProtocol) Wait() {} -func (f *fwdTestNetworkProtocol) LinkAddressRequest(addr, localAddr tcpip.Address, linkEP LinkEndpoint) *tcpip.Error { +func (f *fwdTestNetworkProtocol) LinkAddressRequest(addr, localAddr tcpip.Address, remoteLinkAddr tcpip.LinkAddress, linkEP LinkEndpoint) *tcpip.Error { if f.addrCache != nil && f.onLinkAddressResolved != nil { time.AfterFunc(f.addrResolveDelay, func() { - f.onLinkAddressResolved(f.addrCache, addr) + f.onLinkAddressResolved(f.addrCache, addr, remoteLinkAddr) }) } return nil @@ -301,6 +304,16 @@ func (e *fwdTestLinkEndpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Er // Wait implements stack.LinkEndpoint.Wait. func (*fwdTestLinkEndpoint) Wait() {} +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. +func (*fwdTestLinkEndpoint) ARPHardwareType() header.ARPHardwareType { + panic("not implemented") +} + +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *fwdTestLinkEndpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { + panic("not implemented") +} + func fwdTestNetFactory(t *testing.T, proto *fwdTestNetworkProtocol) (ep1, ep2 *fwdTestLinkEndpoint) { // Create a stack with the network protocol and two NICs. s := New(Options{ @@ -394,7 +407,7 @@ func TestForwardingWithFakeResolver(t *testing.T) { // Create a network protocol with a fake resolver. proto := &fwdTestNetworkProtocol{ addrResolveDelay: 500 * time.Millisecond, - onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address) { + onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address, _ tcpip.LinkAddress) { // Any address will be resolved to the link address "c". cache.add(tcpip.FullAddress{NIC: 2, Addr: addr}, "c") }, @@ -452,7 +465,7 @@ func TestForwardingWithFakeResolverPartialTimeout(t *testing.T) { // Create a network protocol with a fake resolver. proto := &fwdTestNetworkProtocol{ addrResolveDelay: 500 * time.Millisecond, - onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address) { + onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address, _ tcpip.LinkAddress) { // Only packets to address 3 will be resolved to the // link address "c". if addr == "\x03" { @@ -504,7 +517,7 @@ func TestForwardingWithFakeResolverTwoPackets(t *testing.T) { // Create a network protocol with a fake resolver. proto := &fwdTestNetworkProtocol{ addrResolveDelay: 500 * time.Millisecond, - onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address) { + onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address, _ tcpip.LinkAddress) { // Any packets will be resolved to the link address "c". cache.add(tcpip.FullAddress{NIC: 2, Addr: addr}, "c") }, @@ -548,7 +561,7 @@ func TestForwardingWithFakeResolverManyPackets(t *testing.T) { // Create a network protocol with a fake resolver. proto := &fwdTestNetworkProtocol{ addrResolveDelay: 500 * time.Millisecond, - onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address) { + onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address, _ tcpip.LinkAddress) { // Any packets will be resolved to the link address "c". cache.add(tcpip.FullAddress{NIC: 2, Addr: addr}, "c") }, @@ -605,7 +618,7 @@ func TestForwardingWithFakeResolverManyResolutions(t *testing.T) { // Create a network protocol with a fake resolver. proto := &fwdTestNetworkProtocol{ addrResolveDelay: 500 * time.Millisecond, - onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address) { + onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address, _ tcpip.LinkAddress) { // Any packets will be resolved to the link address "c". cache.add(tcpip.FullAddress{NIC: 2, Addr: addr}, "c") }, diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go index 62d4eb1b6..cbbae4224 100644 --- a/pkg/tcpip/stack/iptables.go +++ b/pkg/tcpip/stack/iptables.go @@ -16,39 +16,49 @@ package stack import ( "fmt" + "time" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/header" ) -// Table names. +// tableID is an index into IPTables.tables. +type tableID int + const ( - TablenameNat = "nat" - TablenameMangle = "mangle" - TablenameFilter = "filter" + natID tableID = iota + mangleID + filterID + numTables ) -// Chain names as defined by net/ipv4/netfilter/ip_tables.c. +// Table names. const ( - ChainNamePrerouting = "PREROUTING" - ChainNameInput = "INPUT" - ChainNameForward = "FORWARD" - ChainNameOutput = "OUTPUT" - ChainNamePostrouting = "POSTROUTING" + NATTable = "nat" + MangleTable = "mangle" + FilterTable = "filter" ) +// nameToID is immutable. +var nameToID = map[string]tableID{ + NATTable: natID, + MangleTable: mangleID, + FilterTable: filterID, +} + // HookUnset indicates that there is no hook set for an entrypoint or // underflow. const HookUnset = -1 +// reaperDelay is how long to wait before starting to reap connections. +const reaperDelay = 5 * time.Second + // DefaultTables returns a default set of tables. Each chain is set to accept // all packets. func DefaultTables() *IPTables { - // TODO(gvisor.dev/issue/170): We may be able to swap out some strings for - // iotas. return &IPTables{ - tables: map[string]Table{ - TablenameNat: Table{ + tables: [numTables]Table{ + natID: Table{ Rules: []Rule{ Rule{Target: AcceptTarget{}}, Rule{Target: AcceptTarget{}}, @@ -56,65 +66,71 @@ func DefaultTables() *IPTables { Rule{Target: AcceptTarget{}}, Rule{Target: ErrorTarget{}}, }, - BuiltinChains: map[Hook]int{ + BuiltinChains: [NumHooks]int{ Prerouting: 0, Input: 1, + Forward: HookUnset, Output: 2, Postrouting: 3, }, - Underflows: map[Hook]int{ + Underflows: [NumHooks]int{ Prerouting: 0, Input: 1, + Forward: HookUnset, Output: 2, Postrouting: 3, }, - UserChains: map[string]int{}, }, - TablenameMangle: Table{ + mangleID: Table{ Rules: []Rule{ Rule{Target: AcceptTarget{}}, Rule{Target: AcceptTarget{}}, Rule{Target: ErrorTarget{}}, }, - BuiltinChains: map[Hook]int{ + BuiltinChains: [NumHooks]int{ Prerouting: 0, Output: 1, }, - Underflows: map[Hook]int{ - Prerouting: 0, - Output: 1, + Underflows: [NumHooks]int{ + Prerouting: 0, + Input: HookUnset, + Forward: HookUnset, + Output: 1, + Postrouting: HookUnset, }, - UserChains: map[string]int{}, }, - TablenameFilter: Table{ + filterID: Table{ Rules: []Rule{ Rule{Target: AcceptTarget{}}, Rule{Target: AcceptTarget{}}, Rule{Target: AcceptTarget{}}, Rule{Target: ErrorTarget{}}, }, - BuiltinChains: map[Hook]int{ - Input: 0, - Forward: 1, - Output: 2, + BuiltinChains: [NumHooks]int{ + Prerouting: HookUnset, + Input: 0, + Forward: 1, + Output: 2, + Postrouting: HookUnset, }, - Underflows: map[Hook]int{ - Input: 0, - Forward: 1, - Output: 2, + Underflows: [NumHooks]int{ + Prerouting: HookUnset, + Input: 0, + Forward: 1, + Output: 2, + Postrouting: HookUnset, }, - UserChains: map[string]int{}, }, }, - priorities: map[Hook][]string{ - Input: []string{TablenameNat, TablenameFilter}, - Prerouting: []string{TablenameMangle, TablenameNat}, - Output: []string{TablenameMangle, TablenameNat, TablenameFilter}, + priorities: [NumHooks][]tableID{ + Prerouting: []tableID{mangleID, natID}, + Input: []tableID{natID, filterID}, + Output: []tableID{mangleID, natID, filterID}, }, - connections: ConnTrackTable{ - CtMap: make(map[uint32]ConnTrackTupleHolder), - Seed: generateRandUint32(), + connections: ConnTrack{ + seed: generateRandUint32(), }, + reaperDone: make(chan struct{}, 1), } } @@ -123,62 +139,59 @@ func DefaultTables() *IPTables { func EmptyFilterTable() Table { return Table{ Rules: []Rule{}, - BuiltinChains: map[Hook]int{ - Input: HookUnset, - Forward: HookUnset, - Output: HookUnset, + BuiltinChains: [NumHooks]int{ + Prerouting: HookUnset, + Postrouting: HookUnset, }, - Underflows: map[Hook]int{ - Input: HookUnset, - Forward: HookUnset, - Output: HookUnset, + Underflows: [NumHooks]int{ + Prerouting: HookUnset, + Postrouting: HookUnset, }, - UserChains: map[string]int{}, } } -// EmptyNatTable returns a Table with no rules and the filter table chains +// EmptyNATTable returns a Table with no rules and the filter table chains // mapped to HookUnset. -func EmptyNatTable() Table { +func EmptyNATTable() Table { return Table{ Rules: []Rule{}, - BuiltinChains: map[Hook]int{ - Prerouting: HookUnset, - Input: HookUnset, - Output: HookUnset, - Postrouting: HookUnset, + BuiltinChains: [NumHooks]int{ + Forward: HookUnset, }, - Underflows: map[Hook]int{ - Prerouting: HookUnset, - Input: HookUnset, - Output: HookUnset, - Postrouting: HookUnset, + Underflows: [NumHooks]int{ + Forward: HookUnset, }, - UserChains: map[string]int{}, } } -// GetTable returns table by name. +// GetTable returns a table by name. func (it *IPTables) GetTable(name string) (Table, bool) { + id, ok := nameToID[name] + if !ok { + return Table{}, false + } it.mu.RLock() defer it.mu.RUnlock() - t, ok := it.tables[name] - return t, ok + return it.tables[id], true } // ReplaceTable replaces or inserts table by name. -func (it *IPTables) ReplaceTable(name string, table Table) { +func (it *IPTables) ReplaceTable(name string, table Table) *tcpip.Error { + id, ok := nameToID[name] + if !ok { + return tcpip.ErrInvalidOptionValue + } it.mu.Lock() defer it.mu.Unlock() + // If iptables is being enabled, initialize the conntrack table and + // reaper. + if !it.modified { + it.connections.buckets = make([]bucket, numBuckets) + it.startReaper(reaperDelay) + } it.modified = true - it.tables[name] = table -} - -// GetPriorities returns slice of priorities associated with hook. -func (it *IPTables) GetPriorities(hook Hook) []string { - it.mu.RLock() - defer it.mu.RUnlock() - return it.priorities[hook] + it.tables[id] = table + return nil } // A chainVerdict is what a table decides should be done with a packet. @@ -213,11 +226,19 @@ func (it *IPTables) Check(hook Hook, pkt *PacketBuffer, gso *GSO, r *Route, addr // Packets are manipulated only if connection and matching // NAT rule exists. - it.connections.HandlePacket(pkt, hook, gso, r) + shouldTrack := it.connections.handlePacket(pkt, hook, gso, r) // Go through each table containing the hook. - for _, tablename := range it.GetPriorities(hook) { - table, _ := it.GetTable(tablename) + it.mu.RLock() + defer it.mu.RUnlock() + priorities := it.priorities[hook] + for _, tableID := range priorities { + // If handlePacket already NATed the packet, we don't need to + // check the NAT table. + if tableID == natID && pkt.NatDone { + continue + } + table := it.tables[tableID] ruleIdx := table.BuiltinChains[hook] switch verdict := it.checkChain(hook, pkt, table, ruleIdx, gso, r, address, nicName); verdict { // If the table returns Accept, move on to the next table. @@ -246,17 +267,59 @@ func (it *IPTables) Check(hook Hook, pkt *PacketBuffer, gso *GSO, r *Route, addr } } + // If this connection should be tracked, try to add an entry for it. If + // traversing the nat table didn't end in adding an entry, + // maybeInsertNoop will add a no-op entry for the connection. This is + // needeed when establishing connections so that the SYN/ACK reply to an + // outgoing SYN is delivered to the correct endpoint rather than being + // redirected by a prerouting rule. + // + // From the iptables documentation: "If there is no rule, a `null' + // binding is created: this usually does not map the packet, but exists + // to ensure we don't map another stream over an existing one." + if shouldTrack { + it.connections.maybeInsertNoop(pkt, hook) + } + // Every table returned Accept. return true } +// beforeSave is invoked by stateify. +func (it *IPTables) beforeSave() { + // Ensure the reaper exits cleanly. + it.reaperDone <- struct{}{} + // Prevent others from modifying the connection table. + it.connections.mu.Lock() +} + +// afterLoad is invoked by stateify. +func (it *IPTables) afterLoad() { + it.startReaper(reaperDelay) +} + +// startReaper starts a goroutine that wakes up periodically to reap timed out +// connections. +func (it *IPTables) startReaper(interval time.Duration) { + go func() { // S/R-SAFE: reaperDone is signalled when iptables is saved. + bucket := 0 + for { + select { + case <-it.reaperDone: + return + case <-time.After(interval): + bucket, interval = it.connections.reapUnused(bucket, interval) + } + } + }() +} + // CheckPackets runs pkts through the rules for hook and returns a map of packets that // should not go forward. // -// Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. -// -// TODO(gvisor.dev/issue/170): pk.NetworkHeader will always be set as a -// precondition. +// Preconditions: +// - pkt is a IPv4 packet of at least length header.IPv4MinimumSize. +// - pkt.NetworkHeader is not nil. // // NOTE: unlike the Check API the returned map contains packets that should be // dropped. @@ -280,9 +343,9 @@ func (it *IPTables) CheckPackets(hook Hook, pkts PacketBufferList, gso *GSO, r * return drop, natPkts } -// Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. -// TODO(gvisor.dev/issue/170): pkt.NetworkHeader will always be set as a -// precondition. +// Preconditions: +// - pkt is a IPv4 packet of at least length header.IPv4MinimumSize. +// - pkt.NetworkHeader is not nil. func (it *IPTables) checkChain(hook Hook, pkt *PacketBuffer, table Table, ruleIdx int, gso *GSO, r *Route, address tcpip.Address, nicName string) chainVerdict { // Start from ruleIdx and walk the list of rules until a rule gives us // a verdict. @@ -327,23 +390,12 @@ func (it *IPTables) checkChain(hook Hook, pkt *PacketBuffer, table Table, ruleId return chainDrop } -// Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. -// TODO(gvisor.dev/issue/170): pkt.NetworkHeader will always be set as a -// precondition. +// Preconditions: +// - pkt is a IPv4 packet of at least length header.IPv4MinimumSize. +// - pkt.NetworkHeader is not nil. func (it *IPTables) checkRule(hook Hook, pkt *PacketBuffer, table Table, ruleIdx int, gso *GSO, r *Route, address tcpip.Address, nicName string) (RuleVerdict, int) { rule := table.Rules[ruleIdx] - // If pkt.NetworkHeader hasn't been set yet, it will be contained in - // pkt.Data. - if pkt.NetworkHeader == nil { - var ok bool - pkt.NetworkHeader, ok = pkt.Data.PullUp(header.IPv4MinimumSize) - if !ok { - // Precondition has been violated. - panic(fmt.Sprintf("iptables checks require IPv4 headers of at least %d bytes", header.IPv4MinimumSize)) - } - } - // Check whether the packet matches the IP header filter. if !rule.Filter.match(header.IPv4(pkt.NetworkHeader), hook, nicName) { // Continue on to the next rule. diff --git a/pkg/tcpip/stack/iptables_state.go b/pkg/tcpip/stack/iptables_state.go new file mode 100644 index 000000000..529e02a07 --- /dev/null +++ b/pkg/tcpip/stack/iptables_state.go @@ -0,0 +1,40 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package stack + +import ( + "time" +) + +// +stateify savable +type unixTime struct { + second int64 + nano int64 +} + +// saveLastUsed is invoked by stateify. +func (cn *conn) saveLastUsed() unixTime { + return unixTime{cn.lastUsed.Unix(), cn.lastUsed.UnixNano()} +} + +// loadLastUsed is invoked by stateify. +func (cn *conn) loadLastUsed(unix unixTime) { + cn.lastUsed = time.Unix(unix.second, unix.nano) +} + +// beforeSave is invoked by stateify. +func (ct *ConnTrack) beforeSave() { + ct.mu.Lock() +} diff --git a/pkg/tcpip/stack/iptables_targets.go b/pkg/tcpip/stack/iptables_targets.go index 92e31643e..dc88033c7 100644 --- a/pkg/tcpip/stack/iptables_targets.go +++ b/pkg/tcpip/stack/iptables_targets.go @@ -24,7 +24,7 @@ import ( type AcceptTarget struct{} // Action implements Target.Action. -func (AcceptTarget) Action(*PacketBuffer, *ConnTrackTable, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { +func (AcceptTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { return RuleAccept, 0 } @@ -32,7 +32,7 @@ func (AcceptTarget) Action(*PacketBuffer, *ConnTrackTable, Hook, *GSO, *Route, t type DropTarget struct{} // Action implements Target.Action. -func (DropTarget) Action(*PacketBuffer, *ConnTrackTable, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { +func (DropTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { return RuleDrop, 0 } @@ -41,7 +41,7 @@ func (DropTarget) Action(*PacketBuffer, *ConnTrackTable, Hook, *GSO, *Route, tcp type ErrorTarget struct{} // Action implements Target.Action. -func (ErrorTarget) Action(*PacketBuffer, *ConnTrackTable, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { +func (ErrorTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { log.Debugf("ErrorTarget triggered.") return RuleDrop, 0 } @@ -52,7 +52,7 @@ type UserChainTarget struct { } // Action implements Target.Action. -func (UserChainTarget) Action(*PacketBuffer, *ConnTrackTable, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { +func (UserChainTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { panic("UserChainTarget should never be called.") } @@ -61,7 +61,7 @@ func (UserChainTarget) Action(*PacketBuffer, *ConnTrackTable, Hook, *GSO, *Route type ReturnTarget struct{} // Action implements Target.Action. -func (ReturnTarget) Action(*PacketBuffer, *ConnTrackTable, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { +func (ReturnTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { return RuleReturn, 0 } @@ -92,7 +92,7 @@ type RedirectTarget struct { // TODO(gvisor.dev/issue/170): Parse headers without copying. The current // implementation only works for PREROUTING and calls pkt.Clone(), neither // of which should be the case. -func (rt RedirectTarget) Action(pkt *PacketBuffer, ct *ConnTrackTable, hook Hook, gso *GSO, r *Route, address tcpip.Address) (RuleVerdict, int) { +func (rt RedirectTarget) Action(pkt *PacketBuffer, ct *ConnTrack, hook Hook, gso *GSO, r *Route, address tcpip.Address) (RuleVerdict, int) { // Packet is already manipulated. if pkt.NatDone { return RuleAccept, 0 @@ -150,12 +150,11 @@ func (rt RedirectTarget) Action(pkt *PacketBuffer, ct *ConnTrackTable, hook Hook return RuleAccept, 0 } - // Set up conection for matching NAT rule. - // Only the first packet of the connection comes here. - // Other packets will be manipulated in connection tracking. - if conn, _ := ct.connTrackForPacket(pkt, hook, true); conn != nil { - ct.SetNatInfo(pkt, rt, hook) - ct.HandlePacket(pkt, hook, gso, r) + // Set up conection for matching NAT rule. Only the first + // packet of the connection comes here. Other packets will be + // manipulated in connection tracking. + if conn := ct.insertRedirectConn(pkt, hook, rt); conn != nil { + ct.handlePacket(pkt, hook, gso, r) } default: return RuleDrop, 0 diff --git a/pkg/tcpip/stack/iptables_types.go b/pkg/tcpip/stack/iptables_types.go index 7026990c4..73274ada9 100644 --- a/pkg/tcpip/stack/iptables_types.go +++ b/pkg/tcpip/stack/iptables_types.go @@ -78,50 +78,55 @@ const ( ) // IPTables holds all the tables for a netstack. +// +// +stateify savable type IPTables struct { // mu protects tables, priorities, and modified. mu sync.RWMutex - // tables maps table names to tables. User tables have arbitrary names. - // mu needs to be locked for accessing. - tables map[string]Table + // tables maps tableIDs to tables. Holds builtin tables only, not user + // tables. mu must be locked for accessing. + tables [numTables]Table // priorities maps each hook to a list of table names. The order of the // list is the order in which each table should be visited for that // hook. mu needs to be locked for accessing. - priorities map[Hook][]string + priorities [NumHooks][]tableID // modified is whether tables have been modified at least once. It is // used to elide the iptables performance overhead for workloads that // don't utilize iptables. modified bool - connections ConnTrackTable + connections ConnTrack + + // reaperDone can be signalled to stop the reaper goroutine. + reaperDone chan struct{} } // A Table defines a set of chains and hooks into the network stack. It is // really just a list of rules. +// +// +stateify savable type Table struct { // Rules holds the rules that make up the table. Rules []Rule // BuiltinChains maps builtin chains to their entrypoint rule in Rules. - BuiltinChains map[Hook]int + BuiltinChains [NumHooks]int // Underflows maps builtin chains to their underflow rule in Rules // (i.e. the rule to execute if the chain returns without a verdict). - Underflows map[Hook]int - - // UserChains holds user-defined chains for the keyed by name. Users - // can give their chains arbitrary names. - UserChains map[string]int + Underflows [NumHooks]int } // ValidHooks returns a bitmap of the builtin hooks for the given table. func (table *Table) ValidHooks() uint32 { hooks := uint32(0) - for hook := range table.BuiltinChains { - hooks |= 1 << hook + for hook, ruleIdx := range table.BuiltinChains { + if ruleIdx != HookUnset { + hooks |= 1 << hook + } } return hooks } @@ -130,6 +135,8 @@ func (table *Table) ValidHooks() uint32 { // contains zero or more matchers, each of which is a specification of which // packets this rule applies to. If there are no matchers in the rule, it // applies to any packet. +// +// +stateify savable type Rule struct { // Filter holds basic IP filtering fields common to every rule. Filter IPHeaderFilter @@ -142,6 +149,8 @@ type Rule struct { } // IPHeaderFilter holds basic IP filtering data common to every rule. +// +// +stateify savable type IPHeaderFilter struct { // Protocol matches the transport protocol. Protocol tcpip.TransportProtocolNumber @@ -249,5 +258,5 @@ type Target interface { // Action takes an action on the packet and returns a verdict on how // traversal should (or should not) continue. If the return value is // Jump, it also returns the index of the rule to jump to. - Action(packet *PacketBuffer, connections *ConnTrackTable, hook Hook, gso *GSO, r *Route, address tcpip.Address) (RuleVerdict, int) + Action(packet *PacketBuffer, connections *ConnTrack, hook Hook, gso *GSO, r *Route, address tcpip.Address) (RuleVerdict, int) } diff --git a/pkg/tcpip/stack/linkaddrcache.go b/pkg/tcpip/stack/linkaddrcache.go index 403557fd7..6f73a0ce4 100644 --- a/pkg/tcpip/stack/linkaddrcache.go +++ b/pkg/tcpip/stack/linkaddrcache.go @@ -244,7 +244,7 @@ func (c *linkAddrCache) startAddressResolution(k tcpip.FullAddress, linkRes Link for i := 0; ; i++ { // Send link request, then wait for the timeout limit and check // whether the request succeeded. - linkRes.LinkAddressRequest(k.Addr, localAddr, linkEP) + linkRes.LinkAddressRequest(k.Addr, localAddr, "" /* linkAddr */, linkEP) select { case now := <-time.After(c.resolutionTimeout): diff --git a/pkg/tcpip/stack/linkaddrcache_test.go b/pkg/tcpip/stack/linkaddrcache_test.go index 1baa498d0..b15b8d1cb 100644 --- a/pkg/tcpip/stack/linkaddrcache_test.go +++ b/pkg/tcpip/stack/linkaddrcache_test.go @@ -48,7 +48,7 @@ type testLinkAddressResolver struct { onLinkAddressRequest func() } -func (r *testLinkAddressResolver) LinkAddressRequest(addr, _ tcpip.Address, _ LinkEndpoint) *tcpip.Error { +func (r *testLinkAddressResolver) LinkAddressRequest(addr, _ tcpip.Address, _ tcpip.LinkAddress, _ LinkEndpoint) *tcpip.Error { time.AfterFunc(r.delay, func() { r.fakeRequest(addr) }) if f := r.onLinkAddressRequest; f != nil { f() diff --git a/pkg/tcpip/stack/ndp.go b/pkg/tcpip/stack/ndp.go index e28c23d66..5174e639c 100644 --- a/pkg/tcpip/stack/ndp.go +++ b/pkg/tcpip/stack/ndp.go @@ -33,12 +33,6 @@ const ( // Default = 1 (from RFC 4862 section 5.1) defaultDupAddrDetectTransmits = 1 - // defaultRetransmitTimer is the default amount of time to wait between - // sending NDP Neighbor solicitation messages. - // - // Default = 1s (from RFC 4861 section 10). - defaultRetransmitTimer = time.Second - // defaultMaxRtrSolicitations is the default number of Router // Solicitation messages to send when a NIC becomes enabled. // @@ -79,16 +73,6 @@ const ( // Default = true. defaultAutoGenGlobalAddresses = true - // minimumRetransmitTimer is the minimum amount of time to wait between - // sending NDP Neighbor solicitation messages. Note, RFC 4861 does - // not impose a minimum Retransmit Timer, but we do here to make sure - // the messages are not sent all at once. We also come to this value - // because in the RetransmitTimer field of a Router Advertisement, a - // value of 0 means unspecified, so the smallest valid value is 1. - // Note, the unit of the RetransmitTimer field in the Router - // Advertisement is milliseconds. - minimumRetransmitTimer = time.Millisecond - // minimumRtrSolicitationInterval is the minimum amount of time to wait // between sending Router Solicitation messages. This limit is imposed // to make sure that Router Solicitation messages are not sent all at @@ -469,7 +453,7 @@ type ndpState struct { rtrSolicit struct { // The timer used to send the next router solicitation message. - timer *time.Timer + timer tcpip.Timer // Used to let the Router Solicitation timer know that it has been stopped. // @@ -503,7 +487,7 @@ type ndpState struct { // to the DAD goroutine that DAD should stop. type dadState struct { // The DAD timer to send the next NS message, or resolve the address. - timer *time.Timer + timer tcpip.Timer // Used to let the DAD timer know that it has been stopped. // @@ -515,38 +499,38 @@ type dadState struct { // defaultRouterState holds data associated with a default router discovered by // a Router Advertisement (RA). type defaultRouterState struct { - // Timer to invalidate the default router. + // Job to invalidate the default router. // // Must not be nil. - invalidationTimer *tcpip.CancellableTimer + invalidationJob *tcpip.Job } // onLinkPrefixState holds data associated with an on-link prefix discovered by // a Router Advertisement's Prefix Information option (PI) when the NDP // configurations was configured to do so. type onLinkPrefixState struct { - // Timer to invalidate the on-link prefix. + // Job to invalidate the on-link prefix. // // Must not be nil. - invalidationTimer *tcpip.CancellableTimer + invalidationJob *tcpip.Job } // tempSLAACAddrState holds state associated with a temporary SLAAC address. type tempSLAACAddrState struct { - // Timer to deprecate the temporary SLAAC address. + // Job to deprecate the temporary SLAAC address. // // Must not be nil. - deprecationTimer *tcpip.CancellableTimer + deprecationJob *tcpip.Job - // Timer to invalidate the temporary SLAAC address. + // Job to invalidate the temporary SLAAC address. // // Must not be nil. - invalidationTimer *tcpip.CancellableTimer + invalidationJob *tcpip.Job - // Timer to regenerate the temporary SLAAC address. + // Job to regenerate the temporary SLAAC address. // // Must not be nil. - regenTimer *tcpip.CancellableTimer + regenJob *tcpip.Job createdAt time.Time @@ -561,15 +545,15 @@ type tempSLAACAddrState struct { // slaacPrefixState holds state associated with a SLAAC prefix. type slaacPrefixState struct { - // Timer to deprecate the prefix. + // Job to deprecate the prefix. // // Must not be nil. - deprecationTimer *tcpip.CancellableTimer + deprecationJob *tcpip.Job - // Timer to invalidate the prefix. + // Job to invalidate the prefix. // // Must not be nil. - invalidationTimer *tcpip.CancellableTimer + invalidationJob *tcpip.Job // Nonzero only when the address is not valid forever. validUntil time.Time @@ -651,12 +635,12 @@ func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, ref *ref } var done bool - var timer *time.Timer + var timer tcpip.Timer // We initially start a timer to fire immediately because some of the DAD work // cannot be done while holding the NIC's lock. This is effectively the same // as starting a goroutine but we use a timer that fires immediately so we can // reset it for the next DAD iteration. - timer = time.AfterFunc(0, func() { + timer = ndp.nic.stack.Clock().AfterFunc(0, func() { ndp.nic.mu.Lock() defer ndp.nic.mu.Unlock() @@ -871,9 +855,9 @@ func (ndp *ndpState) handleRA(ip tcpip.Address, ra header.NDPRouterAdvert) { case ok && rl != 0: // This is an already discovered default router. Update - // the invalidation timer. - rtr.invalidationTimer.StopLocked() - rtr.invalidationTimer.Reset(rl) + // the invalidation job. + rtr.invalidationJob.Cancel() + rtr.invalidationJob.Schedule(rl) ndp.defaultRouters[ip] = rtr case ok && rl == 0: @@ -950,7 +934,7 @@ func (ndp *ndpState) invalidateDefaultRouter(ip tcpip.Address) { return } - rtr.invalidationTimer.StopLocked() + rtr.invalidationJob.Cancel() delete(ndp.defaultRouters, ip) // Let the integrator know a discovered default router is invalidated. @@ -979,12 +963,12 @@ func (ndp *ndpState) rememberDefaultRouter(ip tcpip.Address, rl time.Duration) { } state := defaultRouterState{ - invalidationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { ndp.invalidateDefaultRouter(ip) }), } - state.invalidationTimer.Reset(rl) + state.invalidationJob.Schedule(rl) ndp.defaultRouters[ip] = state } @@ -1009,13 +993,13 @@ func (ndp *ndpState) rememberOnLinkPrefix(prefix tcpip.Subnet, l time.Duration) } state := onLinkPrefixState{ - invalidationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { ndp.invalidateOnLinkPrefix(prefix) }), } if l < header.NDPInfiniteLifetime { - state.invalidationTimer.Reset(l) + state.invalidationJob.Schedule(l) } ndp.onLinkPrefixes[prefix] = state @@ -1033,7 +1017,7 @@ func (ndp *ndpState) invalidateOnLinkPrefix(prefix tcpip.Subnet) { return } - s.invalidationTimer.StopLocked() + s.invalidationJob.Cancel() delete(ndp.onLinkPrefixes, prefix) // Let the integrator know a discovered on-link prefix is invalidated. @@ -1082,14 +1066,14 @@ func (ndp *ndpState) handleOnLinkPrefixInformation(pi header.NDPPrefixInformatio // This is an already discovered on-link prefix with a // new non-zero valid lifetime. // - // Update the invalidation timer. + // Update the invalidation job. - prefixState.invalidationTimer.StopLocked() + prefixState.invalidationJob.Cancel() if vl < header.NDPInfiniteLifetime { - // Prefix is valid for a finite lifetime, reset the timer to expire after + // Prefix is valid for a finite lifetime, schedule the job to execute after // the new valid lifetime. - prefixState.invalidationTimer.Reset(vl) + prefixState.invalidationJob.Schedule(vl) } ndp.onLinkPrefixes[prefix] = prefixState @@ -1154,7 +1138,7 @@ func (ndp *ndpState) doSLAAC(prefix tcpip.Subnet, pl, vl time.Duration) { } state := slaacPrefixState{ - deprecationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + deprecationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { state, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for the deprecated SLAAC prefix %s", prefix)) @@ -1162,7 +1146,7 @@ func (ndp *ndpState) doSLAAC(prefix tcpip.Subnet, pl, vl time.Duration) { ndp.deprecateSLAACAddress(state.stableAddr.ref) }), - invalidationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { state, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for the invalidated SLAAC prefix %s", prefix)) @@ -1184,19 +1168,19 @@ func (ndp *ndpState) doSLAAC(prefix tcpip.Subnet, pl, vl time.Duration) { if !ndp.generateSLAACAddr(prefix, &state) { // We were unable to generate an address for the prefix, we do not nothing - // further as there is no reason to maintain state or timers for a prefix we + // further as there is no reason to maintain state or jobs for a prefix we // do not have an address for. return } - // Setup the initial timers to deprecate and invalidate prefix. + // Setup the initial jobs to deprecate and invalidate prefix. if pl < header.NDPInfiniteLifetime && pl != 0 { - state.deprecationTimer.Reset(pl) + state.deprecationJob.Schedule(pl) } if vl < header.NDPInfiniteLifetime { - state.invalidationTimer.Reset(vl) + state.invalidationJob.Schedule(vl) state.validUntil = now.Add(vl) } @@ -1428,7 +1412,7 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla } state := tempSLAACAddrState{ - deprecationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + deprecationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { prefixState, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for %s to deprecate temporary address %s", prefix, generatedAddr)) @@ -1441,7 +1425,7 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla ndp.deprecateSLAACAddress(tempAddrState.ref) }), - invalidationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { prefixState, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for %s to invalidate temporary address %s", prefix, generatedAddr)) @@ -1454,7 +1438,7 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla ndp.invalidateTempSLAACAddr(prefixState.tempAddrs, generatedAddr.Address, tempAddrState) }), - regenTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + regenJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { prefixState, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for %s to regenerate temporary address after %s", prefix, generatedAddr)) @@ -1481,9 +1465,9 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla ref: ref, } - state.deprecationTimer.Reset(pl) - state.invalidationTimer.Reset(vl) - state.regenTimer.Reset(pl - ndp.configs.RegenAdvanceDuration) + state.deprecationJob.Schedule(pl) + state.invalidationJob.Schedule(vl) + state.regenJob.Schedule(pl - ndp.configs.RegenAdvanceDuration) prefixState.generationAttempts++ prefixState.tempAddrs[generatedAddr.Address] = state @@ -1518,16 +1502,16 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat prefixState.stableAddr.ref.deprecated = false } - // If prefix was preferred for some finite lifetime before, stop the - // deprecation timer so it can be reset. - prefixState.deprecationTimer.StopLocked() + // If prefix was preferred for some finite lifetime before, cancel the + // deprecation job so it can be reset. + prefixState.deprecationJob.Cancel() now := time.Now() - // Reset the deprecation timer if prefix has a finite preferred lifetime. + // Schedule the deprecation job if prefix has a finite preferred lifetime. if pl < header.NDPInfiniteLifetime { if !deprecated { - prefixState.deprecationTimer.Reset(pl) + prefixState.deprecationJob.Schedule(pl) } prefixState.preferredUntil = now.Add(pl) } else { @@ -1546,9 +1530,9 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat // 3) Otherwise, reset the valid lifetime of the prefix to 2 hours. if vl >= header.NDPInfiniteLifetime { - // Handle the infinite valid lifetime separately as we do not keep a timer - // in this case. - prefixState.invalidationTimer.StopLocked() + // Handle the infinite valid lifetime separately as we do not schedule a + // job in this case. + prefixState.invalidationJob.Cancel() prefixState.validUntil = time.Time{} } else { var effectiveVl time.Duration @@ -1569,8 +1553,8 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat } if effectiveVl != 0 { - prefixState.invalidationTimer.StopLocked() - prefixState.invalidationTimer.Reset(effectiveVl) + prefixState.invalidationJob.Cancel() + prefixState.invalidationJob.Schedule(effectiveVl) prefixState.validUntil = now.Add(effectiveVl) } } @@ -1582,7 +1566,7 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat } // Note, we do not need to update the entries in the temporary address map - // after updating the timers because the timers are held as pointers. + // after updating the jobs because the jobs are held as pointers. var regenForAddr tcpip.Address allAddressesRegenerated := true for tempAddr, tempAddrState := range prefixState.tempAddrs { @@ -1596,14 +1580,14 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat } // If the address is no longer valid, invalidate it immediately. Otherwise, - // reset the invalidation timer. + // reset the invalidation job. newValidLifetime := validUntil.Sub(now) if newValidLifetime <= 0 { ndp.invalidateTempSLAACAddr(prefixState.tempAddrs, tempAddr, tempAddrState) continue } - tempAddrState.invalidationTimer.StopLocked() - tempAddrState.invalidationTimer.Reset(newValidLifetime) + tempAddrState.invalidationJob.Cancel() + tempAddrState.invalidationJob.Schedule(newValidLifetime) // As per RFC 4941 section 3.3 step 4, the preferred lifetime of a temporary // address is the lower of the preferred lifetime of the stable address or @@ -1616,17 +1600,17 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat } // If the address is no longer preferred, deprecate it immediately. - // Otherwise, reset the deprecation timer. + // Otherwise, schedule the deprecation job again. newPreferredLifetime := preferredUntil.Sub(now) - tempAddrState.deprecationTimer.StopLocked() + tempAddrState.deprecationJob.Cancel() if newPreferredLifetime <= 0 { ndp.deprecateSLAACAddress(tempAddrState.ref) } else { tempAddrState.ref.deprecated = false - tempAddrState.deprecationTimer.Reset(newPreferredLifetime) + tempAddrState.deprecationJob.Schedule(newPreferredLifetime) } - tempAddrState.regenTimer.StopLocked() + tempAddrState.regenJob.Cancel() if tempAddrState.regenerated { } else { allAddressesRegenerated = false @@ -1637,7 +1621,7 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat // immediately after we finish iterating over the temporary addresses. regenForAddr = tempAddr } else { - tempAddrState.regenTimer.Reset(newPreferredLifetime - ndp.configs.RegenAdvanceDuration) + tempAddrState.regenJob.Schedule(newPreferredLifetime - ndp.configs.RegenAdvanceDuration) } } } @@ -1717,7 +1701,7 @@ func (ndp *ndpState) cleanupSLAACAddrResourcesAndNotify(addr tcpip.AddressWithPr ndp.cleanupSLAACPrefixResources(prefix, state) } -// cleanupSLAACPrefixResources cleansup a SLAAC prefix's timers and entry. +// cleanupSLAACPrefixResources cleans up a SLAAC prefix's jobs and entry. // // Panics if the SLAAC prefix is not known. // @@ -1729,8 +1713,8 @@ func (ndp *ndpState) cleanupSLAACPrefixResources(prefix tcpip.Subnet, state slaa } state.stableAddr.ref = nil - state.deprecationTimer.StopLocked() - state.invalidationTimer.StopLocked() + state.deprecationJob.Cancel() + state.invalidationJob.Cancel() delete(ndp.slaacPrefixes, prefix) } @@ -1775,13 +1759,13 @@ func (ndp *ndpState) cleanupTempSLAACAddrResourcesAndNotify(addr tcpip.AddressWi } // cleanupTempSLAACAddrResourcesAndNotify cleans up a temporary SLAAC address's -// timers and entry. +// jobs and entry. // // The NIC that ndp belongs to MUST be locked. func (ndp *ndpState) cleanupTempSLAACAddrResources(tempAddrs map[tcpip.Address]tempSLAACAddrState, tempAddr tcpip.Address, tempAddrState tempSLAACAddrState) { - tempAddrState.deprecationTimer.StopLocked() - tempAddrState.invalidationTimer.StopLocked() - tempAddrState.regenTimer.StopLocked() + tempAddrState.deprecationJob.Cancel() + tempAddrState.invalidationJob.Cancel() + tempAddrState.regenJob.Cancel() delete(tempAddrs, tempAddr) } @@ -1860,7 +1844,7 @@ func (ndp *ndpState) startSolicitingRouters() { var done bool ndp.rtrSolicit.done = &done - ndp.rtrSolicit.timer = time.AfterFunc(delay, func() { + ndp.rtrSolicit.timer = ndp.nic.stack.Clock().AfterFunc(delay, func() { ndp.nic.mu.Lock() if done { // If we reach this point, it means that the RS timer fired after another diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go index 6f86abc98..644ba7c33 100644 --- a/pkg/tcpip/stack/ndp_test.go +++ b/pkg/tcpip/stack/ndp_test.go @@ -1254,7 +1254,7 @@ func TestRouterDiscovery(t *testing.T) { default: } - // Wait for lladdr2's router invalidation timer to fire. The lifetime + // Wait for lladdr2's router invalidation job to execute. The lifetime // of the router should have been updated to the most recent (smaller) // lifetime. // @@ -1271,7 +1271,7 @@ func TestRouterDiscovery(t *testing.T) { e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, 0)) expectRouterEvent(llAddr2, false) - // Wait for lladdr3's router invalidation timer to fire. The lifetime + // Wait for lladdr3's router invalidation job to execute. The lifetime // of the router should have been updated to the most recent (smaller) // lifetime. // @@ -1502,7 +1502,7 @@ func TestPrefixDiscovery(t *testing.T) { default: } - // Wait for prefix2's most recent invalidation timer plus some buffer to + // Wait for prefix2's most recent invalidation job plus some buffer to // expire. select { case e := <-ndpDisp.prefixC: @@ -2395,7 +2395,7 @@ func TestAutoGenTempAddrRegen(t *testing.T) { for _, addr := range tempAddrs { // Wait for a deprecation then invalidation event, or just an invalidation // event. We need to cover both cases but cannot deterministically hit both - // cases because the deprecation and invalidation timers could fire in any + // cases because the deprecation and invalidation jobs could execute in any // order. select { case e := <-ndpDisp.autoGenAddrC: @@ -2432,9 +2432,9 @@ func TestAutoGenTempAddrRegen(t *testing.T) { } } -// TestAutoGenTempAddrRegenTimerUpdates tests that a temporary address's -// regeneration timer gets updated when refreshing the address's lifetimes. -func TestAutoGenTempAddrRegenTimerUpdates(t *testing.T) { +// TestAutoGenTempAddrRegenJobUpdates tests that a temporary address's +// regeneration job gets updated when refreshing the address's lifetimes. +func TestAutoGenTempAddrRegenJobUpdates(t *testing.T) { const ( nicID = 1 regenAfter = 2 * time.Second @@ -2533,7 +2533,7 @@ func TestAutoGenTempAddrRegenTimerUpdates(t *testing.T) { // // A new temporary address should immediately be generated since the // regeneration time has already passed since the last address was generated - // - this regeneration does not depend on a timer. + // - this regeneration does not depend on a job. e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, 100, 100)) expectAutoGenAddrEvent(tempAddr2, newAddr) @@ -2559,11 +2559,11 @@ func TestAutoGenTempAddrRegenTimerUpdates(t *testing.T) { } // Set the maximum lifetimes for temporary addresses such that on the next - // RA, the regeneration timer gets reset. + // RA, the regeneration job gets scheduled again. // // The maximum lifetime is the sum of the minimum lifetimes for temporary // addresses + the time that has already passed since the last address was - // generated so that the regeneration timer is needed to generate the next + // generated so that the regeneration job is needed to generate the next // address. newLifetimes := newMinVLDuration + regenAfter + defaultAsyncNegativeEventTimeout ndpConfigs.MaxTempAddrValidLifetime = newLifetimes @@ -2993,9 +2993,9 @@ func TestAutoGenAddrDeprecateFromPI(t *testing.T) { expectPrimaryAddr(addr2) } -// TestAutoGenAddrTimerDeprecation tests that an address is properly deprecated +// TestAutoGenAddrJobDeprecation tests that an address is properly deprecated // when its preferred lifetime expires. -func TestAutoGenAddrTimerDeprecation(t *testing.T) { +func TestAutoGenAddrJobDeprecation(t *testing.T) { const nicID = 1 const newMinVL = 2 newMinVLDuration := newMinVL * time.Second @@ -3513,8 +3513,8 @@ func TestAutoGenAddrRemoval(t *testing.T) { } expectAutoGenAddrEvent(addr, invalidatedAddr) - // Wait for the original valid lifetime to make sure the original timer - // got stopped/cleaned up. + // Wait for the original valid lifetime to make sure the original job got + // cancelled/cleaned up. select { case <-ndpDisp.autoGenAddrC: t.Fatal("unexpectedly received an auto gen addr event") diff --git a/pkg/tcpip/stack/neighbor_cache.go b/pkg/tcpip/stack/neighbor_cache.go new file mode 100644 index 000000000..1d37716c2 --- /dev/null +++ b/pkg/tcpip/stack/neighbor_cache.go @@ -0,0 +1,335 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package stack + +import ( + "fmt" + "time" + + "gvisor.dev/gvisor/pkg/sleep" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/tcpip" +) + +const neighborCacheSize = 512 // max entries per interface + +// neighborCache maps IP addresses to link addresses. It uses the Least +// Recently Used (LRU) eviction strategy to implement a bounded cache for +// dynmically acquired entries. It contains the state machine and configuration +// for running Neighbor Unreachability Detection (NUD). +// +// There are two types of entries in the neighbor cache: +// 1. Dynamic entries are discovered automatically by neighbor discovery +// protocols (e.g. ARP, NDP). These protocols will attempt to reconfirm +// reachability with the device once the entry's state becomes Stale. +// 2. Static entries are explicitly added by a user and have no expiration. +// Their state is always Static. The amount of static entries stored in the +// cache is unbounded. +// +// neighborCache implements NUDHandler. +type neighborCache struct { + nic *NIC + state *NUDState + + // mu protects the fields below. + mu sync.RWMutex + + cache map[tcpip.Address]*neighborEntry + dynamic struct { + lru neighborEntryList + + // count tracks the amount of dynamic entries in the cache. This is + // needed since static entries do not count towards the LRU cache + // eviction strategy. + count uint16 + } +} + +var _ NUDHandler = (*neighborCache)(nil) + +// getOrCreateEntry retrieves a cache entry associated with addr. The +// returned entry is always refreshed in the cache (it is reachable via the +// map, and its place is bumped in LRU). +// +// If a matching entry exists in the cache, it is returned. If no matching +// entry exists and the cache is full, an existing entry is evicted via LRU, +// reset to state incomplete, and returned. If no matching entry exists and the +// cache is not full, a new entry with state incomplete is allocated and +// returned. +func (n *neighborCache) getOrCreateEntry(remoteAddr, localAddr tcpip.Address, linkRes LinkAddressResolver) *neighborEntry { + n.mu.Lock() + defer n.mu.Unlock() + + if entry, ok := n.cache[remoteAddr]; ok { + entry.mu.RLock() + if entry.neigh.State != Static { + n.dynamic.lru.Remove(entry) + n.dynamic.lru.PushFront(entry) + } + entry.mu.RUnlock() + return entry + } + + // The entry that needs to be created must be dynamic since all static + // entries are directly added to the cache via addStaticEntry. + entry := newNeighborEntry(n.nic, remoteAddr, localAddr, n.state, linkRes) + if n.dynamic.count == neighborCacheSize { + e := n.dynamic.lru.Back() + e.mu.Lock() + + delete(n.cache, e.neigh.Addr) + n.dynamic.lru.Remove(e) + n.dynamic.count-- + + e.dispatchRemoveEventLocked() + e.setStateLocked(Unknown) + e.notifyWakersLocked() + e.mu.Unlock() + } + n.cache[remoteAddr] = entry + n.dynamic.lru.PushFront(entry) + n.dynamic.count++ + return entry +} + +// entry looks up the neighbor cache for translating address to link address +// (e.g. IP -> MAC). If the LinkEndpoint requests address resolution and there +// is a LinkAddressResolver registered with the network protocol, the cache +// attempts to resolve the address and returns ErrWouldBlock. If a Waker is +// provided, it will be notified when address resolution is complete (success +// or not). +// +// If address resolution is required, ErrNoLinkAddress and a notification +// channel is returned for the top level caller to block. Channel is closed +// once address resolution is complete (success or not). +func (n *neighborCache) entry(remoteAddr, localAddr tcpip.Address, linkRes LinkAddressResolver, w *sleep.Waker) (NeighborEntry, <-chan struct{}, *tcpip.Error) { + if linkRes != nil { + if linkAddr, ok := linkRes.ResolveStaticAddress(remoteAddr); ok { + e := NeighborEntry{ + Addr: remoteAddr, + LocalAddr: localAddr, + LinkAddr: linkAddr, + State: Static, + UpdatedAt: time.Now(), + } + return e, nil, nil + } + } + + entry := n.getOrCreateEntry(remoteAddr, localAddr, linkRes) + entry.mu.Lock() + defer entry.mu.Unlock() + + switch s := entry.neigh.State; s { + case Reachable, Static: + return entry.neigh, nil, nil + + case Unknown, Incomplete, Stale, Delay, Probe: + entry.addWakerLocked(w) + + if entry.done == nil { + // Address resolution needs to be initiated. + if linkRes == nil { + return entry.neigh, nil, tcpip.ErrNoLinkAddress + } + entry.done = make(chan struct{}) + } + + entry.handlePacketQueuedLocked() + return entry.neigh, entry.done, tcpip.ErrWouldBlock + + case Failed: + return entry.neigh, nil, tcpip.ErrNoLinkAddress + + default: + panic(fmt.Sprintf("Invalid cache entry state: %s", s)) + } +} + +// removeWaker removes a waker that has been added when link resolution for +// addr was requested. +func (n *neighborCache) removeWaker(addr tcpip.Address, waker *sleep.Waker) { + n.mu.Lock() + if entry, ok := n.cache[addr]; ok { + delete(entry.wakers, waker) + } + n.mu.Unlock() +} + +// entries returns all entries in the neighbor cache. +func (n *neighborCache) entries() []NeighborEntry { + entries := make([]NeighborEntry, 0, len(n.cache)) + n.mu.RLock() + for _, entry := range n.cache { + entry.mu.RLock() + entries = append(entries, entry.neigh) + entry.mu.RUnlock() + } + n.mu.RUnlock() + return entries +} + +// addStaticEntry adds a static entry to the neighbor cache, mapping an IP +// address to a link address. If a dynamic entry exists in the neighbor cache +// with the same address, it will be replaced with this static entry. If a +// static entry exists with the same address but different link address, it +// will be updated with the new link address. If a static entry exists with the +// same address and link address, nothing will happen. +func (n *neighborCache) addStaticEntry(addr tcpip.Address, linkAddr tcpip.LinkAddress) { + n.mu.Lock() + defer n.mu.Unlock() + + if entry, ok := n.cache[addr]; ok { + entry.mu.Lock() + if entry.neigh.State != Static { + // Dynamic entry found with the same address. + n.dynamic.lru.Remove(entry) + n.dynamic.count-- + } else if entry.neigh.LinkAddr == linkAddr { + // Static entry found with the same address and link address. + entry.mu.Unlock() + return + } else { + // Static entry found with the same address but different link address. + entry.neigh.LinkAddr = linkAddr + entry.dispatchChangeEventLocked(entry.neigh.State) + entry.mu.Unlock() + return + } + + // Notify that resolution has been interrupted, just in case the entry was + // in the Incomplete or Probe state. + entry.dispatchRemoveEventLocked() + entry.setStateLocked(Unknown) + entry.notifyWakersLocked() + entry.mu.Unlock() + } + + entry := newStaticNeighborEntry(n.nic, addr, linkAddr, n.state) + n.cache[addr] = entry +} + +// removeEntryLocked removes the specified entry from the neighbor cache. +func (n *neighborCache) removeEntryLocked(entry *neighborEntry) { + if entry.neigh.State != Static { + n.dynamic.lru.Remove(entry) + n.dynamic.count-- + } + if entry.neigh.State != Failed { + entry.dispatchRemoveEventLocked() + } + entry.setStateLocked(Unknown) + entry.notifyWakersLocked() + + delete(n.cache, entry.neigh.Addr) +} + +// removeEntry removes a dynamic or static entry by address from the neighbor +// cache. Returns true if the entry was found and deleted. +func (n *neighborCache) removeEntry(addr tcpip.Address) bool { + n.mu.Lock() + defer n.mu.Unlock() + + entry, ok := n.cache[addr] + if !ok { + return false + } + + entry.mu.Lock() + defer entry.mu.Unlock() + + n.removeEntryLocked(entry) + return true +} + +// clear removes all dynamic and static entries from the neighbor cache. +func (n *neighborCache) clear() { + n.mu.Lock() + defer n.mu.Unlock() + + for _, entry := range n.cache { + entry.mu.Lock() + entry.dispatchRemoveEventLocked() + entry.setStateLocked(Unknown) + entry.notifyWakersLocked() + entry.mu.Unlock() + } + + n.dynamic.lru = neighborEntryList{} + n.cache = make(map[tcpip.Address]*neighborEntry) + n.dynamic.count = 0 +} + +// config returns the NUD configuration. +func (n *neighborCache) config() NUDConfigurations { + return n.state.Config() +} + +// setConfig changes the NUD configuration. +// +// If config contains invalid NUD configuration values, it will be fixed to +// use default values for the erroneous values. +func (n *neighborCache) setConfig(config NUDConfigurations) { + config.resetInvalidFields() + n.state.SetConfig(config) +} + +// HandleProbe implements NUDHandler.HandleProbe by following the logic defined +// in RFC 4861 section 7.2.3. Validation of the probe is expected to be handled +// by the caller. +func (n *neighborCache) HandleProbe(remoteAddr, localAddr tcpip.Address, protocol tcpip.NetworkProtocolNumber, remoteLinkAddr tcpip.LinkAddress) { + entry := n.getOrCreateEntry(remoteAddr, localAddr, nil) + entry.mu.Lock() + entry.handleProbeLocked(remoteLinkAddr) + entry.mu.Unlock() +} + +// HandleConfirmation implements NUDHandler.HandleConfirmation by following the +// logic defined in RFC 4861 section 7.2.5. +// +// TODO(gvisor.dev/issue/2277): To protect against ARP poisoning and other +// attacks against NDP functions, Secure Neighbor Discovery (SEND) Protocol +// should be deployed where preventing access to the broadcast segment might +// not be possible. SEND uses RSA key pairs to produce cryptographically +// generated addresses, as defined in RFC 3972, Cryptographically Generated +// Addresses (CGA). This ensures that the claimed source of an NDP message is +// the owner of the claimed address. +func (n *neighborCache) HandleConfirmation(addr tcpip.Address, linkAddr tcpip.LinkAddress, flags ReachabilityConfirmationFlags) { + n.mu.RLock() + entry, ok := n.cache[addr] + n.mu.RUnlock() + if ok { + entry.mu.Lock() + entry.handleConfirmationLocked(linkAddr, flags) + entry.mu.Unlock() + } + // The confirmation SHOULD be silently discarded if the recipient did not + // initiate any communication with the target. This is indicated if there is + // no matching entry for the remote address. +} + +// HandleUpperLevelConfirmation implements +// NUDHandler.HandleUpperLevelConfirmation by following the logic defined in +// RFC 4861 section 7.3.1. +func (n *neighborCache) HandleUpperLevelConfirmation(addr tcpip.Address) { + n.mu.RLock() + entry, ok := n.cache[addr] + n.mu.RUnlock() + if ok { + entry.mu.Lock() + entry.handleUpperLevelConfirmationLocked() + entry.mu.Unlock() + } +} diff --git a/pkg/tcpip/stack/neighbor_cache_test.go b/pkg/tcpip/stack/neighbor_cache_test.go new file mode 100644 index 000000000..4cb2c9c6b --- /dev/null +++ b/pkg/tcpip/stack/neighbor_cache_test.go @@ -0,0 +1,1752 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package stack + +import ( + "bytes" + "encoding/binary" + "fmt" + "math" + "math/rand" + "strings" + "sync" + "sync/atomic" + "testing" + "time" + + "github.com/google/go-cmp/cmp" + "github.com/google/go-cmp/cmp/cmpopts" + "gvisor.dev/gvisor/pkg/sleep" + "gvisor.dev/gvisor/pkg/tcpip" +) + +const ( + // entryStoreSize is the default number of entries that will be generated and + // added to the entry store. This number needs to be larger than the size of + // the neighbor cache to give ample opportunity for verifying behavior during + // cache overflows. Four times the size of the neighbor cache allows for + // three complete cache overflows. + entryStoreSize = 4 * neighborCacheSize + + // typicalLatency is the typical latency for an ARP or NDP packet to travel + // to a router and back. + typicalLatency = time.Millisecond + + // testEntryBroadcastAddr is a special address that indicates a packet should + // be sent to all nodes. + testEntryBroadcastAddr = tcpip.Address("broadcast") + + // testEntryLocalAddr is the source address of neighbor probes. + testEntryLocalAddr = tcpip.Address("local_addr") + + // testEntryBroadcastLinkAddr is a special link address sent back to + // multicast neighbor probes. + testEntryBroadcastLinkAddr = tcpip.LinkAddress("mac_broadcast") + + // infiniteDuration indicates that a task will not occur in our lifetime. + infiniteDuration = time.Duration(math.MaxInt64) +) + +// entryDiffOpts returns the options passed to cmp.Diff to compare neighbor +// entries. The UpdatedAt field is ignored due to a lack of a deterministic +// method to predict the time that an event will be dispatched. +func entryDiffOpts() []cmp.Option { + return []cmp.Option{ + cmpopts.IgnoreFields(NeighborEntry{}, "UpdatedAt"), + } +} + +// entryDiffOptsWithSort is like entryDiffOpts but also includes an option to +// sort slices of entries for cases where ordering must be ignored. +func entryDiffOptsWithSort() []cmp.Option { + return []cmp.Option{ + cmpopts.IgnoreFields(NeighborEntry{}, "UpdatedAt"), + cmpopts.SortSlices(func(a, b NeighborEntry) bool { + return strings.Compare(string(a.Addr), string(b.Addr)) < 0 + }), + } +} + +func newTestNeighborCache(nudDisp NUDDispatcher, config NUDConfigurations, clock tcpip.Clock) *neighborCache { + config.resetInvalidFields() + rng := rand.New(rand.NewSource(time.Now().UnixNano())) + return &neighborCache{ + nic: &NIC{ + stack: &Stack{ + clock: clock, + nudDisp: nudDisp, + }, + id: 1, + }, + state: NewNUDState(config, rng), + cache: make(map[tcpip.Address]*neighborEntry, neighborCacheSize), + } +} + +// testEntryStore contains a set of IP to NeighborEntry mappings. +type testEntryStore struct { + mu sync.RWMutex + entriesMap map[tcpip.Address]NeighborEntry +} + +func toAddress(i int) tcpip.Address { + buf := new(bytes.Buffer) + binary.Write(buf, binary.BigEndian, uint8(1)) + binary.Write(buf, binary.BigEndian, uint8(0)) + binary.Write(buf, binary.BigEndian, uint16(i)) + return tcpip.Address(buf.String()) +} + +func toLinkAddress(i int) tcpip.LinkAddress { + buf := new(bytes.Buffer) + binary.Write(buf, binary.BigEndian, uint8(1)) + binary.Write(buf, binary.BigEndian, uint8(0)) + binary.Write(buf, binary.BigEndian, uint32(i)) + return tcpip.LinkAddress(buf.String()) +} + +// newTestEntryStore returns a testEntryStore pre-populated with entries. +func newTestEntryStore() *testEntryStore { + store := &testEntryStore{ + entriesMap: make(map[tcpip.Address]NeighborEntry), + } + for i := 0; i < entryStoreSize; i++ { + addr := toAddress(i) + linkAddr := toLinkAddress(i) + + store.entriesMap[addr] = NeighborEntry{ + Addr: addr, + LocalAddr: testEntryLocalAddr, + LinkAddr: linkAddr, + } + } + return store +} + +// size returns the number of entries in the store. +func (s *testEntryStore) size() int { + s.mu.RLock() + defer s.mu.RUnlock() + return len(s.entriesMap) +} + +// entry returns the entry at index i. Returns an empty entry and false if i is +// out of bounds. +func (s *testEntryStore) entry(i int) (NeighborEntry, bool) { + return s.entryByAddr(toAddress(i)) +} + +// entryByAddr returns the entry matching addr for situations when the index is +// not available. Returns an empty entry and false if no entries match addr. +func (s *testEntryStore) entryByAddr(addr tcpip.Address) (NeighborEntry, bool) { + s.mu.RLock() + defer s.mu.RUnlock() + entry, ok := s.entriesMap[addr] + return entry, ok +} + +// entries returns all entries in the store. +func (s *testEntryStore) entries() []NeighborEntry { + entries := make([]NeighborEntry, 0, len(s.entriesMap)) + s.mu.RLock() + defer s.mu.RUnlock() + for i := 0; i < entryStoreSize; i++ { + addr := toAddress(i) + if entry, ok := s.entriesMap[addr]; ok { + entries = append(entries, entry) + } + } + return entries +} + +// set modifies the link addresses of an entry. +func (s *testEntryStore) set(i int, linkAddr tcpip.LinkAddress) { + addr := toAddress(i) + s.mu.Lock() + defer s.mu.Unlock() + if entry, ok := s.entriesMap[addr]; ok { + entry.LinkAddr = linkAddr + s.entriesMap[addr] = entry + } +} + +// testNeighborResolver implements LinkAddressResolver to emulate sending a +// neighbor probe. +type testNeighborResolver struct { + clock tcpip.Clock + neigh *neighborCache + entries *testEntryStore + delay time.Duration + onLinkAddressRequest func() +} + +var _ LinkAddressResolver = (*testNeighborResolver)(nil) + +func (r *testNeighborResolver) LinkAddressRequest(addr, localAddr tcpip.Address, linkAddr tcpip.LinkAddress, linkEP LinkEndpoint) *tcpip.Error { + // Delay handling the request to emulate network latency. + r.clock.AfterFunc(r.delay, func() { + r.fakeRequest(addr) + }) + + // Execute post address resolution action, if available. + if f := r.onLinkAddressRequest; f != nil { + f() + } + return nil +} + +// fakeRequest emulates handling a response for a link address request. +func (r *testNeighborResolver) fakeRequest(addr tcpip.Address) { + if entry, ok := r.entries.entryByAddr(addr); ok { + r.neigh.HandleConfirmation(addr, entry.LinkAddr, ReachabilityConfirmationFlags{ + Solicited: true, + Override: false, + IsRouter: false, + }) + } +} + +func (*testNeighborResolver) ResolveStaticAddress(addr tcpip.Address) (tcpip.LinkAddress, bool) { + if addr == testEntryBroadcastAddr { + return testEntryBroadcastLinkAddr, true + } + return "", false +} + +func (*testNeighborResolver) LinkAddressProtocol() tcpip.NetworkProtocolNumber { + return 0 +} + +type entryEvent struct { + nicID tcpip.NICID + address tcpip.Address + linkAddr tcpip.LinkAddress + state NeighborState +} + +func TestNeighborCacheGetConfig(t *testing.T) { + nudDisp := testNUDDispatcher{} + c := DefaultNUDConfigurations() + clock := newFakeClock() + neigh := newTestNeighborCache(&nudDisp, c, clock) + + if got, want := neigh.config(), c; got != want { + t.Errorf("got neigh.config() = %+v, want = %+v", got, want) + } + + // No events should have been dispatched. + nudDisp.mu.Lock() + defer nudDisp.mu.Unlock() + if diff := cmp.Diff(nudDisp.events, []testEntryEventInfo(nil)); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } +} + +func TestNeighborCacheSetConfig(t *testing.T) { + nudDisp := testNUDDispatcher{} + c := DefaultNUDConfigurations() + clock := newFakeClock() + neigh := newTestNeighborCache(&nudDisp, c, clock) + + c.MinRandomFactor = 1 + c.MaxRandomFactor = 1 + neigh.setConfig(c) + + if got, want := neigh.config(), c; got != want { + t.Errorf("got neigh.config() = %+v, want = %+v", got, want) + } + + // No events should have been dispatched. + nudDisp.mu.Lock() + defer nudDisp.mu.Unlock() + if diff := cmp.Diff(nudDisp.events, []testEntryEventInfo(nil)); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } +} + +func TestNeighborCacheEntry(t *testing.T) { + c := DefaultNUDConfigurations() + nudDisp := testNUDDispatcher{} + clock := newFakeClock() + neigh := newTestNeighborCache(&nudDisp, c, clock) + store := newTestEntryStore() + linkRes := &testNeighborResolver{ + clock: clock, + neigh: neigh, + entries: store, + delay: typicalLatency, + } + + entry, ok := store.entry(0) + if !ok { + t.Fatalf("store.entry(0) not found") + } + _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil) + if err != tcpip.ErrWouldBlock { + t.Errorf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) + } + + clock.advance(typicalLatency) + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: 1, + Addr: entry.Addr, + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: 1, + Addr: entry.Addr, + LinkAddr: entry.LinkAddr, + State: Reachable, + }, + } + nudDisp.mu.Lock() + diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...) + nudDisp.events = nil + nudDisp.mu.Unlock() + if diff != "" { + t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + + if _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil); err != nil { + t.Fatalf("unexpected error from neigh.entry(%s, %s, _, nil): %s", entry.Addr, entry.LocalAddr, err) + } + + // No more events should have been dispatched. + nudDisp.mu.Lock() + defer nudDisp.mu.Unlock() + if diff := cmp.Diff(nudDisp.events, []testEntryEventInfo(nil)); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } +} + +// TestNeighborCacheEntryNoLinkAddress verifies calling entry() without a +// LinkAddressResolver returns ErrNoLinkAddress. +func TestNeighborCacheEntryNoLinkAddress(t *testing.T) { + nudDisp := testNUDDispatcher{} + c := DefaultNUDConfigurations() + clock := newFakeClock() + neigh := newTestNeighborCache(&nudDisp, c, clock) + store := newTestEntryStore() + + entry, ok := store.entry(0) + if !ok { + t.Fatalf("store.entry(0) not found") + } + _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, nil, nil) + if err != tcpip.ErrNoLinkAddress { + t.Errorf("got neigh.entry(%s, %s, nil, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrNoLinkAddress) + } + + // No events should have been dispatched. + nudDisp.mu.Lock() + defer nudDisp.mu.Unlock() + if diff := cmp.Diff(nudDisp.events, []testEntryEventInfo(nil)); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } +} + +func TestNeighborCacheRemoveEntry(t *testing.T) { + config := DefaultNUDConfigurations() + + nudDisp := testNUDDispatcher{} + clock := newFakeClock() + neigh := newTestNeighborCache(&nudDisp, config, clock) + store := newTestEntryStore() + linkRes := &testNeighborResolver{ + clock: clock, + neigh: neigh, + entries: store, + delay: typicalLatency, + } + + entry, ok := store.entry(0) + if !ok { + t.Fatalf("store.entry(0) not found") + } + _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil) + if err != tcpip.ErrWouldBlock { + t.Errorf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) + } + + clock.advance(typicalLatency) + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: 1, + Addr: entry.Addr, + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: 1, + Addr: entry.Addr, + LinkAddr: entry.LinkAddr, + State: Reachable, + }, + } + nudDisp.mu.Lock() + diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...) + nudDisp.events = nil + nudDisp.mu.Unlock() + if diff != "" { + t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + + neigh.removeEntry(entry.Addr) + + { + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestRemoved, + NICID: 1, + Addr: entry.Addr, + LinkAddr: entry.LinkAddr, + State: Reachable, + }, + } + nudDisp.mu.Lock() + diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...) + nudDisp.mu.Unlock() + if diff != "" { + t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + } + + if _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil); err != tcpip.ErrWouldBlock { + t.Errorf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) + } +} + +type testContext struct { + clock *fakeClock + neigh *neighborCache + store *testEntryStore + linkRes *testNeighborResolver + nudDisp *testNUDDispatcher +} + +func newTestContext(c NUDConfigurations) testContext { + nudDisp := &testNUDDispatcher{} + clock := newFakeClock() + neigh := newTestNeighborCache(nudDisp, c, clock) + store := newTestEntryStore() + linkRes := &testNeighborResolver{ + clock: clock, + neigh: neigh, + entries: store, + delay: typicalLatency, + } + + return testContext{ + clock: clock, + neigh: neigh, + store: store, + linkRes: linkRes, + nudDisp: nudDisp, + } +} + +type overflowOptions struct { + startAtEntryIndex int + wantStaticEntries []NeighborEntry +} + +func (c *testContext) overflowCache(opts overflowOptions) error { + // Fill the neighbor cache to capacity to verify the LRU eviction strategy is + // working properly after the entry removal. + for i := opts.startAtEntryIndex; i < c.store.size(); i++ { + // Add a new entry + entry, ok := c.store.entry(i) + if !ok { + return fmt.Errorf("c.store.entry(%d) not found", i) + } + if _, _, err := c.neigh.entry(entry.Addr, entry.LocalAddr, c.linkRes, nil); err != tcpip.ErrWouldBlock { + return fmt.Errorf("got c.neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) + } + c.clock.advance(c.neigh.config().RetransmitTimer) + + var wantEvents []testEntryEventInfo + + // When beyond the full capacity, the cache will evict an entry as per the + // LRU eviction strategy. Note that the number of static entries should not + // affect the total number of dynamic entries that can be added. + if i >= neighborCacheSize+opts.startAtEntryIndex { + removedEntry, ok := c.store.entry(i - neighborCacheSize) + if !ok { + return fmt.Errorf("store.entry(%d) not found", i-neighborCacheSize) + } + wantEvents = append(wantEvents, testEntryEventInfo{ + EventType: entryTestRemoved, + NICID: 1, + Addr: removedEntry.Addr, + LinkAddr: removedEntry.LinkAddr, + State: Reachable, + }) + } + + wantEvents = append(wantEvents, testEntryEventInfo{ + EventType: entryTestAdded, + NICID: 1, + Addr: entry.Addr, + State: Incomplete, + }, testEntryEventInfo{ + EventType: entryTestChanged, + NICID: 1, + Addr: entry.Addr, + LinkAddr: entry.LinkAddr, + State: Reachable, + }) + + c.nudDisp.mu.Lock() + diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...) + c.nudDisp.events = nil + c.nudDisp.mu.Unlock() + if diff != "" { + return fmt.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + } + + // Expect to find only the most recent entries. The order of entries reported + // by entries() is undeterministic, so entries have to be sorted before + // comparison. + wantUnsortedEntries := opts.wantStaticEntries + for i := c.store.size() - neighborCacheSize; i < c.store.size(); i++ { + entry, ok := c.store.entry(i) + if !ok { + return fmt.Errorf("c.store.entry(%d) not found", i) + } + wantEntry := NeighborEntry{ + Addr: entry.Addr, + LocalAddr: entry.LocalAddr, + LinkAddr: entry.LinkAddr, + State: Reachable, + } + wantUnsortedEntries = append(wantUnsortedEntries, wantEntry) + } + + if diff := cmp.Diff(c.neigh.entries(), wantUnsortedEntries, entryDiffOptsWithSort()...); diff != "" { + return fmt.Errorf("neighbor entries mismatch (-got, +want):\n%s", diff) + } + + // No more events should have been dispatched. + c.nudDisp.mu.Lock() + defer c.nudDisp.mu.Unlock() + if diff := cmp.Diff(c.nudDisp.events, []testEntryEventInfo(nil)); diff != "" { + return fmt.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + + return nil +} + +// TestNeighborCacheOverflow verifies that the LRU cache eviction strategy +// respects the dynamic entry count. +func TestNeighborCacheOverflow(t *testing.T) { + config := DefaultNUDConfigurations() + // Stay in Reachable so the cache can overflow + config.BaseReachableTime = infiniteDuration + config.MinRandomFactor = 1 + config.MaxRandomFactor = 1 + + c := newTestContext(config) + opts := overflowOptions{ + startAtEntryIndex: 0, + } + if err := c.overflowCache(opts); err != nil { + t.Errorf("c.overflowCache(%+v): %s", opts, err) + } +} + +// TestNeighborCacheRemoveEntryThenOverflow verifies that the LRU cache +// eviction strategy respects the dynamic entry count when an entry is removed. +func TestNeighborCacheRemoveEntryThenOverflow(t *testing.T) { + config := DefaultNUDConfigurations() + // Stay in Reachable so the cache can overflow + config.BaseReachableTime = infiniteDuration + config.MinRandomFactor = 1 + config.MaxRandomFactor = 1 + + c := newTestContext(config) + + // Add a dynamic entry + entry, ok := c.store.entry(0) + if !ok { + t.Fatalf("c.store.entry(0) not found") + } + _, _, err := c.neigh.entry(entry.Addr, entry.LocalAddr, c.linkRes, nil) + if err != tcpip.ErrWouldBlock { + t.Errorf("got c.neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) + } + c.clock.advance(c.neigh.config().RetransmitTimer) + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: 1, + Addr: entry.Addr, + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: 1, + Addr: entry.Addr, + LinkAddr: entry.LinkAddr, + State: Reachable, + }, + } + c.nudDisp.mu.Lock() + diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...) + c.nudDisp.events = nil + c.nudDisp.mu.Unlock() + if diff != "" { + t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + + // Remove the entry + c.neigh.removeEntry(entry.Addr) + + { + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestRemoved, + NICID: 1, + Addr: entry.Addr, + LinkAddr: entry.LinkAddr, + State: Reachable, + }, + } + c.nudDisp.mu.Lock() + diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...) + c.nudDisp.events = nil + c.nudDisp.mu.Unlock() + if diff != "" { + t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + } + + opts := overflowOptions{ + startAtEntryIndex: 0, + } + if err := c.overflowCache(opts); err != nil { + t.Errorf("c.overflowCache(%+v): %s", opts, err) + } +} + +// TestNeighborCacheDuplicateStaticEntryWithSameLinkAddress verifies that +// adding a duplicate static entry with the same link address does not dispatch +// any events. +func TestNeighborCacheDuplicateStaticEntryWithSameLinkAddress(t *testing.T) { + config := DefaultNUDConfigurations() + c := newTestContext(config) + + // Add a static entry + entry, ok := c.store.entry(0) + if !ok { + t.Fatalf("c.store.entry(0) not found") + } + staticLinkAddr := entry.LinkAddr + "static" + c.neigh.addStaticEntry(entry.Addr, staticLinkAddr) + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: 1, + Addr: entry.Addr, + LinkAddr: staticLinkAddr, + State: Static, + }, + } + c.nudDisp.mu.Lock() + diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...) + c.nudDisp.events = nil + c.nudDisp.mu.Unlock() + if diff != "" { + t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + + // Remove the static entry that was just added + c.neigh.addStaticEntry(entry.Addr, staticLinkAddr) + + // No more events should have been dispatched. + c.nudDisp.mu.Lock() + defer c.nudDisp.mu.Unlock() + if diff := cmp.Diff(c.nudDisp.events, []testEntryEventInfo(nil)); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } +} + +// TestNeighborCacheDuplicateStaticEntryWithDifferentLinkAddress verifies that +// adding a duplicate static entry with a different link address dispatches a +// change event. +func TestNeighborCacheDuplicateStaticEntryWithDifferentLinkAddress(t *testing.T) { + config := DefaultNUDConfigurations() + c := newTestContext(config) + + // Add a static entry + entry, ok := c.store.entry(0) + if !ok { + t.Fatalf("c.store.entry(0) not found") + } + staticLinkAddr := entry.LinkAddr + "static" + c.neigh.addStaticEntry(entry.Addr, staticLinkAddr) + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: 1, + Addr: entry.Addr, + LinkAddr: staticLinkAddr, + State: Static, + }, + } + c.nudDisp.mu.Lock() + diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...) + c.nudDisp.events = nil + c.nudDisp.mu.Unlock() + if diff != "" { + t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + + // Add a duplicate entry with a different link address + staticLinkAddr += "duplicate" + c.neigh.addStaticEntry(entry.Addr, staticLinkAddr) + { + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestChanged, + NICID: 1, + Addr: entry.Addr, + LinkAddr: staticLinkAddr, + State: Static, + }, + } + c.nudDisp.mu.Lock() + defer c.nudDisp.mu.Unlock() + if diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + } +} + +// TestNeighborCacheRemoveStaticEntryThenOverflow verifies that the LRU cache +// eviction strategy respects the dynamic entry count when a static entry is +// added then removed. In this case, the dynamic entry count shouldn't have +// been touched. +func TestNeighborCacheRemoveStaticEntryThenOverflow(t *testing.T) { + config := DefaultNUDConfigurations() + // Stay in Reachable so the cache can overflow + config.BaseReachableTime = infiniteDuration + config.MinRandomFactor = 1 + config.MaxRandomFactor = 1 + + c := newTestContext(config) + + // Add a static entry + entry, ok := c.store.entry(0) + if !ok { + t.Fatalf("c.store.entry(0) not found") + } + staticLinkAddr := entry.LinkAddr + "static" + c.neigh.addStaticEntry(entry.Addr, staticLinkAddr) + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: 1, + Addr: entry.Addr, + LinkAddr: staticLinkAddr, + State: Static, + }, + } + c.nudDisp.mu.Lock() + diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...) + c.nudDisp.events = nil + c.nudDisp.mu.Unlock() + if diff != "" { + t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + + // Remove the static entry that was just added + c.neigh.removeEntry(entry.Addr) + { + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestRemoved, + NICID: 1, + Addr: entry.Addr, + LinkAddr: staticLinkAddr, + State: Static, + }, + } + c.nudDisp.mu.Lock() + diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...) + c.nudDisp.events = nil + c.nudDisp.mu.Unlock() + if diff != "" { + t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + } + + opts := overflowOptions{ + startAtEntryIndex: 0, + } + if err := c.overflowCache(opts); err != nil { + t.Errorf("c.overflowCache(%+v): %s", opts, err) + } +} + +// TestNeighborCacheOverwriteWithStaticEntryThenOverflow verifies that the LRU +// cache eviction strategy keeps count of the dynamic entry count when an entry +// is overwritten by a static entry. Static entries should not count towards +// the size of the LRU cache. +func TestNeighborCacheOverwriteWithStaticEntryThenOverflow(t *testing.T) { + config := DefaultNUDConfigurations() + // Stay in Reachable so the cache can overflow + config.BaseReachableTime = infiniteDuration + config.MinRandomFactor = 1 + config.MaxRandomFactor = 1 + + c := newTestContext(config) + + // Add a dynamic entry + entry, ok := c.store.entry(0) + if !ok { + t.Fatalf("c.store.entry(0) not found") + } + _, _, err := c.neigh.entry(entry.Addr, entry.LocalAddr, c.linkRes, nil) + if err != tcpip.ErrWouldBlock { + t.Errorf("got c.neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) + } + c.clock.advance(typicalLatency) + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: 1, + Addr: entry.Addr, + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: 1, + Addr: entry.Addr, + LinkAddr: entry.LinkAddr, + State: Reachable, + }, + } + c.nudDisp.mu.Lock() + diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...) + c.nudDisp.events = nil + c.nudDisp.mu.Unlock() + if diff != "" { + t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + + // Override the entry with a static one using the same address + staticLinkAddr := entry.LinkAddr + "static" + c.neigh.addStaticEntry(entry.Addr, staticLinkAddr) + { + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestRemoved, + NICID: 1, + Addr: entry.Addr, + LinkAddr: entry.LinkAddr, + State: Reachable, + }, + { + EventType: entryTestAdded, + NICID: 1, + Addr: entry.Addr, + LinkAddr: staticLinkAddr, + State: Static, + }, + } + c.nudDisp.mu.Lock() + diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...) + c.nudDisp.events = nil + c.nudDisp.mu.Unlock() + if diff != "" { + t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + } + + opts := overflowOptions{ + startAtEntryIndex: 1, + wantStaticEntries: []NeighborEntry{ + { + Addr: entry.Addr, + LocalAddr: "", // static entries don't need a local address + LinkAddr: staticLinkAddr, + State: Static, + }, + }, + } + if err := c.overflowCache(opts); err != nil { + t.Errorf("c.overflowCache(%+v): %s", opts, err) + } +} + +func TestNeighborCacheNotifiesWaker(t *testing.T) { + config := DefaultNUDConfigurations() + + nudDisp := testNUDDispatcher{} + clock := newFakeClock() + neigh := newTestNeighborCache(&nudDisp, config, clock) + store := newTestEntryStore() + linkRes := &testNeighborResolver{ + clock: clock, + neigh: neigh, + entries: store, + delay: typicalLatency, + } + + w := sleep.Waker{} + s := sleep.Sleeper{} + const wakerID = 1 + s.AddWaker(&w, wakerID) + + entry, ok := store.entry(0) + if !ok { + t.Fatalf("store.entry(0) not found") + } + _, doneCh, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, &w) + if err != tcpip.ErrWouldBlock { + t.Fatalf("got neigh.entry(%s, %s, _, _ = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) + } + if doneCh == nil { + t.Fatalf("expected done channel from neigh.entry(%s, %s, _, _)", entry.Addr, entry.LocalAddr) + } + clock.advance(typicalLatency) + + select { + case <-doneCh: + default: + t.Fatal("expected notification from done channel") + } + + id, ok := s.Fetch(false /* block */) + if !ok { + t.Errorf("expected waker to be notified after neigh.entry(%s, %s, _, _)", entry.Addr, entry.LocalAddr) + } + if id != wakerID { + t.Errorf("got s.Fetch(false) = %d, want = %d", id, wakerID) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: 1, + Addr: entry.Addr, + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: 1, + Addr: entry.Addr, + LinkAddr: entry.LinkAddr, + State: Reachable, + }, + } + nudDisp.mu.Lock() + defer nudDisp.mu.Unlock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } +} + +func TestNeighborCacheRemoveWaker(t *testing.T) { + config := DefaultNUDConfigurations() + + nudDisp := testNUDDispatcher{} + clock := newFakeClock() + neigh := newTestNeighborCache(&nudDisp, config, clock) + store := newTestEntryStore() + linkRes := &testNeighborResolver{ + clock: clock, + neigh: neigh, + entries: store, + delay: typicalLatency, + } + + w := sleep.Waker{} + s := sleep.Sleeper{} + const wakerID = 1 + s.AddWaker(&w, wakerID) + + entry, ok := store.entry(0) + if !ok { + t.Fatalf("store.entry(0) not found") + } + _, doneCh, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, &w) + if err != tcpip.ErrWouldBlock { + t.Fatalf("got neigh.entry(%s, %s, _, _) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) + } + if doneCh == nil { + t.Fatalf("expected done channel from neigh.entry(%s, %s, _, _)", entry.Addr, entry.LocalAddr) + } + + // Remove the waker before the neighbor cache has the opportunity to send a + // notification. + neigh.removeWaker(entry.Addr, &w) + clock.advance(typicalLatency) + + select { + case <-doneCh: + default: + t.Fatal("expected notification from done channel") + } + + if id, ok := s.Fetch(false /* block */); ok { + t.Errorf("unexpected notification from waker with id %d", id) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: 1, + Addr: entry.Addr, + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: 1, + Addr: entry.Addr, + LinkAddr: entry.LinkAddr, + State: Reachable, + }, + } + nudDisp.mu.Lock() + defer nudDisp.mu.Unlock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } +} + +func TestNeighborCacheAddStaticEntryThenOverflow(t *testing.T) { + config := DefaultNUDConfigurations() + // Stay in Reachable so the cache can overflow + config.BaseReachableTime = infiniteDuration + config.MinRandomFactor = 1 + config.MaxRandomFactor = 1 + + c := newTestContext(config) + + entry, ok := c.store.entry(0) + if !ok { + t.Fatalf("c.store.entry(0) not found") + } + c.neigh.addStaticEntry(entry.Addr, entry.LinkAddr) + e, _, err := c.neigh.entry(entry.Addr, "", nil, nil) + if err != nil { + t.Errorf("unexpected error from c.neigh.entry(%s, \"\", nil nil): %s", entry.Addr, err) + } + want := NeighborEntry{ + Addr: entry.Addr, + LocalAddr: "", // static entries don't need a local address + LinkAddr: entry.LinkAddr, + State: Static, + } + if diff := cmp.Diff(e, want, entryDiffOpts()...); diff != "" { + t.Errorf("c.neigh.entry(%s, \"\", nil, nil) mismatch (-got, +want):\n%s", entry.Addr, diff) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: 1, + Addr: entry.Addr, + LinkAddr: entry.LinkAddr, + State: Static, + }, + } + c.nudDisp.mu.Lock() + diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...) + c.nudDisp.events = nil + c.nudDisp.mu.Unlock() + if diff != "" { + t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + + opts := overflowOptions{ + startAtEntryIndex: 1, + wantStaticEntries: []NeighborEntry{ + { + Addr: entry.Addr, + LocalAddr: "", // static entries don't need a local address + LinkAddr: entry.LinkAddr, + State: Static, + }, + }, + } + if err := c.overflowCache(opts); err != nil { + t.Errorf("c.overflowCache(%+v): %s", opts, err) + } +} + +func TestNeighborCacheClear(t *testing.T) { + config := DefaultNUDConfigurations() + + nudDisp := testNUDDispatcher{} + clock := newFakeClock() + neigh := newTestNeighborCache(&nudDisp, config, clock) + store := newTestEntryStore() + linkRes := &testNeighborResolver{ + clock: clock, + neigh: neigh, + entries: store, + delay: typicalLatency, + } + + // Add a dynamic entry. + entry, ok := store.entry(0) + if !ok { + t.Fatalf("store.entry(0) not found") + } + _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil) + if err != tcpip.ErrWouldBlock { + t.Errorf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) + } + clock.advance(typicalLatency) + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: 1, + Addr: entry.Addr, + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: 1, + Addr: entry.Addr, + LinkAddr: entry.LinkAddr, + State: Reachable, + }, + } + nudDisp.mu.Lock() + diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...) + nudDisp.events = nil + nudDisp.mu.Unlock() + if diff != "" { + t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + + // Add a static entry. + neigh.addStaticEntry(entryTestAddr1, entryTestLinkAddr1) + + { + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: 1, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Static, + }, + } + nudDisp.mu.Lock() + diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...) + nudDisp.events = nil + nudDisp.mu.Unlock() + if diff != "" { + t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + } + + // Clear shoud remove both dynamic and static entries. + neigh.clear() + + // Remove events dispatched from clear() have no deterministic order so they + // need to be sorted beforehand. + wantUnsortedEvents := []testEntryEventInfo{ + { + EventType: entryTestRemoved, + NICID: 1, + Addr: entry.Addr, + LinkAddr: entry.LinkAddr, + State: Reachable, + }, + { + EventType: entryTestRemoved, + NICID: 1, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Static, + }, + } + nudDisp.mu.Lock() + defer nudDisp.mu.Unlock() + if diff := cmp.Diff(nudDisp.events, wantUnsortedEvents, eventDiffOptsWithSort()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } +} + +// TestNeighborCacheClearThenOverflow verifies that the LRU cache eviction +// strategy keeps count of the dynamic entry count when all entries are +// cleared. +func TestNeighborCacheClearThenOverflow(t *testing.T) { + config := DefaultNUDConfigurations() + // Stay in Reachable so the cache can overflow + config.BaseReachableTime = infiniteDuration + config.MinRandomFactor = 1 + config.MaxRandomFactor = 1 + + c := newTestContext(config) + + // Add a dynamic entry + entry, ok := c.store.entry(0) + if !ok { + t.Fatalf("c.store.entry(0) not found") + } + _, _, err := c.neigh.entry(entry.Addr, entry.LocalAddr, c.linkRes, nil) + if err != tcpip.ErrWouldBlock { + t.Errorf("got c.neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) + } + c.clock.advance(typicalLatency) + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: 1, + Addr: entry.Addr, + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: 1, + Addr: entry.Addr, + LinkAddr: entry.LinkAddr, + State: Reachable, + }, + } + c.nudDisp.mu.Lock() + diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...) + c.nudDisp.events = nil + c.nudDisp.mu.Unlock() + if diff != "" { + t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + + // Clear the cache. + c.neigh.clear() + { + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestRemoved, + NICID: 1, + Addr: entry.Addr, + LinkAddr: entry.LinkAddr, + State: Reachable, + }, + } + c.nudDisp.mu.Lock() + diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...) + c.nudDisp.events = nil + c.nudDisp.mu.Unlock() + if diff != "" { + t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + } + + opts := overflowOptions{ + startAtEntryIndex: 0, + } + if err := c.overflowCache(opts); err != nil { + t.Errorf("c.overflowCache(%+v): %s", opts, err) + } +} + +func TestNeighborCacheKeepFrequentlyUsed(t *testing.T) { + config := DefaultNUDConfigurations() + // Stay in Reachable so the cache can overflow + config.BaseReachableTime = infiniteDuration + config.MinRandomFactor = 1 + config.MaxRandomFactor = 1 + + nudDisp := testNUDDispatcher{} + clock := newFakeClock() + neigh := newTestNeighborCache(&nudDisp, config, clock) + store := newTestEntryStore() + linkRes := &testNeighborResolver{ + clock: clock, + neigh: neigh, + entries: store, + delay: typicalLatency, + } + + frequentlyUsedEntry, ok := store.entry(0) + if !ok { + t.Fatalf("store.entry(0) not found") + } + + // The following logic is very similar to overflowCache, but + // periodically refreshes the frequently used entry. + + // Fill the neighbor cache to capacity + for i := 0; i < neighborCacheSize; i++ { + entry, ok := store.entry(i) + if !ok { + t.Fatalf("store.entry(%d) not found", i) + } + _, doneCh, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil) + if err != tcpip.ErrWouldBlock { + t.Errorf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) + } + clock.advance(typicalLatency) + select { + case <-doneCh: + default: + t.Fatalf("expected notification from done channel returned by neigh.entry(%s, %s, _, nil)", entry.Addr, entry.LocalAddr) + } + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: 1, + Addr: entry.Addr, + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: 1, + Addr: entry.Addr, + LinkAddr: entry.LinkAddr, + State: Reachable, + }, + } + nudDisp.mu.Lock() + diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...) + nudDisp.events = nil + nudDisp.mu.Unlock() + if diff != "" { + t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + } + + // Keep adding more entries + for i := neighborCacheSize; i < store.size(); i++ { + // Periodically refresh the frequently used entry + if i%(neighborCacheSize/2) == 0 { + _, _, err := neigh.entry(frequentlyUsedEntry.Addr, frequentlyUsedEntry.LocalAddr, linkRes, nil) + if err != nil { + t.Errorf("unexpected error from neigh.entry(%s, %s, _, nil): %s", frequentlyUsedEntry.Addr, frequentlyUsedEntry.LocalAddr, err) + } + } + + entry, ok := store.entry(i) + if !ok { + t.Fatalf("store.entry(%d) not found", i) + } + _, doneCh, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil) + if err != tcpip.ErrWouldBlock { + t.Errorf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) + } + clock.advance(typicalLatency) + select { + case <-doneCh: + default: + t.Fatalf("expected notification from done channel returned by neigh.entry(%s, %s, _, nil)", entry.Addr, entry.LocalAddr) + } + + // An entry should have been removed, as per the LRU eviction strategy + removedEntry, ok := store.entry(i - neighborCacheSize + 1) + if !ok { + t.Fatalf("store.entry(%d) not found", i-neighborCacheSize+1) + } + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestRemoved, + NICID: 1, + Addr: removedEntry.Addr, + LinkAddr: removedEntry.LinkAddr, + State: Reachable, + }, + { + EventType: entryTestAdded, + NICID: 1, + Addr: entry.Addr, + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: 1, + Addr: entry.Addr, + LinkAddr: entry.LinkAddr, + State: Reachable, + }, + } + nudDisp.mu.Lock() + diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...) + nudDisp.events = nil + nudDisp.mu.Unlock() + if diff != "" { + t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + } + + // Expect to find only the frequently used entry and the most recent entries. + // The order of entries reported by entries() is undeterministic, so entries + // have to be sorted before comparison. + wantUnsortedEntries := []NeighborEntry{ + { + Addr: frequentlyUsedEntry.Addr, + LocalAddr: frequentlyUsedEntry.LocalAddr, + LinkAddr: frequentlyUsedEntry.LinkAddr, + State: Reachable, + }, + } + + for i := store.size() - neighborCacheSize + 1; i < store.size(); i++ { + entry, ok := store.entry(i) + if !ok { + t.Fatalf("store.entry(%d) not found", i) + } + wantEntry := NeighborEntry{ + Addr: entry.Addr, + LocalAddr: entry.LocalAddr, + LinkAddr: entry.LinkAddr, + State: Reachable, + } + wantUnsortedEntries = append(wantUnsortedEntries, wantEntry) + } + + if diff := cmp.Diff(neigh.entries(), wantUnsortedEntries, entryDiffOptsWithSort()...); diff != "" { + t.Errorf("neighbor entries mismatch (-got, +want):\n%s", diff) + } + + // No more events should have been dispatched. + nudDisp.mu.Lock() + defer nudDisp.mu.Unlock() + if diff := cmp.Diff(nudDisp.events, []testEntryEventInfo(nil)); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } +} + +func TestNeighborCacheConcurrent(t *testing.T) { + const concurrentProcesses = 16 + + config := DefaultNUDConfigurations() + + nudDisp := testNUDDispatcher{} + clock := newFakeClock() + neigh := newTestNeighborCache(&nudDisp, config, clock) + store := newTestEntryStore() + linkRes := &testNeighborResolver{ + clock: clock, + neigh: neigh, + entries: store, + delay: typicalLatency, + } + + storeEntries := store.entries() + for _, entry := range storeEntries { + var wg sync.WaitGroup + for r := 0; r < concurrentProcesses; r++ { + wg.Add(1) + go func(entry NeighborEntry) { + defer wg.Done() + e, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil) + if err != nil && err != tcpip.ErrWouldBlock { + t.Errorf("got neigh.entry(%s, %s, _, nil) = (%+v, _, %s), want (_, _, nil) or (_, _, %s)", entry.Addr, entry.LocalAddr, e, err, tcpip.ErrWouldBlock) + } + }(entry) + } + + // Wait for all gorountines to send a request + wg.Wait() + + // Process all the requests for a single entry concurrently + clock.advance(typicalLatency) + } + + // All goroutines add in the same order and add more values than can fit in + // the cache. Our eviction strategy requires that the last entries are + // present, up to the size of the neighbor cache, and the rest are missing. + // The order of entries reported by entries() is undeterministic, so entries + // have to be sorted before comparison. + var wantUnsortedEntries []NeighborEntry + for i := store.size() - neighborCacheSize; i < store.size(); i++ { + entry, ok := store.entry(i) + if !ok { + t.Errorf("store.entry(%d) not found", i) + } + wantEntry := NeighborEntry{ + Addr: entry.Addr, + LocalAddr: entry.LocalAddr, + LinkAddr: entry.LinkAddr, + State: Reachable, + } + wantUnsortedEntries = append(wantUnsortedEntries, wantEntry) + } + + if diff := cmp.Diff(neigh.entries(), wantUnsortedEntries, entryDiffOptsWithSort()...); diff != "" { + t.Errorf("neighbor entries mismatch (-got, +want):\n%s", diff) + } +} + +func TestNeighborCacheReplace(t *testing.T) { + config := DefaultNUDConfigurations() + + nudDisp := testNUDDispatcher{} + clock := newFakeClock() + neigh := newTestNeighborCache(&nudDisp, config, clock) + store := newTestEntryStore() + linkRes := &testNeighborResolver{ + clock: clock, + neigh: neigh, + entries: store, + delay: typicalLatency, + } + + // Add an entry + entry, ok := store.entry(0) + if !ok { + t.Fatalf("store.entry(0) not found") + } + _, doneCh, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil) + if err != tcpip.ErrWouldBlock { + t.Fatalf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) + } + clock.advance(typicalLatency) + select { + case <-doneCh: + default: + t.Fatalf("expected notification from done channel returned by neigh.entry(%s, %s, _, nil)", entry.Addr, entry.LocalAddr) + } + + // Verify the entry exists + e, doneCh, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil) + if err != nil { + t.Errorf("unexpected error from neigh.entry(%s, %s, _, nil): %s", entry.Addr, entry.LocalAddr, err) + } + if doneCh != nil { + t.Errorf("unexpected done channel from neigh.entry(%s, %s, _, nil): %v", entry.Addr, entry.LocalAddr, doneCh) + } + if t.Failed() { + t.FailNow() + } + want := NeighborEntry{ + Addr: entry.Addr, + LocalAddr: entry.LocalAddr, + LinkAddr: entry.LinkAddr, + State: Reachable, + } + if diff := cmp.Diff(e, want, entryDiffOpts()...); diff != "" { + t.Errorf("neigh.entry(%s, %s, _, nil) mismatch (-got, +want):\n%s", entry.Addr, entry.LinkAddr, diff) + } + + // Notify of a link address change + var updatedLinkAddr tcpip.LinkAddress + { + entry, ok := store.entry(1) + if !ok { + t.Fatalf("store.entry(1) not found") + } + updatedLinkAddr = entry.LinkAddr + } + store.set(0, updatedLinkAddr) + neigh.HandleConfirmation(entry.Addr, updatedLinkAddr, ReachabilityConfirmationFlags{ + Solicited: false, + Override: true, + IsRouter: false, + }) + + // Requesting the entry again should start address resolution + { + _, doneCh, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil) + if err != tcpip.ErrWouldBlock { + t.Fatalf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) + } + clock.advance(config.DelayFirstProbeTime + typicalLatency) + select { + case <-doneCh: + default: + t.Fatalf("expected notification from done channel returned by neigh.entry(%s, %s, _, nil)", entry.Addr, entry.LocalAddr) + } + } + + // Verify the entry's new link address + { + e, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil) + clock.advance(typicalLatency) + if err != nil { + t.Errorf("unexpected error from neigh.entry(%s, %s, _, nil): %s", entry.Addr, entry.LocalAddr, err) + } + want = NeighborEntry{ + Addr: entry.Addr, + LocalAddr: entry.LocalAddr, + LinkAddr: updatedLinkAddr, + State: Reachable, + } + if diff := cmp.Diff(e, want, entryDiffOpts()...); diff != "" { + t.Errorf("neigh.entry(%s, %s, _, nil) mismatch (-got, +want):\n%s", entry.Addr, entry.LocalAddr, diff) + } + } +} + +func TestNeighborCacheResolutionFailed(t *testing.T) { + config := DefaultNUDConfigurations() + + nudDisp := testNUDDispatcher{} + clock := newFakeClock() + neigh := newTestNeighborCache(&nudDisp, config, clock) + store := newTestEntryStore() + + var requestCount uint32 + linkRes := &testNeighborResolver{ + clock: clock, + neigh: neigh, + entries: store, + delay: typicalLatency, + onLinkAddressRequest: func() { + atomic.AddUint32(&requestCount, 1) + }, + } + + // First, sanity check that resolution is working + entry, ok := store.entry(0) + if !ok { + t.Fatalf("store.entry(0) not found") + } + if _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil); err != tcpip.ErrWouldBlock { + t.Fatalf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) + } + clock.advance(typicalLatency) + got, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil) + if err != nil { + t.Fatalf("unexpected error from neigh.entry(%s, %s, _, nil): %s", entry.Addr, entry.LocalAddr, err) + } + want := NeighborEntry{ + Addr: entry.Addr, + LocalAddr: entry.LocalAddr, + LinkAddr: entry.LinkAddr, + State: Reachable, + } + if diff := cmp.Diff(got, want, entryDiffOpts()...); diff != "" { + t.Errorf("neigh.entry(%s, %s, _, nil) mismatch (-got, +want):\n%s", entry.Addr, entry.LocalAddr, diff) + } + + // Verify that address resolution for an unknown address returns ErrNoLinkAddress + before := atomic.LoadUint32(&requestCount) + + entry.Addr += "2" + if _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil); err != tcpip.ErrWouldBlock { + t.Fatalf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) + } + waitFor := config.DelayFirstProbeTime + typicalLatency*time.Duration(config.MaxMulticastProbes) + clock.advance(waitFor) + if _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil); err != tcpip.ErrNoLinkAddress { + t.Fatalf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrNoLinkAddress) + } + + maxAttempts := neigh.config().MaxUnicastProbes + if got, want := atomic.LoadUint32(&requestCount)-before, maxAttempts; got != want { + t.Errorf("got link address request count = %d, want = %d", got, want) + } +} + +// TestNeighborCacheResolutionTimeout simulates sending MaxMulticastProbes +// probes and not retrieving a confirmation before the duration defined by +// MaxMulticastProbes * RetransmitTimer. +func TestNeighborCacheResolutionTimeout(t *testing.T) { + config := DefaultNUDConfigurations() + config.RetransmitTimer = time.Millisecond // small enough to cause timeout + + clock := newFakeClock() + neigh := newTestNeighborCache(nil, config, clock) + store := newTestEntryStore() + linkRes := &testNeighborResolver{ + clock: clock, + neigh: neigh, + entries: store, + delay: time.Minute, // large enough to cause timeout + } + + entry, ok := store.entry(0) + if !ok { + t.Fatalf("store.entry(0) not found") + } + if _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil); err != tcpip.ErrWouldBlock { + t.Fatalf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) + } + waitFor := config.RetransmitTimer * time.Duration(config.MaxMulticastProbes) + clock.advance(waitFor) + if _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil); err != tcpip.ErrNoLinkAddress { + t.Fatalf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrNoLinkAddress) + } +} + +// TestNeighborCacheStaticResolution checks that static link addresses are +// resolved immediately and don't send resolution requests. +func TestNeighborCacheStaticResolution(t *testing.T) { + config := DefaultNUDConfigurations() + clock := newFakeClock() + neigh := newTestNeighborCache(nil, config, clock) + store := newTestEntryStore() + linkRes := &testNeighborResolver{ + clock: clock, + neigh: neigh, + entries: store, + delay: typicalLatency, + } + + got, _, err := neigh.entry(testEntryBroadcastAddr, testEntryLocalAddr, linkRes, nil) + if err != nil { + t.Fatalf("unexpected error from neigh.entry(%s, %s, _, nil): %s", testEntryBroadcastAddr, testEntryLocalAddr, err) + } + want := NeighborEntry{ + Addr: testEntryBroadcastAddr, + LocalAddr: testEntryLocalAddr, + LinkAddr: testEntryBroadcastLinkAddr, + State: Static, + } + if diff := cmp.Diff(got, want, entryDiffOpts()...); diff != "" { + t.Errorf("neigh.entry(%s, %s, _, nil) mismatch (-got, +want):\n%s", testEntryBroadcastAddr, testEntryLocalAddr, diff) + } +} + +func BenchmarkCacheClear(b *testing.B) { + b.StopTimer() + config := DefaultNUDConfigurations() + clock := &tcpip.StdClock{} + neigh := newTestNeighborCache(nil, config, clock) + store := newTestEntryStore() + linkRes := &testNeighborResolver{ + clock: clock, + neigh: neigh, + entries: store, + delay: 0, + } + + // Clear for every possible size of the cache + for cacheSize := 0; cacheSize < neighborCacheSize; cacheSize++ { + // Fill the neighbor cache to capacity. + for i := 0; i < cacheSize; i++ { + entry, ok := store.entry(i) + if !ok { + b.Fatalf("store.entry(%d) not found", i) + } + _, doneCh, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil) + if err != tcpip.ErrWouldBlock { + b.Fatalf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) + } + if doneCh != nil { + <-doneCh + } + } + + b.StartTimer() + neigh.clear() + b.StopTimer() + } +} diff --git a/pkg/tcpip/stack/neighbor_entry.go b/pkg/tcpip/stack/neighbor_entry.go new file mode 100644 index 000000000..0068cacb8 --- /dev/null +++ b/pkg/tcpip/stack/neighbor_entry.go @@ -0,0 +1,482 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package stack + +import ( + "fmt" + "sync" + "time" + + "gvisor.dev/gvisor/pkg/sleep" + "gvisor.dev/gvisor/pkg/tcpip" +) + +// NeighborEntry describes a neighboring device in the local network. +type NeighborEntry struct { + Addr tcpip.Address + LocalAddr tcpip.Address + LinkAddr tcpip.LinkAddress + State NeighborState + UpdatedAt time.Time +} + +// NeighborState defines the state of a NeighborEntry within the Neighbor +// Unreachability Detection state machine, as per RFC 4861 section 7.3.2. +type NeighborState uint8 + +const ( + // Unknown means reachability has not been verified yet. This is the initial + // state of entries that have been created automatically by the Neighbor + // Unreachability Detection state machine. + Unknown NeighborState = iota + // Incomplete means that there is an outstanding request to resolve the + // address. + Incomplete + // Reachable means the path to the neighbor is functioning properly for both + // receive and transmit paths. + Reachable + // Stale means reachability to the neighbor is unknown, but packets are still + // able to be transmitted to the possibly stale link address. + Stale + // Delay means reachability to the neighbor is unknown and pending + // confirmation from an upper-level protocol like TCP, but packets are still + // able to be transmitted to the possibly stale link address. + Delay + // Probe means a reachability confirmation is actively being sought by + // periodically retransmitting reachability probes until a reachability + // confirmation is received, or until the max amount of probes has been sent. + Probe + // Static describes entries that have been explicitly added by the user. They + // do not expire and are not deleted until explicitly removed. + Static + // Failed means traffic should not be sent to this neighbor since attempts of + // reachability have returned inconclusive. + Failed +) + +// neighborEntry implements a neighbor entry's individual node behavior, as per +// RFC 4861 section 7.3.3. Neighbor Unreachability Detection operates in +// parallel with the sending of packets to a neighbor, necessitating the +// entry's lock to be acquired for all operations. +type neighborEntry struct { + neighborEntryEntry + + nic *NIC + protocol tcpip.NetworkProtocolNumber + + // linkRes provides the functionality to send reachability probes, used in + // Neighbor Unreachability Detection. + linkRes LinkAddressResolver + + // nudState points to the Neighbor Unreachability Detection configuration. + nudState *NUDState + + // mu protects the fields below. + mu sync.RWMutex + + neigh NeighborEntry + + // wakers is a set of waiters for address resolution result. Anytime state + // transitions out of incomplete these waiters are notified. It is nil iff + // address resolution is ongoing and no clients are waiting for the result. + wakers map[*sleep.Waker]struct{} + + // done is used to allow callers to wait on address resolution. It is nil + // iff nudState is not Reachable and address resolution is not yet in + // progress. + done chan struct{} + + isRouter bool + job *tcpip.Job +} + +// newNeighborEntry creates a neighbor cache entry starting at the default +// state, Unknown. Transition out of Unknown by calling either +// `handlePacketQueuedLocked` or `handleProbeLocked` on the newly created +// neighborEntry. +func newNeighborEntry(nic *NIC, remoteAddr tcpip.Address, localAddr tcpip.Address, nudState *NUDState, linkRes LinkAddressResolver) *neighborEntry { + return &neighborEntry{ + nic: nic, + linkRes: linkRes, + nudState: nudState, + neigh: NeighborEntry{ + Addr: remoteAddr, + LocalAddr: localAddr, + State: Unknown, + }, + } +} + +// newStaticNeighborEntry creates a neighbor cache entry starting at the Static +// state. The entry can only transition out of Static by directly calling +// `setStateLocked`. +func newStaticNeighborEntry(nic *NIC, addr tcpip.Address, linkAddr tcpip.LinkAddress, state *NUDState) *neighborEntry { + if nic.stack.nudDisp != nil { + nic.stack.nudDisp.OnNeighborAdded(nic.id, addr, linkAddr, Static, time.Now()) + } + return &neighborEntry{ + nic: nic, + nudState: state, + neigh: NeighborEntry{ + Addr: addr, + LinkAddr: linkAddr, + State: Static, + UpdatedAt: time.Now(), + }, + } +} + +// addWaker adds w to the list of wakers waiting for address resolution. +// Assumes the entry has already been appropriately locked. +func (e *neighborEntry) addWakerLocked(w *sleep.Waker) { + if w == nil { + return + } + if e.wakers == nil { + e.wakers = make(map[*sleep.Waker]struct{}) + } + e.wakers[w] = struct{}{} +} + +// notifyWakersLocked notifies those waiting for address resolution, whether it +// succeeded or failed. Assumes the entry has already been appropriately locked. +func (e *neighborEntry) notifyWakersLocked() { + for w := range e.wakers { + w.Assert() + } + e.wakers = nil + if ch := e.done; ch != nil { + close(ch) + e.done = nil + } +} + +// dispatchAddEventLocked signals to stack's NUD Dispatcher that the entry has +// been added. +func (e *neighborEntry) dispatchAddEventLocked(nextState NeighborState) { + if nudDisp := e.nic.stack.nudDisp; nudDisp != nil { + nudDisp.OnNeighborAdded(e.nic.id, e.neigh.Addr, e.neigh.LinkAddr, nextState, time.Now()) + } +} + +// dispatchChangeEventLocked signals to stack's NUD Dispatcher that the entry +// has changed state or link-layer address. +func (e *neighborEntry) dispatchChangeEventLocked(nextState NeighborState) { + if nudDisp := e.nic.stack.nudDisp; nudDisp != nil { + nudDisp.OnNeighborChanged(e.nic.id, e.neigh.Addr, e.neigh.LinkAddr, nextState, time.Now()) + } +} + +// dispatchRemoveEventLocked signals to stack's NUD Dispatcher that the entry +// has been removed. +func (e *neighborEntry) dispatchRemoveEventLocked() { + if nudDisp := e.nic.stack.nudDisp; nudDisp != nil { + nudDisp.OnNeighborRemoved(e.nic.id, e.neigh.Addr, e.neigh.LinkAddr, e.neigh.State, time.Now()) + } +} + +// setStateLocked transitions the entry to the specified state immediately. +// +// Follows the logic defined in RFC 4861 section 7.3.3. +// +// e.mu MUST be locked. +func (e *neighborEntry) setStateLocked(next NeighborState) { + // Cancel the previously scheduled action, if there is one. Entries in + // Unknown, Stale, or Static state do not have scheduled actions. + if timer := e.job; timer != nil { + timer.Cancel() + } + + prev := e.neigh.State + e.neigh.State = next + e.neigh.UpdatedAt = time.Now() + config := e.nudState.Config() + + switch next { + case Incomplete: + var retryCounter uint32 + var sendMulticastProbe func() + + sendMulticastProbe = func() { + if retryCounter == config.MaxMulticastProbes { + // "If no Neighbor Advertisement is received after + // MAX_MULTICAST_SOLICIT solicitations, address resolution has failed. + // The sender MUST return ICMP destination unreachable indications with + // code 3 (Address Unreachable) for each packet queued awaiting address + // resolution." - RFC 4861 section 7.2.2 + // + // There is no need to send an ICMP destination unreachable indication + // since the failure to resolve the address is expected to only occur + // on this node. Thus, redirecting traffic is currently not supported. + // + // "If the error occurs on a node other than the node originating the + // packet, an ICMP error message is generated. If the error occurs on + // the originating node, an implementation is not required to actually + // create and send an ICMP error packet to the source, as long as the + // upper-layer sender is notified through an appropriate mechanism + // (e.g. return value from a procedure call). Note, however, that an + // implementation may find it convenient in some cases to return errors + // to the sender by taking the offending packet, generating an ICMP + // error message, and then delivering it (locally) through the generic + // error-handling routines.' - RFC 4861 section 2.1 + e.dispatchRemoveEventLocked() + e.setStateLocked(Failed) + return + } + + if err := e.linkRes.LinkAddressRequest(e.neigh.Addr, e.neigh.LocalAddr, "", e.nic.linkEP); err != nil { + // There is no need to log the error here; the NUD implementation may + // assume a working link. A valid link should be the responsibility of + // the NIC/stack.LinkEndpoint. + e.dispatchRemoveEventLocked() + e.setStateLocked(Failed) + return + } + + retryCounter++ + e.job = e.nic.stack.newJob(&e.mu, sendMulticastProbe) + e.job.Schedule(config.RetransmitTimer) + } + + sendMulticastProbe() + + case Reachable: + e.job = e.nic.stack.newJob(&e.mu, func() { + e.dispatchChangeEventLocked(Stale) + e.setStateLocked(Stale) + }) + e.job.Schedule(e.nudState.ReachableTime()) + + case Delay: + e.job = e.nic.stack.newJob(&e.mu, func() { + e.dispatchChangeEventLocked(Probe) + e.setStateLocked(Probe) + }) + e.job.Schedule(config.DelayFirstProbeTime) + + case Probe: + var retryCounter uint32 + var sendUnicastProbe func() + + sendUnicastProbe = func() { + if retryCounter == config.MaxUnicastProbes { + e.dispatchRemoveEventLocked() + e.setStateLocked(Failed) + return + } + + if err := e.linkRes.LinkAddressRequest(e.neigh.Addr, e.neigh.LocalAddr, e.neigh.LinkAddr, e.nic.linkEP); err != nil { + e.dispatchRemoveEventLocked() + e.setStateLocked(Failed) + return + } + + retryCounter++ + if retryCounter == config.MaxUnicastProbes { + e.dispatchRemoveEventLocked() + e.setStateLocked(Failed) + return + } + + e.job = e.nic.stack.newJob(&e.mu, sendUnicastProbe) + e.job.Schedule(config.RetransmitTimer) + } + + sendUnicastProbe() + + case Failed: + e.notifyWakersLocked() + e.job = e.nic.stack.newJob(&e.mu, func() { + e.nic.neigh.removeEntryLocked(e) + }) + e.job.Schedule(config.UnreachableTime) + + case Unknown, Stale, Static: + // Do nothing + + default: + panic(fmt.Sprintf("Invalid state transition from %q to %q", prev, next)) + } +} + +// handlePacketQueuedLocked advances the state machine according to a packet +// being queued for outgoing transmission. +// +// Follows the logic defined in RFC 4861 section 7.3.3. +func (e *neighborEntry) handlePacketQueuedLocked() { + switch e.neigh.State { + case Unknown: + e.dispatchAddEventLocked(Incomplete) + e.setStateLocked(Incomplete) + + case Stale: + e.dispatchChangeEventLocked(Delay) + e.setStateLocked(Delay) + + case Incomplete, Reachable, Delay, Probe, Static, Failed: + // Do nothing + + default: + panic(fmt.Sprintf("Invalid cache entry state: %s", e.neigh.State)) + } +} + +// handleProbeLocked processes an incoming neighbor probe (e.g. ARP request or +// Neighbor Solicitation for ARP or NDP, respectively). +// +// Follows the logic defined in RFC 4861 section 7.2.3. +func (e *neighborEntry) handleProbeLocked(remoteLinkAddr tcpip.LinkAddress) { + // Probes MUST be silently discarded if the target address is tentative, does + // not exist, or not bound to the NIC as per RFC 4861 section 7.2.3. These + // checks MUST be done by the NetworkEndpoint. + + switch e.neigh.State { + case Unknown, Incomplete, Failed: + e.neigh.LinkAddr = remoteLinkAddr + e.dispatchAddEventLocked(Stale) + e.setStateLocked(Stale) + e.notifyWakersLocked() + + case Reachable, Delay, Probe: + if e.neigh.LinkAddr != remoteLinkAddr { + e.neigh.LinkAddr = remoteLinkAddr + e.dispatchChangeEventLocked(Stale) + e.setStateLocked(Stale) + } + + case Stale: + if e.neigh.LinkAddr != remoteLinkAddr { + e.neigh.LinkAddr = remoteLinkAddr + e.dispatchChangeEventLocked(Stale) + } + + case Static: + // Do nothing + + default: + panic(fmt.Sprintf("Invalid cache entry state: %s", e.neigh.State)) + } +} + +// handleConfirmationLocked processes an incoming neighbor confirmation +// (e.g. ARP reply or Neighbor Advertisement for ARP or NDP, respectively). +// +// Follows the state machine defined by RFC 4861 section 7.2.5. +// +// TODO(gvisor.dev/issue/2277): To protect against ARP poisoning and other +// attacks against NDP functions, Secure Neighbor Discovery (SEND) Protocol +// should be deployed where preventing access to the broadcast segment might +// not be possible. SEND uses RSA key pairs to produce Cryptographically +// Generated Addresses (CGA), as defined in RFC 3972. This ensures that the +// claimed source of an NDP message is the owner of the claimed address. +func (e *neighborEntry) handleConfirmationLocked(linkAddr tcpip.LinkAddress, flags ReachabilityConfirmationFlags) { + switch e.neigh.State { + case Incomplete: + if len(linkAddr) == 0 { + // "If the link layer has addresses and no Target Link-Layer Address + // option is included, the receiving node SHOULD silently discard the + // received advertisement." - RFC 4861 section 7.2.5 + break + } + + e.neigh.LinkAddr = linkAddr + if flags.Solicited { + e.dispatchChangeEventLocked(Reachable) + e.setStateLocked(Reachable) + } else { + e.dispatchChangeEventLocked(Stale) + e.setStateLocked(Stale) + } + e.isRouter = flags.IsRouter + e.notifyWakersLocked() + + // "Note that the Override flag is ignored if the entry is in the + // INCOMPLETE state." - RFC 4861 section 7.2.5 + + case Reachable, Stale, Delay, Probe: + sameLinkAddr := e.neigh.LinkAddr == linkAddr + + if !sameLinkAddr { + if !flags.Override { + if e.neigh.State == Reachable { + e.dispatchChangeEventLocked(Stale) + e.setStateLocked(Stale) + } + break + } + + e.neigh.LinkAddr = linkAddr + + if !flags.Solicited { + if e.neigh.State != Stale { + e.dispatchChangeEventLocked(Stale) + e.setStateLocked(Stale) + } else { + // Notify the LinkAddr change, even though NUD state hasn't changed. + e.dispatchChangeEventLocked(e.neigh.State) + } + break + } + } + + if flags.Solicited && (flags.Override || sameLinkAddr) { + if e.neigh.State != Reachable { + e.dispatchChangeEventLocked(Reachable) + } + // Set state to Reachable again to refresh timers. + e.setStateLocked(Reachable) + e.notifyWakersLocked() + } + + if e.isRouter && !flags.IsRouter { + // "In those cases where the IsRouter flag changes from TRUE to FALSE as + // a result of this update, the node MUST remove that router from the + // Default Router List and update the Destination Cache entries for all + // destinations using that neighbor as a router as specified in Section + // 7.3.3. This is needed to detect when a node that is used as a router + // stops forwarding packets due to being configured as a host." + // - RFC 4861 section 7.2.5 + e.nic.mu.Lock() + e.nic.mu.ndp.invalidateDefaultRouter(e.neigh.Addr) + e.nic.mu.Unlock() + } + e.isRouter = flags.IsRouter + + case Unknown, Failed, Static: + // Do nothing + + default: + panic(fmt.Sprintf("Invalid cache entry state: %s", e.neigh.State)) + } +} + +// handleUpperLevelConfirmationLocked processes an incoming upper-level protocol +// (e.g. TCP acknowledgements) reachability confirmation. +func (e *neighborEntry) handleUpperLevelConfirmationLocked() { + switch e.neigh.State { + case Reachable, Stale, Delay, Probe: + if e.neigh.State != Reachable { + e.dispatchChangeEventLocked(Reachable) + // Set state to Reachable again to refresh timers. + } + e.setStateLocked(Reachable) + + case Unknown, Incomplete, Failed, Static: + // Do nothing + + default: + panic(fmt.Sprintf("Invalid cache entry state: %s", e.neigh.State)) + } +} diff --git a/pkg/tcpip/stack/neighbor_entry_test.go b/pkg/tcpip/stack/neighbor_entry_test.go new file mode 100644 index 000000000..08c9ccd25 --- /dev/null +++ b/pkg/tcpip/stack/neighbor_entry_test.go @@ -0,0 +1,2770 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package stack + +import ( + "fmt" + "math" + "math/rand" + "strings" + "sync" + "testing" + "time" + + "github.com/google/go-cmp/cmp" + "github.com/google/go-cmp/cmp/cmpopts" + "gvisor.dev/gvisor/pkg/sleep" + "gvisor.dev/gvisor/pkg/tcpip" +) + +const ( + entryTestNetNumber tcpip.NetworkProtocolNumber = math.MaxUint32 + + entryTestNICID tcpip.NICID = 1 + entryTestAddr1 = tcpip.Address("\x00\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01") + entryTestAddr2 = tcpip.Address("\x00\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02") + + entryTestLinkAddr1 = tcpip.LinkAddress("\x0a\x00\x00\x00\x00\x01") + entryTestLinkAddr2 = tcpip.LinkAddress("\x0a\x00\x00\x00\x00\x02") + + // entryTestNetDefaultMTU is the MTU, in bytes, used throughout the tests, + // except where another value is explicitly used. It is chosen to match the + // MTU of loopback interfaces on Linux systems. + entryTestNetDefaultMTU = 65536 +) + +// eventDiffOpts are the options passed to cmp.Diff to compare entry events. +// The UpdatedAt field is ignored due to a lack of a deterministic method to +// predict the time that an event will be dispatched. +func eventDiffOpts() []cmp.Option { + return []cmp.Option{ + cmpopts.IgnoreFields(testEntryEventInfo{}, "UpdatedAt"), + } +} + +// eventDiffOptsWithSort is like eventDiffOpts but also includes an option to +// sort slices of events for cases where ordering must be ignored. +func eventDiffOptsWithSort() []cmp.Option { + return []cmp.Option{ + cmpopts.IgnoreFields(testEntryEventInfo{}, "UpdatedAt"), + cmpopts.SortSlices(func(a, b testEntryEventInfo) bool { + return strings.Compare(string(a.Addr), string(b.Addr)) < 0 + }), + } +} + +// The following unit tests exercise every state transition and verify its +// behavior with RFC 4681. +// +// | From | To | Cause | Action | Event | +// | ========== | ========== | ========================================== | =============== | ======= | +// | Unknown | Unknown | Confirmation w/ unknown address | | Added | +// | Unknown | Incomplete | Packet queued to unknown address | Send probe | Added | +// | Unknown | Stale | Probe w/ unknown address | | Added | +// | Incomplete | Incomplete | Retransmit timer expired | Send probe | Changed | +// | Incomplete | Reachable | Solicited confirmation | Notify wakers | Changed | +// | Incomplete | Stale | Unsolicited confirmation | Notify wakers | Changed | +// | Incomplete | Failed | Max probes sent without reply | Notify wakers | Removed | +// | Reachable | Reachable | Confirmation w/ different isRouter flag | Update IsRouter | | +// | Reachable | Stale | Reachable timer expired | | Changed | +// | Reachable | Stale | Probe or confirmation w/ different address | | Changed | +// | Stale | Reachable | Solicited override confirmation | Update LinkAddr | Changed | +// | Stale | Stale | Override confirmation | Update LinkAddr | Changed | +// | Stale | Stale | Probe w/ different address | Update LinkAddr | Changed | +// | Stale | Delay | Packet sent | | Changed | +// | Delay | Reachable | Upper-layer confirmation | | Changed | +// | Delay | Reachable | Solicited override confirmation | Update LinkAddr | Changed | +// | Delay | Stale | Probe or confirmation w/ different address | | Changed | +// | Delay | Probe | Delay timer expired | Send probe | Changed | +// | Probe | Reachable | Solicited override confirmation | Update LinkAddr | Changed | +// | Probe | Reachable | Solicited confirmation w/ same address | Notify wakers | Changed | +// | Probe | Stale | Probe or confirmation w/ different address | | Changed | +// | Probe | Probe | Retransmit timer expired | Send probe | Changed | +// | Probe | Failed | Max probes sent without reply | Notify wakers | Removed | +// | Failed | | Unreachability timer expired | Delete entry | | + +type testEntryEventType uint8 + +const ( + entryTestAdded testEntryEventType = iota + entryTestChanged + entryTestRemoved +) + +func (t testEntryEventType) String() string { + switch t { + case entryTestAdded: + return "add" + case entryTestChanged: + return "change" + case entryTestRemoved: + return "remove" + default: + return fmt.Sprintf("unknown (%d)", t) + } +} + +// Fields are exported for use with cmp.Diff. +type testEntryEventInfo struct { + EventType testEntryEventType + NICID tcpip.NICID + Addr tcpip.Address + LinkAddr tcpip.LinkAddress + State NeighborState + UpdatedAt time.Time +} + +func (e testEntryEventInfo) String() string { + return fmt.Sprintf("%s event for NIC #%d, addr=%q, linkAddr=%q, state=%q", e.EventType, e.NICID, e.Addr, e.LinkAddr, e.State) +} + +// testNUDDispatcher implements NUDDispatcher to validate the dispatching of +// events upon certain NUD state machine events. +type testNUDDispatcher struct { + mu sync.Mutex + events []testEntryEventInfo +} + +var _ NUDDispatcher = (*testNUDDispatcher)(nil) + +func (d *testNUDDispatcher) queueEvent(e testEntryEventInfo) { + d.mu.Lock() + defer d.mu.Unlock() + d.events = append(d.events, e) +} + +func (d *testNUDDispatcher) OnNeighborAdded(nicID tcpip.NICID, addr tcpip.Address, linkAddr tcpip.LinkAddress, state NeighborState, updatedAt time.Time) { + d.queueEvent(testEntryEventInfo{ + EventType: entryTestAdded, + NICID: nicID, + Addr: addr, + LinkAddr: linkAddr, + State: state, + UpdatedAt: updatedAt, + }) +} + +func (d *testNUDDispatcher) OnNeighborChanged(nicID tcpip.NICID, addr tcpip.Address, linkAddr tcpip.LinkAddress, state NeighborState, updatedAt time.Time) { + d.queueEvent(testEntryEventInfo{ + EventType: entryTestChanged, + NICID: nicID, + Addr: addr, + LinkAddr: linkAddr, + State: state, + UpdatedAt: updatedAt, + }) +} + +func (d *testNUDDispatcher) OnNeighborRemoved(nicID tcpip.NICID, addr tcpip.Address, linkAddr tcpip.LinkAddress, state NeighborState, updatedAt time.Time) { + d.queueEvent(testEntryEventInfo{ + EventType: entryTestRemoved, + NICID: nicID, + Addr: addr, + LinkAddr: linkAddr, + State: state, + UpdatedAt: updatedAt, + }) +} + +type entryTestLinkResolver struct { + mu sync.Mutex + probes []entryTestProbeInfo +} + +var _ LinkAddressResolver = (*entryTestLinkResolver)(nil) + +type entryTestProbeInfo struct { + RemoteAddress tcpip.Address + RemoteLinkAddress tcpip.LinkAddress + LocalAddress tcpip.Address +} + +func (p entryTestProbeInfo) String() string { + return fmt.Sprintf("probe with RemoteAddress=%q, RemoteLinkAddress=%q, LocalAddress=%q", p.RemoteAddress, p.RemoteLinkAddress, p.LocalAddress) +} + +// LinkAddressRequest sends a request for the LinkAddress of addr. Broadcasts +// to the local network if linkAddr is the zero value. +func (r *entryTestLinkResolver) LinkAddressRequest(addr, localAddr tcpip.Address, linkAddr tcpip.LinkAddress, linkEP LinkEndpoint) *tcpip.Error { + p := entryTestProbeInfo{ + RemoteAddress: addr, + RemoteLinkAddress: linkAddr, + LocalAddress: localAddr, + } + r.mu.Lock() + defer r.mu.Unlock() + r.probes = append(r.probes, p) + return nil +} + +// ResolveStaticAddress attempts to resolve address without sending requests. +// It either resolves the name immediately or returns the empty LinkAddress. +func (r *entryTestLinkResolver) ResolveStaticAddress(addr tcpip.Address) (tcpip.LinkAddress, bool) { + return "", false +} + +// LinkAddressProtocol returns the network protocol of the addresses this +// resolver can resolve. +func (r *entryTestLinkResolver) LinkAddressProtocol() tcpip.NetworkProtocolNumber { + return entryTestNetNumber +} + +func entryTestSetup(c NUDConfigurations) (*neighborEntry, *testNUDDispatcher, *entryTestLinkResolver, *fakeClock) { + clock := newFakeClock() + disp := testNUDDispatcher{} + nic := NIC{ + id: entryTestNICID, + linkEP: nil, // entryTestLinkResolver doesn't use a LinkEndpoint + stack: &Stack{ + clock: clock, + nudDisp: &disp, + }, + } + + rng := rand.New(rand.NewSource(time.Now().UnixNano())) + nudState := NewNUDState(c, rng) + linkRes := entryTestLinkResolver{} + entry := newNeighborEntry(&nic, entryTestAddr1, entryTestAddr2, nudState, &linkRes) + + // Stub out ndpState to verify modification of default routers. + nic.mu.ndp = ndpState{ + nic: &nic, + defaultRouters: make(map[tcpip.Address]defaultRouterState), + } + + // Stub out the neighbor cache to verify deletion from the cache. + nic.neigh = &neighborCache{ + nic: &nic, + state: nudState, + cache: make(map[tcpip.Address]*neighborEntry, neighborCacheSize), + } + nic.neigh.cache[entryTestAddr1] = entry + + return entry, &disp, &linkRes, clock +} + +// TestEntryInitiallyUnknown verifies that the state of a newly created +// neighborEntry is Unknown. +func TestEntryInitiallyUnknown(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, clock := entryTestSetup(c) + + e.mu.Lock() + if got, want := e.neigh.State, Unknown; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() + + clock.advance(c.RetransmitTimer) + + // No probes should have been sent. + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, []entryTestProbeInfo(nil)) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + // No events should have been dispatched. + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, []testEntryEventInfo(nil)); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() +} + +func TestEntryUnknownToUnknownWhenConfirmationWithUnknownAddress(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, clock := entryTestSetup(c) + + e.mu.Lock() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: false, + IsRouter: false, + }) + if got, want := e.neigh.State, Unknown; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() + + clock.advance(time.Hour) + + // No probes should have been sent. + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, []entryTestProbeInfo(nil)) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + // No events should have been dispatched. + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, []testEntryEventInfo(nil)); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() +} + +func TestEntryUnknownToIncomplete(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, _ := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + if got, want := e.neigh.State, Incomplete; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() + + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + } + { + nudDisp.mu.Lock() + diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...) + nudDisp.mu.Unlock() + if diff != "" { + t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + } +} + +func TestEntryUnknownToStale(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, _ := entryTestSetup(c) + + e.mu.Lock() + e.handleProbeLocked(entryTestLinkAddr1) + if got, want := e.neigh.State, Stale; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() + + // No probes should have been sent. + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, []entryTestProbeInfo(nil)) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() +} + +func TestEntryIncompleteToIncompleteDoesNotChangeUpdatedAt(t *testing.T) { + c := DefaultNUDConfigurations() + c.MaxMulticastProbes = 3 + e, nudDisp, linkRes, clock := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + if got, want := e.neigh.State, Incomplete; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + updatedAt := e.neigh.UpdatedAt + e.mu.Unlock() + + clock.advance(c.RetransmitTimer) + + // UpdatedAt should remain the same during address resolution. + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.probes = nil + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + e.mu.Lock() + if got, want := e.neigh.UpdatedAt, updatedAt; got != want { + t.Errorf("got e.neigh.UpdatedAt = %q, want = %q", got, want) + } + e.mu.Unlock() + + clock.advance(c.RetransmitTimer) + + // UpdatedAt should change after failing address resolution. Timing out after + // sending the last probe transitions the entry to Failed. + { + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + } + + clock.advance(c.RetransmitTimer) + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestRemoved, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() + + e.mu.Lock() + if got, notWant := e.neigh.UpdatedAt, updatedAt; got == notWant { + t.Errorf("expected e.neigh.UpdatedAt to change, got = %q", got) + } + e.mu.Unlock() +} + +func TestEntryIncompleteToReachable(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, _ := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + if got, want := e.neigh.State, Incomplete; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: true, + Override: false, + IsRouter: false, + }) + if got, want := e.neigh.State, Reachable; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() + + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Reachable, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() +} + +// TestEntryAddsAndClearsWakers verifies that wakers are added when +// addWakerLocked is called and cleared when address resolution finishes. In +// this case, address resolution will finish when transitioning from Incomplete +// to Reachable. +func TestEntryAddsAndClearsWakers(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, _ := entryTestSetup(c) + + w := sleep.Waker{} + s := sleep.Sleeper{} + s.AddWaker(&w, 123) + defer s.Done() + + e.mu.Lock() + e.handlePacketQueuedLocked() + if got := e.wakers; got != nil { + t.Errorf("got e.wakers = %v, want = nil", got) + } + e.addWakerLocked(&w) + if got, want := w.IsAsserted(), false; got != want { + t.Errorf("waker.IsAsserted() = %t, want = %t", got, want) + } + if e.wakers == nil { + t.Error("expected e.wakers to be non-nil") + } + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: true, + Override: false, + IsRouter: false, + }) + if e.wakers != nil { + t.Errorf("got e.wakers = %v, want = nil", e.wakers) + } + if got, want := w.IsAsserted(), true; got != want { + t.Errorf("waker.IsAsserted() = %t, want = %t", got, want) + } + e.mu.Unlock() + + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Reachable, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() +} + +func TestEntryIncompleteToReachableWithRouterFlag(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, _ := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + if got, want := e.neigh.State, Incomplete; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: true, + Override: false, + IsRouter: true, + }) + if got, want := e.neigh.State, Reachable; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + if got, want := e.isRouter, true; got != want { + t.Errorf("got e.isRouter = %t, want = %t", got, want) + } + e.mu.Unlock() + + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + if diff := cmp.Diff(linkRes.probes, wantProbes); diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + linkRes.mu.Unlock() + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Reachable, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() +} + +func TestEntryIncompleteToStale(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, _ := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + if got, want := e.neigh.State, Incomplete; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: false, + IsRouter: false, + }) + if got, want := e.neigh.State, Stale; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() + + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() +} + +func TestEntryIncompleteToFailed(t *testing.T) { + c := DefaultNUDConfigurations() + c.MaxMulticastProbes = 3 + e, nudDisp, linkRes, clock := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + if got, want := e.neigh.State, Incomplete; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() + + waitFor := c.RetransmitTimer * time.Duration(c.MaxMulticastProbes) + clock.advance(waitFor) + + wantProbes := []entryTestProbeInfo{ + // The Incomplete-to-Incomplete state transition is tested here by + // verifying that 3 reachability probes were sent. + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestRemoved, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() + + e.mu.Lock() + if got, want := e.neigh.State, Failed; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() +} + +type testLocker struct{} + +var _ sync.Locker = (*testLocker)(nil) + +func (*testLocker) Lock() {} +func (*testLocker) Unlock() {} + +func TestEntryStaysReachableWhenConfirmationWithRouterFlag(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, _ := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: true, + Override: false, + IsRouter: true, + }) + if got, want := e.neigh.State, Reachable; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + if got, want := e.isRouter, true; got != want { + t.Errorf("got e.isRouter = %t, want = %t", got, want) + } + e.nic.mu.ndp.defaultRouters[entryTestAddr1] = defaultRouterState{ + invalidationJob: e.nic.stack.newJob(&testLocker{}, func() {}), + } + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: false, + IsRouter: false, + }) + if got, want := e.isRouter, false; got != want { + t.Errorf("got e.isRouter = %t, want = %t", got, want) + } + if _, ok := e.nic.mu.ndp.defaultRouters[entryTestAddr1]; ok { + t.Errorf("unexpected defaultRouter for %s", entryTestAddr1) + } + e.mu.Unlock() + + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Reachable, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() + + e.mu.Lock() + if got, want := e.neigh.State, Reachable; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() +} + +func TestEntryStaysReachableWhenProbeWithSameAddress(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, _ := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: true, + Override: false, + IsRouter: false, + }) + if got, want := e.neigh.State, Reachable; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.handleProbeLocked(entryTestLinkAddr1) + if got, want := e.neigh.State, Reachable; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + if got, want := e.neigh.LinkAddr, entryTestLinkAddr1; got != want { + t.Errorf("got e.neigh.LinkAddr = %q, want = %q", got, want) + } + e.mu.Unlock() + + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Reachable, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() +} + +func TestEntryReachableToStaleWhenTimeout(t *testing.T) { + c := DefaultNUDConfigurations() + // Eliminate random factors from ReachableTime computation so the transition + // from Stale to Reachable will only take BaseReachableTime duration. + c.MinRandomFactor = 1 + c.MaxRandomFactor = 1 + + e, nudDisp, linkRes, clock := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: true, + Override: false, + IsRouter: false, + }) + if got, want := e.neigh.State, Reachable; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() + + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + clock.advance(c.BaseReachableTime) + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Reachable, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() + + e.mu.Lock() + if got, want := e.neigh.State, Stale; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() +} + +func TestEntryReachableToStaleWhenProbeWithDifferentAddress(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, _ := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: true, + Override: false, + IsRouter: false, + }) + if got, want := e.neigh.State, Reachable; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.handleProbeLocked(entryTestLinkAddr2) + if got, want := e.neigh.State, Stale; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() + + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Reachable, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr2, + State: Stale, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() + + e.mu.Lock() + if got, want := e.neigh.State, Stale; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() +} + +func TestEntryReachableToStaleWhenConfirmationWithDifferentAddress(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, _ := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: true, + Override: false, + IsRouter: false, + }) + if got, want := e.neigh.State, Reachable; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.handleConfirmationLocked(entryTestLinkAddr2, ReachabilityConfirmationFlags{ + Solicited: false, + Override: false, + IsRouter: false, + }) + if got, want := e.neigh.State, Stale; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() + + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Reachable, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() + + e.mu.Lock() + if got, want := e.neigh.State, Stale; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() +} + +func TestEntryReachableToStaleWhenConfirmationWithDifferentAddressAndOverride(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, _ := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: true, + Override: false, + IsRouter: false, + }) + if got, want := e.neigh.State, Reachable; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.handleConfirmationLocked(entryTestLinkAddr2, ReachabilityConfirmationFlags{ + Solicited: false, + Override: true, + IsRouter: false, + }) + if got, want := e.neigh.State, Stale; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() + + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Reachable, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr2, + State: Stale, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() + + e.mu.Lock() + if got, want := e.neigh.State, Stale; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() +} + +func TestEntryStaysStaleWhenProbeWithSameAddress(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, _ := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: false, + IsRouter: false, + }) + if got, want := e.neigh.State, Stale; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.handleProbeLocked(entryTestLinkAddr1) + if got, want := e.neigh.State, Stale; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + if got, want := e.neigh.LinkAddr, entryTestLinkAddr1; got != want { + t.Errorf("got e.neigh.LinkAddr = %q, want = %q", got, want) + } + e.mu.Unlock() + + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() +} + +func TestEntryStaleToReachableWhenSolicitedOverrideConfirmation(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, _ := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: false, + IsRouter: false, + }) + if got, want := e.neigh.State, Stale; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.handleConfirmationLocked(entryTestLinkAddr2, ReachabilityConfirmationFlags{ + Solicited: true, + Override: true, + IsRouter: false, + }) + if got, want := e.neigh.State, Reachable; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + if got, want := e.neigh.LinkAddr, entryTestLinkAddr2; got != want { + t.Errorf("got e.neigh.LinkAddr = %q, want = %q", got, want) + } + e.mu.Unlock() + + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr2, + State: Reachable, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() +} + +func TestEntryStaleToStaleWhenOverrideConfirmation(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, _ := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: false, + IsRouter: false, + }) + if got, want := e.neigh.State, Stale; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.handleConfirmationLocked(entryTestLinkAddr2, ReachabilityConfirmationFlags{ + Solicited: false, + Override: true, + IsRouter: false, + }) + if got, want := e.neigh.State, Stale; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + if got, want := e.neigh.LinkAddr, entryTestLinkAddr2; got != want { + t.Errorf("got e.neigh.LinkAddr = %q, want = %q", got, want) + } + e.mu.Unlock() + + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr2, + State: Stale, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() +} + +func TestEntryStaleToStaleWhenProbeUpdateAddress(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, _ := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: false, + IsRouter: false, + }) + if got, want := e.neigh.State, Stale; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.handleProbeLocked(entryTestLinkAddr2) + if got, want := e.neigh.State, Stale; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + if got, want := e.neigh.LinkAddr, entryTestLinkAddr2; got != want { + t.Errorf("got e.neigh.LinkAddr = %q, want = %q", got, want) + } + e.mu.Unlock() + + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr2, + State: Stale, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() +} + +func TestEntryStaleToDelay(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, _ := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: false, + IsRouter: false, + }) + if got, want := e.neigh.State, Stale; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.handlePacketQueuedLocked() + if got, want := e.neigh.State, Delay; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() + + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Delay, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() +} + +func TestEntryDelayToReachableWhenUpperLevelConfirmation(t *testing.T) { + c := DefaultNUDConfigurations() + // Eliminate random factors from ReachableTime computation so the transition + // from Stale to Reachable will only take BaseReachableTime duration. + c.MinRandomFactor = 1 + c.MaxRandomFactor = 1 + + e, nudDisp, linkRes, clock := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: false, + IsRouter: false, + }) + e.handlePacketQueuedLocked() + if got, want := e.neigh.State, Delay; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.handleUpperLevelConfirmationLocked() + if got, want := e.neigh.State, Reachable; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() + + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + clock.advance(c.BaseReachableTime) + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Delay, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Reachable, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() +} + +func TestEntryDelayToReachableWhenSolicitedOverrideConfirmation(t *testing.T) { + c := DefaultNUDConfigurations() + c.MaxMulticastProbes = 1 + // Eliminate random factors from ReachableTime computation so the transition + // from Stale to Reachable will only take BaseReachableTime duration. + c.MinRandomFactor = 1 + c.MaxRandomFactor = 1 + + e, nudDisp, linkRes, clock := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: false, + IsRouter: false, + }) + e.handlePacketQueuedLocked() + if got, want := e.neigh.State, Delay; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.handleConfirmationLocked(entryTestLinkAddr2, ReachabilityConfirmationFlags{ + Solicited: true, + Override: true, + IsRouter: false, + }) + if got, want := e.neigh.State, Reachable; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + if got, want := e.neigh.LinkAddr, entryTestLinkAddr2; got != want { + t.Errorf("got e.neigh.LinkAddr = %q, want = %q", got, want) + } + e.mu.Unlock() + + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + clock.advance(c.BaseReachableTime) + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Delay, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr2, + State: Reachable, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr2, + State: Stale, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() +} + +func TestEntryStaysDelayWhenOverrideConfirmationWithSameAddress(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, _ := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: false, + IsRouter: false, + }) + e.handlePacketQueuedLocked() + if got, want := e.neigh.State, Delay; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: true, + IsRouter: false, + }) + if got, want := e.neigh.State, Delay; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + if got, want := e.neigh.LinkAddr, entryTestLinkAddr1; got != want { + t.Errorf("got e.neigh.LinkAddr = %q, want = %q", got, want) + } + e.mu.Unlock() + + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Delay, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() +} + +func TestEntryDelayToStaleWhenProbeWithDifferentAddress(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, _ := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: false, + IsRouter: false, + }) + e.handlePacketQueuedLocked() + if got, want := e.neigh.State, Delay; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.handleProbeLocked(entryTestLinkAddr2) + if got, want := e.neigh.State, Stale; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() + + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Delay, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr2, + State: Stale, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() +} + +func TestEntryDelayToStaleWhenConfirmationWithDifferentAddress(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, _ := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: false, + IsRouter: false, + }) + e.handlePacketQueuedLocked() + if got, want := e.neigh.State, Delay; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.handleConfirmationLocked(entryTestLinkAddr2, ReachabilityConfirmationFlags{ + Solicited: false, + Override: true, + IsRouter: false, + }) + if got, want := e.neigh.State, Stale; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() + + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Delay, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr2, + State: Stale, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() +} + +func TestEntryDelayToProbe(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, clock := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: false, + IsRouter: false, + }) + e.handlePacketQueuedLocked() + if got, want := e.neigh.State, Delay; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() + + clock.advance(c.DelayFirstProbeTime) + + wantProbes := []entryTestProbeInfo{ + // The first probe is caused by the Unknown-to-Incomplete transition. + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + // The second probe is caused by the Delay-to-Probe transition. + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: entryTestLinkAddr1, + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Delay, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Probe, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() + + e.mu.Lock() + if got, want := e.neigh.State, Probe; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() +} + +func TestEntryProbeToStaleWhenProbeWithDifferentAddress(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, clock := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: false, + IsRouter: false, + }) + e.handlePacketQueuedLocked() + e.mu.Unlock() + + clock.advance(c.DelayFirstProbeTime) + + wantProbes := []entryTestProbeInfo{ + // The first probe is caused by the Unknown-to-Incomplete transition. + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + // The second probe is caused by the Delay-to-Probe transition. + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: entryTestLinkAddr1, + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + e.mu.Lock() + if got, want := e.neigh.State, Probe; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.handleProbeLocked(entryTestLinkAddr2) + if got, want := e.neigh.State, Stale; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Delay, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Probe, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr2, + State: Stale, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() + + e.mu.Lock() + if got, want := e.neigh.State, Stale; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() +} + +func TestEntryProbeToStaleWhenConfirmationWithDifferentAddress(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, clock := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: false, + IsRouter: false, + }) + e.handlePacketQueuedLocked() + e.mu.Unlock() + + clock.advance(c.DelayFirstProbeTime) + + wantProbes := []entryTestProbeInfo{ + // The first probe is caused by the Unknown-to-Incomplete transition. + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + // The second probe is caused by the Delay-to-Probe transition. + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: entryTestLinkAddr1, + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + e.mu.Lock() + if got, want := e.neigh.State, Probe; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.handleConfirmationLocked(entryTestLinkAddr2, ReachabilityConfirmationFlags{ + Solicited: false, + Override: true, + IsRouter: false, + }) + if got, want := e.neigh.State, Stale; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Delay, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Probe, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr2, + State: Stale, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() + + e.mu.Lock() + if got, want := e.neigh.State, Stale; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() +} + +func TestEntryStaysProbeWhenOverrideConfirmationWithSameAddress(t *testing.T) { + c := DefaultNUDConfigurations() + e, nudDisp, linkRes, clock := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: false, + IsRouter: false, + }) + e.handlePacketQueuedLocked() + e.mu.Unlock() + + clock.advance(c.DelayFirstProbeTime) + + wantProbes := []entryTestProbeInfo{ + // The first probe is caused by the Unknown-to-Incomplete transition. + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + // The second probe is caused by the Delay-to-Probe transition. + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: entryTestLinkAddr1, + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + e.mu.Lock() + if got, want := e.neigh.State, Probe; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: true, + IsRouter: false, + }) + if got, want := e.neigh.State, Probe; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + if got, want := e.neigh.LinkAddr, entryTestLinkAddr1; got != want { + t.Errorf("got e.neigh.LinkAddr = %q, want = %q", got, want) + } + e.mu.Unlock() + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Delay, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Probe, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() +} + +func TestEntryProbeToReachableWhenSolicitedOverrideConfirmation(t *testing.T) { + c := DefaultNUDConfigurations() + // Eliminate random factors from ReachableTime computation so the transition + // from Stale to Reachable will only take BaseReachableTime duration. + c.MinRandomFactor = 1 + c.MaxRandomFactor = 1 + + e, nudDisp, linkRes, clock := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: false, + IsRouter: false, + }) + e.handlePacketQueuedLocked() + e.mu.Unlock() + + clock.advance(c.DelayFirstProbeTime) + + wantProbes := []entryTestProbeInfo{ + // The first probe is caused by the Unknown-to-Incomplete transition. + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + // The second probe is caused by the Delay-to-Probe transition. + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: entryTestLinkAddr1, + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + e.mu.Lock() + if got, want := e.neigh.State, Probe; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.handleConfirmationLocked(entryTestLinkAddr2, ReachabilityConfirmationFlags{ + Solicited: true, + Override: true, + IsRouter: false, + }) + if got, want := e.neigh.State, Reachable; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + if got, want := e.neigh.LinkAddr, entryTestLinkAddr2; got != want { + t.Errorf("got e.neigh.LinkAddr = %q, want = %q", got, want) + } + e.mu.Unlock() + + clock.advance(c.BaseReachableTime) + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Delay, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Probe, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr2, + State: Reachable, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr2, + State: Stale, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() +} + +func TestEntryProbeToReachableWhenSolicitedConfirmationWithSameAddress(t *testing.T) { + c := DefaultNUDConfigurations() + // Eliminate random factors from ReachableTime computation so the transition + // from Stale to Reachable will only take BaseReachableTime duration. + c.MinRandomFactor = 1 + c.MaxRandomFactor = 1 + + e, nudDisp, linkRes, clock := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: false, + IsRouter: false, + }) + e.handlePacketQueuedLocked() + e.mu.Unlock() + + clock.advance(c.DelayFirstProbeTime) + + wantProbes := []entryTestProbeInfo{ + // The first probe is caused by the Unknown-to-Incomplete transition. + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + // The second probe is caused by the Delay-to-Probe transition. + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: entryTestLinkAddr1, + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + e.mu.Lock() + if got, want := e.neigh.State, Probe; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: true, + Override: false, + IsRouter: false, + }) + if got, want := e.neigh.State, Reachable; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() + + clock.advance(c.BaseReachableTime) + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Delay, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Probe, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Reachable, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() +} + +func TestEntryProbeToFailed(t *testing.T) { + c := DefaultNUDConfigurations() + c.MaxMulticastProbes = 3 + c.MaxUnicastProbes = 3 + e, nudDisp, linkRes, clock := entryTestSetup(c) + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: false, + IsRouter: false, + }) + e.handlePacketQueuedLocked() + e.mu.Unlock() + + waitFor := c.DelayFirstProbeTime + c.RetransmitTimer*time.Duration(c.MaxUnicastProbes) + clock.advance(waitFor) + + wantProbes := []entryTestProbeInfo{ + // The first probe is caused by the Unknown-to-Incomplete transition. + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + // The next three probe are caused by the Delay-to-Probe transition. + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: entryTestLinkAddr1, + LocalAddress: entryTestAddr2, + }, + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: entryTestLinkAddr1, + LocalAddress: entryTestAddr2, + }, + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: entryTestLinkAddr1, + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Delay, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Probe, + }, + { + EventType: entryTestRemoved, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Probe, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() + + e.mu.Lock() + if got, want := e.neigh.State, Failed; got != want { + t.Errorf("got e.neigh.State = %q, want = %q", got, want) + } + e.mu.Unlock() +} + +func TestEntryFailedGetsDeleted(t *testing.T) { + c := DefaultNUDConfigurations() + c.MaxMulticastProbes = 3 + c.MaxUnicastProbes = 3 + e, nudDisp, linkRes, clock := entryTestSetup(c) + + // Verify the cache contains the entry. + if _, ok := e.nic.neigh.cache[entryTestAddr1]; !ok { + t.Errorf("expected entry %q to exist in the neighbor cache", entryTestAddr1) + } + + e.mu.Lock() + e.handlePacketQueuedLocked() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: false, + IsRouter: false, + }) + e.handlePacketQueuedLocked() + e.mu.Unlock() + + waitFor := c.DelayFirstProbeTime + c.RetransmitTimer*time.Duration(c.MaxUnicastProbes) + c.UnreachableTime + clock.advance(waitFor) + + wantProbes := []entryTestProbeInfo{ + // The first probe is caused by the Unknown-to-Incomplete transition. + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: tcpip.LinkAddress(""), + LocalAddress: entryTestAddr2, + }, + // The next three probe are caused by the Delay-to-Probe transition. + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: entryTestLinkAddr1, + LocalAddress: entryTestAddr2, + }, + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: entryTestLinkAddr1, + LocalAddress: entryTestAddr2, + }, + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: entryTestLinkAddr1, + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Delay, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Probe, + }, + { + EventType: entryTestRemoved, + NICID: entryTestNICID, + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Probe, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() + + // Verify the cache no longer contains the entry. + if _, ok := e.nic.neigh.cache[entryTestAddr1]; ok { + t.Errorf("entry %q should have been deleted from the neighbor cache", entryTestAddr1) + } +} diff --git a/pkg/tcpip/stack/neighborstate_string.go b/pkg/tcpip/stack/neighborstate_string.go new file mode 100644 index 000000000..aa7311ec6 --- /dev/null +++ b/pkg/tcpip/stack/neighborstate_string.go @@ -0,0 +1,44 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Code generated by "stringer -type NeighborState"; DO NOT EDIT. + +package stack + +import "strconv" + +func _() { + // An "invalid array index" compiler error signifies that the constant values have changed. + // Re-run the stringer command to generate them again. + var x [1]struct{} + _ = x[Unknown-0] + _ = x[Incomplete-1] + _ = x[Reachable-2] + _ = x[Stale-3] + _ = x[Delay-4] + _ = x[Probe-5] + _ = x[Static-6] + _ = x[Failed-7] +} + +const _NeighborState_name = "UnknownIncompleteReachableStaleDelayProbeStaticFailed" + +var _NeighborState_index = [...]uint8{0, 7, 17, 26, 31, 36, 41, 47, 53} + +func (i NeighborState) String() string { + if i >= NeighborState(len(_NeighborState_index)-1) { + return "NeighborState(" + strconv.FormatInt(int64(i), 10) + ")" + } + return _NeighborState_name[_NeighborState_index[i]:_NeighborState_index[i+1]] +} diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index afb7dfeaf..f21066fce 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -16,6 +16,7 @@ package stack import ( "fmt" + "math/rand" "reflect" "sort" "strings" @@ -45,6 +46,7 @@ type NIC struct { context NICContext stats NICStats + neigh *neighborCache mu struct { sync.RWMutex @@ -141,6 +143,16 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, ctx NICC nic.mu.packetEPs[netProto.Number()] = []PacketEndpoint{} } + // Check for Neighbor Unreachability Detection support. + if ep.Capabilities()&CapabilityResolutionRequired != 0 && len(stack.linkAddrResolvers) != 0 { + rng := rand.New(rand.NewSource(stack.clock.NowNanoseconds())) + nic.neigh = &neighborCache{ + nic: nic, + state: NewNUDState(stack.nudConfigs, rng), + cache: make(map[tcpip.Address]*neighborEntry, neighborCacheSize), + } + } + nic.linkEP.Attach(nic) return nic @@ -181,7 +193,7 @@ func (n *NIC) disableLocked() *tcpip.Error { return nil } - // TODO(b/147015577): Should Routes that are currently bound to n be + // TODO(gvisor.dev/issue/1491): Should Routes that are currently bound to n be // invalidated? Currently, Routes will continue to work when a NIC is enabled // again, and applications may not know that the underlying NIC was ever // disabled. @@ -1200,15 +1212,13 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp // Are any packet sockets listening for this network protocol? packetEPs := n.mu.packetEPs[protocol] - // Check whether there are packet sockets listening for every protocol. - // If we received a packet with protocol EthernetProtocolAll, then the - // previous for loop will have handled it. - if protocol != header.EthernetProtocolAll { - packetEPs = append(packetEPs, n.mu.packetEPs[header.EthernetProtocolAll]...) - } + // Add any other packet sockets that maybe listening for all protocols. + packetEPs = append(packetEPs, n.mu.packetEPs[header.EthernetProtocolAll]...) n.mu.RUnlock() for _, ep := range packetEPs { - ep.HandlePacket(n.id, local, protocol, pkt.Clone()) + p := pkt.Clone() + p.PktType = tcpip.PacketHost + ep.HandlePacket(n.id, local, protocol, p) } if netProto.Number() == header.IPv4ProtocolNumber || netProto.Number() == header.IPv6ProtocolNumber { @@ -1311,6 +1321,24 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp } } +// DeliverOutboundPacket implements NetworkDispatcher.DeliverOutboundPacket. +func (n *NIC) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { + n.mu.RLock() + // We do not deliver to protocol specific packet endpoints as on Linux + // only ETH_P_ALL endpoints get outbound packets. + // Add any other packet sockets that maybe listening for all protocols. + packetEPs := n.mu.packetEPs[header.EthernetProtocolAll] + n.mu.RUnlock() + for _, ep := range packetEPs { + p := pkt.Clone() + p.PktType = tcpip.PacketOutgoing + // Add the link layer header as outgoing packets are intercepted + // before the link layer header is created. + n.linkEP.AddHeader(local, remote, protocol, p) + ep.HandlePacket(n.id, local, protocol, p) + } +} + func (n *NIC) forwardPacket(r *Route, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { // TODO(b/143425874) Decrease the TTL field in forwarded packets. // TODO(b/151227689): Avoid copying the packet when forwarding. We can do this @@ -1358,16 +1386,19 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN // TransportHeader is nil only when pkt is an ICMP packet or was reassembled // from fragments. if pkt.TransportHeader == nil { - // TODO(gvisor.dev/issue/170): ICMP packets don't have their - // TransportHeader fields set. See icmp/protocol.go:protocol.Parse for a + // TODO(gvisor.dev/issue/170): ICMP packets don't have their TransportHeader + // fields set yet, parse it here. See icmp/protocol.go:protocol.Parse for a // full explanation. if protocol == header.ICMPv4ProtocolNumber || protocol == header.ICMPv6ProtocolNumber { + // ICMP packets may be longer, but until icmp.Parse is implemented, here + // we parse it using the minimum size. transHeader, ok := pkt.Data.PullUp(transProto.MinimumPacketSize()) if !ok { n.stack.stats.MalformedRcvdPackets.Increment() return } pkt.TransportHeader = transHeader + pkt.Data.TrimFront(len(pkt.TransportHeader)) } else { // This is either a bad packet or was re-assembled from fragments. transProto.Parse(pkt) @@ -1521,6 +1552,27 @@ func (n *NIC) setNDPConfigs(c NDPConfigurations) { n.mu.Unlock() } +// NUDConfigs gets the NUD configurations for n. +func (n *NIC) NUDConfigs() (NUDConfigurations, *tcpip.Error) { + if n.neigh == nil { + return NUDConfigurations{}, tcpip.ErrNotSupported + } + return n.neigh.config(), nil +} + +// setNUDConfigs sets the NUD configurations for n. +// +// Note, if c contains invalid NUD configuration values, it will be fixed to +// use default values for the erroneous values. +func (n *NIC) setNUDConfigs(c NUDConfigurations) *tcpip.Error { + if n.neigh == nil { + return tcpip.ErrNotSupported + } + c.resetInvalidFields() + n.neigh.setConfig(c) + return nil +} + // handleNDPRA handles an NDP Router Advertisement message that arrived on n. func (n *NIC) handleNDPRA(ip tcpip.Address, ra header.NDPRouterAdvert) { n.mu.Lock() diff --git a/pkg/tcpip/stack/nic_test.go b/pkg/tcpip/stack/nic_test.go index 31f865260..a70792b50 100644 --- a/pkg/tcpip/stack/nic_test.go +++ b/pkg/tcpip/stack/nic_test.go @@ -84,6 +84,16 @@ func (e *testLinkEndpoint) WriteRawPacket(buffer.VectorisedView) *tcpip.Error { return tcpip.ErrNotSupported } +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. +func (*testLinkEndpoint) ARPHardwareType() header.ARPHardwareType { + panic("not implemented") +} + +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *testLinkEndpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { + panic("not implemented") +} + var _ NetworkEndpoint = (*testIPv6Endpoint)(nil) // An IPv6 NetworkEndpoint that throws away outgoing packets. @@ -233,7 +243,7 @@ func (*testIPv6Protocol) LinkAddressProtocol() tcpip.NetworkProtocolNumber { } // LinkAddressRequest implements LinkAddressResolver. -func (*testIPv6Protocol) LinkAddressRequest(_, _ tcpip.Address, _ LinkEndpoint) *tcpip.Error { +func (*testIPv6Protocol) LinkAddressRequest(_, _ tcpip.Address, _ tcpip.LinkAddress, _ LinkEndpoint) *tcpip.Error { return nil } diff --git a/pkg/tcpip/stack/nud.go b/pkg/tcpip/stack/nud.go new file mode 100644 index 000000000..f848d50ad --- /dev/null +++ b/pkg/tcpip/stack/nud.go @@ -0,0 +1,466 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package stack + +import ( + "math" + "sync" + "time" + + "gvisor.dev/gvisor/pkg/tcpip" +) + +const ( + // defaultBaseReachableTime is the default base duration for computing the + // random reachable time. + // + // Reachable time is the duration for which a neighbor is considered + // reachable after a positive reachability confirmation is received. It is a + // function of a uniformly distributed random value between the minimum and + // maximum random factors, multiplied by the base reachable time. Using a + // random component eliminates the possibility that Neighbor Unreachability + // Detection messages will synchronize with each other. + // + // Default taken from REACHABLE_TIME of RFC 4861 section 10. + defaultBaseReachableTime = 30 * time.Second + + // minimumBaseReachableTime is the minimum base duration for computing the + // random reachable time. + // + // Minimum = 1ms + minimumBaseReachableTime = time.Millisecond + + // defaultMinRandomFactor is the default minimum value of the random factor + // used for computing reachable time. + // + // Default taken from MIN_RANDOM_FACTOR of RFC 4861 section 10. + defaultMinRandomFactor = 0.5 + + // defaultMaxRandomFactor is the default maximum value of the random factor + // used for computing reachable time. + // + // The default value depends on the value of MinRandomFactor. + // If MinRandomFactor is less than MAX_RANDOM_FACTOR of RFC 4861 section 10, + // the value from the RFC will be used; otherwise, the default is + // MinRandomFactor multiplied by three. + defaultMaxRandomFactor = 1.5 + + // defaultRetransmitTimer is the default amount of time to wait between + // sending reachability probes. + // + // Default taken from RETRANS_TIMER of RFC 4861 section 10. + defaultRetransmitTimer = time.Second + + // minimumRetransmitTimer is the minimum amount of time to wait between + // sending reachability probes. + // + // Note, RFC 4861 does not impose a minimum Retransmit Timer, but we do here + // to make sure the messages are not sent all at once. We also come to this + // value because in the RetransmitTimer field of a Router Advertisement, a + // value of 0 means unspecified, so the smallest valid value is 1. Note, the + // unit of the RetransmitTimer field in the Router Advertisement is + // milliseconds. + minimumRetransmitTimer = time.Millisecond + + // defaultDelayFirstProbeTime is the default duration to wait for a + // non-Neighbor-Discovery related protocol to reconfirm reachability after + // entering the DELAY state. After this time, a reachability probe will be + // sent and the entry will transition to the PROBE state. + // + // Default taken from DELAY_FIRST_PROBE_TIME of RFC 4861 section 10. + defaultDelayFirstProbeTime = 5 * time.Second + + // defaultMaxMulticastProbes is the default number of reachabililty probes + // to send before concluding negative reachability and deleting the neighbor + // entry from the INCOMPLETE state. + // + // Default taken from MAX_MULTICAST_SOLICIT of RFC 4861 section 10. + defaultMaxMulticastProbes = 3 + + // defaultMaxUnicastProbes is the default number of reachability probes to + // send before concluding retransmission from within the PROBE state should + // cease and the entry SHOULD be deleted. + // + // Default taken from MAX_UNICASE_SOLICIT of RFC 4861 section 10. + defaultMaxUnicastProbes = 3 + + // defaultMaxAnycastDelayTime is the default time in which the stack SHOULD + // delay sending a response for a random time between 0 and this time, if the + // target address is an anycast address. + // + // Default taken from MAX_ANYCAST_DELAY_TIME of RFC 4861 section 10. + defaultMaxAnycastDelayTime = time.Second + + // defaultMaxReachbilityConfirmations is the default amount of unsolicited + // reachability confirmation messages a node MAY send to all-node multicast + // address when it determines its link-layer address has changed. + // + // Default taken from MAX_NEIGHBOR_ADVERTISEMENT of RFC 4861 section 10. + defaultMaxReachbilityConfirmations = 3 + + // defaultUnreachableTime is the default duration for how long an entry will + // remain in the FAILED state before being removed from the neighbor cache. + // + // Note, there is no equivalent protocol constant defined in RFC 4861. It + // leaves the specifics of any garbage collection mechanism up to the + // implementation. + defaultUnreachableTime = 5 * time.Second +) + +// NUDDispatcher is the interface integrators of netstack must implement to +// receive and handle NUD related events. +type NUDDispatcher interface { + // OnNeighborAdded will be called when a new entry is added to a NIC's (with + // ID nicID) neighbor table. + // + // This function is permitted to block indefinitely without interfering with + // the stack's operation. + // + // May be called concurrently. + OnNeighborAdded(nicID tcpip.NICID, ipAddr tcpip.Address, linkAddr tcpip.LinkAddress, state NeighborState, updatedAt time.Time) + + // OnNeighborChanged will be called when an entry in a NIC's (with ID nicID) + // neighbor table changes state and/or link address. + // + // This function is permitted to block indefinitely without interfering with + // the stack's operation. + // + // May be called concurrently. + OnNeighborChanged(nicID tcpip.NICID, ipAddr tcpip.Address, linkAddr tcpip.LinkAddress, state NeighborState, updatedAt time.Time) + + // OnNeighborRemoved will be called when an entry is removed from a NIC's + // (with ID nicID) neighbor table. + // + // This function is permitted to block indefinitely without interfering with + // the stack's operation. + // + // May be called concurrently. + OnNeighborRemoved(nicID tcpip.NICID, ipAddr tcpip.Address, linkAddr tcpip.LinkAddress, state NeighborState, updatedAt time.Time) +} + +// ReachabilityConfirmationFlags describes the flags used within a reachability +// confirmation (e.g. ARP reply or Neighbor Advertisement for ARP or NDP, +// respectively). +type ReachabilityConfirmationFlags struct { + // Solicited indicates that the advertisement was sent in response to a + // reachability probe. + Solicited bool + + // Override indicates that the reachability confirmation should override an + // existing neighbor cache entry and update the cached link-layer address. + // When Override is not set the confirmation will not update a cached + // link-layer address, but will update an existing neighbor cache entry for + // which no link-layer address is known. + Override bool + + // IsRouter indicates that the sender is a router. + IsRouter bool +} + +// NUDHandler communicates external events to the Neighbor Unreachability +// Detection state machine, which is implemented per-interface. This is used by +// network endpoints to inform the Neighbor Cache of probes and confirmations. +type NUDHandler interface { + // HandleProbe processes an incoming neighbor probe (e.g. ARP request or + // Neighbor Solicitation for ARP or NDP, respectively). Validation of the + // probe needs to be performed before calling this function since the + // Neighbor Cache doesn't have access to view the NIC's assigned addresses. + HandleProbe(remoteAddr, localAddr tcpip.Address, protocol tcpip.NetworkProtocolNumber, remoteLinkAddr tcpip.LinkAddress) + + // HandleConfirmation processes an incoming neighbor confirmation (e.g. ARP + // reply or Neighbor Advertisement for ARP or NDP, respectively). + HandleConfirmation(addr tcpip.Address, linkAddr tcpip.LinkAddress, flags ReachabilityConfirmationFlags) + + // HandleUpperLevelConfirmation processes an incoming upper-level protocol + // (e.g. TCP acknowledgements) reachability confirmation. + HandleUpperLevelConfirmation(addr tcpip.Address) +} + +// NUDConfigurations is the NUD configurations for the netstack. This is used +// by the neighbor cache to operate the NUD state machine on each device in the +// local network. +type NUDConfigurations struct { + // BaseReachableTime is the base duration for computing the random reachable + // time. + // + // Reachable time is the duration for which a neighbor is considered + // reachable after a positive reachability confirmation is received. It is a + // function of uniformly distributed random value between minRandomFactor and + // maxRandomFactor multiplied by baseReachableTime. Using a random component + // eliminates the possibility that Neighbor Unreachability Detection messages + // will synchronize with each other. + // + // After this time, a neighbor entry will transition from REACHABLE to STALE + // state. + // + // Must be greater than 0. + BaseReachableTime time.Duration + + // LearnBaseReachableTime enables learning BaseReachableTime during runtime + // from the neighbor discovery protocol, if supported. + // + // TODO(gvisor.dev/issue/2240): Implement this NUD configuration option. + LearnBaseReachableTime bool + + // MinRandomFactor is the minimum value of the random factor used for + // computing reachable time. + // + // See BaseReachbleTime for more information on computing the reachable time. + // + // Must be greater than 0. + MinRandomFactor float32 + + // MaxRandomFactor is the maximum value of the random factor used for + // computing reachabile time. + // + // See BaseReachbleTime for more information on computing the reachable time. + // + // Must be great than or equal to MinRandomFactor. + MaxRandomFactor float32 + + // RetransmitTimer is the duration between retransmission of reachability + // probes in the PROBE state. + RetransmitTimer time.Duration + + // LearnRetransmitTimer enables learning RetransmitTimer during runtime from + // the neighbor discovery protocol, if supported. + // + // TODO(gvisor.dev/issue/2241): Implement this NUD configuration option. + LearnRetransmitTimer bool + + // DelayFirstProbeTime is the duration to wait for a non-Neighbor-Discovery + // related protocol to reconfirm reachability after entering the DELAY state. + // After this time, a reachability probe will be sent and the entry will + // transition to the PROBE state. + // + // Must be greater than 0. + DelayFirstProbeTime time.Duration + + // MaxMulticastProbes is the number of reachability probes to send before + // concluding negative reachability and deleting the neighbor entry from the + // INCOMPLETE state. + // + // Must be greater than 0. + MaxMulticastProbes uint32 + + // MaxUnicastProbes is the number of reachability probes to send before + // concluding retransmission from within the PROBE state should cease and + // entry SHOULD be deleted. + // + // Must be greater than 0. + MaxUnicastProbes uint32 + + // MaxAnycastDelayTime is the time in which the stack SHOULD delay sending a + // response for a random time between 0 and this time, if the target address + // is an anycast address. + // + // TODO(gvisor.dev/issue/2242): Use this option when sending solicited + // neighbor confirmations to anycast addresses and proxying neighbor + // confirmations. + MaxAnycastDelayTime time.Duration + + // MaxReachabilityConfirmations is the number of unsolicited reachability + // confirmation messages a node MAY send to all-node multicast address when + // it determines its link-layer address has changed. + // + // TODO(gvisor.dev/issue/2246): Discuss if implementation of this NUD + // configuration option is necessary. + MaxReachabilityConfirmations uint32 + + // UnreachableTime describes how long an entry will remain in the FAILED + // state before being removed from the neighbor cache. + UnreachableTime time.Duration +} + +// DefaultNUDConfigurations returns a NUDConfigurations populated with default +// values defined by RFC 4861 section 10. +func DefaultNUDConfigurations() NUDConfigurations { + return NUDConfigurations{ + BaseReachableTime: defaultBaseReachableTime, + LearnBaseReachableTime: true, + MinRandomFactor: defaultMinRandomFactor, + MaxRandomFactor: defaultMaxRandomFactor, + RetransmitTimer: defaultRetransmitTimer, + LearnRetransmitTimer: true, + DelayFirstProbeTime: defaultDelayFirstProbeTime, + MaxMulticastProbes: defaultMaxMulticastProbes, + MaxUnicastProbes: defaultMaxUnicastProbes, + MaxAnycastDelayTime: defaultMaxAnycastDelayTime, + MaxReachabilityConfirmations: defaultMaxReachbilityConfirmations, + UnreachableTime: defaultUnreachableTime, + } +} + +// resetInvalidFields modifies an invalid NDPConfigurations with valid values. +// If invalid values are present in c, the corresponding default values will be +// used instead. This is needed to check, and conditionally fix, user-specified +// NUDConfigurations. +func (c *NUDConfigurations) resetInvalidFields() { + if c.BaseReachableTime < minimumBaseReachableTime { + c.BaseReachableTime = defaultBaseReachableTime + } + if c.MinRandomFactor <= 0 { + c.MinRandomFactor = defaultMinRandomFactor + } + if c.MaxRandomFactor < c.MinRandomFactor { + c.MaxRandomFactor = calcMaxRandomFactor(c.MinRandomFactor) + } + if c.RetransmitTimer < minimumRetransmitTimer { + c.RetransmitTimer = defaultRetransmitTimer + } + if c.DelayFirstProbeTime == 0 { + c.DelayFirstProbeTime = defaultDelayFirstProbeTime + } + if c.MaxMulticastProbes == 0 { + c.MaxMulticastProbes = defaultMaxMulticastProbes + } + if c.MaxUnicastProbes == 0 { + c.MaxUnicastProbes = defaultMaxUnicastProbes + } + if c.UnreachableTime == 0 { + c.UnreachableTime = defaultUnreachableTime + } +} + +// calcMaxRandomFactor calculates the maximum value of the random factor used +// for computing reachable time. This function is necessary for when the +// default specified in RFC 4861 section 10 is less than the current +// MinRandomFactor. +// +// Assumes minRandomFactor is positive since validation of the minimum value +// should come before the validation of the maximum. +func calcMaxRandomFactor(minRandomFactor float32) float32 { + if minRandomFactor > defaultMaxRandomFactor { + return minRandomFactor * 3 + } + return defaultMaxRandomFactor +} + +// A Rand is a source of random numbers. +type Rand interface { + // Float32 returns, as a float32, a pseudo-random number in [0.0,1.0). + Float32() float32 +} + +// NUDState stores states needed for calculating reachable time. +type NUDState struct { + rng Rand + + // mu protects the fields below. + // + // It is necessary for NUDState to handle its own locking since neighbor + // entries may access the NUD state from within the goroutine spawned by + // time.AfterFunc(). This goroutine may run concurrently with the main + // process for controlling the neighbor cache and would otherwise introduce + // race conditions if NUDState was not locked properly. + mu sync.RWMutex + + config NUDConfigurations + + // reachableTime is the duration to wait for a REACHABLE entry to + // transition into STALE after inactivity. This value is calculated with + // the algorithm defined in RFC 4861 section 6.3.2. + reachableTime time.Duration + + expiration time.Time + prevBaseReachableTime time.Duration + prevMinRandomFactor float32 + prevMaxRandomFactor float32 +} + +// NewNUDState returns new NUDState using c as configuration and the specified +// random number generator for use in recomputing ReachableTime. +func NewNUDState(c NUDConfigurations, rng Rand) *NUDState { + s := &NUDState{ + rng: rng, + } + s.config = c + return s +} + +// Config returns the NUD configuration. +func (s *NUDState) Config() NUDConfigurations { + s.mu.RLock() + defer s.mu.RUnlock() + return s.config +} + +// SetConfig replaces the existing NUD configurations with c. +func (s *NUDState) SetConfig(c NUDConfigurations) { + s.mu.Lock() + defer s.mu.Unlock() + s.config = c +} + +// ReachableTime returns the duration to wait for a REACHABLE entry to +// transition into STALE after inactivity. This value is recalculated for new +// values of BaseReachableTime, MinRandomFactor, and MaxRandomFactor using the +// algorithm defined in RFC 4861 section 6.3.2. +func (s *NUDState) ReachableTime() time.Duration { + s.mu.Lock() + defer s.mu.Unlock() + + if time.Now().After(s.expiration) || + s.config.BaseReachableTime != s.prevBaseReachableTime || + s.config.MinRandomFactor != s.prevMinRandomFactor || + s.config.MaxRandomFactor != s.prevMaxRandomFactor { + return s.recomputeReachableTimeLocked() + } + return s.reachableTime +} + +// recomputeReachableTimeLocked forces a recalculation of ReachableTime using +// the algorithm defined in RFC 4861 section 6.3.2. +// +// This SHOULD automatically be invoked during certain situations, as per +// RFC 4861 section 6.3.4: +// +// If the received Reachable Time value is non-zero, the host SHOULD set its +// BaseReachableTime variable to the received value. If the new value +// differs from the previous value, the host SHOULD re-compute a new random +// ReachableTime value. ReachableTime is computed as a uniformly +// distributed random value between MIN_RANDOM_FACTOR and MAX_RANDOM_FACTOR +// times the BaseReachableTime. Using a random component eliminates the +// possibility that Neighbor Unreachability Detection messages will +// synchronize with each other. +// +// In most cases, the advertised Reachable Time value will be the same in +// consecutive Router Advertisements, and a host's BaseReachableTime rarely +// changes. In such cases, an implementation SHOULD ensure that a new +// random value gets re-computed at least once every few hours. +// +// s.mu MUST be locked for writing. +func (s *NUDState) recomputeReachableTimeLocked() time.Duration { + s.prevBaseReachableTime = s.config.BaseReachableTime + s.prevMinRandomFactor = s.config.MinRandomFactor + s.prevMaxRandomFactor = s.config.MaxRandomFactor + + randomFactor := s.config.MinRandomFactor + s.rng.Float32()*(s.config.MaxRandomFactor-s.config.MinRandomFactor) + + // Check for overflow, given that minRandomFactor and maxRandomFactor are + // guaranteed to be positive numbers. + if float32(math.MaxInt64)/randomFactor < float32(s.config.BaseReachableTime) { + s.reachableTime = time.Duration(math.MaxInt64) + } else if randomFactor == 1 { + // Avoid loss of precision when a large base reachable time is used. + s.reachableTime = s.config.BaseReachableTime + } else { + reachableTime := int64(float32(s.config.BaseReachableTime) * randomFactor) + s.reachableTime = time.Duration(reachableTime) + } + + s.expiration = time.Now().Add(2 * time.Hour) + return s.reachableTime +} diff --git a/pkg/tcpip/stack/nud_test.go b/pkg/tcpip/stack/nud_test.go new file mode 100644 index 000000000..2494ee610 --- /dev/null +++ b/pkg/tcpip/stack/nud_test.go @@ -0,0 +1,795 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package stack_test + +import ( + "math" + "testing" + "time" + + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/link/channel" + "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" + "gvisor.dev/gvisor/pkg/tcpip/stack" +) + +const ( + defaultBaseReachableTime = 30 * time.Second + minimumBaseReachableTime = time.Millisecond + defaultMinRandomFactor = 0.5 + defaultMaxRandomFactor = 1.5 + defaultRetransmitTimer = time.Second + minimumRetransmitTimer = time.Millisecond + defaultDelayFirstProbeTime = 5 * time.Second + defaultMaxMulticastProbes = 3 + defaultMaxUnicastProbes = 3 + defaultMaxAnycastDelayTime = time.Second + defaultMaxReachbilityConfirmations = 3 + defaultUnreachableTime = 5 * time.Second + + defaultFakeRandomNum = 0.5 +) + +// fakeRand is a deterministic random number generator. +type fakeRand struct { + num float32 +} + +var _ stack.Rand = (*fakeRand)(nil) + +func (f *fakeRand) Float32() float32 { + return f.num +} + +// TestSetNUDConfigurationFailsForBadNICID tests to make sure we get an error if +// we attempt to update NUD configurations using an invalid NICID. +func TestSetNUDConfigurationFailsForBadNICID(t *testing.T) { + s := stack.New(stack.Options{ + // A neighbor cache is required to store NUDConfigurations. The networking + // stack will only allocate neighbor caches if a protocol providing link + // address resolution is specified (e.g. ARP or IPv6). + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + }) + + // No NIC with ID 1 yet. + config := stack.NUDConfigurations{} + if err := s.SetNUDConfigurations(1, config); err != tcpip.ErrUnknownNICID { + t.Fatalf("got s.SetNDPConfigurations(1, %+v) = %v, want = %s", config, err, tcpip.ErrUnknownNICID) + } +} + +// TestNUDConfigurationFailsForNotSupported tests to make sure we get a +// NotSupported error if we attempt to retrieve NUD configurations when the +// stack doesn't support NUD. +// +// The stack will report to not support NUD if a neighbor cache for a given NIC +// is not allocated. The networking stack will only allocate neighbor caches if +// a protocol providing link address resolution is specified (e.g. ARP, IPv6). +func TestNUDConfigurationFailsForNotSupported(t *testing.T) { + const nicID = 1 + + e := channel.New(0, 1280, linkAddr1) + e.LinkEPCapabilities |= stack.CapabilityResolutionRequired + + s := stack.New(stack.Options{ + NUDConfigs: stack.DefaultNUDConfigurations(), + }) + if err := s.CreateNIC(nicID, e); err != nil { + t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) + } + if _, err := s.NUDConfigurations(nicID); err != tcpip.ErrNotSupported { + t.Fatalf("got s.NDPConfigurations(%d) = %v, want = %s", nicID, err, tcpip.ErrNotSupported) + } +} + +// TestNUDConfigurationFailsForNotSupported tests to make sure we get a +// NotSupported error if we attempt to set NUD configurations when the stack +// doesn't support NUD. +// +// The stack will report to not support NUD if a neighbor cache for a given NIC +// is not allocated. The networking stack will only allocate neighbor caches if +// a protocol providing link address resolution is specified (e.g. ARP, IPv6). +func TestSetNUDConfigurationFailsForNotSupported(t *testing.T) { + const nicID = 1 + + e := channel.New(0, 1280, linkAddr1) + e.LinkEPCapabilities |= stack.CapabilityResolutionRequired + + s := stack.New(stack.Options{ + NUDConfigs: stack.DefaultNUDConfigurations(), + }) + if err := s.CreateNIC(nicID, e); err != nil { + t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) + } + + config := stack.NUDConfigurations{} + if err := s.SetNUDConfigurations(nicID, config); err != tcpip.ErrNotSupported { + t.Fatalf("got s.SetNDPConfigurations(%d, %+v) = %v, want = %s", nicID, config, err, tcpip.ErrNotSupported) + } +} + +// TestDefaultNUDConfigurationIsValid verifies that calling +// resetInvalidFields() on the result of DefaultNUDConfigurations() does not +// change anything. DefaultNUDConfigurations() should return a valid +// NUDConfigurations. +func TestDefaultNUDConfigurations(t *testing.T) { + const nicID = 1 + + e := channel.New(0, 1280, linkAddr1) + e.LinkEPCapabilities |= stack.CapabilityResolutionRequired + + s := stack.New(stack.Options{ + // A neighbor cache is required to store NUDConfigurations. The networking + // stack will only allocate neighbor caches if a protocol providing link + // address resolution is specified (e.g. ARP or IPv6). + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NUDConfigs: stack.DefaultNUDConfigurations(), + }) + if err := s.CreateNIC(nicID, e); err != nil { + t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) + } + c, err := s.NUDConfigurations(nicID) + if err != nil { + t.Fatalf("got stack.NUDConfigurations(%d) = %s", nicID, err) + } + if got, want := c, stack.DefaultNUDConfigurations(); got != want { + t.Errorf("got stack.NUDConfigurations(%d) = %+v, want = %+v", nicID, got, want) + } +} + +func TestNUDConfigurationsBaseReachableTime(t *testing.T) { + tests := []struct { + name string + baseReachableTime time.Duration + want time.Duration + }{ + // Invalid cases + { + name: "EqualToZero", + baseReachableTime: 0, + want: defaultBaseReachableTime, + }, + // Valid cases + { + name: "MoreThanZero", + baseReachableTime: time.Millisecond, + want: time.Millisecond, + }, + { + name: "MoreThanDefaultBaseReachableTime", + baseReachableTime: 2 * defaultBaseReachableTime, + want: 2 * defaultBaseReachableTime, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + const nicID = 1 + + c := stack.DefaultNUDConfigurations() + c.BaseReachableTime = test.baseReachableTime + + e := channel.New(0, 1280, linkAddr1) + e.LinkEPCapabilities |= stack.CapabilityResolutionRequired + + s := stack.New(stack.Options{ + // A neighbor cache is required to store NUDConfigurations. The + // networking stack will only allocate neighbor caches if a protocol + // providing link address resolution is specified (e.g. ARP or IPv6). + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NUDConfigs: c, + }) + if err := s.CreateNIC(nicID, e); err != nil { + t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) + } + sc, err := s.NUDConfigurations(nicID) + if err != nil { + t.Fatalf("got stack.NUDConfigurations(%d) = %s", nicID, err) + } + if got := sc.BaseReachableTime; got != test.want { + t.Errorf("got BaseReachableTime = %q, want = %q", got, test.want) + } + }) + } +} + +func TestNUDConfigurationsMinRandomFactor(t *testing.T) { + tests := []struct { + name string + minRandomFactor float32 + want float32 + }{ + // Invalid cases + { + name: "LessThanZero", + minRandomFactor: -1, + want: defaultMinRandomFactor, + }, + { + name: "EqualToZero", + minRandomFactor: 0, + want: defaultMinRandomFactor, + }, + // Valid cases + { + name: "MoreThanZero", + minRandomFactor: 1, + want: 1, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + const nicID = 1 + + c := stack.DefaultNUDConfigurations() + c.MinRandomFactor = test.minRandomFactor + + e := channel.New(0, 1280, linkAddr1) + e.LinkEPCapabilities |= stack.CapabilityResolutionRequired + + s := stack.New(stack.Options{ + // A neighbor cache is required to store NUDConfigurations. The + // networking stack will only allocate neighbor caches if a protocol + // providing link address resolution is specified (e.g. ARP or IPv6). + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NUDConfigs: c, + }) + if err := s.CreateNIC(nicID, e); err != nil { + t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) + } + sc, err := s.NUDConfigurations(nicID) + if err != nil { + t.Fatalf("got stack.NUDConfigurations(%d) = %s", nicID, err) + } + if got := sc.MinRandomFactor; got != test.want { + t.Errorf("got MinRandomFactor = %f, want = %f", got, test.want) + } + }) + } +} + +func TestNUDConfigurationsMaxRandomFactor(t *testing.T) { + tests := []struct { + name string + minRandomFactor float32 + maxRandomFactor float32 + want float32 + }{ + // Invalid cases + { + name: "LessThanZero", + minRandomFactor: defaultMinRandomFactor, + maxRandomFactor: -1, + want: defaultMaxRandomFactor, + }, + { + name: "EqualToZero", + minRandomFactor: defaultMinRandomFactor, + maxRandomFactor: 0, + want: defaultMaxRandomFactor, + }, + { + name: "LessThanMinRandomFactor", + minRandomFactor: defaultMinRandomFactor, + maxRandomFactor: defaultMinRandomFactor * 0.99, + want: defaultMaxRandomFactor, + }, + { + name: "MoreThanMinRandomFactorWhenMinRandomFactorIsLargerThanMaxRandomFactorDefault", + minRandomFactor: defaultMaxRandomFactor * 2, + maxRandomFactor: defaultMaxRandomFactor, + want: defaultMaxRandomFactor * 6, + }, + // Valid cases + { + name: "EqualToMinRandomFactor", + minRandomFactor: defaultMinRandomFactor, + maxRandomFactor: defaultMinRandomFactor, + want: defaultMinRandomFactor, + }, + { + name: "MoreThanMinRandomFactor", + minRandomFactor: defaultMinRandomFactor, + maxRandomFactor: defaultMinRandomFactor * 1.1, + want: defaultMinRandomFactor * 1.1, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + const nicID = 1 + + c := stack.DefaultNUDConfigurations() + c.MinRandomFactor = test.minRandomFactor + c.MaxRandomFactor = test.maxRandomFactor + + e := channel.New(0, 1280, linkAddr1) + e.LinkEPCapabilities |= stack.CapabilityResolutionRequired + + s := stack.New(stack.Options{ + // A neighbor cache is required to store NUDConfigurations. The + // networking stack will only allocate neighbor caches if a protocol + // providing link address resolution is specified (e.g. ARP or IPv6). + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NUDConfigs: c, + }) + if err := s.CreateNIC(nicID, e); err != nil { + t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) + } + sc, err := s.NUDConfigurations(nicID) + if err != nil { + t.Fatalf("got stack.NUDConfigurations(%d) = %s", nicID, err) + } + if got := sc.MaxRandomFactor; got != test.want { + t.Errorf("got MaxRandomFactor = %f, want = %f", got, test.want) + } + }) + } +} + +func TestNUDConfigurationsRetransmitTimer(t *testing.T) { + tests := []struct { + name string + retransmitTimer time.Duration + want time.Duration + }{ + // Invalid cases + { + name: "EqualToZero", + retransmitTimer: 0, + want: defaultRetransmitTimer, + }, + { + name: "LessThanMinimumRetransmitTimer", + retransmitTimer: minimumRetransmitTimer - time.Nanosecond, + want: defaultRetransmitTimer, + }, + // Valid cases + { + name: "EqualToMinimumRetransmitTimer", + retransmitTimer: minimumRetransmitTimer, + want: minimumBaseReachableTime, + }, + { + name: "LargetThanMinimumRetransmitTimer", + retransmitTimer: 2 * minimumBaseReachableTime, + want: 2 * minimumBaseReachableTime, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + const nicID = 1 + + c := stack.DefaultNUDConfigurations() + c.RetransmitTimer = test.retransmitTimer + + e := channel.New(0, 1280, linkAddr1) + e.LinkEPCapabilities |= stack.CapabilityResolutionRequired + + s := stack.New(stack.Options{ + // A neighbor cache is required to store NUDConfigurations. The + // networking stack will only allocate neighbor caches if a protocol + // providing link address resolution is specified (e.g. ARP or IPv6). + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NUDConfigs: c, + }) + if err := s.CreateNIC(nicID, e); err != nil { + t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) + } + sc, err := s.NUDConfigurations(nicID) + if err != nil { + t.Fatalf("got stack.NUDConfigurations(%d) = %s", nicID, err) + } + if got := sc.RetransmitTimer; got != test.want { + t.Errorf("got RetransmitTimer = %q, want = %q", got, test.want) + } + }) + } +} + +func TestNUDConfigurationsDelayFirstProbeTime(t *testing.T) { + tests := []struct { + name string + delayFirstProbeTime time.Duration + want time.Duration + }{ + // Invalid cases + { + name: "EqualToZero", + delayFirstProbeTime: 0, + want: defaultDelayFirstProbeTime, + }, + // Valid cases + { + name: "MoreThanZero", + delayFirstProbeTime: time.Millisecond, + want: time.Millisecond, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + const nicID = 1 + + c := stack.DefaultNUDConfigurations() + c.DelayFirstProbeTime = test.delayFirstProbeTime + + e := channel.New(0, 1280, linkAddr1) + e.LinkEPCapabilities |= stack.CapabilityResolutionRequired + + s := stack.New(stack.Options{ + // A neighbor cache is required to store NUDConfigurations. The + // networking stack will only allocate neighbor caches if a protocol + // providing link address resolution is specified (e.g. ARP or IPv6). + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NUDConfigs: c, + }) + if err := s.CreateNIC(nicID, e); err != nil { + t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) + } + sc, err := s.NUDConfigurations(nicID) + if err != nil { + t.Fatalf("got stack.NUDConfigurations(%d) = %s", nicID, err) + } + if got := sc.DelayFirstProbeTime; got != test.want { + t.Errorf("got DelayFirstProbeTime = %q, want = %q", got, test.want) + } + }) + } +} + +func TestNUDConfigurationsMaxMulticastProbes(t *testing.T) { + tests := []struct { + name string + maxMulticastProbes uint32 + want uint32 + }{ + // Invalid cases + { + name: "EqualToZero", + maxMulticastProbes: 0, + want: defaultMaxMulticastProbes, + }, + // Valid cases + { + name: "MoreThanZero", + maxMulticastProbes: 1, + want: 1, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + const nicID = 1 + + c := stack.DefaultNUDConfigurations() + c.MaxMulticastProbes = test.maxMulticastProbes + + e := channel.New(0, 1280, linkAddr1) + e.LinkEPCapabilities |= stack.CapabilityResolutionRequired + + s := stack.New(stack.Options{ + // A neighbor cache is required to store NUDConfigurations. The + // networking stack will only allocate neighbor caches if a protocol + // providing link address resolution is specified (e.g. ARP or IPv6). + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NUDConfigs: c, + }) + if err := s.CreateNIC(nicID, e); err != nil { + t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) + } + sc, err := s.NUDConfigurations(nicID) + if err != nil { + t.Fatalf("got stack.NUDConfigurations(%d) = %s", nicID, err) + } + if got := sc.MaxMulticastProbes; got != test.want { + t.Errorf("got MaxMulticastProbes = %q, want = %q", got, test.want) + } + }) + } +} + +func TestNUDConfigurationsMaxUnicastProbes(t *testing.T) { + tests := []struct { + name string + maxUnicastProbes uint32 + want uint32 + }{ + // Invalid cases + { + name: "EqualToZero", + maxUnicastProbes: 0, + want: defaultMaxUnicastProbes, + }, + // Valid cases + { + name: "MoreThanZero", + maxUnicastProbes: 1, + want: 1, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + const nicID = 1 + + c := stack.DefaultNUDConfigurations() + c.MaxUnicastProbes = test.maxUnicastProbes + + e := channel.New(0, 1280, linkAddr1) + e.LinkEPCapabilities |= stack.CapabilityResolutionRequired + + s := stack.New(stack.Options{ + // A neighbor cache is required to store NUDConfigurations. The + // networking stack will only allocate neighbor caches if a protocol + // providing link address resolution is specified (e.g. ARP or IPv6). + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NUDConfigs: c, + }) + if err := s.CreateNIC(nicID, e); err != nil { + t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) + } + sc, err := s.NUDConfigurations(nicID) + if err != nil { + t.Fatalf("got stack.NUDConfigurations(%d) = %s", nicID, err) + } + if got := sc.MaxUnicastProbes; got != test.want { + t.Errorf("got MaxUnicastProbes = %q, want = %q", got, test.want) + } + }) + } +} + +func TestNUDConfigurationsUnreachableTime(t *testing.T) { + tests := []struct { + name string + unreachableTime time.Duration + want time.Duration + }{ + // Invalid cases + { + name: "EqualToZero", + unreachableTime: 0, + want: defaultUnreachableTime, + }, + // Valid cases + { + name: "MoreThanZero", + unreachableTime: time.Millisecond, + want: time.Millisecond, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + const nicID = 1 + + c := stack.DefaultNUDConfigurations() + c.UnreachableTime = test.unreachableTime + + e := channel.New(0, 1280, linkAddr1) + e.LinkEPCapabilities |= stack.CapabilityResolutionRequired + + s := stack.New(stack.Options{ + // A neighbor cache is required to store NUDConfigurations. The + // networking stack will only allocate neighbor caches if a protocol + // providing link address resolution is specified (e.g. ARP or IPv6). + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NUDConfigs: c, + }) + if err := s.CreateNIC(nicID, e); err != nil { + t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) + } + sc, err := s.NUDConfigurations(nicID) + if err != nil { + t.Fatalf("got stack.NUDConfigurations(%d) = %s", nicID, err) + } + if got := sc.UnreachableTime; got != test.want { + t.Errorf("got UnreachableTime = %q, want = %q", got, test.want) + } + }) + } +} + +// TestNUDStateReachableTime verifies the correctness of the ReachableTime +// computation. +func TestNUDStateReachableTime(t *testing.T) { + tests := []struct { + name string + baseReachableTime time.Duration + minRandomFactor float32 + maxRandomFactor float32 + want time.Duration + }{ + { + name: "AllZeros", + baseReachableTime: 0, + minRandomFactor: 0, + maxRandomFactor: 0, + want: 0, + }, + { + name: "ZeroMaxRandomFactor", + baseReachableTime: time.Second, + minRandomFactor: 0, + maxRandomFactor: 0, + want: 0, + }, + { + name: "ZeroMinRandomFactor", + baseReachableTime: time.Second, + minRandomFactor: 0, + maxRandomFactor: 1, + want: time.Duration(defaultFakeRandomNum * float32(time.Second)), + }, + { + name: "FractionalRandomFactor", + baseReachableTime: time.Duration(math.MaxInt64), + minRandomFactor: 0.001, + maxRandomFactor: 0.002, + want: time.Duration((0.001 + (0.001 * defaultFakeRandomNum)) * float32(math.MaxInt64)), + }, + { + name: "MinAndMaxRandomFactorsEqual", + baseReachableTime: time.Second, + minRandomFactor: 1, + maxRandomFactor: 1, + want: time.Second, + }, + { + name: "MinAndMaxRandomFactorsDifferent", + baseReachableTime: time.Second, + minRandomFactor: 1, + maxRandomFactor: 2, + want: time.Duration((1.0 + defaultFakeRandomNum) * float32(time.Second)), + }, + { + name: "MaxInt64", + baseReachableTime: time.Duration(math.MaxInt64), + minRandomFactor: 1, + maxRandomFactor: 1, + want: time.Duration(math.MaxInt64), + }, + { + name: "Overflow", + baseReachableTime: time.Duration(math.MaxInt64), + minRandomFactor: 1.5, + maxRandomFactor: 1.5, + want: time.Duration(math.MaxInt64), + }, + { + name: "DoubleOverflow", + baseReachableTime: time.Duration(math.MaxInt64), + minRandomFactor: 2.5, + maxRandomFactor: 2.5, + want: time.Duration(math.MaxInt64), + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + c := stack.NUDConfigurations{ + BaseReachableTime: test.baseReachableTime, + MinRandomFactor: test.minRandomFactor, + MaxRandomFactor: test.maxRandomFactor, + } + // A fake random number generator is used to ensure deterministic + // results. + rng := fakeRand{ + num: defaultFakeRandomNum, + } + s := stack.NewNUDState(c, &rng) + if got, want := s.ReachableTime(), test.want; got != want { + t.Errorf("got ReachableTime = %q, want = %q", got, want) + } + }) + } +} + +// TestNUDStateRecomputeReachableTime exercises the ReachableTime function +// twice to verify recomputation of reachable time when the min random factor, +// max random factor, or base reachable time changes. +func TestNUDStateRecomputeReachableTime(t *testing.T) { + const defaultBase = time.Second + const defaultMin = 2.0 * defaultMaxRandomFactor + const defaultMax = 3.0 * defaultMaxRandomFactor + + tests := []struct { + name string + baseReachableTime time.Duration + minRandomFactor float32 + maxRandomFactor float32 + want time.Duration + }{ + { + name: "BaseReachableTime", + baseReachableTime: 2 * defaultBase, + minRandomFactor: defaultMin, + maxRandomFactor: defaultMax, + want: time.Duration((defaultMin + (defaultMax-defaultMin)*defaultFakeRandomNum) * float32(2*defaultBase)), + }, + { + name: "MinRandomFactor", + baseReachableTime: defaultBase, + minRandomFactor: defaultMax, + maxRandomFactor: defaultMax, + want: time.Duration(defaultMax * float32(defaultBase)), + }, + { + name: "MaxRandomFactor", + baseReachableTime: defaultBase, + minRandomFactor: defaultMin, + maxRandomFactor: defaultMin, + want: time.Duration(defaultMin * float32(defaultBase)), + }, + { + name: "BothRandomFactor", + baseReachableTime: defaultBase, + minRandomFactor: 2 * defaultMin, + maxRandomFactor: 2 * defaultMax, + want: time.Duration((2*defaultMin + (2*defaultMax-2*defaultMin)*defaultFakeRandomNum) * float32(defaultBase)), + }, + { + name: "BaseReachableTimeAndBothRandomFactors", + baseReachableTime: 2 * defaultBase, + minRandomFactor: 2 * defaultMin, + maxRandomFactor: 2 * defaultMax, + want: time.Duration((2*defaultMin + (2*defaultMax-2*defaultMin)*defaultFakeRandomNum) * float32(2*defaultBase)), + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + c := stack.DefaultNUDConfigurations() + c.BaseReachableTime = defaultBase + c.MinRandomFactor = defaultMin + c.MaxRandomFactor = defaultMax + + // A fake random number generator is used to ensure deterministic + // results. + rng := fakeRand{ + num: defaultFakeRandomNum, + } + s := stack.NewNUDState(c, &rng) + old := s.ReachableTime() + + if got, want := s.ReachableTime(), old; got != want { + t.Errorf("got ReachableTime = %q, want = %q", got, want) + } + + // Check for recomputation when changing the min random factor, the max + // random factor, the base reachability time, or any permutation of those + // three options. + c.BaseReachableTime = test.baseReachableTime + c.MinRandomFactor = test.minRandomFactor + c.MaxRandomFactor = test.maxRandomFactor + s.SetConfig(c) + + if got, want := s.ReachableTime(), test.want; got != want { + t.Errorf("got ReachableTime = %q, want = %q", got, want) + } + + // Verify that ReachableTime isn't recomputed when none of the + // configuration options change. The random factor is changed so that if + // a recompution were to occur, ReachableTime would change. + rng.num = defaultFakeRandomNum / 2.0 + if got, want := s.ReachableTime(), test.want; got != want { + t.Errorf("got ReachableTime = %q, want = %q", got, want) + } + }) + } +} diff --git a/pkg/tcpip/stack/packet_buffer.go b/pkg/tcpip/stack/packet_buffer.go index 1b5da6017..5d6865e35 100644 --- a/pkg/tcpip/stack/packet_buffer.go +++ b/pkg/tcpip/stack/packet_buffer.go @@ -14,6 +14,7 @@ package stack import ( + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" ) @@ -24,7 +25,7 @@ import ( // multiple endpoints. Clone() should be called in such cases so that // modifications to the Data field do not affect other copies. type PacketBuffer struct { - _ noCopy + _ sync.NoCopy // PacketBufferEntry is used to build an intrusive list of // PacketBuffers. @@ -78,6 +79,10 @@ type PacketBuffer struct { // NatDone indicates if the packet has been manipulated as per NAT // iptables rule. NatDone bool + + // PktType indicates the SockAddrLink.PacketType of the packet as defined in + // https://www.man7.org/linux/man-pages/man7/packet.7.html. + PktType tcpip.PacketType } // Clone makes a copy of pk. It clones the Data field, which creates a new @@ -102,14 +107,3 @@ func (pk *PacketBuffer) Clone() *PacketBuffer { NatDone: pk.NatDone, } } - -// noCopy may be embedded into structs which must not be copied -// after the first use. -// -// See https://golang.org/issues/8005#issuecomment-190753527 -// for details. -type noCopy struct{} - -// Lock is a no-op used by -copylocks checker from `go vet`. -func (*noCopy) Lock() {} -func (*noCopy) Unlock() {} diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go index 5cbc946b6..8604c4259 100644 --- a/pkg/tcpip/stack/registration.go +++ b/pkg/tcpip/stack/registration.go @@ -18,6 +18,7 @@ import ( "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/waiter" ) @@ -51,8 +52,11 @@ type TransportEndpointID struct { type ControlType int // The following are the allowed values for ControlType values. +// TODO(http://gvisor.dev/issue/3210): Support time exceeded messages. const ( - ControlPacketTooBig ControlType = iota + ControlNetworkUnreachable ControlType = iota + ControlNoRoute + ControlPacketTooBig ControlPortUnreachable ControlUnknown ) @@ -329,8 +333,7 @@ type NetworkProtocol interface { } // NetworkDispatcher contains the methods used by the network stack to deliver -// packets to the appropriate network endpoint after it has been handled by -// the data link layer. +// inbound/outbound packets to the appropriate network/packet(if any) endpoints. type NetworkDispatcher interface { // DeliverNetworkPacket finds the appropriate network protocol endpoint // and hands the packet over for further processing. @@ -341,6 +344,16 @@ type NetworkDispatcher interface { // // DeliverNetworkPacket takes ownership of pkt. DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) + + // DeliverOutboundPacket is called by link layer when a packet is being + // sent out. + // + // pkt.LinkHeader may or may not be set before calling + // DeliverOutboundPacket. Some packets do not have link headers (e.g. + // packets sent via loopback), and won't have the field set. + // + // DeliverOutboundPacket takes ownership of pkt. + DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) } // LinkEndpointCapabilities is the type associated with the capabilities @@ -436,6 +449,15 @@ type LinkEndpoint interface { // Wait will not block if the endpoint hasn't started any goroutines // yet, even if it might later. Wait() + + // ARPHardwareType returns the ARPHRD_TYPE of the link endpoint. + // + // See: + // https://github.com/torvalds/linux/blob/aa0c9086b40c17a7ad94425b3b70dd1fdd7497bf/include/uapi/linux/if_arp.h#L30 + ARPHardwareType() header.ARPHardwareType + + // AddHeader adds a link layer header to pkt if required. + AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) } // InjectableLinkEndpoint is a LinkEndpoint where inbound packets are @@ -456,12 +478,13 @@ type InjectableLinkEndpoint interface { // A LinkAddressResolver is an extension to a NetworkProtocol that // can resolve link addresses. type LinkAddressResolver interface { - // LinkAddressRequest sends a request for the LinkAddress of addr. - // The request is sent on linkEP with localAddr as the source. + // LinkAddressRequest sends a request for the LinkAddress of addr. Broadcasts + // the request on the local network if remoteLinkAddr is the zero value. The + // request is sent on linkEP with localAddr as the source. // // A valid response will cause the discovery protocol's network // endpoint to call AddLinkAddress. - LinkAddressRequest(addr, localAddr tcpip.Address, linkEP LinkEndpoint) *tcpip.Error + LinkAddressRequest(addr, localAddr tcpip.Address, remoteLinkAddr tcpip.LinkAddress, linkEP LinkEndpoint) *tcpip.Error // ResolveStaticAddress attempts to resolve address without sending // requests. It either resolves the name immediately or returns the diff --git a/pkg/tcpip/stack/route.go b/pkg/tcpip/stack/route.go index d65f8049e..91e0110f1 100644 --- a/pkg/tcpip/stack/route.go +++ b/pkg/tcpip/stack/route.go @@ -48,6 +48,10 @@ type Route struct { // Loop controls where WritePacket should send packets. Loop PacketLooping + + // directedBroadcast indicates whether this route is sending a directed + // broadcast packet. + directedBroadcast bool } // makeRoute initializes a new route. It takes ownership of the provided @@ -275,6 +279,12 @@ func (r *Route) Stack() *Stack { return r.ref.stack() } +// IsBroadcast returns true if the route is to send a broadcast packet. +func (r *Route) IsBroadcast() bool { + // Only IPv4 has a notion of broadcast. + return r.directedBroadcast || r.RemoteAddress == header.IPv4Broadcast +} + // ReverseRoute returns new route with given source and destination address. func (r *Route) ReverseRoute(src tcpip.Address, dst tcpip.Address) Route { return Route{ diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 51abe32a7..3f07e4159 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -425,6 +425,7 @@ type Stack struct { handleLocal bool // tables are the iptables packet filtering and manipulation rules. + // TODO(gvisor.dev/issue/170): S/R this field. tables *IPTables // resumableEndpoints is a list of endpoints that need to be resumed if the @@ -444,6 +445,9 @@ type Stack struct { // ndpConfigs is the default NDP configurations used by interfaces. ndpConfigs NDPConfigurations + // nudConfigs is the default NUD configurations used by interfaces. + nudConfigs NUDConfigurations + // autoGenIPv6LinkLocal determines whether or not the stack will attempt // to auto-generate an IPv6 link-local address for newly enabled non-loopback // NICs. See the AutoGenIPv6LinkLocal field of Options for more details. @@ -453,6 +457,10 @@ type Stack struct { // integrator NDP related events. ndpDisp NDPDispatcher + // nudDisp is the NUD event dispatcher that is used to send the netstack + // integrator NUD related events. + nudDisp NUDDispatcher + // uniqueIDGenerator is a generator of unique identifiers. uniqueIDGenerator UniqueID @@ -471,6 +479,14 @@ type Stack struct { // randomGenerator is an injectable pseudo random generator that can be // used when a random number is required. randomGenerator *mathrand.Rand + + // sendBufferSize holds the min/default/max send buffer sizes for + // endpoints other than TCP. + sendBufferSize SendBufferSizeOption + + // receiveBufferSize holds the min/default/max receive buffer sizes for + // endpoints other than TCP. + receiveBufferSize ReceiveBufferSizeOption } // UniqueID is an abstract generator of unique identifiers. @@ -509,6 +525,9 @@ type Options struct { // before assigning an address to a NIC. NDPConfigs NDPConfigurations + // NUDConfigs is the default NUD configurations used by interfaces. + NUDConfigs NUDConfigurations + // AutoGenIPv6LinkLocal determines whether or not the stack will attempt to // auto-generate an IPv6 link-local address for newly enabled non-loopback // NICs. @@ -527,6 +546,10 @@ type Options struct { // receive NDP related events. NDPDisp NDPDispatcher + // NUDDisp is the NUD event dispatcher that an integrator can provide to + // receive NUD related events. + NUDDisp NUDDispatcher + // RawFactory produces raw endpoints. Raw endpoints are enabled only if // this is non-nil. RawFactory RawFactory @@ -661,6 +684,8 @@ func New(opts Options) *Stack { // Make sure opts.NDPConfigs contains valid values only. opts.NDPConfigs.validate() + opts.NUDConfigs.resetInvalidFields() + s := &Stack{ transportProtocols: make(map[tcpip.TransportProtocolNumber]*transportProtocolState), networkProtocols: make(map[tcpip.NetworkProtocolNumber]NetworkProtocol), @@ -676,13 +701,25 @@ func New(opts Options) *Stack { icmpRateLimiter: NewICMPRateLimiter(), seed: generateRandUint32(), ndpConfigs: opts.NDPConfigs, + nudConfigs: opts.NUDConfigs, autoGenIPv6LinkLocal: opts.AutoGenIPv6LinkLocal, uniqueIDGenerator: opts.UniqueID, ndpDisp: opts.NDPDisp, + nudDisp: opts.NUDDisp, opaqueIIDOpts: opts.OpaqueIIDOpts, tempIIDSeed: opts.TempIIDSeed, forwarder: newForwardQueue(), randomGenerator: mathrand.New(randSrc), + sendBufferSize: SendBufferSizeOption{ + Min: MinBufferSize, + Default: DefaultBufferSize, + Max: DefaultMaxBufferSize, + }, + receiveBufferSize: ReceiveBufferSizeOption{ + Min: MinBufferSize, + Default: DefaultBufferSize, + Max: DefaultMaxBufferSize, + }, } // Add specified network protocols. @@ -709,6 +746,11 @@ func New(opts Options) *Stack { return s } +// newJob returns a tcpip.Job using the Stack clock. +func (s *Stack) newJob(l sync.Locker, f func()) *tcpip.Job { + return tcpip.NewJob(s.clock, l, f) +} + // UniqueID returns a unique identifier. func (s *Stack) UniqueID() uint64 { return s.uniqueIDGenerator.UniqueID() @@ -782,9 +824,10 @@ func (s *Stack) SetTransportProtocolHandler(p tcpip.TransportProtocolNumber, h f } } -// NowNanoseconds implements tcpip.Clock.NowNanoseconds. -func (s *Stack) NowNanoseconds() int64 { - return s.clock.NowNanoseconds() +// Clock returns the Stack's clock for retrieving the current time and +// scheduling work. +func (s *Stack) Clock() tcpip.Clock { + return s.clock } // Stats returns a mutable copy of the current stats. @@ -1076,6 +1119,11 @@ type NICInfo struct { // Context is user-supplied data optionally supplied in CreateNICWithOptions. // See type NICOptions for more details. Context NICContext + + // ARPHardwareType holds the ARP Hardware type of the NIC. This is the + // value sent in haType field of an ARP Request sent by this NIC and the + // value expected in the haType field of an ARP response. + ARPHardwareType header.ARPHardwareType } // HasNIC returns true if the NICID is defined in the stack. @@ -1107,6 +1155,7 @@ func (s *Stack) NICInfo() map[tcpip.NICID]NICInfo { MTU: nic.linkEP.MTU(), Stats: nic.stats, Context: nic.context, + ARPHardwareType: nic.linkEP.ARPHardwareType(), } } return nics @@ -1253,9 +1302,9 @@ func (s *Stack) FindRoute(id tcpip.NICID, localAddr, remoteAddr tcpip.Address, n s.mu.RLock() defer s.mu.RUnlock() - isBroadcast := remoteAddr == header.IPv4Broadcast + isLocalBroadcast := remoteAddr == header.IPv4Broadcast isMulticast := header.IsV4MulticastAddress(remoteAddr) || header.IsV6MulticastAddress(remoteAddr) - needRoute := !(isBroadcast || isMulticast || header.IsV6LinkLocalAddress(remoteAddr)) + needRoute := !(isLocalBroadcast || isMulticast || header.IsV6LinkLocalAddress(remoteAddr)) if id != 0 && !needRoute { if nic, ok := s.nics[id]; ok && nic.enabled() { if ref := s.getRefEP(nic, localAddr, remoteAddr, netProto); ref != nil { @@ -1276,9 +1325,16 @@ func (s *Stack) FindRoute(id tcpip.NICID, localAddr, remoteAddr tcpip.Address, n } r := makeRoute(netProto, ref.ep.ID().LocalAddress, remoteAddr, nic.linkEP.LinkAddress(), ref, s.handleLocal && !nic.isLoopback(), multicastLoop && !nic.isLoopback()) - if needRoute { - r.NextHop = route.Gateway + r.directedBroadcast = route.Destination.IsBroadcast(remoteAddr) + + if len(route.Gateway) > 0 { + if needRoute { + r.NextHop = route.Gateway + } + } else if r.directedBroadcast { + r.RemoteLinkAddress = header.EthernetBroadcastAddress } + return r, nil } } @@ -1408,6 +1464,12 @@ func (s *Stack) RegisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.N return s.demux.registerEndpoint(netProtos, protocol, id, ep, flags, bindToDevice) } +// CheckRegisterTransportEndpoint checks if an endpoint can be registered with +// the stack transport dispatcher. +func (s *Stack) CheckRegisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, flags ports.Flags, bindToDevice tcpip.NICID) *tcpip.Error { + return s.demux.checkEndpoint(netProtos, protocol, id, flags, bindToDevice) +} + // UnregisterTransportEndpoint removes the endpoint with the given id from the // stack transport dispatcher. func (s *Stack) UnregisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, flags ports.Flags, bindToDevice tcpip.NICID) { @@ -1825,10 +1887,38 @@ func (s *Stack) SetNDPConfigurations(id tcpip.NICID, c NDPConfigurations) *tcpip } nic.setNDPConfigs(c) - return nil } +// NUDConfigurations gets the per-interface NUD configurations. +func (s *Stack) NUDConfigurations(id tcpip.NICID) (NUDConfigurations, *tcpip.Error) { + s.mu.RLock() + nic, ok := s.nics[id] + s.mu.RUnlock() + + if !ok { + return NUDConfigurations{}, tcpip.ErrUnknownNICID + } + + return nic.NUDConfigs() +} + +// SetNUDConfigurations sets the per-interface NUD configurations. +// +// Note, if c contains invalid NUD configuration values, it will be fixed to +// use default values for the erroneous values. +func (s *Stack) SetNUDConfigurations(id tcpip.NICID, c NUDConfigurations) *tcpip.Error { + s.mu.RLock() + nic, ok := s.nics[id] + s.mu.RUnlock() + + if !ok { + return tcpip.ErrUnknownNICID + } + + return nic.setNUDConfigs(c) +} + // HandleNDPRA provides a NIC with ID id a validated NDP Router Advertisement // message that it needs to handle. func (s *Stack) HandleNDPRA(id tcpip.NICID, ip tcpip.Address, ra header.NDPRouterAdvert) *tcpip.Error { diff --git a/pkg/tcpip/stack/stack_options.go b/pkg/tcpip/stack/stack_options.go new file mode 100644 index 000000000..0b093e6c5 --- /dev/null +++ b/pkg/tcpip/stack/stack_options.go @@ -0,0 +1,106 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package stack + +import "gvisor.dev/gvisor/pkg/tcpip" + +const ( + // MinBufferSize is the smallest size of a receive or send buffer. + MinBufferSize = 4 << 10 // 4 KiB + + // DefaultBufferSize is the default size of the send/recv buffer for a + // transport endpoint. + DefaultBufferSize = 212 << 10 // 212 KiB + + // DefaultMaxBufferSize is the default maximum permitted size of a + // send/receive buffer. + DefaultMaxBufferSize = 4 << 20 // 4 MiB +) + +// SendBufferSizeOption is used by stack.(Stack*).Option/SetOption to +// get/set the default, min and max send buffer sizes. +type SendBufferSizeOption struct { + Min int + Default int + Max int +} + +// ReceiveBufferSizeOption is used by stack.(Stack*).Option/SetOption to +// get/set the default, min and max receive buffer sizes. +type ReceiveBufferSizeOption struct { + Min int + Default int + Max int +} + +// SetOption allows setting stack wide options. +func (s *Stack) SetOption(option interface{}) *tcpip.Error { + switch v := option.(type) { + case SendBufferSizeOption: + // Make sure we don't allow lowering the buffer below minimum + // required for stack to work. + if v.Min < MinBufferSize { + return tcpip.ErrInvalidOptionValue + } + + if v.Default < v.Min || v.Default > v.Max { + return tcpip.ErrInvalidOptionValue + } + + s.mu.Lock() + s.sendBufferSize = v + s.mu.Unlock() + return nil + + case ReceiveBufferSizeOption: + // Make sure we don't allow lowering the buffer below minimum + // required for stack to work. + if v.Min < MinBufferSize { + return tcpip.ErrInvalidOptionValue + } + + if v.Default < v.Min || v.Default > v.Max { + return tcpip.ErrInvalidOptionValue + } + + s.mu.Lock() + s.receiveBufferSize = v + s.mu.Unlock() + return nil + + default: + return tcpip.ErrUnknownProtocolOption + } +} + +// Option allows retrieving stack wide options. +func (s *Stack) Option(option interface{}) *tcpip.Error { + switch v := option.(type) { + case *SendBufferSizeOption: + s.mu.RLock() + *v = s.sendBufferSize + s.mu.RUnlock() + return nil + + case *ReceiveBufferSizeOption: + s.mu.RLock() + *v = s.receiveBufferSize + s.mu.RUnlock() + return nil + + default: + return tcpip.ErrUnknownProtocolOption + } +} diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index 5aacbf53e..f22062889 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -27,6 +27,7 @@ import ( "time" "github.com/google/go-cmp/cmp" + "github.com/google/go-cmp/cmp/cmpopts" "gvisor.dev/gvisor/pkg/rand" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" @@ -867,9 +868,9 @@ func TestRouteWithDownNIC(t *testing.T) { // Writes with Routes that use NIC1 after being brought up should // succeed. // - // TODO(b/147015577): Should we instead completely invalidate all - // Routes that were bound to a NIC that was brought down at some - // point? + // TODO(gvisor.dev/issue/1491): Should we instead completely + // invalidate all Routes that were bound to a NIC that was brought + // down at some point? if err := upFn(s, nicID1); err != nil { t.Fatalf("test.upFn(_, %d): %s", nicID1, err) } @@ -3338,3 +3339,305 @@ func TestDoDADWhenNICEnabled(t *testing.T) { t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (%s, nil), want = (%s, nil)", nicID, header.IPv6ProtocolNumber, got, addr.AddressWithPrefix) } } + +func TestStackReceiveBufferSizeOption(t *testing.T) { + const sMin = stack.MinBufferSize + testCases := []struct { + name string + rs stack.ReceiveBufferSizeOption + err *tcpip.Error + }{ + // Invalid configurations. + {"min_below_zero", stack.ReceiveBufferSizeOption{Min: -1, Default: sMin, Max: sMin}, tcpip.ErrInvalidOptionValue}, + {"min_zero", stack.ReceiveBufferSizeOption{Min: 0, Default: sMin, Max: sMin}, tcpip.ErrInvalidOptionValue}, + {"default_below_min", stack.ReceiveBufferSizeOption{Min: sMin, Default: sMin - 1, Max: sMin - 1}, tcpip.ErrInvalidOptionValue}, + {"default_above_max", stack.ReceiveBufferSizeOption{Min: sMin, Default: sMin + 1, Max: sMin}, tcpip.ErrInvalidOptionValue}, + {"max_below_min", stack.ReceiveBufferSizeOption{Min: sMin, Default: sMin + 1, Max: sMin - 1}, tcpip.ErrInvalidOptionValue}, + + // Valid Configurations + {"in_ascending_order", stack.ReceiveBufferSizeOption{Min: sMin, Default: sMin + 1, Max: sMin + 2}, nil}, + {"all_equal", stack.ReceiveBufferSizeOption{Min: sMin, Default: sMin, Max: sMin}, nil}, + {"min_default_equal", stack.ReceiveBufferSizeOption{Min: sMin, Default: sMin, Max: sMin + 1}, nil}, + {"default_max_equal", stack.ReceiveBufferSizeOption{Min: sMin, Default: sMin + 1, Max: sMin + 1}, nil}, + } + for _, tc := range testCases { + t.Run(tc.name, func(t *testing.T) { + s := stack.New(stack.Options{}) + defer s.Close() + if err := s.SetOption(tc.rs); err != tc.err { + t.Fatalf("s.SetOption(%#v) = %v, want: %v", tc.rs, err, tc.err) + } + var rs stack.ReceiveBufferSizeOption + if tc.err == nil { + if err := s.Option(&rs); err != nil { + t.Fatalf("s.Option(%#v) = %v, want: nil", rs, err) + } + if got, want := rs, tc.rs; got != want { + t.Fatalf("s.Option(..) returned unexpected value got: %#v, want: %#v", got, want) + } + } + }) + } +} + +func TestStackSendBufferSizeOption(t *testing.T) { + const sMin = stack.MinBufferSize + testCases := []struct { + name string + ss stack.SendBufferSizeOption + err *tcpip.Error + }{ + // Invalid configurations. + {"min_below_zero", stack.SendBufferSizeOption{Min: -1, Default: sMin, Max: sMin}, tcpip.ErrInvalidOptionValue}, + {"min_zero", stack.SendBufferSizeOption{Min: 0, Default: sMin, Max: sMin}, tcpip.ErrInvalidOptionValue}, + {"default_below_min", stack.SendBufferSizeOption{Min: 0, Default: sMin - 1, Max: sMin - 1}, tcpip.ErrInvalidOptionValue}, + {"default_above_max", stack.SendBufferSizeOption{Min: 0, Default: sMin + 1, Max: sMin}, tcpip.ErrInvalidOptionValue}, + {"max_below_min", stack.SendBufferSizeOption{Min: sMin, Default: sMin + 1, Max: sMin - 1}, tcpip.ErrInvalidOptionValue}, + + // Valid Configurations + {"in_ascending_order", stack.SendBufferSizeOption{Min: sMin, Default: sMin + 1, Max: sMin + 2}, nil}, + {"all_equal", stack.SendBufferSizeOption{Min: sMin, Default: sMin, Max: sMin}, nil}, + {"min_default_equal", stack.SendBufferSizeOption{Min: sMin, Default: sMin, Max: sMin + 1}, nil}, + {"default_max_equal", stack.SendBufferSizeOption{Min: sMin, Default: sMin + 1, Max: sMin + 1}, nil}, + } + for _, tc := range testCases { + t.Run(tc.name, func(t *testing.T) { + s := stack.New(stack.Options{}) + defer s.Close() + if err := s.SetOption(tc.ss); err != tc.err { + t.Fatalf("s.SetOption(%+v) = %v, want: %v", tc.ss, err, tc.err) + } + var ss stack.SendBufferSizeOption + if tc.err == nil { + if err := s.Option(&ss); err != nil { + t.Fatalf("s.Option(%+v) = %v, want: nil", ss, err) + } + if got, want := ss, tc.ss; got != want { + t.Fatalf("s.Option(..) returned unexpected value got: %#v, want: %#v", got, want) + } + } + }) + } +} + +func TestOutgoingSubnetBroadcast(t *testing.T) { + const ( + unspecifiedNICID = 0 + nicID1 = 1 + ) + + defaultAddr := tcpip.AddressWithPrefix{ + Address: header.IPv4Any, + PrefixLen: 0, + } + defaultSubnet := defaultAddr.Subnet() + ipv4Addr := tcpip.AddressWithPrefix{ + Address: "\xc0\xa8\x01\x3a", + PrefixLen: 24, + } + ipv4Subnet := ipv4Addr.Subnet() + ipv4SubnetBcast := ipv4Subnet.Broadcast() + ipv4Gateway := tcpip.Address("\xc0\xa8\x01\x01") + ipv4AddrPrefix31 := tcpip.AddressWithPrefix{ + Address: "\xc0\xa8\x01\x3a", + PrefixLen: 31, + } + ipv4Subnet31 := ipv4AddrPrefix31.Subnet() + ipv4Subnet31Bcast := ipv4Subnet31.Broadcast() + ipv4AddrPrefix32 := tcpip.AddressWithPrefix{ + Address: "\xc0\xa8\x01\x3a", + PrefixLen: 32, + } + ipv4Subnet32 := ipv4AddrPrefix32.Subnet() + ipv4Subnet32Bcast := ipv4Subnet32.Broadcast() + ipv6Addr := tcpip.AddressWithPrefix{ + Address: "\x20\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01", + PrefixLen: 64, + } + ipv6Subnet := ipv6Addr.Subnet() + ipv6SubnetBcast := ipv6Subnet.Broadcast() + remNetAddr := tcpip.AddressWithPrefix{ + Address: "\x64\x0a\x7b\x18", + PrefixLen: 24, + } + remNetSubnet := remNetAddr.Subnet() + remNetSubnetBcast := remNetSubnet.Broadcast() + + tests := []struct { + name string + nicAddr tcpip.ProtocolAddress + routes []tcpip.Route + remoteAddr tcpip.Address + expectedRoute stack.Route + }{ + // Broadcast to a locally attached subnet populates the broadcast MAC. + { + name: "IPv4 Broadcast to local subnet", + nicAddr: tcpip.ProtocolAddress{ + Protocol: header.IPv4ProtocolNumber, + AddressWithPrefix: ipv4Addr, + }, + routes: []tcpip.Route{ + { + Destination: ipv4Subnet, + NIC: nicID1, + }, + }, + remoteAddr: ipv4SubnetBcast, + expectedRoute: stack.Route{ + LocalAddress: ipv4Addr.Address, + RemoteAddress: ipv4SubnetBcast, + RemoteLinkAddress: header.EthernetBroadcastAddress, + NetProto: header.IPv4ProtocolNumber, + Loop: stack.PacketOut, + }, + }, + // Broadcast to a locally attached /31 subnet does not populate the + // broadcast MAC. + { + name: "IPv4 Broadcast to local /31 subnet", + nicAddr: tcpip.ProtocolAddress{ + Protocol: header.IPv4ProtocolNumber, + AddressWithPrefix: ipv4AddrPrefix31, + }, + routes: []tcpip.Route{ + { + Destination: ipv4Subnet31, + NIC: nicID1, + }, + }, + remoteAddr: ipv4Subnet31Bcast, + expectedRoute: stack.Route{ + LocalAddress: ipv4AddrPrefix31.Address, + RemoteAddress: ipv4Subnet31Bcast, + NetProto: header.IPv4ProtocolNumber, + Loop: stack.PacketOut, + }, + }, + // Broadcast to a locally attached /32 subnet does not populate the + // broadcast MAC. + { + name: "IPv4 Broadcast to local /32 subnet", + nicAddr: tcpip.ProtocolAddress{ + Protocol: header.IPv4ProtocolNumber, + AddressWithPrefix: ipv4AddrPrefix32, + }, + routes: []tcpip.Route{ + { + Destination: ipv4Subnet32, + NIC: nicID1, + }, + }, + remoteAddr: ipv4Subnet32Bcast, + expectedRoute: stack.Route{ + LocalAddress: ipv4AddrPrefix32.Address, + RemoteAddress: ipv4Subnet32Bcast, + NetProto: header.IPv4ProtocolNumber, + Loop: stack.PacketOut, + }, + }, + // IPv6 has no notion of a broadcast. + { + name: "IPv6 'Broadcast' to local subnet", + nicAddr: tcpip.ProtocolAddress{ + Protocol: header.IPv6ProtocolNumber, + AddressWithPrefix: ipv6Addr, + }, + routes: []tcpip.Route{ + { + Destination: ipv6Subnet, + NIC: nicID1, + }, + }, + remoteAddr: ipv6SubnetBcast, + expectedRoute: stack.Route{ + LocalAddress: ipv6Addr.Address, + RemoteAddress: ipv6SubnetBcast, + NetProto: header.IPv6ProtocolNumber, + Loop: stack.PacketOut, + }, + }, + // Broadcast to a remote subnet in the route table is send to the next-hop + // gateway. + { + name: "IPv4 Broadcast to remote subnet", + nicAddr: tcpip.ProtocolAddress{ + Protocol: header.IPv4ProtocolNumber, + AddressWithPrefix: ipv4Addr, + }, + routes: []tcpip.Route{ + { + Destination: remNetSubnet, + Gateway: ipv4Gateway, + NIC: nicID1, + }, + }, + remoteAddr: remNetSubnetBcast, + expectedRoute: stack.Route{ + LocalAddress: ipv4Addr.Address, + RemoteAddress: remNetSubnetBcast, + NextHop: ipv4Gateway, + NetProto: header.IPv4ProtocolNumber, + Loop: stack.PacketOut, + }, + }, + // Broadcast to an unknown subnet follows the default route. Note that this + // is essentially just routing an unknown destination IP, because w/o any + // subnet prefix information a subnet broadcast address is just a normal IP. + { + name: "IPv4 Broadcast to unknown subnet", + nicAddr: tcpip.ProtocolAddress{ + Protocol: header.IPv4ProtocolNumber, + AddressWithPrefix: ipv4Addr, + }, + routes: []tcpip.Route{ + { + Destination: defaultSubnet, + Gateway: ipv4Gateway, + NIC: nicID1, + }, + }, + remoteAddr: remNetSubnetBcast, + expectedRoute: stack.Route{ + LocalAddress: ipv4Addr.Address, + RemoteAddress: remNetSubnetBcast, + NextHop: ipv4Gateway, + NetProto: header.IPv4ProtocolNumber, + Loop: stack.PacketOut, + }, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, + }) + ep := channel.New(0, defaultMTU, "") + if err := s.CreateNIC(nicID1, ep); err != nil { + t.Fatalf("CreateNIC(%d, _): %s", nicID1, err) + } + if err := s.AddProtocolAddress(nicID1, test.nicAddr); err != nil { + t.Fatalf("AddProtocolAddress(%d, %+v): %s", nicID1, test.nicAddr, err) + } + + s.SetRouteTable(test.routes) + + var netProto tcpip.NetworkProtocolNumber + switch l := len(test.remoteAddr); l { + case header.IPv4AddressSize: + netProto = header.IPv4ProtocolNumber + case header.IPv6AddressSize: + netProto = header.IPv6ProtocolNumber + default: + t.Fatalf("got unexpected address length = %d bytes", l) + } + + if r, err := s.FindRoute(unspecifiedNICID, "" /* localAddr */, test.remoteAddr, netProto, false /* multicastLoop */); err != nil { + t.Fatalf("FindRoute(%d, '', %s, %d): %s", unspecifiedNICID, test.remoteAddr, netProto, err) + } else if diff := cmp.Diff(r, test.expectedRoute, cmpopts.IgnoreUnexported(r)); diff != "" { + t.Errorf("route mismatch (-want +got):\n%s", diff) + } + }) + } +} diff --git a/pkg/tcpip/stack/transport_demuxer.go b/pkg/tcpip/stack/transport_demuxer.go index 118b449d5..b902c6ca9 100644 --- a/pkg/tcpip/stack/transport_demuxer.go +++ b/pkg/tcpip/stack/transport_demuxer.go @@ -221,6 +221,18 @@ func (epsByNIC *endpointsByNIC) registerEndpoint(d *transportDemuxer, netProto t return multiPortEp.singleRegisterEndpoint(t, flags) } +func (epsByNIC *endpointsByNIC) checkEndpoint(d *transportDemuxer, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, flags ports.Flags, bindToDevice tcpip.NICID) *tcpip.Error { + epsByNIC.mu.RLock() + defer epsByNIC.mu.RUnlock() + + multiPortEp, ok := epsByNIC.endpoints[bindToDevice] + if !ok { + return nil + } + + return multiPortEp.singleCheckEndpoint(flags) +} + // unregisterEndpoint returns true if endpointsByNIC has to be unregistered. func (epsByNIC *endpointsByNIC) unregisterEndpoint(bindToDevice tcpip.NICID, t TransportEndpoint, flags ports.Flags) bool { epsByNIC.mu.Lock() @@ -289,6 +301,17 @@ func (d *transportDemuxer) registerEndpoint(netProtos []tcpip.NetworkProtocolNum return nil } +// checkEndpoint checks if an endpoint can be registered with the dispatcher. +func (d *transportDemuxer) checkEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, flags ports.Flags, bindToDevice tcpip.NICID) *tcpip.Error { + for _, n := range netProtos { + if err := d.singleCheckEndpoint(n, protocol, id, flags, bindToDevice); err != nil { + return err + } + } + + return nil +} + // multiPortEndpoint is a container for TransportEndpoints which are bound to // the same pair of address and port. endpointsArr always has at least one // element. @@ -380,7 +403,7 @@ func (ep *multiPortEndpoint) singleRegisterEndpoint(t TransportEndpoint, flags p ep.mu.Lock() defer ep.mu.Unlock() - bits := flags.Bits() + bits := flags.Bits() & ports.MultiBindFlagMask if len(ep.endpoints) != 0 { // If it was previously bound, we need to check if we can bind again. @@ -395,6 +418,22 @@ func (ep *multiPortEndpoint) singleRegisterEndpoint(t TransportEndpoint, flags p return nil } +func (ep *multiPortEndpoint) singleCheckEndpoint(flags ports.Flags) *tcpip.Error { + ep.mu.RLock() + defer ep.mu.RUnlock() + + bits := flags.Bits() & ports.MultiBindFlagMask + + if len(ep.endpoints) != 0 { + // If it was previously bound, we need to check if we can bind again. + if ep.flags.TotalRefs() > 0 && bits&ep.flags.IntersectionRefs() == 0 { + return tcpip.ErrPortInUse + } + } + + return nil +} + // unregisterEndpoint returns true if multiPortEndpoint has to be unregistered. func (ep *multiPortEndpoint) unregisterEndpoint(t TransportEndpoint, flags ports.Flags) bool { ep.mu.Lock() @@ -406,7 +445,7 @@ func (ep *multiPortEndpoint) unregisterEndpoint(t TransportEndpoint, flags ports ep.endpoints[len(ep.endpoints)-1] = nil ep.endpoints = ep.endpoints[:len(ep.endpoints)-1] - ep.flags.DropRef(flags.Bits()) + ep.flags.DropRef(flags.Bits() & ports.MultiBindFlagMask) break } } @@ -439,6 +478,28 @@ func (d *transportDemuxer) singleRegisterEndpoint(netProto tcpip.NetworkProtocol return epsByNIC.registerEndpoint(d, netProto, protocol, ep, flags, bindToDevice) } +func (d *transportDemuxer) singleCheckEndpoint(netProto tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, flags ports.Flags, bindToDevice tcpip.NICID) *tcpip.Error { + if id.RemotePort != 0 { + // SO_REUSEPORT only applies to bound/listening endpoints. + flags.LoadBalanced = false + } + + eps, ok := d.protocol[protocolIDs{netProto, protocol}] + if !ok { + return tcpip.ErrUnknownProtocol + } + + eps.mu.RLock() + defer eps.mu.RUnlock() + + epsByNIC, ok := eps.endpoints[id] + if !ok { + return nil + } + + return epsByNIC.checkEndpoint(d, netProto, protocol, flags, bindToDevice) +} + // unregisterEndpoint unregisters the endpoint with the given id such that it // won't receive any more packets. func (d *transportDemuxer) unregisterEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, flags ports.Flags, bindToDevice tcpip.NICID) { diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 3ad130b23..a634b9b60 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -43,6 +43,9 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// Using header.IPv4AddressSize would cause an import cycle. +const ipv4AddressSize = 4 + // Error represents an error in the netstack error space. Using a special type // ensures that errors outside of this space are not accidentally introduced. // @@ -192,7 +195,7 @@ func (e ErrSaveRejection) Error() string { return "save rejected due to unsupported networking state: " + e.Err.Error() } -// A Clock provides the current time. +// A Clock provides the current time and schedules work for execution. // // Times returned by a Clock should always be used for application-visible // time. Only monotonic times should be used for netstack internal timekeeping. @@ -203,6 +206,31 @@ type Clock interface { // NowMonotonic returns a monotonic time value. NowMonotonic() int64 + + // AfterFunc waits for the duration to elapse and then calls f in its own + // goroutine. It returns a Timer that can be used to cancel the call using + // its Stop method. + AfterFunc(d time.Duration, f func()) Timer +} + +// Timer represents a single event. A Timer must be created with +// Clock.AfterFunc. +type Timer interface { + // Stop prevents the Timer from firing. It returns true if the call stops the + // timer, false if the timer has already expired or been stopped. + // + // If Stop returns false, then the timer has already expired and the function + // f of Clock.AfterFunc(d, f) has been started in its own goroutine; Stop + // does not wait for f to complete before returning. If the caller needs to + // know whether f is completed, it must coordinate with f explicitly. + Stop() bool + + // Reset changes the timer to expire after duration d. + // + // Reset should be invoked only on stopped or expired timers. If the timer is + // known to have expired, Reset can be used directly. Otherwise, the caller + // must coordinate with the function f of Clock.AfterFunc(d, f). + Reset(d time.Duration) } // Address is a byte slice cast as a string that represents the address of a @@ -295,6 +323,29 @@ func (s *Subnet) Broadcast() Address { return Address(addr) } +// IsBroadcast returns true if the address is considered a broadcast address. +func (s *Subnet) IsBroadcast(address Address) bool { + // Only IPv4 supports the notion of a broadcast address. + if len(address) != ipv4AddressSize { + return false + } + + // Normally, we would just compare address with the subnet's broadcast + // address but there is an exception where a simple comparison is not + // correct. This exception is for /31 and /32 IPv4 subnets where all + // addresses are considered valid host addresses. + // + // For /31 subnets, the case is easy. RFC 3021 Section 2.1 states that + // both addresses in a /31 subnet "MUST be interpreted as host addresses." + // + // For /32, the case is a bit more vague. RFC 3021 makes no mention of /32 + // subnets. However, the same reasoning applies - if an exception is not + // made, then there do not exist any host addresses in a /32 subnet. RFC + // 4632 Section 3.1 also vaguely implies this interpretation by referring + // to addresses in /32 subnets as "host routes." + return s.Prefix() <= 30 && s.Broadcast() == address +} + // Equal returns true if s equals o. // // Needed to use cmp.Equal on Subnet as its fields are unexported. @@ -316,6 +367,28 @@ const ( ShutdownWrite ) +// PacketType is used to indicate the destination of the packet. +type PacketType uint8 + +const ( + // PacketHost indicates a packet addressed to the local host. + PacketHost PacketType = iota + + // PacketOtherHost indicates an outgoing packet addressed to + // another host caught by a NIC in promiscuous mode. + PacketOtherHost + + // PacketOutgoing for a packet originating from the local host + // that is looped back to a packet socket. + PacketOutgoing + + // PacketBroadcast indicates a link layer broadcast packet. + PacketBroadcast + + // PacketMulticast indicates a link layer multicast packet. + PacketMulticast +) + // FullAddress represents a full transport node address, as required by the // Connect() and Bind() methods. // @@ -549,6 +622,28 @@ type Endpoint interface { SetOwner(owner PacketOwner) } +// LinkPacketInfo holds Link layer information for a received packet. +// +// +stateify savable +type LinkPacketInfo struct { + // Protocol is the NetworkProtocolNumber for the packet. + Protocol NetworkProtocolNumber + + // PktType is used to indicate the destination of the packet. + PktType PacketType +} + +// PacketEndpoint are additional methods that are only implemented by Packet +// endpoints. +type PacketEndpoint interface { + // ReadPacket reads a datagram/packet from the endpoint and optionally + // returns the sender and additional LinkPacketInfo. + // + // This method does not block if there is no data pending. It will also + // either return an error or data, never both. + ReadPacket(*FullAddress, *LinkPacketInfo) (buffer.View, ControlMessages, *Error) +} + // EndpointInfo is the interface implemented by each endpoint info struct. type EndpointInfo interface { // IsEndpointInfo is an empty method to implement the tcpip.EndpointInfo @@ -585,85 +680,108 @@ type WriteOptions struct { type SockOptBool int const ( - // BroadcastOption is used by SetSockOpt/GetSockOpt to specify whether - // datagram sockets are allowed to send packets to a broadcast address. + // BroadcastOption is used by SetSockOptBool/GetSockOptBool to specify + // whether datagram sockets are allowed to send packets to a broadcast + // address. BroadcastOption SockOptBool = iota - // CorkOption is used by SetSockOpt/GetSockOpt to specify if data should be - // held until segments are full by the TCP transport protocol. + // CorkOption is used by SetSockOptBool/GetSockOptBool to specify if + // data should be held until segments are full by the TCP transport + // protocol. CorkOption - // DelayOption is used by SetSockOpt/GetSockOpt to specify if data - // should be sent out immediately by the transport protocol. For TCP, - // it determines if the Nagle algorithm is on or off. + // DelayOption is used by SetSockOptBool/GetSockOptBool to specify if + // data should be sent out immediately by the transport protocol. For + // TCP, it determines if the Nagle algorithm is on or off. DelayOption - // KeepaliveEnabledOption is used by SetSockOpt/GetSockOpt to specify whether - // TCP keepalive is enabled for this socket. + // KeepaliveEnabledOption is used by SetSockOptBool/GetSockOptBool to + // specify whether TCP keepalive is enabled for this socket. KeepaliveEnabledOption - // MulticastLoopOption is used by SetSockOpt/GetSockOpt to specify whether - // multicast packets sent over a non-loopback interface will be looped back. + // MulticastLoopOption is used by SetSockOptBool/GetSockOptBool to + // specify whether multicast packets sent over a non-loopback interface + // will be looped back. MulticastLoopOption - // PasscredOption is used by SetSockOpt/GetSockOpt to specify whether - // SCM_CREDENTIALS socket control messages are enabled. + // NoChecksumOption is used by SetSockOptBool/GetSockOptBool to specify + // whether UDP checksum is disabled for this socket. + NoChecksumOption + + // PasscredOption is used by SetSockOptBool/GetSockOptBool to specify + // whether SCM_CREDENTIALS socket control messages are enabled. // // Only supported on Unix sockets. PasscredOption - // QuickAckOption is stubbed out in SetSockOpt/GetSockOpt. + // QuickAckOption is stubbed out in SetSockOptBool/GetSockOptBool. QuickAckOption - // ReceiveTClassOption is used by SetSockOpt/GetSockOpt to specify if the - // IPV6_TCLASS ancillary message is passed with incoming packets. + // ReceiveTClassOption is used by SetSockOptBool/GetSockOptBool to + // specify if the IPV6_TCLASS ancillary message is passed with incoming + // packets. ReceiveTClassOption - // ReceiveTOSOption is used by SetSockOpt/GetSockOpt to specify if the TOS - // ancillary message is passed with incoming packets. + // ReceiveTOSOption is used by SetSockOptBool/GetSockOptBool to specify + // if the TOS ancillary message is passed with incoming packets. ReceiveTOSOption - // ReceiveIPPacketInfoOption is used by {G,S}etSockOptBool to specify - // if more inforamtion is provided with incoming packets such - // as interface index and address. + // ReceiveIPPacketInfoOption is used by SetSockOptBool/GetSockOptBool to + // specify if more inforamtion is provided with incoming packets such as + // interface index and address. ReceiveIPPacketInfoOption - // ReuseAddressOption is used by SetSockOpt/GetSockOpt to specify whether Bind() - // should allow reuse of local address. + // ReuseAddressOption is used by SetSockOptBool/GetSockOptBool to + // specify whether Bind() should allow reuse of local address. ReuseAddressOption - // ReusePortOption is used by SetSockOpt/GetSockOpt to permit multiple sockets - // to be bound to an identical socket address. + // ReusePortOption is used by SetSockOptBool/GetSockOptBool to permit + // multiple sockets to be bound to an identical socket address. ReusePortOption - // V6OnlyOption is used by {G,S}etSockOptBool to specify whether an IPv6 - // socket is to be restricted to sending and receiving IPv6 packets only. + // V6OnlyOption is used by SetSockOptBool/GetSockOptBool to specify + // whether an IPv6 socket is to be restricted to sending and receiving + // IPv6 packets only. V6OnlyOption + + // IPHdrIncludedOption is used by SetSockOpt to indicate for a raw + // endpoint that all packets being written have an IP header and the + // endpoint should not attach an IP header. + IPHdrIncludedOption ) // SockOptInt represents socket options which values have the int type. type SockOptInt int const ( - // KeepaliveCountOption is used by SetSockOpt/GetSockOpt to specify the number - // of un-ACKed TCP keepalives that will be sent before the connection is - // closed. + // KeepaliveCountOption is used by SetSockOptInt/GetSockOptInt to + // specify the number of un-ACKed TCP keepalives that will be sent + // before the connection is closed. KeepaliveCountOption SockOptInt = iota - // IPv4TOSOption is used by SetSockOpt/GetSockOpt to specify TOS + // IPv4TOSOption is used by SetSockOptInt/GetSockOptInt to specify TOS // for all subsequent outgoing IPv4 packets from the endpoint. IPv4TOSOption - // IPv6TrafficClassOption is used by SetSockOpt/GetSockOpt to specify TOS - // for all subsequent outgoing IPv6 packets from the endpoint. + // IPv6TrafficClassOption is used by SetSockOptInt/GetSockOptInt to + // specify TOS for all subsequent outgoing IPv6 packets from the + // endpoint. IPv6TrafficClassOption - // MaxSegOption is used by SetSockOpt/GetSockOpt to set/get the current - // Maximum Segment Size(MSS) value as specified using the TCP_MAXSEG option. + // MaxSegOption is used by SetSockOptInt/GetSockOptInt to set/get the + // current Maximum Segment Size(MSS) value as specified using the + // TCP_MAXSEG option. MaxSegOption - // MulticastTTLOption is used by SetSockOpt/GetSockOpt to control the default - // TTL value for multicast messages. The default is 1. + // MTUDiscoverOption is used to set/get the path MTU discovery setting. + // + // NOTE: Setting this option to any other value than PMTUDiscoveryDont + // is not supported and will fail as such, and getting this option will + // always return PMTUDiscoveryDont. + MTUDiscoverOption + + // MulticastTTLOption is used by SetSockOptInt/GetSockOptInt to control + // the default TTL value for multicast messages. The default is 1. MulticastTTLOption // ReceiveQueueSizeOption is used in GetSockOptInt to specify that the @@ -682,26 +800,45 @@ const ( // number of unread bytes in the output buffer should be returned. SendQueueSizeOption - // TTLOption is used by SetSockOpt/GetSockOpt to control the default TTL/hop - // limit value for unicast messages. The default is protocol specific. + // TTLOption is used by SetSockOptInt/GetSockOptInt to control the + // default TTL/hop limit value for unicast messages. The default is + // protocol specific. // // A zero value indicates the default. TTLOption - // TCPSynCountOption is used by SetSockOpt/GetSockOpt to specify the number of - // SYN retransmits that TCP should send before aborting the attempt to - // connect. It cannot exceed 255. + // TCPSynCountOption is used by SetSockOptInt/GetSockOptInt to specify + // the number of SYN retransmits that TCP should send before aborting + // the attempt to connect. It cannot exceed 255. // // NOTE: This option is currently only stubbed out and is no-op. TCPSynCountOption - // TCPWindowClampOption is used by SetSockOpt/GetSockOpt to bound the size - // of the advertised window to this value. + // TCPWindowClampOption is used by SetSockOptInt/GetSockOptInt to bound + // the size of the advertised window to this value. // // NOTE: This option is currently only stubed out and is a no-op TCPWindowClampOption ) +const ( + // PMTUDiscoveryWant is a setting of the MTUDiscoverOption to use + // per-route settings. + PMTUDiscoveryWant int = iota + + // PMTUDiscoveryDont is a setting of the MTUDiscoverOption to disable + // path MTU discovery. + PMTUDiscoveryDont + + // PMTUDiscoveryDo is a setting of the MTUDiscoverOption to always do + // path MTU discovery. + PMTUDiscoveryDo + + // PMTUDiscoveryProbe is a setting of the MTUDiscoverOption to set DF + // but ignore path MTU. + PMTUDiscoveryProbe +) + // ErrorOption is used in GetSockOpt to specify that the last error reported by // the endpoint should be cleared and returned. type ErrorOption struct{} @@ -740,7 +877,7 @@ type CongestionControlOption string // control algorithms. type AvailableCongestionControlOption string -// buffer moderation. +// ModerateReceiveBufferOption is used by buffer moderation. type ModerateReceiveBufferOption bool // TCPLingerTimeoutOption is used by SetSockOpt/GetSockOpt to set/get the @@ -813,33 +950,11 @@ type OutOfBandInlineOption int // a default TTL. type DefaultTTLOption uint8 -// StackSACKEnabled is used by stack.(*Stack).TransportProtocolOption to -// enable/disable SACK support in TCP. See: https://tools.ietf.org/html/rfc2018. -type StackSACKEnabled bool - -// StackDelayEnabled is used by stack.(Stack*).TransportProtocolOption to -// enable/disable Nagle's algorithm in TCP. -type StackDelayEnabled bool +// SocketDetachFilterOption is used by SetSockOpt to detach a previously attached +// classic BPF filter on a given endpoint. +type SocketDetachFilterOption int -// StackSendBufferSizeOption is used by stack.(Stack*).TransportProtocolOption -// to get/set the default, min and max send buffer sizes. -type StackSendBufferSizeOption struct { - Min int - Default int - Max int -} - -// StackReceiveBufferSizeOption is used by -// stack.(Stack*).TransportProtocolOption to get/set the default, min and max -// receive buffer sizes. -type StackReceiveBufferSizeOption struct { - Min int - Default int - Max int -} - -// -// IPPacketInfo is the message struture for IP_PKTINFO. +// IPPacketInfo is the message structure for IP_PKTINFO. // // +stateify savable type IPPacketInfo struct { @@ -1224,6 +1339,12 @@ type UDPStats struct { // PacketSendErrors is the number of datagrams failed to be sent. PacketSendErrors *StatCounter + + // ChecksumErrors is the number of datagrams dropped due to bad checksums. + ChecksumErrors *StatCounter + + // InvalidSourceAddress is the number of invalid sourced datagrams dropped. + InvalidSourceAddress *StatCounter } // Stats holds statistics about the networking stack. @@ -1267,6 +1388,9 @@ type ReceiveErrors struct { // ClosedReceiver is the number of received packets dropped because // of receiving endpoint state being closed. ClosedReceiver StatCounter + + // ChecksumErrors is the number of packets dropped due to bad checksums. + ChecksumErrors StatCounter } // SendErrors collects packet send errors within the transport layer for diff --git a/pkg/tcpip/time_unsafe.go b/pkg/tcpip/time_unsafe.go index 7f172f978..f32d58091 100644 --- a/pkg/tcpip/time_unsafe.go +++ b/pkg/tcpip/time_unsafe.go @@ -20,7 +20,7 @@ package tcpip import ( - _ "time" // Used with go:linkname. + "time" // Used with go:linkname. _ "unsafe" // Required for go:linkname. ) @@ -45,3 +45,31 @@ func (*StdClock) NowMonotonic() int64 { _, _, mono := now() return mono } + +// AfterFunc implements Clock.AfterFunc. +func (*StdClock) AfterFunc(d time.Duration, f func()) Timer { + return &stdTimer{ + t: time.AfterFunc(d, f), + } +} + +type stdTimer struct { + t *time.Timer +} + +var _ Timer = (*stdTimer)(nil) + +// Stop implements Timer.Stop. +func (st *stdTimer) Stop() bool { + return st.t.Stop() +} + +// Reset implements Timer.Reset. +func (st *stdTimer) Reset(d time.Duration) { + st.t.Reset(d) +} + +// NewStdTimer returns a Timer implemented with the time package. +func NewStdTimer(t *time.Timer) Timer { + return &stdTimer{t: t} +} diff --git a/pkg/tcpip/timer.go b/pkg/tcpip/timer.go index 59f3b391f..f1dd7c310 100644 --- a/pkg/tcpip/timer.go +++ b/pkg/tcpip/timer.go @@ -15,54 +15,54 @@ package tcpip import ( - "sync" "time" + + "gvisor.dev/gvisor/pkg/sync" ) -// cancellableTimerInstance is a specific instance of CancellableTimer. +// jobInstance is a specific instance of Job. // -// Different instances are created each time CancellableTimer is Reset so each -// timer has its own earlyReturn signal. This is to address a bug when a -// CancellableTimer is stopped and reset in quick succession resulting in a -// timer instance's earlyReturn signal being affected or seen by another timer -// instance. +// Different instances are created each time Job is scheduled so each timer has +// its own earlyReturn signal. This is to address a bug when a Job is stopped +// and reset in quick succession resulting in a timer instance's earlyReturn +// signal being affected or seen by another timer instance. // // Consider the following sceneario where timer instances share a common // earlyReturn signal (T1 creates, stops and resets a Cancellable timer under a // lock L; T2, T3, T4 and T5 are goroutines that handle the first (A), second // (B), third (C), and fourth (D) instance of the timer firing, respectively): // T1: Obtain L -// T1: Create a new CancellableTimer w/ lock L (create instance A) +// T1: Create a new Job w/ lock L (create instance A) // T2: instance A fires, blocked trying to obtain L. // T1: Attempt to stop instance A (set earlyReturn = true) -// T1: Reset timer (create instance B) +// T1: Schedule timer (create instance B) // T3: instance B fires, blocked trying to obtain L. // T1: Attempt to stop instance B (set earlyReturn = true) -// T1: Reset timer (create instance C) +// T1: Schedule timer (create instance C) // T4: instance C fires, blocked trying to obtain L. // T1: Attempt to stop instance C (set earlyReturn = true) -// T1: Reset timer (create instance D) +// T1: Schedule timer (create instance D) // T5: instance D fires, blocked trying to obtain L. // T1: Release L // -// Now that T1 has released L, any of the 4 timer instances can take L and check -// earlyReturn. If the timers simply check earlyReturn and then do nothing -// further, then instance D will never early return even though it was not -// requested to stop. If the timers reset earlyReturn before early returning, -// then all but one of the timers will do work when only one was expected to. -// If CancellableTimer resets earlyReturn when resetting, then all the timers +// Now that T1 has released L, any of the 4 timer instances can take L and +// check earlyReturn. If the timers simply check earlyReturn and then do +// nothing further, then instance D will never early return even though it was +// not requested to stop. If the timers reset earlyReturn before early +// returning, then all but one of the timers will do work when only one was +// expected to. If Job resets earlyReturn when resetting, then all the timers // will fire (again, when only one was expected to). // // To address the above concerns the simplest solution was to give each timer // its own earlyReturn signal. -type cancellableTimerInstance struct { - timer *time.Timer +type jobInstance struct { + timer Timer // Used to inform the timer to early return when it gets stopped while the // lock the timer tries to obtain when fired is held (T1 is a goroutine that // tries to cancel the timer and T2 is the goroutine that handles the timer // firing): - // T1: Obtain the lock, then call StopLocked() + // T1: Obtain the lock, then call Cancel() // T2: timer fires, and gets blocked on obtaining the lock // T1: Releases lock // T2: Obtains lock does unintended work @@ -73,27 +73,33 @@ type cancellableTimerInstance struct { earlyReturn *bool } -// stop stops the timer instance t from firing if it hasn't fired already. If it +// stop stops the job instance j from firing if it hasn't fired already. If it // has fired and is blocked at obtaining the lock, earlyReturn will be set to // true so that it will early return when it obtains the lock. -func (t *cancellableTimerInstance) stop() { - if t.timer != nil { - t.timer.Stop() - *t.earlyReturn = true +func (j *jobInstance) stop() { + if j.timer != nil { + j.timer.Stop() + *j.earlyReturn = true } } -// CancellableTimer is a timer that does some work and can be safely cancelled -// when it fires at the same time some "related work" is being done. +// Job represents some work that can be scheduled for execution. The work can +// be safely cancelled when it fires at the same time some "related work" is +// being done. // // The term "related work" is defined as some work that needs to be done while // holding some lock that the timer must also hold while doing some work. // -// Note, it is not safe to copy a CancellableTimer as its timer instance creates -// a closure over the address of the CancellableTimer. -type CancellableTimer struct { +// Note, it is not safe to copy a Job as its timer instance creates +// a closure over the address of the Job. +type Job struct { + _ sync.NoCopy + + // The clock used to schedule the backing timer + clock Clock + // The active instance of a cancellable timer. - instance cancellableTimerInstance + instance jobInstance // locker is the lock taken by the timer immediately after it fires and must // be held when attempting to stop the timer. @@ -110,75 +116,91 @@ type CancellableTimer struct { fn func() } -// StopLocked prevents the Timer from firing if it has not fired already. +// Cancel prevents the Job from executing if it has not executed already. // -// If the timer is blocked on obtaining the t.locker lock when StopLocked is -// called, it will early return instead of calling t.fn. +// Cancel requires appropriate locking to be in place for any resources managed +// by the Job. If the Job is blocked on obtaining the lock when Cancel is +// called, it will early return. // // Note, t will be modified. // -// t.locker MUST be locked. -func (t *CancellableTimer) StopLocked() { - t.instance.stop() +// j.locker MUST be locked. +func (j *Job) Cancel() { + j.instance.stop() // Nothing to do with the stopped instance anymore. - t.instance = cancellableTimerInstance{} + j.instance = jobInstance{} } -// Reset changes the timer to expire after duration d. +// Schedule schedules the Job for execution after duration d. This can be +// called on cancelled or completed Jobs to schedule them again. // -// Note, t will be modified. +// Schedule should be invoked only on unscheduled, cancelled, or completed +// Jobs. To be safe, callers should always call Cancel before calling Schedule. // -// Reset should only be called on stopped or expired timers. To be safe, callers -// should always call StopLocked before calling Reset. -func (t *CancellableTimer) Reset(d time.Duration) { +// Note, j will be modified. +func (j *Job) Schedule(d time.Duration) { // Create a new instance. earlyReturn := false // Capture the locker so that updating the timer does not cause a data race // when a timer fires and tries to obtain the lock (read the timer's locker). - locker := t.locker - t.instance = cancellableTimerInstance{ - timer: time.AfterFunc(d, func() { + locker := j.locker + j.instance = jobInstance{ + timer: j.clock.AfterFunc(d, func() { locker.Lock() defer locker.Unlock() if earlyReturn { // If we reach this point, it means that the timer fired while another - // goroutine called StopLocked while it had the lock. Simply return - // here and do nothing further. + // goroutine called Cancel while it had the lock. Simply return here + // and do nothing further. earlyReturn = false return } - t.fn() + j.fn() }), earlyReturn: &earlyReturn, } } -// Lock is a no-op used by the copylocks checker from go vet. -// -// See CancellableTimer for details about why it shouldn't be copied. -// -// See https://github.com/golang/go/issues/8005#issuecomment-190753527 for more -// details about the copylocks checker. -func (*CancellableTimer) Lock() {} - -// Unlock is a no-op used by the copylocks checker from go vet. -// -// See CancellableTimer for details about why it shouldn't be copied. -// -// See https://github.com/golang/go/issues/8005#issuecomment-190753527 for more -// details about the copylocks checker. -func (*CancellableTimer) Unlock() {} - -// NewCancellableTimer returns an unscheduled CancellableTimer with the given -// locker and fn. -// -// fn MUST NOT attempt to lock locker. -// -// Callers must call Reset to schedule the timer to fire. -func NewCancellableTimer(locker sync.Locker, fn func()) *CancellableTimer { - return &CancellableTimer{locker: locker, fn: fn} +// NewJob returns a new Job that can be used to schedule f to run in its own +// gorountine. l will be locked before calling f then unlocked after f returns. +// +// var clock tcpip.StdClock +// var mu sync.Mutex +// message := "foo" +// job := tcpip.NewJob(&clock, &mu, func() { +// fmt.Println(message) +// }) +// job.Schedule(time.Second) +// +// mu.Lock() +// message = "bar" +// mu.Unlock() +// +// // Output: bar +// +// f MUST NOT attempt to lock l. +// +// l MUST be locked prior to calling the returned job's Cancel(). +// +// var clock tcpip.StdClock +// var mu sync.Mutex +// message := "foo" +// job := tcpip.NewJob(&clock, &mu, func() { +// fmt.Println(message) +// }) +// job.Schedule(time.Second) +// +// mu.Lock() +// job.Cancel() +// mu.Unlock() +func NewJob(c Clock, l sync.Locker, f func()) *Job { + return &Job{ + clock: c, + locker: l, + fn: f, + } } diff --git a/pkg/tcpip/timer_test.go b/pkg/tcpip/timer_test.go index b4940e397..a82384c49 100644 --- a/pkg/tcpip/timer_test.go +++ b/pkg/tcpip/timer_test.go @@ -28,8 +28,8 @@ const ( longDuration = 1 * time.Second ) -func TestCancellableTimerReassignment(t *testing.T) { - var timer tcpip.CancellableTimer +func TestJobReschedule(t *testing.T) { + var clock tcpip.StdClock var wg sync.WaitGroup var lock sync.Mutex @@ -43,26 +43,27 @@ func TestCancellableTimerReassignment(t *testing.T) { // that has an active timer (even if it has been stopped as a stopped // timer may be blocked on a lock before it can check if it has been // stopped while another goroutine holds the same lock). - timer = *tcpip.NewCancellableTimer(&lock, func() { + job := tcpip.NewJob(&clock, &lock, func() { wg.Done() }) - timer.Reset(shortDuration) + job.Schedule(shortDuration) lock.Unlock() }() } wg.Wait() } -func TestCancellableTimerFire(t *testing.T) { +func TestJobExecution(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) - timer := tcpip.NewCancellableTimer(&lock, func() { + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) + job.Schedule(shortDuration) // Wait for timer to fire. select { @@ -82,17 +83,18 @@ func TestCancellableTimerFire(t *testing.T) { func TestCancellableTimerResetFromLongDuration(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(middleDuration) + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(middleDuration) lock.Lock() - timer.StopLocked() + job.Cancel() lock.Unlock() - timer.Reset(shortDuration) + job.Schedule(shortDuration) // Wait for timer to fire. select { @@ -109,16 +111,17 @@ func TestCancellableTimerResetFromLongDuration(t *testing.T) { } } -func TestCancellableTimerResetFromShortDuration(t *testing.T) { +func TestJobRescheduleFromShortDuration(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) lock.Lock() - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) - timer.StopLocked() + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(shortDuration) + job.Cancel() lock.Unlock() // Wait for timer to fire if it wasn't correctly stopped. @@ -128,7 +131,7 @@ func TestCancellableTimerResetFromShortDuration(t *testing.T) { case <-time.After(middleDuration): } - timer.Reset(shortDuration) + job.Schedule(shortDuration) // Wait for timer to fire. select { @@ -145,17 +148,18 @@ func TestCancellableTimerResetFromShortDuration(t *testing.T) { } } -func TestCancellableTimerImmediatelyStop(t *testing.T) { +func TestJobImmediatelyCancel(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) for i := 0; i < 1000; i++ { lock.Lock() - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) - timer.StopLocked() + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(shortDuration) + job.Cancel() lock.Unlock() } @@ -167,25 +171,26 @@ func TestCancellableTimerImmediatelyStop(t *testing.T) { } } -func TestCancellableTimerStoppedResetWithoutLock(t *testing.T) { +func TestJobCancelledRescheduleWithoutLock(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) lock.Lock() - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) - timer.StopLocked() + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(shortDuration) + job.Cancel() lock.Unlock() for i := 0; i < 10; i++ { - timer.Reset(middleDuration) + job.Schedule(middleDuration) lock.Lock() // Sleep until the timer fires and gets blocked trying to take the lock. time.Sleep(middleDuration * 2) - timer.StopLocked() + job.Cancel() lock.Unlock() } @@ -201,17 +206,18 @@ func TestCancellableTimerStoppedResetWithoutLock(t *testing.T) { func TestManyCancellableTimerResetAfterBlockedOnLock(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) lock.Lock() - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(shortDuration) for i := 0; i < 10; i++ { // Sleep until the timer fires and gets blocked trying to take the lock. time.Sleep(middleDuration) - timer.StopLocked() - timer.Reset(shortDuration) + job.Cancel() + job.Schedule(shortDuration) } lock.Unlock() @@ -230,18 +236,19 @@ func TestManyCancellableTimerResetAfterBlockedOnLock(t *testing.T) { } } -func TestManyCancellableTimerResetUnderLock(t *testing.T) { +func TestManyJobReschedulesUnderLock(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) lock.Lock() - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(shortDuration) for i := 0; i < 10; i++ { - timer.StopLocked() - timer.Reset(shortDuration) + job.Cancel() + job.Schedule(shortDuration) } lock.Unlock() diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index 8ce294002..4612be4e7 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -344,6 +344,10 @@ func (e *endpoint) Peek([][]byte) (int64, tcpip.ControlMessages, *tcpip.Error) { // SetSockOpt sets a socket option. func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { + switch opt.(type) { + case tcpip.SocketDetachFilterOption: + return nil + } return nil } @@ -744,15 +748,15 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk // Only accept echo replies. switch e.NetProto { case header.IPv4ProtocolNumber: - h, ok := pkt.Data.PullUp(header.ICMPv4MinimumSize) - if !ok || header.ICMPv4(h).Type() != header.ICMPv4EchoReply { + h := header.ICMPv4(pkt.TransportHeader) + if len(h) < header.ICMPv4MinimumSize || h.Type() != header.ICMPv4EchoReply { e.stack.Stats().DroppedPackets.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() return } case header.IPv6ProtocolNumber: - h, ok := pkt.Data.PullUp(header.ICMPv6MinimumSize) - if !ok || header.ICMPv6(h).Type() != header.ICMPv6EchoReply { + h := header.ICMPv6(pkt.TransportHeader) + if len(h) < header.ICMPv6MinimumSize || h.Type() != header.ICMPv6EchoReply { e.stack.Stats().DroppedPackets.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() return @@ -786,12 +790,14 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk }, } - packet.data = pkt.Data + // ICMP socket's data includes ICMP header. + packet.data = pkt.TransportHeader.ToVectorisedView() + packet.data.Append(pkt.Data) e.rcvList.PushBack(packet) e.rcvBufSize += packet.data.Size() - packet.timestamp = e.stack.NowNanoseconds() + packet.timestamp = e.stack.Clock().NowNanoseconds() e.rcvMu.Unlock() e.stats.PacketsReceived.Increment() diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go index baf08eda6..df478115d 100644 --- a/pkg/tcpip/transport/packet/endpoint.go +++ b/pkg/tcpip/transport/packet/endpoint.go @@ -25,6 +25,8 @@ package packet import ( + "fmt" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" @@ -43,6 +45,9 @@ type packet struct { timestampNS int64 // senderAddr is the network address of the sender. senderAddr tcpip.FullAddress + // packetInfo holds additional information like the protocol + // of the packet etc. + packetInfo tcpip.LinkPacketInfo } // endpoint is the packet socket implementation of tcpip.Endpoint. It is legal @@ -71,11 +76,17 @@ type endpoint struct { rcvClosed bool // The following fields are protected by mu. - mu sync.RWMutex `state:"nosave"` - sndBufSize int - closed bool - stats tcpip.TransportEndpointStats `state:"nosave"` - bound bool + mu sync.RWMutex `state:"nosave"` + sndBufSize int + sndBufSizeMax int + closed bool + stats tcpip.TransportEndpointStats `state:"nosave"` + bound bool + boundNIC tcpip.NICID + + // lastErrorMu protects lastError. + lastErrorMu sync.Mutex `state:"nosave"` + lastError *tcpip.Error `state:".(string)"` } // NewEndpoint returns a new packet endpoint. @@ -92,6 +103,17 @@ func NewEndpoint(s *stack.Stack, cooked bool, netProto tcpip.NetworkProtocolNumb sndBufSize: 32 * 1024, } + // Override with stack defaults. + var ss stack.SendBufferSizeOption + if err := s.Option(&ss); err == nil { + ep.sndBufSizeMax = ss.Default + } + + var rs stack.ReceiveBufferSizeOption + if err := s.Option(&rs); err == nil { + ep.rcvBufSizeMax = rs.Default + } + if err := s.RegisterPacketEndpoint(0, netProto, ep); err != nil { return nil, err } @@ -132,8 +154,8 @@ func (ep *endpoint) Close() { // ModerateRecvBuf implements tcpip.Endpoint.ModerateRecvBuf. func (ep *endpoint) ModerateRecvBuf(copied int) {} -// Read implements tcpip.Endpoint.Read. -func (ep *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) { +// Read implements tcpip.PacketEndpoint.ReadPacket. +func (ep *endpoint) ReadPacket(addr *tcpip.FullAddress, info *tcpip.LinkPacketInfo) (buffer.View, tcpip.ControlMessages, *tcpip.Error) { ep.rcvMu.Lock() // If there's no data to read, return that read would block or that the @@ -158,11 +180,20 @@ func (ep *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMes *addr = packet.senderAddr } + if info != nil { + *info = packet.packetInfo + } + return packet.data.ToView(), tcpip.ControlMessages{HasTimestamp: true, Timestamp: packet.timestampNS}, nil } +// Read implements tcpip.Endpoint.Read. +func (ep *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) { + return ep.ReadPacket(addr, nil) +} + func (ep *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { - // TODO(b/129292371): Implement. + // TODO(gvisor.dev/issue/173): Implement. return 0, nil, tcpip.ErrInvalidOptionValue } @@ -215,12 +246,14 @@ func (ep *endpoint) Bind(addr tcpip.FullAddress) *tcpip.Error { ep.mu.Lock() defer ep.mu.Unlock() - if ep.bound { - return tcpip.ErrAlreadyBound + if ep.bound && ep.boundNIC == addr.NIC { + // If the NIC being bound is the same then just return success. + return nil } // Unregister endpoint with all the nics. ep.stack.UnregisterPacketEndpoint(0, ep.netProto, ep) + ep.bound = false // Bind endpoint to receive packets from specific interface. if err := ep.stack.RegisterPacketEndpoint(addr.NIC, ep.netProto, ep); err != nil { @@ -228,6 +261,7 @@ func (ep *endpoint) Bind(addr tcpip.FullAddress) *tcpip.Error { } ep.bound = true + ep.boundNIC = addr.NIC return nil } @@ -264,7 +298,13 @@ func (ep *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { // used with SetSockOpt, and this function always returns // tcpip.ErrNotSupported. func (ep *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { - return tcpip.ErrUnknownProtocolOption + switch opt.(type) { + case tcpip.SocketDetachFilterOption: + return nil + + default: + return tcpip.ErrUnknownProtocolOption + } } // SetSockOptBool implements tcpip.Endpoint.SetSockOptBool. @@ -274,11 +314,63 @@ func (ep *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { // SetSockOptInt implements tcpip.Endpoint.SetSockOptInt. func (ep *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { - return tcpip.ErrUnknownProtocolOption + switch opt { + case tcpip.SendBufferSizeOption: + // Make sure the send buffer size is within the min and max + // allowed. + var ss stack.SendBufferSizeOption + if err := ep.stack.Option(&ss); err != nil { + panic(fmt.Sprintf("s.Option(%#v) = %s", ss, err)) + } + if v > ss.Max { + v = ss.Max + } + if v < ss.Min { + v = ss.Min + } + ep.mu.Lock() + ep.sndBufSizeMax = v + ep.mu.Unlock() + return nil + + case tcpip.ReceiveBufferSizeOption: + // Make sure the receive buffer size is within the min and max + // allowed. + var rs stack.ReceiveBufferSizeOption + if err := ep.stack.Option(&rs); err != nil { + panic(fmt.Sprintf("s.Option(%#v) = %s", rs, err)) + } + if v > rs.Max { + v = rs.Max + } + if v < rs.Min { + v = rs.Min + } + ep.rcvMu.Lock() + ep.rcvBufSizeMax = v + ep.rcvMu.Unlock() + return nil + + default: + return tcpip.ErrUnknownProtocolOption + } +} + +func (ep *endpoint) takeLastError() *tcpip.Error { + ep.lastErrorMu.Lock() + defer ep.lastErrorMu.Unlock() + + err := ep.lastError + ep.lastError = nil + return err } // GetSockOpt implements tcpip.Endpoint.GetSockOpt. func (ep *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { + switch opt.(type) { + case tcpip.ErrorOption: + return ep.takeLastError() + } return tcpip.ErrNotSupported } @@ -289,7 +381,32 @@ func (ep *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. func (ep *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { - return 0, tcpip.ErrNotSupported + switch opt { + case tcpip.ReceiveQueueSizeOption: + v := 0 + ep.rcvMu.Lock() + if !ep.rcvList.Empty() { + p := ep.rcvList.Front() + v = p.data.Size() + } + ep.rcvMu.Unlock() + return v, nil + + case tcpip.SendBufferSizeOption: + ep.mu.Lock() + v := ep.sndBufSizeMax + ep.mu.Unlock() + return v, nil + + case tcpip.ReceiveBufferSizeOption: + ep.rcvMu.Lock() + v := ep.rcvBufSizeMax + ep.rcvMu.Unlock() + return v, nil + + default: + return -1, tcpip.ErrUnknownProtocolOption + } } // HandlePacket implements stack.PacketEndpoint.HandlePacket. @@ -315,7 +432,7 @@ func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, // Push new packet into receive list and increment the buffer size. var packet packet - // TODO(b/129292371): Return network protocol. + // TODO(gvisor.dev/issue/173): Return network protocol. if len(pkt.LinkHeader) > 0 { // Get info directly from the ethernet header. hdr := header.Ethernet(pkt.LinkHeader) @@ -323,40 +440,66 @@ func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, NIC: nicID, Addr: tcpip.Address(hdr.SourceAddress()), } + packet.packetInfo.Protocol = netProto + packet.packetInfo.PktType = pkt.PktType } else { // Guess the would-be ethernet header. packet.senderAddr = tcpip.FullAddress{ NIC: nicID, Addr: tcpip.Address(localAddr), } + packet.packetInfo.Protocol = netProto + packet.packetInfo.PktType = pkt.PktType } if ep.cooked { // Cooked packets can simply be queued. - packet.data = pkt.Data + switch pkt.PktType { + case tcpip.PacketHost: + packet.data = pkt.Data + case tcpip.PacketOutgoing: + // Strip Link Header from the Header. + pkt.Header = buffer.NewPrependableFromView(pkt.Header.View()[len(pkt.LinkHeader):]) + combinedVV := pkt.Header.View().ToVectorisedView() + combinedVV.Append(pkt.Data) + packet.data = combinedVV + default: + panic(fmt.Sprintf("unexpected PktType in pkt: %+v", pkt)) + } + } else { // Raw packets need their ethernet headers prepended before // queueing. var linkHeader buffer.View - if len(pkt.LinkHeader) == 0 { - // We weren't provided with an actual ethernet header, - // so fake one. - ethFields := header.EthernetFields{ - SrcAddr: tcpip.LinkAddress([]byte{0x00, 0x00, 0x00, 0x00, 0x00, 0x00}), - DstAddr: localAddr, - Type: netProto, + var combinedVV buffer.VectorisedView + if pkt.PktType != tcpip.PacketOutgoing { + if len(pkt.LinkHeader) == 0 { + // We weren't provided with an actual ethernet header, + // so fake one. + ethFields := header.EthernetFields{ + SrcAddr: tcpip.LinkAddress([]byte{0x00, 0x00, 0x00, 0x00, 0x00, 0x00}), + DstAddr: localAddr, + Type: netProto, + } + fakeHeader := make(header.Ethernet, header.EthernetMinimumSize) + fakeHeader.Encode(ðFields) + linkHeader = buffer.View(fakeHeader) + } else { + linkHeader = append(buffer.View(nil), pkt.LinkHeader...) } - fakeHeader := make(header.Ethernet, header.EthernetMinimumSize) - fakeHeader.Encode(ðFields) - linkHeader = buffer.View(fakeHeader) - } else { - linkHeader = append(buffer.View(nil), pkt.LinkHeader...) + combinedVV = linkHeader.ToVectorisedView() + } + if pkt.PktType == tcpip.PacketOutgoing { + // For outgoing packets the Link, Network and Transport + // headers are in the pkt.Header fields normally unless + // a Raw socket is in use in which case pkt.Header could + // be nil. + combinedVV.AppendView(pkt.Header.View()) } - combinedVV := linkHeader.ToVectorisedView() combinedVV.Append(pkt.Data) packet.data = combinedVV } - packet.timestampNS = ep.stack.NowNanoseconds() + packet.timestampNS = ep.stack.Clock().NowNanoseconds() ep.rcvList.PushBack(&packet) ep.rcvBufSize += packet.data.Size() diff --git a/pkg/tcpip/transport/packet/endpoint_state.go b/pkg/tcpip/transport/packet/endpoint_state.go index 9b88f17e4..e2fa96d17 100644 --- a/pkg/tcpip/transport/packet/endpoint_state.go +++ b/pkg/tcpip/transport/packet/endpoint_state.go @@ -15,6 +15,7 @@ package packet import ( + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -70,3 +71,21 @@ func (ep *endpoint) afterLoad() { panic(*err) } } + +// saveLastError is invoked by stateify. +func (ep *endpoint) saveLastError() string { + if ep.lastError == nil { + return "" + } + + return ep.lastError.String() +} + +// loadLastError is invoked by stateify. +func (ep *endpoint) loadLastError(s string) { + if s == "" { + return + } + + ep.lastError = tcpip.StringToError(s) +} diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index 6a7977259..f85a68554 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -63,6 +63,7 @@ type endpoint struct { stack *stack.Stack `state:"manual"` waiterQueue *waiter.Queue associated bool + hdrIncluded bool // The following fields are used to manage the receive queue and are // protected by rcvMu. @@ -94,7 +95,7 @@ func NewEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, trans } func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue, associated bool) (tcpip.Endpoint, *tcpip.Error) { - if netProto != header.IPv4ProtocolNumber { + if netProto != header.IPv4ProtocolNumber && netProto != header.IPv6ProtocolNumber { return nil, tcpip.ErrUnknownProtocol } @@ -108,16 +109,17 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProt rcvBufSizeMax: 32 * 1024, sndBufSizeMax: 32 * 1024, associated: associated, + hdrIncluded: !associated, } // Override with stack defaults. - var ss tcpip.StackSendBufferSizeOption - if err := s.TransportProtocolOption(transProto, &ss); err == nil { + var ss stack.SendBufferSizeOption + if err := s.Option(&ss); err == nil { e.sndBufSizeMax = ss.Default } - var rs tcpip.StackReceiveBufferSizeOption - if err := s.TransportProtocolOption(transProto, &rs); err == nil { + var rs stack.ReceiveBufferSizeOption + if err := s.Option(&rs); err == nil { e.rcvBufSizeMax = rs.Default } @@ -182,10 +184,6 @@ func (e *endpoint) SetOwner(owner tcpip.PacketOwner) { // Read implements tcpip.Endpoint.Read. func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) { - if !e.associated { - return buffer.View{}, tcpip.ControlMessages{}, tcpip.ErrInvalidOptionValue - } - e.rcvMu.Lock() // If there's no data to read, return that read would block or that the @@ -215,6 +213,11 @@ func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMess // Write implements tcpip.Endpoint.Write. func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { + // We can create, but not write to, unassociated IPv6 endpoints. + if !e.associated && e.TransportEndpointInfo.NetProto == header.IPv6ProtocolNumber { + return 0, nil, tcpip.ErrInvalidOptionValue + } + n, ch, err := e.write(p, opts) switch err { case nil: @@ -258,7 +261,7 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c // If this is an unassociated socket and callee provided a nonzero // destination address, route using that address. - if !e.associated { + if e.hdrIncluded { ip := header.IPv4(payloadBytes) if !ip.IsValid(len(payloadBytes)) { e.mu.RUnlock() @@ -319,12 +322,6 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c return 0, nil, tcpip.ErrNoRoute } - // We don't support IPv6 yet, so this has to be an IPv4 address. - if len(opts.To.Addr) != header.IPv4AddressSize { - e.mu.RUnlock() - return 0, nil, tcpip.ErrInvalidEndpointState - } - // Find the route to the destination. If BindAddress is 0, // FindRoute will choose an appropriate source address. route, err := e.stack.FindRoute(nic, e.BindAddr, opts.To.Addr, e.NetProto, false) @@ -354,17 +351,13 @@ func (e *endpoint) finishWrite(payloadBytes []byte, route *stack.Route) (int64, } } - switch e.NetProto { - case header.IPv4ProtocolNumber: - if !e.associated { - if err := route.WriteHeaderIncludedPacket(&stack.PacketBuffer{ - Data: buffer.View(payloadBytes).ToVectorisedView(), - }); err != nil { - return 0, nil, err - } - break + if e.hdrIncluded { + if err := route.WriteHeaderIncludedPacket(&stack.PacketBuffer{ + Data: buffer.View(payloadBytes).ToVectorisedView(), + }); err != nil { + return 0, nil, err } - + } else { hdr := buffer.NewPrependable(len(payloadBytes) + int(route.MaxHeaderLength())) if err := route.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: e.TransProto, TTL: route.DefaultTTL(), TOS: stack.DefaultTOS}, &stack.PacketBuffer{ Header: hdr, @@ -373,9 +366,6 @@ func (e *endpoint) finishWrite(payloadBytes []byte, route *stack.Route) (int64, }); err != nil { return 0, nil, err } - - default: - return 0, nil, tcpip.ErrUnknownProtocol } return int64(len(payloadBytes)), nil, nil @@ -400,11 +390,6 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { return tcpip.ErrInvalidEndpointState } - // We don't support IPv6 yet. - if len(addr.Addr) != header.IPv4AddressSize { - return tcpip.ErrInvalidEndpointState - } - nic := addr.NIC if e.bound { if e.BindNICID == 0 { @@ -470,14 +455,8 @@ func (e *endpoint) Bind(addr tcpip.FullAddress) *tcpip.Error { e.mu.Lock() defer e.mu.Unlock() - // Callers must provide an IPv4 address or no network address (for - // binding to a NIC, but not an address). - if len(addr.Addr) != 0 && len(addr.Addr) != 4 { - return tcpip.ErrInvalidEndpointState - } - // If a local address was specified, verify that it's valid. - if len(addr.Addr) == header.IPv4AddressSize && e.stack.CheckLocalAddress(addr.NIC, e.NetProto, addr.Addr) == 0 { + if len(addr.Addr) != 0 && e.stack.CheckLocalAddress(addr.NIC, e.NetProto, addr.Addr) == 0 { return tcpip.ErrBadLocalAddress } @@ -527,11 +506,24 @@ func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { // SetSockOpt implements tcpip.Endpoint.SetSockOpt. func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { - return tcpip.ErrUnknownProtocolOption + switch opt.(type) { + case tcpip.SocketDetachFilterOption: + return nil + + default: + return tcpip.ErrUnknownProtocolOption + } } // SetSockOptBool implements tcpip.Endpoint.SetSockOptBool. func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { + switch opt { + case tcpip.IPHdrIncludedOption: + e.mu.Lock() + e.hdrIncluded = v + e.mu.Unlock() + return nil + } return tcpip.ErrUnknownProtocolOption } @@ -541,9 +533,9 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { case tcpip.SendBufferSizeOption: // Make sure the send buffer size is within the min and max // allowed. - var ss tcpip.StackSendBufferSizeOption - if err := e.stack.TransportProtocolOption(e.TransProto, &ss); err != nil { - panic(fmt.Sprintf("s.TransportProtocolOption(%d, %+v) = %s", e.TransProto, ss, err)) + var ss stack.SendBufferSizeOption + if err := e.stack.Option(&ss); err != nil { + panic(fmt.Sprintf("s.Option(%#v) = %s", ss, err)) } if v > ss.Max { v = ss.Max @@ -559,9 +551,9 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { case tcpip.ReceiveBufferSizeOption: // Make sure the receive buffer size is within the min and max // allowed. - var rs tcpip.StackReceiveBufferSizeOption - if err := e.stack.TransportProtocolOption(e.TransProto, &rs); err != nil { - panic(fmt.Sprintf("s.TransportProtocolOption(%d, %+v) = %s", e.TransProto, rs, err)) + var rs stack.ReceiveBufferSizeOption + if err := e.stack.Option(&rs); err != nil { + panic(fmt.Sprintf("s.Option(%#v) = %s", rs, err)) } if v > rs.Max { v = rs.Max @@ -596,6 +588,12 @@ func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { case tcpip.KeepaliveEnabledOption: return false, nil + case tcpip.IPHdrIncludedOption: + e.mu.Lock() + v := e.hdrIncluded + e.mu.Unlock() + return v, nil + default: return false, tcpip.ErrUnknownProtocolOption } @@ -635,8 +633,15 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { func (e *endpoint) HandlePacket(route *stack.Route, pkt *stack.PacketBuffer) { e.rcvMu.Lock() - // Drop the packet if our buffer is currently full. - if e.rcvClosed { + // Drop the packet if our buffer is currently full or if this is an unassociated + // endpoint (i.e endpoint created w/ IPPROTO_RAW). Such endpoints are send only + // See: https://man7.org/linux/man-pages/man7/raw.7.html + // + // An IPPROTO_RAW socket is send only. If you really want to receive + // all IP packets, use a packet(7) socket with the ETH_P_IP protocol. + // Note that packet sockets don't reassemble IP fragments, unlike raw + // sockets. + if e.rcvClosed || !e.associated { e.rcvMu.Unlock() e.stack.Stats().DroppedPackets.Increment() e.stats.ReceiveErrors.ClosedReceiver.Increment() @@ -680,12 +685,22 @@ func (e *endpoint) HandlePacket(route *stack.Route, pkt *stack.PacketBuffer) { }, } - headers := append(buffer.View(nil), pkt.NetworkHeader...) - headers = append(headers, pkt.TransportHeader...) - combinedVV := headers.ToVectorisedView() + // Raw IPv4 endpoints return the IP header, but IPv6 endpoints do not. + // We copy headers' underlying bytes because pkt.*Header may point to + // the middle of a slice, and another struct may point to the "outer" + // slice. Save/restore doesn't support overlapping slices and will fail. + var combinedVV buffer.VectorisedView + if e.TransportEndpointInfo.NetProto == header.IPv4ProtocolNumber { + headers := make(buffer.View, 0, len(pkt.NetworkHeader)+len(pkt.TransportHeader)) + headers = append(headers, pkt.NetworkHeader...) + headers = append(headers, pkt.TransportHeader...) + combinedVV = headers.ToVectorisedView() + } else { + combinedVV = append(buffer.View(nil), pkt.TransportHeader...).ToVectorisedView() + } combinedVV.Append(pkt.Data) packet.data = combinedVV - packet.timestampNS = e.stack.NowNanoseconds() + packet.timestampNS = e.stack.Clock().NowNanoseconds() e.rcvList.PushBack(packet) e.rcvBufSize += packet.data.Size() diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD index 6baeda8e4..e860ee484 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -49,6 +49,7 @@ go_library( "segment_heap.go", "segment_queue.go", "segment_state.go", + "segment_unsafe.go", "snd.go", "snd_state.go", "tcp_endpoint_list.go", @@ -86,6 +87,7 @@ go_test( "tcp_test.go", "tcp_timestamp_test.go", ], + shard_count = 10, deps = [ ":tcp", "//pkg/sync", diff --git a/pkg/tcpip/transport/tcp/accept.go b/pkg/tcpip/transport/tcp/accept.go index 7679fe169..6e00e5526 100644 --- a/pkg/tcpip/transport/tcp/accept.go +++ b/pkg/tcpip/transport/tcp/accept.go @@ -27,7 +27,6 @@ import ( "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/header" - "gvisor.dev/gvisor/pkg/tcpip/ports" "gvisor.dev/gvisor/pkg/tcpip/seqnum" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/waiter" @@ -199,9 +198,8 @@ func (l *listenContext) isCookieValid(id stack.TransportEndpointID, cookie seqnu } // createConnectingEndpoint creates a new endpoint in a connecting state, with -// the connection parameters given by the arguments. The endpoint is returned -// with n.mu held. -func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, irs seqnum.Value, rcvdSynOpts *header.TCPSynOptions, queue *waiter.Queue) (*endpoint, *tcpip.Error) { +// the connection parameters given by the arguments. +func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, irs seqnum.Value, rcvdSynOpts *header.TCPSynOptions, queue *waiter.Queue) *endpoint { // Create a new endpoint. netProto := l.netProto if netProto == 0 { @@ -227,22 +225,7 @@ func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, i // window to grow to a really large value. n.rcvAutoParams.prevCopied = n.initialReceiveWindow() - // Lock the endpoint before registering to ensure that no out of - // band changes are possible due to incoming packets etc till - // the endpoint is done initializing. - n.mu.Lock() - - // Register new endpoint so that packets are routed to it. - if err := n.stack.RegisterTransportEndpoint(n.boundNICID, n.effectiveNetProtos, ProtocolNumber, n.ID, n, ports.Flags{LoadBalanced: n.reusePort}, n.boundBindToDevice); err != nil { - n.mu.Unlock() - n.Close() - return nil, err - } - - n.isRegistered = true - n.registeredReusePort = n.reusePort - - return n, nil + return n } // createEndpointAndPerformHandshake creates a new endpoint in connected state @@ -253,10 +236,12 @@ func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *head // Create new endpoint. irs := s.sequenceNumber isn := generateSecureISN(s.id, l.stack.Seed()) - ep, err := l.createConnectingEndpoint(s, isn, irs, opts, queue) - if err != nil { - return nil, err - } + ep := l.createConnectingEndpoint(s, isn, irs, opts, queue) + + // Lock the endpoint before registering to ensure that no out of + // band changes are possible due to incoming packets etc till + // the endpoint is done initializing. + ep.mu.Lock() ep.owner = owner // listenEP is nil when listenContext is used by tcp.Forwarder. @@ -264,18 +249,13 @@ func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *head if l.listenEP != nil { l.listenEP.mu.Lock() if l.listenEP.EndpointState() != StateListen { + l.listenEP.mu.Unlock() // Ensure we release any registrations done by the newly // created endpoint. ep.mu.Unlock() ep.Close() - // Wake up any waiters. This is strictly not required normally - // as a socket that was never accepted can't really have any - // registered waiters except when stack.Wait() is called which - // waits for all registered endpoints to stop and expects an - // EventHUp. - ep.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut) return nil, tcpip.ErrConnectionAborted } l.addPendingEndpoint(ep) @@ -284,21 +264,44 @@ func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *head // to the newly created endpoint. l.listenEP.propagateInheritableOptionsLocked(ep) + if !ep.reserveTupleLocked() { + ep.mu.Unlock() + ep.Close() + + if l.listenEP != nil { + l.removePendingEndpoint(ep) + l.listenEP.mu.Unlock() + } + + return nil, tcpip.ErrConnectionAborted + } + deferAccept = l.listenEP.deferAccept l.listenEP.mu.Unlock() } + // Register new endpoint so that packets are routed to it. + if err := ep.stack.RegisterTransportEndpoint(ep.boundNICID, ep.effectiveNetProtos, ProtocolNumber, ep.ID, ep, ep.boundPortFlags, ep.boundBindToDevice); err != nil { + ep.mu.Unlock() + ep.Close() + + if l.listenEP != nil { + l.removePendingEndpoint(ep) + } + + ep.drainClosingSegmentQueue() + + return nil, err + } + + ep.isRegistered = true + // Perform the 3-way handshake. h := newPassiveHandshake(ep, seqnum.Size(ep.initialReceiveWindow()), isn, irs, opts, deferAccept) if err := h.execute(); err != nil { ep.mu.Unlock() ep.Close() - // Wake up any waiters. This is strictly not required normally - // as a socket that was never accepted can't really have any - // registered waiters except when stack.Wait() is called which - // waits for all registered endpoints to stop and expects an - // EventHUp. - ep.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut) + ep.notifyAborted() if l.listenEP != nil { l.removePendingEndpoint(ep) @@ -374,6 +377,43 @@ func (e *endpoint) deliverAccepted(n *endpoint) { // Precondition: e.mu and n.mu must be held. func (e *endpoint) propagateInheritableOptionsLocked(n *endpoint) { n.userTimeout = e.userTimeout + n.portFlags = e.portFlags + n.boundBindToDevice = e.boundBindToDevice + n.boundPortFlags = e.boundPortFlags +} + +// reserveTupleLocked reserves an accepted endpoint's tuple. +// +// Preconditions: +// * propagateInheritableOptionsLocked has been called. +// * e.mu is held. +func (e *endpoint) reserveTupleLocked() bool { + dest := tcpip.FullAddress{Addr: e.ID.RemoteAddress, Port: e.ID.RemotePort} + if !e.stack.ReserveTuple( + e.effectiveNetProtos, + ProtocolNumber, + e.ID.LocalAddress, + e.ID.LocalPort, + e.boundPortFlags, + e.boundBindToDevice, + dest, + ) { + return false + } + + e.isPortReserved = true + e.boundDest = dest + return true +} + +// notifyAborted wakes up any waiters on registered, but not accepted +// endpoints. +// +// This is strictly not required normally as a socket that was never accepted +// can't really have any registered waiters except when stack.Wait() is called +// which waits for all registered endpoints to stop and expects an EventHUp. +func (e *endpoint) notifyAborted() { + e.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut) } // handleSynSegment is called in its own goroutine once the listening endpoint @@ -568,16 +608,34 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { rcvdSynOptions.TSEcr = s.parsedOptions.TSEcr } - n, err := ctx.createConnectingEndpoint(s, iss, irs, rcvdSynOptions, &waiter.Queue{}) - if err != nil { + n := ctx.createConnectingEndpoint(s, iss, irs, rcvdSynOptions, &waiter.Queue{}) + + n.mu.Lock() + + // Propagate any inheritable options from the listening endpoint + // to the newly created endpoint. + e.propagateInheritableOptionsLocked(n) + + if !n.reserveTupleLocked() { + n.mu.Unlock() + n.Close() + e.stack.Stats().TCP.FailedConnectionAttempts.Increment() e.stats.FailedConnectionAttempts.Increment() return } - // Propagate any inheritable options from the listening endpoint - // to the newly created endpoint. - e.propagateInheritableOptionsLocked(n) + // Register new endpoint so that packets are routed to it. + if err := n.stack.RegisterTransportEndpoint(n.boundNICID, n.effectiveNetProtos, ProtocolNumber, n.ID, n, n.boundPortFlags, n.boundBindToDevice); err != nil { + n.mu.Unlock() + n.Close() + + e.stack.Stats().TCP.FailedConnectionAttempts.Increment() + e.stats.FailedConnectionAttempts.Increment() + return + } + + n.isRegistered = true // clear the tsOffset for the newly created // endpoint as the Timestamp was already diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 377643b82..1798510bc 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -490,6 +490,9 @@ func (h *handshake) resolveRoute() *tcpip.Error { <-h.ep.undrain h.ep.mu.Lock() } + if n¬ifyError != 0 { + return h.ep.takeLastError() + } } // Wait for notification. @@ -509,9 +512,7 @@ func (h *handshake) execute() *tcpip.Error { // Initialize the resend timer. resendWaker := sleep.Waker{} timeOut := time.Duration(time.Second) - rt := time.AfterFunc(timeOut, func() { - resendWaker.Assert() - }) + rt := time.AfterFunc(timeOut, resendWaker.Assert) defer rt.Stop() // Set up the wakers. @@ -521,7 +522,7 @@ func (h *handshake) execute() *tcpip.Error { s.AddWaker(&h.ep.newSegmentWaker, wakerForNewSegment) defer s.Done() - var sackEnabled tcpip.StackSACKEnabled + var sackEnabled SACKEnabled if err := h.ep.stack.TransportProtocolOption(ProtocolNumber, &sackEnabled); err != nil { // If stack returned an error when checking for SACKEnabled // status then just default to switching off SACK negotiation. @@ -618,6 +619,9 @@ func (h *handshake) execute() *tcpip.Error { <-h.ep.undrain h.ep.mu.Lock() } + if n¬ifyError != 0 { + return h.ep.takeLastError() + } case wakerForNewSegment: if err := h.processSegments(); err != nil { @@ -1050,8 +1054,8 @@ func (e *endpoint) tryDeliverSegmentFromClosedEndpoint(s *segment) { panic("current endpoint not removed from demuxer, enqueing segments to itself") } - if ep.(*endpoint).enqueueSegment(s) { - ep.(*endpoint).newSegmentWaker.Assert() + if ep := ep.(*endpoint); ep.enqueueSegment(s) { + ep.newSegmentWaker.Assert() } } @@ -1120,7 +1124,7 @@ func (e *endpoint) handleReset(s *segment) (ok bool, err *tcpip.Error) { func (e *endpoint) handleSegments(fastPath bool) *tcpip.Error { checkRequeue := true for i := 0; i < maxSegmentsPerWake; i++ { - if e.EndpointState() == StateClose || e.EndpointState() == StateError { + if e.EndpointState().closed() { return nil } s := e.segmentQueue.dequeue() @@ -1440,9 +1444,7 @@ func (e *endpoint) protocolMainLoop(handshake bool, wakerInitDone chan<- struct{ if e.EndpointState() == StateFinWait2 && e.closed { // The socket has been closed and we are in FIN_WAIT2 // so start the FIN_WAIT2 timer. - closeTimer = time.AfterFunc(e.tcpLingerTimeout, func() { - closeWaker.Assert() - }) + closeTimer = time.AfterFunc(e.tcpLingerTimeout, closeWaker.Assert) e.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut) } } @@ -1460,7 +1462,7 @@ func (e *endpoint) protocolMainLoop(handshake bool, wakerInitDone chan<- struct{ return err } } - if e.EndpointState() != StateClose && e.EndpointState() != StateError { + if !e.EndpointState().closed() { // Only block the worker if the endpoint // is not in closed state or error state. close(e.drainDone) @@ -1526,7 +1528,12 @@ func (e *endpoint) protocolMainLoop(handshake bool, wakerInitDone chan<- struct{ } loop: - for e.EndpointState() != StateTimeWait && e.EndpointState() != StateClose && e.EndpointState() != StateError { + for { + switch e.EndpointState() { + case StateTimeWait, StateClose, StateError: + break loop + } + e.mu.Unlock() v, _ := s.Fetch(true) e.mu.Lock() diff --git a/pkg/tcpip/transport/tcp/dispatcher.go b/pkg/tcpip/transport/tcp/dispatcher.go index 047704c80..98aecab9e 100644 --- a/pkg/tcpip/transport/tcp/dispatcher.go +++ b/pkg/tcpip/transport/tcp/dispatcher.go @@ -15,6 +15,8 @@ package tcp import ( + "encoding/binary" + "gvisor.dev/gvisor/pkg/rand" "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/sync" @@ -66,89 +68,68 @@ func (q *epQueue) empty() bool { // processor is responsible for processing packets queued to a tcp endpoint. type processor struct { epQ epQueue + sleeper sleep.Sleeper newEndpointWaker sleep.Waker closeWaker sleep.Waker - id int - wg sync.WaitGroup -} - -func newProcessor(id int) *processor { - p := &processor{ - id: id, - } - p.wg.Add(1) - go p.handleSegments() - return p } func (p *processor) close() { p.closeWaker.Assert() } -func (p *processor) wait() { - p.wg.Wait() -} - func (p *processor) queueEndpoint(ep *endpoint) { // Queue an endpoint for processing by the processor goroutine. p.epQ.enqueue(ep) p.newEndpointWaker.Assert() } -func (p *processor) handleSegments() { - const newEndpointWaker = 1 - const closeWaker = 2 - s := sleep.Sleeper{} - s.AddWaker(&p.newEndpointWaker, newEndpointWaker) - s.AddWaker(&p.closeWaker, closeWaker) - defer s.Done() +const ( + newEndpointWaker = 1 + closeWaker = 2 +) + +func (p *processor) start(wg *sync.WaitGroup) { + defer wg.Done() + defer p.sleeper.Done() + for { - id, ok := s.Fetch(true) - if ok && id == closeWaker { - p.wg.Done() - return + if id, _ := p.sleeper.Fetch(true); id == closeWaker { + break } - for ep := p.epQ.dequeue(); ep != nil; ep = p.epQ.dequeue() { + for { + ep := p.epQ.dequeue() + if ep == nil { + break + } if ep.segmentQueue.empty() { continue } - // If socket has transitioned out of connected state - // then just let the worker handle the packet. + // If socket has transitioned out of connected state then just let the + // worker handle the packet. // - // NOTE: We read this outside of e.mu lock which means - // that by the time we get to handleSegments the - // endpoint may not be in ESTABLISHED. But this should - // be fine as all normal shutdown states are handled by - // handleSegments and if the endpoint moves to a - // CLOSED/ERROR state then handleSegments is a noop. - if ep.EndpointState() != StateEstablished { - ep.newSegmentWaker.Assert() - continue - } - - if !ep.mu.TryLock() { - ep.newSegmentWaker.Assert() - continue - } - // If the endpoint is in a connected state then we do - // direct delivery to ensure low latency and avoid - // scheduler interactions. - if err := ep.handleSegments(true /* fastPath */); err != nil || ep.EndpointState() == StateClose { - // Send any active resets if required. - if err != nil { + // NOTE: We read this outside of e.mu lock which means that by the time + // we get to handleSegments the endpoint may not be in ESTABLISHED. But + // this should be fine as all normal shutdown states are handled by + // handleSegments and if the endpoint moves to a CLOSED/ERROR state + // then handleSegments is a noop. + if ep.EndpointState() == StateEstablished && ep.mu.TryLock() { + // If the endpoint is in a connected state then we do direct delivery + // to ensure low latency and avoid scheduler interactions. + switch err := ep.handleSegments(true /* fastPath */); { + case err != nil: + // Send any active resets if required. ep.resetConnectionLocked(err) + fallthrough + case ep.EndpointState() == StateClose: + ep.notifyProtocolGoroutine(notifyTickleWorker) + case !ep.segmentQueue.empty(): + p.epQ.enqueue(ep) } - ep.notifyProtocolGoroutine(notifyTickleWorker) ep.mu.Unlock() - continue - } - - if !ep.segmentQueue.empty() { - p.epQ.enqueue(ep) + } else { + ep.newSegmentWaker.Assert() } - - ep.mu.Unlock() } } } @@ -159,31 +140,36 @@ func (p *processor) handleSegments() { // hash of the endpoint id to ensure that delivery for the same endpoint happens // in-order. type dispatcher struct { - processors []*processor + processors []processor seed uint32 -} - -func newDispatcher(nProcessors int) *dispatcher { - processors := []*processor{} - for i := 0; i < nProcessors; i++ { - processors = append(processors, newProcessor(i)) - } - return &dispatcher{ - processors: processors, - seed: generateRandUint32(), + wg sync.WaitGroup +} + +func (d *dispatcher) init(nProcessors int) { + d.close() + d.wait() + d.processors = make([]processor, nProcessors) + d.seed = generateRandUint32() + for i := range d.processors { + p := &d.processors[i] + p.sleeper.AddWaker(&p.newEndpointWaker, newEndpointWaker) + p.sleeper.AddWaker(&p.closeWaker, closeWaker) + d.wg.Add(1) + // NB: sleeper-waker registration must happen synchronously to avoid races + // with `close`. It's possible to pull all this logic into `start`, but + // that results in a heap-allocated function literal. + go p.start(&d.wg) } } func (d *dispatcher) close() { - for _, p := range d.processors { - p.close() + for i := range d.processors { + d.processors[i].close() } } func (d *dispatcher) wait() { - for _, p := range d.processors { - p.wait() - } + d.wg.Wait() } func (d *dispatcher) queuePacket(r *stack.Route, stackEP stack.TransportEndpoint, id stack.TransportEndpointID, pkt *stack.PacketBuffer) { @@ -231,20 +217,18 @@ func generateRandUint32() uint32 { if _, err := rand.Read(b); err != nil { panic(err) } - return uint32(b[0]) | uint32(b[1])<<8 | uint32(b[2])<<16 | uint32(b[3])<<24 + return binary.LittleEndian.Uint32(b) } func (d *dispatcher) selectProcessor(id stack.TransportEndpointID) *processor { - payload := []byte{ - byte(id.LocalPort), - byte(id.LocalPort >> 8), - byte(id.RemotePort), - byte(id.RemotePort >> 8)} + var payload [4]byte + binary.LittleEndian.PutUint16(payload[0:], id.LocalPort) + binary.LittleEndian.PutUint16(payload[2:], id.RemotePort) h := jenkins.Sum32(d.seed) - h.Write(payload) + h.Write(payload[:]) h.Write([]byte(id.LocalAddress)) h.Write([]byte(id.RemoteAddress)) - return d.processors[h.Sum32()%uint32(len(d.processors))] + return &d.processors[h.Sum32()%uint32(len(d.processors))] } diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 10df2bcd5..0f7487963 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -396,7 +396,8 @@ type endpoint struct { mu sync.Mutex `state:"nosave"` ownedByUser uint32 - // state must be read/set using the EndpointState()/setEndpointState() methods. + // state must be read/set using the EndpointState()/setEndpointState() + // methods. state EndpointState `state:".(EndpointState)"` // origEndpointState is only used during a restore phase to save the @@ -405,8 +406,8 @@ type endpoint struct { origEndpointState EndpointState `state:"nosave"` isPortReserved bool `state:"manual"` - isRegistered bool - boundNICID tcpip.NICID `state:"manual"` + isRegistered bool `state:"manual"` + boundNICID tcpip.NICID route stack.Route `state:"manual"` ttl uint8 v6only bool @@ -415,10 +416,14 @@ type endpoint struct { // disabling SO_BROADCAST, albeit as a NOOP. broadcast bool + // portFlags stores the current values of port related flags. + portFlags ports.Flags + // Values used to reserve a port or register a transport endpoint // (which ever happens first). boundBindToDevice tcpip.NICID boundPortFlags ports.Flags + boundDest tcpip.FullAddress // effectiveNetProtos contains the network protocols actually in use. In // most cases it will only contain "netProto", but in cases like IPv6 @@ -426,7 +431,7 @@ type endpoint struct { // protocols (e.g., IPv6 and IPv4) or a single different protocol (e.g., // IPv4 when IPv6 endpoint is bound or connected to an IPv4 mapped // address). - effectiveNetProtos []tcpip.NetworkProtocolNumber `state:"manual"` + effectiveNetProtos []tcpip.NetworkProtocolNumber // workerRunning specifies if a worker goroutine is running. workerRunning bool @@ -462,13 +467,6 @@ type endpoint struct { // sack holds TCP SACK related information for this endpoint. sack SACKInfo - // reusePort is set to true if SO_REUSEPORT is enabled. - reusePort bool - - // registeredReusePort is set if the current endpoint registration was - // done with SO_REUSEPORT enabled. - registeredReusePort bool - // bindToDevice is set to the NIC on which to bind or disabled if 0. bindToDevice tcpip.NICID @@ -488,7 +486,6 @@ type endpoint struct { // The options below aren't implemented, but we remember the user // settings because applications expect to be able to set/query these // options. - reuseAddr bool // slowAck holds the negated state of quick ack. It is stubbed out and // does nothing. @@ -838,7 +835,6 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue rcvBufSize: DefaultReceiveBufferSize, sndBufSize: DefaultSendBufferSize, sndMTU: int(math.MaxInt32), - reuseAddr: true, keepalive: keepalive{ // Linux defaults. idle: 2 * time.Hour, @@ -851,12 +847,12 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue maxSynRetries: DefaultSynRetries, } - var ss tcpip.StackSendBufferSizeOption + var ss SendBufferSizeOption if err := s.TransportProtocolOption(ProtocolNumber, &ss); err == nil { e.sndBufSize = ss.Default } - var rs tcpip.StackReceiveBufferSizeOption + var rs ReceiveBufferSizeOption if err := s.TransportProtocolOption(ProtocolNumber, &rs); err == nil { e.rcvBufSize = rs.Default } @@ -871,7 +867,7 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue e.rcvAutoParams.disabled = !bool(mrb) } - var de tcpip.StackDelayEnabled + var de DelayEnabled if err := s.TransportProtocolOption(ProtocolNumber, &de); err == nil && de { e.SetSockOptBool(tcpip.DelayOption, true) } @@ -1025,15 +1021,15 @@ func (e *endpoint) closeNoShutdownLocked() { // in Listen() when trying to register. if e.EndpointState() == StateListen && e.isPortReserved { if e.isRegistered { - e.stack.StartTransportEndpointCleanup(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, ports.Flags{LoadBalanced: e.registeredReusePort}, e.boundBindToDevice) + e.stack.StartTransportEndpointCleanup(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.boundPortFlags, e.boundBindToDevice) e.isRegistered = false - e.registeredReusePort = false } - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.boundPortFlags, e.boundBindToDevice) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.boundPortFlags, e.boundBindToDevice, e.boundDest) e.isPortReserved = false e.boundBindToDevice = 0 e.boundPortFlags = ports.Flags{} + e.boundDest = tcpip.FullAddress{} } // Mark endpoint as closed. @@ -1091,17 +1087,17 @@ func (e *endpoint) cleanupLocked() { e.workerCleanup = false if e.isRegistered { - e.stack.StartTransportEndpointCleanup(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, ports.Flags{LoadBalanced: e.registeredReusePort}, e.boundBindToDevice) + e.stack.StartTransportEndpointCleanup(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.boundPortFlags, e.boundBindToDevice) e.isRegistered = false - e.registeredReusePort = false } if e.isPortReserved { - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.boundPortFlags, e.boundBindToDevice) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.boundPortFlags, e.boundBindToDevice, e.boundDest) e.isPortReserved = false } e.boundBindToDevice = 0 e.boundPortFlags = ports.Flags{} + e.boundDest = tcpip.FullAddress{} e.route.Release() e.stack.CompleteTransportEndpointCleanup(e) @@ -1213,9 +1209,27 @@ func (e *endpoint) SetOwner(owner tcpip.PacketOwner) { e.owner = owner } +func (e *endpoint) takeLastError() *tcpip.Error { + e.lastErrorMu.Lock() + defer e.lastErrorMu.Unlock() + err := e.lastError + e.lastError = nil + return err +} + // Read reads data from the endpoint. func (e *endpoint) Read(*tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) { e.LockUser() + defer e.UnlockUser() + + // When in SYN-SENT state, let the caller block on the receive. + // An application can initiate a non-blocking connect and then block + // on a receive. It can expect to read any data after the handshake + // is complete. RFC793, section 3.9, p58. + if e.EndpointState() == StateSynSent { + return buffer.View{}, tcpip.ControlMessages{}, tcpip.ErrWouldBlock + } + // The endpoint can be read if it's connected, or if it's already closed // but has some pending unread data. Also note that a RST being received // would cause the state to become StateError so we should allow the @@ -1225,7 +1239,6 @@ func (e *endpoint) Read(*tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, if s := e.EndpointState(); !s.connected() && s != StateClose && bufUsed == 0 { e.rcvListMu.Unlock() he := e.HardError - e.UnlockUser() if s == StateError { return buffer.View{}, tcpip.ControlMessages{}, he } @@ -1235,7 +1248,6 @@ func (e *endpoint) Read(*tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, v, err := e.readLocked() e.rcvListMu.Unlock() - e.UnlockUser() if err == tcpip.ErrClosedForReceive { e.stats.ReadErrors.ReadClosed.Increment() @@ -1522,12 +1534,12 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { case tcpip.ReuseAddressOption: e.LockUser() - e.reuseAddr = v + e.portFlags.TupleOnly = v e.UnlockUser() case tcpip.ReusePortOption: e.LockUser() - e.reusePort = v + e.portFlags.LoadBalanced = v e.UnlockUser() case tcpip.V6OnlyOption: @@ -1585,10 +1597,17 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { e.UnlockUser() e.notifyProtocolGoroutine(notifyMSSChanged) + case tcpip.MTUDiscoverOption: + // Return not supported if attempting to set this option to + // anything other than path MTU discovery disabled. + if v != tcpip.PMTUDiscoveryDont { + return tcpip.ErrNotSupported + } + case tcpip.ReceiveBufferSizeOption: // Make sure the receive buffer size is within the min and max // allowed. - var rs tcpip.StackReceiveBufferSizeOption + var rs ReceiveBufferSizeOption if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err == nil { if v < rs.Min { v = rs.Min @@ -1638,7 +1657,7 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { case tcpip.SendBufferSizeOption: // Make sure the send buffer size is within the min and max // allowed. - var ss tcpip.StackSendBufferSizeOption + var ss SendBufferSizeOption if err := e.stack.TransportProtocolOption(ProtocolNumber, &ss); err == nil { if v < ss.Min { v = ss.Min @@ -1678,7 +1697,7 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { return tcpip.ErrInvalidOptionValue } } - var rs tcpip.StackReceiveBufferSizeOption + var rs ReceiveBufferSizeOption if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err == nil { if v < rs.Min/2 { v = rs.Min / 2 @@ -1781,6 +1800,9 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.deferAccept = time.Duration(v) e.UnlockUser() + case tcpip.SocketDetachFilterOption: + return nil + default: return nil } @@ -1831,14 +1853,14 @@ func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { case tcpip.ReuseAddressOption: e.LockUser() - v := e.reuseAddr + v := e.portFlags.TupleOnly e.UnlockUser() return v, nil case tcpip.ReusePortOption: e.LockUser() - v := e.reusePort + v := e.portFlags.LoadBalanced e.UnlockUser() return v, nil @@ -1892,6 +1914,11 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { v := header.TCPDefaultMSS return v, nil + case tcpip.MTUDiscoverOption: + // Always return the path MTU discovery disabled setting since + // it's the only one supported. + return tcpip.PMTUDiscoveryDont, nil + case tcpip.ReceiveQueueSizeOption: return e.readyReceiveSize() @@ -1937,11 +1964,7 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { switch o := opt.(type) { case tcpip.ErrorOption: - e.lastErrorMu.Lock() - err := e.lastError - e.lastError = nil - e.lastErrorMu.Unlock() - return err + return e.takeLastError() case *tcpip.BindToDeviceOption: e.LockUser() @@ -2091,8 +2114,6 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tc } defer r.Release() - origID := e.ID - netProtos := []tcpip.NetworkProtocolNumber{netProto} e.ID.LocalAddress = r.LocalAddress e.ID.RemoteAddress = r.RemoteAddress @@ -2100,11 +2121,10 @@ 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, ports.Flags{LoadBalanced: e.reusePort}, e.boundBindToDevice) + err := e.stack.RegisterTransportEndpoint(nicID, netProtos, ProtocolNumber, e.ID, e, e.boundPortFlags, e.boundBindToDevice) if err != nil { return err } - e.registeredReusePort = e.reusePort } else { // The endpoint doesn't have a local port yet, so try to get // one. Make sure that it isn't one that will result in the same @@ -2128,40 +2148,33 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tc if sameAddr && p == e.ID.RemotePort { return false, nil } - // reusePort is false below because connect cannot reuse a port even if - // reusePort was set. - if !e.stack.IsPortAvailable(netProtos, ProtocolNumber, e.ID.LocalAddress, p, ports.Flags{LoadBalanced: false}, e.bindToDevice) { + if _, err := e.stack.ReservePort(netProtos, ProtocolNumber, e.ID.LocalAddress, p, e.portFlags, e.bindToDevice, addr); err != nil { return false, nil } id := e.ID id.LocalPort = p - switch e.stack.RegisterTransportEndpoint(nicID, netProtos, ProtocolNumber, id, e, ports.Flags{LoadBalanced: e.reusePort}, e.bindToDevice) { - case nil: - // Port picking successful. Save the details of - // the selected port. - e.ID = id - e.boundBindToDevice = e.bindToDevice - e.registeredReusePort = e.reusePort - return true, nil - case tcpip.ErrPortInUse: - return false, nil - default: + if err := e.stack.RegisterTransportEndpoint(nicID, netProtos, ProtocolNumber, id, e, e.portFlags, e.bindToDevice); err != nil { + e.stack.ReleasePort(netProtos, ProtocolNumber, e.ID.LocalAddress, p, e.portFlags, e.bindToDevice, addr) + if err == tcpip.ErrPortInUse { + return false, nil + } return false, err } + + // Port picking successful. Save the details of + // the selected port. + e.ID = id + e.isPortReserved = true + e.boundBindToDevice = e.bindToDevice + e.boundPortFlags = e.portFlags + e.boundDest = addr + return true, nil }); err != nil { return err } } - // Remove the port reservation. This can happen when Bind is called - // 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.boundPortFlags, e.boundBindToDevice) - e.isPortReserved = false - } - e.isRegistered = true e.setEndpointState(StateConnecting) e.route = r.Clone() @@ -2340,13 +2353,12 @@ func (e *endpoint) listen(backlog int) *tcpip.Error { } // Register the endpoint. - if err := e.stack.RegisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, ports.Flags{LoadBalanced: e.reusePort}, e.boundBindToDevice); err != nil { + if err := e.stack.RegisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.boundPortFlags, e.boundBindToDevice); err != nil { return err } e.isRegistered = true e.setEndpointState(StateListen) - e.registeredReusePort = e.reusePort // The channel may be non-nil when we're restoring the endpoint, and it // may be pre-populated with some previously accepted (but not Accepted) @@ -2433,16 +2445,13 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) (err *tcpip.Error) { } } - flags := ports.Flags{ - LoadBalanced: e.reusePort, - } - port, err := e.stack.ReservePort(netProtos, ProtocolNumber, addr.Addr, addr.Port, flags, e.bindToDevice) + port, err := e.stack.ReservePort(netProtos, ProtocolNumber, addr.Addr, addr.Port, e.portFlags, e.bindToDevice, tcpip.FullAddress{}) if err != nil { return err } e.boundBindToDevice = e.bindToDevice - e.boundPortFlags = flags + e.boundPortFlags = e.portFlags e.isPortReserved = true e.effectiveNetProtos = netProtos e.ID.LocalPort = port @@ -2450,7 +2459,7 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) (err *tcpip.Error) { // Any failures beyond this point must remove the port registration. defer func(portFlags ports.Flags, bindToDevice tcpip.NICID) { if err != nil { - e.stack.ReleasePort(netProtos, ProtocolNumber, addr.Addr, port, portFlags, bindToDevice) + e.stack.ReleasePort(netProtos, ProtocolNumber, addr.Addr, port, portFlags, bindToDevice, tcpip.FullAddress{}) e.isPortReserved = false e.effectiveNetProtos = nil e.ID.LocalPort = 0 @@ -2473,6 +2482,10 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) (err *tcpip.Error) { e.ID.LocalAddress = addr.Addr } + if err := e.stack.CheckRegisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e.boundPortFlags, e.boundBindToDevice); err != nil { + return err + } + // Mark endpoint as bound. e.setEndpointState(StateBound) @@ -2537,6 +2550,18 @@ func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.C e.sndBufMu.Unlock() e.notifyProtocolGoroutine(notifyMTUChanged) + + case stack.ControlNoRoute: + e.lastErrorMu.Lock() + e.lastError = tcpip.ErrNoRoute + e.lastErrorMu.Unlock() + e.notifyProtocolGoroutine(notifyError) + + case stack.ControlNetworkUnreachable: + e.lastErrorMu.Lock() + e.lastError = tcpip.ErrNetworkUnreachable + e.lastErrorMu.Unlock() + e.notifyProtocolGoroutine(notifyError) } } @@ -2609,7 +2634,7 @@ func (e *endpoint) receiveBufferSize() int { } func (e *endpoint) maxReceiveBufferSize() int { - var rs tcpip.StackReceiveBufferSizeOption + var rs ReceiveBufferSizeOption if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err != nil { // As a fallback return the hardcoded max buffer size. return MaxBufferSize @@ -2690,7 +2715,7 @@ func timeStampOffset() uint32 { // if the SYN options indicate that the SACK option was negotiated and the TCP // stack is configured to enable TCP SACK option. func (e *endpoint) maybeEnableSACKPermitted(synOpts *header.TCPSynOptions) { - var v tcpip.StackSACKEnabled + var v SACKEnabled if err := e.stack.TransportProtocolOption(ProtocolNumber, &v); err != nil { // Stack doesn't support SACK. So just return. return diff --git a/pkg/tcpip/transport/tcp/endpoint_state.go b/pkg/tcpip/transport/tcp/endpoint_state.go index 0bebec2d1..abf1ac5c9 100644 --- a/pkg/tcpip/transport/tcp/endpoint_state.go +++ b/pkg/tcpip/transport/tcp/endpoint_state.go @@ -93,10 +93,6 @@ func (e *endpoint) beforeSave() { if e.waiterQueue != nil && !e.waiterQueue.IsEmpty() { panic("endpoint still has waiters upon save") } - - if e.EndpointState() != StateClose && !((e.EndpointState() == StateBound || e.EndpointState() == StateListen) == e.isPortReserved) { - panic("endpoints which are not in the closed state must have a reserved port IFF they are in bound or listen state") - } } // saveAcceptedChan is invoked by stateify. @@ -186,26 +182,33 @@ func (e *endpoint) Resume(s *stack.Stack) { epState := e.origEndpointState switch epState { case StateInitial, StateBound, StateListen, StateConnecting, StateEstablished: - var ss tcpip.StackSendBufferSizeOption + var ss SendBufferSizeOption if err := e.stack.TransportProtocolOption(ProtocolNumber, &ss); err == nil { if e.sndBufSize < ss.Min || e.sndBufSize > ss.Max { panic(fmt.Sprintf("endpoint.sndBufSize %d is outside the min and max allowed [%d, %d]", e.sndBufSize, ss.Min, ss.Max)) } - if e.rcvBufSize < ss.Min || e.rcvBufSize > ss.Max { - panic(fmt.Sprintf("endpoint.rcvBufSize %d is outside the min and max allowed [%d, %d]", e.rcvBufSize, ss.Min, ss.Max)) + } + + var rs ReceiveBufferSizeOption + if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err == nil { + if e.rcvBufSize < rs.Min || e.rcvBufSize > rs.Max { + panic(fmt.Sprintf("endpoint.rcvBufSize %d is outside the min and max allowed [%d, %d]", e.rcvBufSize, rs.Min, rs.Max)) } } } bind := func() { - if len(e.BindAddr) == 0 { - e.BindAddr = e.ID.LocalAddress + addr, _, err := e.checkV4MappedLocked(tcpip.FullAddress{Addr: e.BindAddr, Port: e.ID.LocalPort}) + if err != nil { + panic("unable to parse BindAddr: " + err.String()) } - addr := e.BindAddr - port := e.ID.LocalPort - if err := e.Bind(tcpip.FullAddress{Addr: addr, Port: port}); err != nil { - panic(fmt.Sprintf("endpoint binding [%v]:%d failed: %v", addr, port, err)) + if ok := e.stack.ReserveTuple(e.effectiveNetProtos, ProtocolNumber, addr.Addr, addr.Port, e.boundPortFlags, e.boundBindToDevice, e.boundDest); !ok { + panic(fmt.Sprintf("unable to re-reserve tuple (%v, %q, %d, %+v, %d, %v)", e.effectiveNetProtos, addr.Addr, addr.Port, e.boundPortFlags, e.boundBindToDevice, e.boundDest)) } + e.isPortReserved = true + + // Mark endpoint as bound. + e.setEndpointState(StateBound) } switch { @@ -277,17 +280,7 @@ func (e *endpoint) Resume(s *stack.Stack) { tcpip.AsyncLoading.Done() }() case epState == StateClose: - if e.isPortReserved { - tcpip.AsyncLoading.Add(1) - go func() { - connectedLoading.Wait() - listenLoading.Wait() - connectingLoading.Wait() - bind() - e.setEndpointState(StateClose) - tcpip.AsyncLoading.Done() - }() - } + e.isPortReserved = false e.state = StateClose e.stack.CompleteTransportEndpointCleanup(e) tcpip.DeleteDanglingEndpoint(e) diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index 3cff55afa..b34e47bbd 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -76,6 +76,31 @@ const ( ccCubic = "cubic" ) +// SACKEnabled is used by stack.(*Stack).TransportProtocolOption to +// enable/disable SACK support in TCP. See: https://tools.ietf.org/html/rfc2018. +type SACKEnabled bool + +// DelayEnabled is used by stack.(Stack*).TransportProtocolOption to +// enable/disable Nagle's algorithm in TCP. +type DelayEnabled bool + +// SendBufferSizeOption is used by stack.(Stack*).TransportProtocolOption +// to get/set the default, min and max TCP send buffer sizes. +type SendBufferSizeOption struct { + Min int + Default int + Max int +} + +// ReceiveBufferSizeOption is used by +// stack.(Stack*).TransportProtocolOption to get/set the default, min and max +// TCP receive buffer sizes. +type ReceiveBufferSizeOption struct { + Min int + Default int + Max int +} + // syncRcvdCounter tracks the number of endpoints in the SYN-RCVD state. The // value is protected by a mutex so that we can increment only when it's // guaranteed not to go above a threshold. @@ -137,8 +162,8 @@ type protocol struct { mu sync.RWMutex sackEnabled bool delayEnabled bool - sendBufferSize tcpip.StackSendBufferSizeOption - recvBufferSize tcpip.StackReceiveBufferSizeOption + sendBufferSize SendBufferSizeOption + recvBufferSize ReceiveBufferSizeOption congestionControl string availableCongestionControl []string moderateReceiveBuffer bool @@ -149,7 +174,7 @@ type protocol struct { maxRetries uint32 synRcvdCount synRcvdCounter synRetries uint8 - dispatcher *dispatcher + dispatcher dispatcher } // Number returns the tcp protocol number. @@ -249,19 +274,19 @@ func replyWithReset(s *segment, tos, ttl uint8) { // SetOption implements stack.TransportProtocol.SetOption. func (p *protocol) SetOption(option interface{}) *tcpip.Error { switch v := option.(type) { - case tcpip.StackSACKEnabled: + case SACKEnabled: p.mu.Lock() p.sackEnabled = bool(v) p.mu.Unlock() return nil - case tcpip.StackDelayEnabled: + case DelayEnabled: p.mu.Lock() p.delayEnabled = bool(v) p.mu.Unlock() return nil - case tcpip.StackSendBufferSizeOption: + case SendBufferSizeOption: if v.Min <= 0 || v.Default < v.Min || v.Default > v.Max { return tcpip.ErrInvalidOptionValue } @@ -270,7 +295,7 @@ func (p *protocol) SetOption(option interface{}) *tcpip.Error { p.mu.Unlock() return nil - case tcpip.StackReceiveBufferSizeOption: + case ReceiveBufferSizeOption: if v.Min <= 0 || v.Default < v.Min || v.Default > v.Max { return tcpip.ErrInvalidOptionValue } @@ -363,25 +388,25 @@ func (p *protocol) SetOption(option interface{}) *tcpip.Error { // Option implements stack.TransportProtocol.Option. func (p *protocol) Option(option interface{}) *tcpip.Error { switch v := option.(type) { - case *tcpip.StackSACKEnabled: + case *SACKEnabled: p.mu.RLock() - *v = tcpip.StackSACKEnabled(p.sackEnabled) + *v = SACKEnabled(p.sackEnabled) p.mu.RUnlock() return nil - case *tcpip.StackDelayEnabled: + case *DelayEnabled: p.mu.RLock() - *v = tcpip.StackDelayEnabled(p.delayEnabled) + *v = DelayEnabled(p.delayEnabled) p.mu.RUnlock() return nil - case *tcpip.StackSendBufferSizeOption: + case *SendBufferSizeOption: p.mu.RLock() *v = p.sendBufferSize p.mu.RUnlock() return nil - case *tcpip.StackReceiveBufferSizeOption: + case *ReceiveBufferSizeOption: p.mu.RLock() *v = p.recvBufferSize p.mu.RUnlock() @@ -490,13 +515,13 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) bool { // NewProtocol returns a TCP transport protocol. func NewProtocol() stack.TransportProtocol { - return &protocol{ - sendBufferSize: tcpip.StackSendBufferSizeOption{ + p := protocol{ + sendBufferSize: SendBufferSizeOption{ Min: MinBufferSize, Default: DefaultSendBufferSize, Max: MaxBufferSize, }, - recvBufferSize: tcpip.StackReceiveBufferSizeOption{ + recvBufferSize: ReceiveBufferSizeOption{ Min: MinBufferSize, Default: DefaultReceiveBufferSize, Max: MaxBufferSize, @@ -506,10 +531,11 @@ func NewProtocol() stack.TransportProtocol { tcpLingerTimeout: DefaultTCPLingerTimeout, tcpTimeWaitTimeout: DefaultTCPTimeWaitTimeout, synRcvdCount: synRcvdCounter{threshold: SynRcvdCountThreshold}, - dispatcher: newDispatcher(runtime.GOMAXPROCS(0)), synRetries: DefaultSynRetries, minRTO: MinRTO, maxRTO: MaxRTO, maxRetries: MaxRetries, } + p.dispatcher.init(runtime.GOMAXPROCS(0)) + return &p } diff --git a/pkg/tcpip/transport/tcp/rcv.go b/pkg/tcpip/transport/tcp/rcv.go index dd89a292a..5e0bfe585 100644 --- a/pkg/tcpip/transport/tcp/rcv.go +++ b/pkg/tcpip/transport/tcp/rcv.go @@ -372,7 +372,7 @@ func (r *receiver) handleRcvdSegment(s *segment) (drop bool, err *tcpip.Error) { // We only store the segment if it's within our buffer // size limit. if r.pendingBufUsed < r.pendingBufSize { - r.pendingBufUsed += s.logicalLen() + r.pendingBufUsed += seqnum.Size(s.segMemSize()) s.incRef() heap.Push(&r.pendingRcvdSegments, s) UpdateSACKBlocks(&r.ep.sack, segSeq, segSeq.Add(segLen), r.rcvNxt) @@ -406,7 +406,7 @@ func (r *receiver) handleRcvdSegment(s *segment) (drop bool, err *tcpip.Error) { } heap.Pop(&r.pendingRcvdSegments) - r.pendingBufUsed -= s.logicalLen() + r.pendingBufUsed -= seqnum.Size(s.segMemSize()) s.decRef() } return false, nil diff --git a/pkg/tcpip/transport/tcp/segment.go b/pkg/tcpip/transport/tcp/segment.go index 0280892a8..bb60dc29d 100644 --- a/pkg/tcpip/transport/tcp/segment.go +++ b/pkg/tcpip/transport/tcp/segment.go @@ -138,6 +138,12 @@ func (s *segment) logicalLen() seqnum.Size { return l } +// segMemSize is the amount of memory used to hold the segment data and +// the associated metadata. +func (s *segment) segMemSize() int { + return segSize + s.data.Size() +} + // parse populates the sequence & ack numbers, flags, and window fields of the // segment from the TCP header stored in the data. It then updates the view to // skip the header. diff --git a/pkg/tcpip/transport/tcp/segment_unsafe.go b/pkg/tcpip/transport/tcp/segment_unsafe.go new file mode 100644 index 000000000..0ab7b8f56 --- /dev/null +++ b/pkg/tcpip/transport/tcp/segment_unsafe.go @@ -0,0 +1,23 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tcp + +import ( + "unsafe" +) + +const ( + segSize = int(unsafe.Sizeof(segment{})) +) diff --git a/pkg/tcpip/transport/tcp/snd.go b/pkg/tcpip/transport/tcp/snd.go index acacb42e4..5862c32f2 100644 --- a/pkg/tcpip/transport/tcp/snd.go +++ b/pkg/tcpip/transport/tcp/snd.go @@ -833,25 +833,6 @@ func (s *sender) maybeSendSegment(seg *segment, limit int, end seqnum.Value) (se panic("Netstack queues FIN segments without data.") } - segEnd = seg.sequenceNumber.Add(seqnum.Size(seg.data.Size())) - // If the entire segment cannot be accomodated in the receiver - // advertized window, skip splitting and sending of the segment. - // ref: net/ipv4/tcp_output.c::tcp_snd_wnd_test() - // - // Linux checks this for all segment transmits not triggered - // by a probe timer. On this condition, it defers the segment - // split and transmit to a short probe timer. - // ref: include/net/tcp.h::tcp_check_probe_timer() - // ref: net/ipv4/tcp_output.c::tcp_write_wakeup() - // - // Instead of defining a new transmit timer, we attempt to split the - // segment right here if there are no pending segments. - // If there are pending segments, segment transmits are deferred - // to the retransmit timer handler. - if s.sndUna != s.sndNxt && !segEnd.LessThan(end) { - return false - } - if !seg.sequenceNumber.LessThan(end) { return false } @@ -861,14 +842,48 @@ func (s *sender) maybeSendSegment(seg *segment, limit int, end seqnum.Value) (se return false } - // The segment size limit is computed as a function of sender congestion - // window and MSS. When sender congestion window is > 1, this limit can - // be larger than MSS. Ensure that the currently available send space - // is not greater than minimum of this limit and MSS. + // If the whole segment or at least 1MSS sized segment cannot + // be accomodated in the receiver advertized window, skip + // splitting and sending of the segment. ref: + // net/ipv4/tcp_output.c::tcp_snd_wnd_test() + // + // Linux checks this for all segment transmits not triggered by + // a probe timer. On this condition, it defers the segment split + // and transmit to a short probe timer. + // + // ref: include/net/tcp.h::tcp_check_probe_timer() + // ref: net/ipv4/tcp_output.c::tcp_write_wakeup() + // + // Instead of defining a new transmit timer, we attempt to split + // the segment right here if there are no pending segments. If + // there are pending segments, segment transmits are deferred to + // the retransmit timer handler. + if s.sndUna != s.sndNxt { + switch { + case available >= seg.data.Size(): + // OK to send, the whole segments fits in the + // receiver's advertised window. + case available >= s.maxPayloadSize: + // OK to send, at least 1 MSS sized segment fits + // in the receiver's advertised window. + default: + return false + } + } + + // The segment size limit is computed as a function of sender + // congestion window and MSS. When sender congestion window is > + // 1, this limit can be larger than MSS. Ensure that the + // currently available send space is not greater than minimum of + // this limit and MSS. if available > limit { available = limit } - if available > s.maxPayloadSize { + + // If GSO is not in use then cap available to + // maxPayloadSize. When GSO is in use the gVisor GSO logic or + // the host GSO logic will cap the segment to the correct size. + if s.ep.gso == nil && available > s.maxPayloadSize { available = s.maxPayloadSize } diff --git a/pkg/tcpip/transport/tcp/tcp_sack_test.go b/pkg/tcpip/transport/tcp/tcp_sack_test.go index 812e503bc..99521f0c1 100644 --- a/pkg/tcpip/transport/tcp/tcp_sack_test.go +++ b/pkg/tcpip/transport/tcp/tcp_sack_test.go @@ -46,8 +46,8 @@ func createConnectedWithSACKAndTS(c *context.Context) *context.RawEndpoint { func setStackSACKPermitted(t *testing.T, c *context.Context, enable bool) { t.Helper() - if err := c.Stack().SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackSACKEnabled(enable)); err != nil { - t.Fatalf("c.s.SetTransportProtocolOption(tcp.ProtocolNumber, StackSACKEnabled(%t) = %s", enable, err) + if err := c.Stack().SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SACKEnabled(enable)); err != nil { + t.Fatalf("c.s.SetTransportProtocolOption(tcp.ProtocolNumber, SACKEnabled(%t) = %s", enable, err) } } diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index aca6a7951..e67ec42b1 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -3095,6 +3095,63 @@ func TestMaxRTO(t *testing.T) { } } +// TestRetransmitIPv4IDUniqueness tests that the IPv4 Identification field is +// unique on retransmits. +func TestRetransmitIPv4IDUniqueness(t *testing.T) { + for _, tc := range []struct { + name string + size int + }{ + {"1Byte", 1}, + {"512Bytes", 512}, + } { + t.Run(tc.name, func(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.CreateConnected(789 /* iss */, 30000 /* rcvWnd */, -1 /* epRcvBuf */) + + // Disabling PMTU discovery causes all packets sent from this socket to + // have DF=0. This needs to be done because the IPv4 ID uniqueness + // applies only to non-atomic IPv4 datagrams as defined in RFC 6864 + // Section 4, and datagrams with DF=0 are non-atomic. + if err := c.EP.SetSockOptInt(tcpip.MTUDiscoverOption, tcpip.PMTUDiscoveryDont); err != nil { + t.Fatalf("disabling PMTU discovery via sockopt to force DF=0 failed: %s", err) + } + + if _, _, err := c.EP.Write(tcpip.SlicePayload(buffer.NewView(tc.size)), tcpip.WriteOptions{}); err != nil { + t.Fatalf("Write failed: %s", err) + } + pkt := c.GetPacket() + checker.IPv4(t, pkt, + checker.FragmentFlags(0), + checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), + ), + ) + idSet := map[uint16]struct{}{header.IPv4(pkt).ID(): struct{}{}} + // Expect two retransmitted packets, and that all packets received have + // unique IPv4 ID values. + for i := 0; i <= 2; i++ { + pkt := c.GetPacket() + checker.IPv4(t, pkt, + checker.FragmentFlags(0), + checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), + ), + ) + id := header.IPv4(pkt).ID() + if _, exists := idSet[id]; exists { + t.Fatalf("duplicate IPv4 ID=%d found in retransmitted packet", id) + } + idSet[id] = struct{}{} + } + }) + } +} + func TestFinImmediately(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() @@ -3879,6 +3936,9 @@ func TestReusePort(t *testing.T) { if err != nil { t.Fatalf("NewEndpoint failed; %s", err) } + if err := c.EP.SetSockOptBool(tcpip.ReuseAddressOption, true); err != nil { + t.Fatalf("SetSockOptBool ReuseAddressOption failed: %s", err) + } if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { t.Fatalf("Bind failed: %s", err) } @@ -3888,6 +3948,9 @@ func TestReusePort(t *testing.T) { if err != nil { t.Fatalf("NewEndpoint failed; %s", err) } + if err := c.EP.SetSockOptBool(tcpip.ReuseAddressOption, true); err != nil { + t.Fatalf("SetSockOptBool ReuseAddressOption failed: %s", err) + } if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { t.Fatalf("Bind failed: %s", err) } @@ -3898,6 +3961,9 @@ func TestReusePort(t *testing.T) { if err != nil { t.Fatalf("NewEndpoint failed; %s", err) } + if err := c.EP.SetSockOptBool(tcpip.ReuseAddressOption, true); err != nil { + t.Fatalf("SetSockOptBool ReuseAddressOption failed: %s", err) + } if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { t.Fatalf("Bind failed: %s", err) } @@ -3910,6 +3976,9 @@ func TestReusePort(t *testing.T) { if err != nil { t.Fatalf("NewEndpoint failed; %s", err) } + if err := c.EP.SetSockOptBool(tcpip.ReuseAddressOption, true); err != nil { + t.Fatalf("SetSockOptBool ReuseAddressOption failed: %s", err) + } if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { t.Fatalf("Bind failed: %s", err) } @@ -3920,6 +3989,9 @@ func TestReusePort(t *testing.T) { if err != nil { t.Fatalf("NewEndpoint failed; %s", err) } + if err := c.EP.SetSockOptBool(tcpip.ReuseAddressOption, true); err != nil { + t.Fatalf("SetSockOptBool ReuseAddressOption failed: %s", err) + } if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { t.Fatalf("Bind failed: %s", err) } @@ -3932,6 +4004,9 @@ func TestReusePort(t *testing.T) { if err != nil { t.Fatalf("NewEndpoint failed; %s", err) } + if err := c.EP.SetSockOptBool(tcpip.ReuseAddressOption, true); err != nil { + t.Fatalf("SetSockOptBool ReuseAddressOption failed: %s", err) + } if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { t.Fatalf("Bind failed: %s", err) } @@ -3987,7 +4062,7 @@ func TestDefaultBufferSizes(t *testing.T) { checkRecvBufferSize(t, ep, tcp.DefaultReceiveBufferSize) // Change the default send buffer size. - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackSendBufferSizeOption{ + if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SendBufferSizeOption{ Min: 1, Default: tcp.DefaultSendBufferSize * 2, Max: tcp.DefaultSendBufferSize * 20}); err != nil { @@ -4004,7 +4079,7 @@ func TestDefaultBufferSizes(t *testing.T) { checkRecvBufferSize(t, ep, tcp.DefaultReceiveBufferSize) // Change the default receive buffer size. - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackReceiveBufferSizeOption{ + if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.ReceiveBufferSizeOption{ Min: 1, Default: tcp.DefaultReceiveBufferSize * 3, Max: tcp.DefaultReceiveBufferSize * 30}); err != nil { @@ -4035,11 +4110,11 @@ func TestMinMaxBufferSizes(t *testing.T) { defer ep.Close() // Change the min/max values for send/receive - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackReceiveBufferSizeOption{Min: 200, Default: tcp.DefaultReceiveBufferSize * 2, Max: tcp.DefaultReceiveBufferSize * 20}); err != nil { + if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.ReceiveBufferSizeOption{Min: 200, Default: tcp.DefaultReceiveBufferSize * 2, Max: tcp.DefaultReceiveBufferSize * 20}); err != nil { t.Fatalf("SetTransportProtocolOption failed: %s", err) } - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackSendBufferSizeOption{Min: 300, Default: tcp.DefaultSendBufferSize * 3, Max: tcp.DefaultSendBufferSize * 30}); err != nil { + if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SendBufferSizeOption{Min: 300, Default: tcp.DefaultSendBufferSize * 3, Max: tcp.DefaultSendBufferSize * 30}); err != nil { t.Fatalf("SetTransportProtocolOption failed: %s", err) } @@ -5678,7 +5753,7 @@ func TestReceiveBufferAutoTuningApplicationLimited(t *testing.T) { // the segment queue holding unprocessed packets is limited to 500. const receiveBufferSize = 80 << 10 // 80KB. const maxReceiveBufferSize = receiveBufferSize * 10 - if err := stk.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackReceiveBufferSizeOption{Min: 1, Default: receiveBufferSize, Max: maxReceiveBufferSize}); err != nil { + if err := stk.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.ReceiveBufferSizeOption{Min: 1, Default: receiveBufferSize, Max: maxReceiveBufferSize}); err != nil { t.Fatalf("SetTransportProtocolOption failed: %s", err) } @@ -5799,7 +5874,7 @@ func TestReceiveBufferAutoTuning(t *testing.T) { // the segment queue holding unprocessed packets is limited to 300. const receiveBufferSize = 80 << 10 // 80KB. const maxReceiveBufferSize = receiveBufferSize * 10 - if err := stk.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackReceiveBufferSizeOption{Min: 1, Default: receiveBufferSize, Max: maxReceiveBufferSize}); err != nil { + if err := stk.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.ReceiveBufferSizeOption{Min: 1, Default: receiveBufferSize, Max: maxReceiveBufferSize}); err != nil { t.Fatalf("SetTransportProtocolOption failed: %s", err) } @@ -5941,7 +6016,7 @@ func TestDelayEnabled(t *testing.T) { checkDelayOption(t, c, false, false) // Delay is disabled by default. for _, v := range []struct { - delayEnabled tcpip.StackDelayEnabled + delayEnabled tcp.DelayEnabled wantDelayOption bool }{ {delayEnabled: false, wantDelayOption: false}, @@ -5956,10 +6031,10 @@ func TestDelayEnabled(t *testing.T) { } } -func checkDelayOption(t *testing.T, c *context.Context, wantDelayEnabled tcpip.StackDelayEnabled, wantDelayOption bool) { +func checkDelayOption(t *testing.T, c *context.Context, wantDelayEnabled tcp.DelayEnabled, wantDelayOption bool) { t.Helper() - var gotDelayEnabled tcpip.StackDelayEnabled + var gotDelayEnabled tcp.DelayEnabled if err := c.Stack().TransportProtocolOption(tcp.ProtocolNumber, &gotDelayEnabled); err != nil { t.Fatalf("TransportProtocolOption(tcp, &gotDelayEnabled) failed: %s", err) } diff --git a/pkg/tcpip/transport/tcp/testing/context/context.go b/pkg/tcpip/transport/tcp/testing/context/context.go index 9e262c272..37e7767d6 100644 --- a/pkg/tcpip/transport/tcp/testing/context/context.go +++ b/pkg/tcpip/transport/tcp/testing/context/context.go @@ -143,12 +143,14 @@ func New(t *testing.T, mtu uint32) *Context { TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, }) + const sendBufferSize = 1 << 20 // 1 MiB + const recvBufferSize = 1 << 20 // 1 MiB // Allow minimum send/receive buffer sizes to be 1 during tests. - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackSendBufferSizeOption{Min: 1, Default: tcp.DefaultSendBufferSize, Max: 10 * tcp.DefaultSendBufferSize}); err != nil { + if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SendBufferSizeOption{Min: 1, Default: sendBufferSize, Max: 10 * sendBufferSize}); err != nil { t.Fatalf("SetTransportProtocolOption failed: %s", err) } - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackReceiveBufferSizeOption{Min: 1, Default: tcp.DefaultReceiveBufferSize, Max: 10 * tcp.DefaultReceiveBufferSize}); err != nil { + if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.ReceiveBufferSizeOption{Min: 1, Default: recvBufferSize, Max: 10 * recvBufferSize}); err != nil { t.Fatalf("SetTransportProtocolOption failed: %s", err) } @@ -202,7 +204,7 @@ func New(t *testing.T, mtu uint32) *Context { t: t, s: s, linkEP: ep, - WindowScale: uint8(tcp.FindWndScale(tcp.DefaultReceiveBufferSize)), + WindowScale: uint8(tcp.FindWndScale(recvBufferSize)), } } @@ -1091,7 +1093,7 @@ func (c *Context) PassiveConnectWithOptions(maxPayload, wndScale int, synOptions // SACKEnabled returns true if the TCP Protocol option SACKEnabled is set to true // for the Stack in the context. func (c *Context) SACKEnabled() bool { - var v tcpip.StackSACKEnabled + var v tcp.SACKEnabled if err := c.Stack().TransportProtocolOption(tcp.ProtocolNumber, &v); err != nil { // Stack doesn't support SACK. So just return. return false diff --git a/pkg/tcpip/transport/tcpconntrack/tcp_conntrack.go b/pkg/tcpip/transport/tcpconntrack/tcp_conntrack.go index 12bc1b5b5..558b06df0 100644 --- a/pkg/tcpip/transport/tcpconntrack/tcp_conntrack.go +++ b/pkg/tcpip/transport/tcpconntrack/tcp_conntrack.go @@ -106,6 +106,11 @@ func (t *TCB) UpdateStateOutbound(tcp header.TCP) Result { return st } +// State returns the current state of the TCB. +func (t *TCB) State() Result { + return t.state +} + // IsAlive returns true as long as the connection is established(Alive) // or connecting state. func (t *TCB) IsAlive() bool { diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index f51988047..b7d735889 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -109,6 +109,7 @@ type endpoint struct { portFlags ports.Flags bindToDevice tcpip.NICID broadcast bool + noChecksum bool lastErrorMu sync.Mutex `state:"nosave"` lastError *tcpip.Error `state:".(string)"` @@ -190,13 +191,13 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue } // Override with stack defaults. - var ss tcpip.StackSendBufferSizeOption - if err := s.TransportProtocolOption(ProtocolNumber, &ss); err == nil { + var ss stack.SendBufferSizeOption + if err := s.Option(&ss); err == nil { e.sndBufSizeMax = ss.Default } - var rs tcpip.StackReceiveBufferSizeOption - if err := s.TransportProtocolOption(ProtocolNumber, &rs); err == nil { + var rs stack.ReceiveBufferSizeOption + if err := s.Option(&rs); err == nil { e.rcvBufSizeMax = rs.Default } @@ -231,7 +232,7 @@ func (e *endpoint) Close() { switch e.state { case StateBound, StateConnected: e.stack.UnregisterTransportEndpoint(e.RegisterNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.boundPortFlags, e.boundBindToDevice) - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.boundPortFlags, e.boundBindToDevice) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.boundPortFlags, e.boundBindToDevice, tcpip.FullAddress{}) e.boundBindToDevice = 0 e.boundPortFlags = ports.Flags{} } @@ -482,10 +483,6 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c nicID = e.BindNICID } - if to.Addr == header.IPv4Broadcast && !e.broadcast { - return 0, nil, tcpip.ErrBroadcastDisabled - } - dst, netProto, err := e.checkV4MappedLocked(*to) if err != nil { return 0, nil, err @@ -502,6 +499,10 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c resolve = route.Resolve } + if !e.broadcast && route.IsBroadcast() { + return 0, nil, tcpip.ErrBroadcastDisabled + } + if route.IsResolutionRequired() { if ch, err := resolve(nil); err != nil { if err == tcpip.ErrWouldBlock { @@ -529,7 +530,7 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c useDefaultTTL = false } - if err := sendUDP(route, buffer.View(v).ToVectorisedView(), e.ID.LocalPort, dstPort, ttl, useDefaultTTL, e.sendTOS, e.owner); err != nil { + if err := sendUDP(route, buffer.View(v).ToVectorisedView(), e.ID.LocalPort, dstPort, ttl, useDefaultTTL, e.sendTOS, e.owner, e.noChecksum); err != nil { return 0, nil, err } return int64(len(v)), nil, nil @@ -553,6 +554,11 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { e.multicastLoop = v e.mu.Unlock() + case tcpip.NoChecksumOption: + e.mu.Lock() + e.noChecksum = v + e.mu.Unlock() + case tcpip.ReceiveTOSOption: e.mu.Lock() e.receiveTOS = v @@ -606,6 +612,13 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { // SetSockOptInt implements tcpip.Endpoint.SetSockOptInt. func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { switch opt { + case tcpip.MTUDiscoverOption: + // Return not supported if the value is not disabling path + // MTU discovery. + if v != tcpip.PMTUDiscoveryDont { + return tcpip.ErrNotSupported + } + case tcpip.MulticastTTLOption: e.mu.Lock() e.multicastTTL = uint8(v) @@ -629,9 +642,9 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { case tcpip.ReceiveBufferSizeOption: // Make sure the receive buffer size is within the min and max // allowed. - var rs tcpip.StackReceiveBufferSizeOption - if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err != nil { - panic(fmt.Sprintf("e.stack.TransportProtocolOption(%d, %+v) = %s", ProtocolNumber, rs, err)) + var rs stack.ReceiveBufferSizeOption + if err := e.stack.Option(&rs); err != nil { + panic(fmt.Sprintf("e.stack.Option(%#v) = %s", rs, err)) } if v < rs.Min { @@ -648,9 +661,9 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { case tcpip.SendBufferSizeOption: // Make sure the send buffer size is within the min and max // allowed. - var ss tcpip.StackSendBufferSizeOption - if err := e.stack.TransportProtocolOption(ProtocolNumber, &ss); err != nil { - panic(fmt.Sprintf("e.stack.TransportProtocolOption(%d, %+v) = %s", ProtocolNumber, ss, err)) + var ss stack.SendBufferSizeOption + if err := e.stack.Option(&ss); err != nil { + panic(fmt.Sprintf("e.stack.Option(%#v) = %s", ss, err)) } if v < ss.Min { @@ -803,6 +816,9 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.mu.Lock() e.bindToDevice = id e.mu.Unlock() + + case tcpip.SocketDetachFilterOption: + return nil } return nil } @@ -825,6 +841,12 @@ func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { e.mu.RUnlock() return v, nil + case tcpip.NoChecksumOption: + e.mu.RLock() + v := e.noChecksum + e.mu.RUnlock() + return v, nil + case tcpip.ReceiveTOSOption: e.mu.RLock() v := e.receiveTOS @@ -894,6 +916,10 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { e.mu.RUnlock() return v, nil + case tcpip.MTUDiscoverOption: + // The only supported setting is path MTU discovery disabled. + return tcpip.PMTUDiscoveryDont, nil + case tcpip.MulticastTTLOption: e.mu.Lock() v := int(e.multicastTTL) @@ -959,7 +985,7 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { // sendUDP sends a UDP segment via the provided network endpoint and under the // provided identity. -func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort uint16, ttl uint8, useDefaultTTL bool, tos uint8, owner tcpip.PacketOwner) *tcpip.Error { +func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort uint16, ttl uint8, useDefaultTTL bool, tos uint8, owner tcpip.PacketOwner, noChecksum bool) *tcpip.Error { // Allocate a buffer for the UDP header. hdr := buffer.NewPrependable(header.UDPMinimumSize + int(r.MaxHeaderLength())) @@ -973,8 +999,12 @@ func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort u Length: length, }) - // Only calculate the checksum if offloading isn't supported. - if r.Capabilities()&stack.CapabilityTXChecksumOffload == 0 { + // Set the checksum field unless TX checksum offload is enabled. + // On IPv4, UDP checksum is optional, and a zero value indicates the + // transmitter skipped the checksum generation (RFC768). + // On IPv6, UDP checksum is not optional (RFC2460 Section 8.1). + if r.Capabilities()&stack.CapabilityTXChecksumOffload == 0 && + (!noChecksum || r.NetProto == header.IPv6ProtocolNumber) { xsum := r.PseudoHeaderChecksum(ProtocolNumber, length) for _, v := range data.Views() { xsum = header.Checksum(v, xsum) @@ -1047,7 +1077,7 @@ 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, boundPortFlags, e.boundBindToDevice) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, boundPortFlags, e.boundBindToDevice, tcpip.FullAddress{}) e.boundPortFlags = ports.Flags{} } e.state = StateInitial @@ -1198,7 +1228,7 @@ func (*endpoint) Accept() (tcpip.Endpoint, *waiter.Queue, *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.portFlags, e.bindToDevice) + port, err := e.stack.ReservePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, e.portFlags, e.bindToDevice, tcpip.FullAddress{}) if err != nil { return id, e.bindToDevice, err } @@ -1208,7 +1238,7 @@ func (e *endpoint) registerWithStack(nicID tcpip.NICID, netProtos []tcpip.Networ err := e.stack.RegisterTransportEndpoint(nicID, netProtos, ProtocolNumber, id, e, e.boundPortFlags, e.bindToDevice) if err != nil { - e.stack.ReleasePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, e.boundPortFlags, e.bindToDevice) + e.stack.ReleasePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, e.boundPortFlags, e.bindToDevice, tcpip.FullAddress{}) e.boundPortFlags = ports.Flags{} } return id, e.bindToDevice, err @@ -1350,10 +1380,37 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk return } - e.rcvMu.Lock() + // Never receive from a multicast address. + if header.IsV4MulticastAddress(id.RemoteAddress) || + header.IsV6MulticastAddress(id.RemoteAddress) { + e.stack.Stats().UDP.InvalidSourceAddress.Increment() + e.stack.Stats().IP.InvalidSourceAddressesReceived.Increment() + e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() + return + } + + // Verify checksum unless RX checksum offload is enabled. + // On IPv4, UDP checksum is optional, and a zero value means + // the transmitter omitted the checksum generation (RFC768). + // On IPv6, UDP checksum is not optional (RFC2460 Section 8.1). + if r.Capabilities()&stack.CapabilityRXChecksumOffload == 0 && + (hdr.Checksum() != 0 || r.NetProto == header.IPv6ProtocolNumber) { + xsum := r.PseudoHeaderChecksum(ProtocolNumber, hdr.Length()) + for _, v := range pkt.Data.Views() { + xsum = header.Checksum(v, xsum) + } + if hdr.CalculateChecksum(xsum) != 0xffff { + // Checksum Error. + e.stack.Stats().UDP.ChecksumErrors.Increment() + e.stats.ReceiveErrors.ChecksumErrors.Increment() + return + } + } + e.stack.Stats().UDP.PacketsReceived.Increment() e.stats.PacketsReceived.Increment() + e.rcvMu.Lock() // Drop the packet if our buffer is currently full. if !e.rcvReady || e.rcvClosed { e.rcvMu.Unlock() @@ -1394,7 +1451,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk packet.tos, _ = header.IPv6(pkt.NetworkHeader).TOS() } - packet.timestamp = e.stack.NowNanoseconds() + packet.timestamp = e.stack.Clock().NowNanoseconds() e.rcvMu.Unlock() diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go index fc93f93c0..0e7464e3a 100644 --- a/pkg/tcpip/transport/udp/protocol.go +++ b/pkg/tcpip/transport/udp/protocol.go @@ -21,7 +21,6 @@ package udp import ( - "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -50,9 +49,6 @@ const ( ) type protocol struct { - mu sync.RWMutex - sendBufferSize tcpip.StackSendBufferSizeOption - recvBufferSize tcpip.StackReceiveBufferSizeOption } // Number returns the udp protocol number. @@ -203,48 +199,12 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans // SetOption implements stack.TransportProtocol.SetOption. func (p *protocol) SetOption(option interface{}) *tcpip.Error { - switch v := option.(type) { - case tcpip.StackSendBufferSizeOption: - if v.Min <= 0 || v.Default < v.Min || v.Default > v.Max { - return tcpip.ErrInvalidOptionValue - } - p.mu.Lock() - p.sendBufferSize = v - p.mu.Unlock() - return nil - - case tcpip.StackReceiveBufferSizeOption: - if v.Min <= 0 || v.Default < v.Min || v.Default > v.Max { - return tcpip.ErrInvalidOptionValue - } - p.mu.Lock() - p.recvBufferSize = v - p.mu.Unlock() - return nil - - default: - return tcpip.ErrUnknownProtocolOption - } + return tcpip.ErrUnknownProtocolOption } // Option implements stack.TransportProtocol.Option. func (p *protocol) Option(option interface{}) *tcpip.Error { - switch v := option.(type) { - case *tcpip.StackSendBufferSizeOption: - p.mu.RLock() - *v = p.sendBufferSize - p.mu.RUnlock() - return nil - - case *tcpip.StackReceiveBufferSizeOption: - p.mu.RLock() - *v = p.recvBufferSize - p.mu.RUnlock() - return nil - - default: - return tcpip.ErrUnknownProtocolOption - } + return tcpip.ErrUnknownProtocolOption } // Close implements stack.TransportProtocol.Close. @@ -267,8 +227,5 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) bool { // NewProtocol returns a UDP transport protocol. func NewProtocol() stack.TransportProtocol { - return &protocol{ - sendBufferSize: tcpip.StackSendBufferSizeOption{Min: MinBufferSize, Default: DefaultSendBufferSize, Max: MaxBufferSize}, - recvBufferSize: tcpip.StackReceiveBufferSizeOption{Min: MinBufferSize, Default: DefaultReceiveBufferSize, Max: MaxBufferSize}, - } + return &protocol{} } diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index 313a3f117..66e8911c8 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -83,16 +83,18 @@ type header4Tuple struct { type testFlow int const ( - unicastV4 testFlow = iota // V4 unicast on a V4 socket - unicastV4in6 // V4-mapped unicast on a V6-dual socket - unicastV6 // V6 unicast on a V6 socket - unicastV6Only // V6 unicast on a V6-only socket - multicastV4 // V4 multicast on a V4 socket - multicastV4in6 // V4-mapped multicast on a V6-dual socket - multicastV6 // V6 multicast on a V6 socket - multicastV6Only // V6 multicast on a V6-only socket - broadcast // V4 broadcast on a V4 socket - broadcastIn6 // V4-mapped broadcast on a V6-dual socket + unicastV4 testFlow = iota // V4 unicast on a V4 socket + unicastV4in6 // V4-mapped unicast on a V6-dual socket + unicastV6 // V6 unicast on a V6 socket + unicastV6Only // V6 unicast on a V6-only socket + multicastV4 // V4 multicast on a V4 socket + multicastV4in6 // V4-mapped multicast on a V6-dual socket + multicastV6 // V6 multicast on a V6 socket + multicastV6Only // V6 multicast on a V6-only socket + broadcast // V4 broadcast on a V4 socket + broadcastIn6 // V4-mapped broadcast on a V6-dual socket + reverseMulticast4 // V4 multicast src. Must fail. + reverseMulticast6 // V6 multicast src. Must fail. ) func (flow testFlow) String() string { @@ -117,6 +119,10 @@ func (flow testFlow) String() string { return "broadcast" case broadcastIn6: return "broadcastIn6" + case reverseMulticast4: + return "reverseMulticast4" + case reverseMulticast6: + return "reverseMulticast6" default: return "unknown" } @@ -168,6 +174,9 @@ func (flow testFlow) header4Tuple(d packetDirection) header4Tuple { h.dstAddr.Addr = multicastV6Addr } } + if flow.isReverseMulticast() { + h.srcAddr.Addr = flow.getMcastAddr() + } return h } @@ -199,9 +208,9 @@ func (flow testFlow) netProto() tcpip.NetworkProtocolNumber { // endpoint for this flow. func (flow testFlow) sockProto() tcpip.NetworkProtocolNumber { switch flow { - case unicastV4in6, unicastV6, unicastV6Only, multicastV4in6, multicastV6, multicastV6Only, broadcastIn6: + case unicastV4in6, unicastV6, unicastV6Only, multicastV4in6, multicastV6, multicastV6Only, broadcastIn6, reverseMulticast6: return ipv6.ProtocolNumber - case unicastV4, multicastV4, broadcast: + case unicastV4, multicastV4, broadcast, reverseMulticast4: return ipv4.ProtocolNumber default: panic(fmt.Sprintf("invalid testFlow given: %d", flow)) @@ -224,7 +233,7 @@ func (flow testFlow) isV6Only() bool { switch flow { case unicastV6Only, multicastV6Only: return true - case unicastV4, unicastV4in6, unicastV6, multicastV4, multicastV4in6, multicastV6, broadcast, broadcastIn6: + case unicastV4, unicastV4in6, unicastV6, multicastV4, multicastV4in6, multicastV6, broadcast, broadcastIn6, reverseMulticast4, reverseMulticast6: return false default: panic(fmt.Sprintf("invalid testFlow given: %d", flow)) @@ -235,7 +244,7 @@ func (flow testFlow) isMulticast() bool { switch flow { case multicastV4, multicastV4in6, multicastV6, multicastV6Only: return true - case unicastV4, unicastV4in6, unicastV6, unicastV6Only, broadcast, broadcastIn6: + case unicastV4, unicastV4in6, unicastV6, unicastV6Only, broadcast, broadcastIn6, reverseMulticast4, reverseMulticast6: return false default: panic(fmt.Sprintf("invalid testFlow given: %d", flow)) @@ -246,7 +255,7 @@ func (flow testFlow) isBroadcast() bool { switch flow { case broadcast, broadcastIn6: return true - case unicastV4, unicastV4in6, unicastV6, unicastV6Only, multicastV4, multicastV4in6, multicastV6, multicastV6Only: + case unicastV4, unicastV4in6, unicastV6, unicastV6Only, multicastV4, multicastV4in6, multicastV6, multicastV6Only, reverseMulticast4, reverseMulticast6: return false default: panic(fmt.Sprintf("invalid testFlow given: %d", flow)) @@ -257,13 +266,22 @@ func (flow testFlow) isMapped() bool { switch flow { case unicastV4in6, multicastV4in6, broadcastIn6: return true - case unicastV4, unicastV6, unicastV6Only, multicastV4, multicastV6, multicastV6Only, broadcast: + case unicastV4, unicastV6, unicastV6Only, multicastV4, multicastV6, multicastV6Only, broadcast, reverseMulticast4, reverseMulticast6: return false default: panic(fmt.Sprintf("invalid testFlow given: %d", flow)) } } +func (flow testFlow) isReverseMulticast() bool { + switch flow { + case reverseMulticast4, reverseMulticast6: + return true + default: + return false + } +} + type testContext struct { t *testing.T linkEP *channel.Endpoint @@ -292,15 +310,15 @@ func newDualTestContextWithOptions(t *testing.T, mtu uint32, options stack.Optio wep = sniffer.New(ep) } if err := s.CreateNIC(1, wep); err != nil { - t.Fatalf("CreateNIC failed: %v", err) + t.Fatalf("CreateNIC failed: %s", err) } if err := s.AddAddress(1, ipv4.ProtocolNumber, stackAddr); err != nil { - t.Fatalf("AddAddress failed: %v", err) + t.Fatalf("AddAddress failed: %s", err) } if err := s.AddAddress(1, ipv6.ProtocolNumber, stackV6Addr); err != nil { - t.Fatalf("AddAddress failed: %v", err) + t.Fatalf("AddAddress failed: %s", err) } s.SetRouteTable([]tcpip.Route{ @@ -391,17 +409,21 @@ func (c *testContext) injectPacket(flow testFlow, payload []byte) { h := flow.header4Tuple(incoming) if flow.isV4() { - c.injectV4Packet(payload, &h, true /* valid */) + buf := c.buildV4Packet(payload, &h) + c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) } else { - c.injectV6Packet(payload, &h, true /* valid */) + buf := c.buildV6Packet(payload, &h) + c.linkEP.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) } } -// injectV6Packet creates a V6 test packet with the given payload and header -// values, and injects it into the link endpoint. valid indicates if the -// caller intends to inject a packet with a valid or an invalid UDP header. -// We can invalidate the header by corrupting the UDP payload length. -func (c *testContext) injectV6Packet(payload []byte, h *header4Tuple, valid bool) { +// buildV6Packet creates a V6 test packet with the given payload and header +// values in a buffer. +func (c *testContext) buildV6Packet(payload []byte, h *header4Tuple) buffer.View { // Allocate a buffer for data and headers. buf := buffer.NewView(header.UDPMinimumSize + header.IPv6MinimumSize + len(payload)) payloadStart := len(buf) - len(payload) @@ -420,16 +442,10 @@ func (c *testContext) injectV6Packet(payload []byte, h *header4Tuple, valid bool // Initialize the UDP header. u := header.UDP(buf[header.IPv6MinimumSize:]) - l := uint16(header.UDPMinimumSize + len(payload)) - if !valid { - // Change the UDP payload length to corrupt the header - // as requested by the caller. - l++ - } u.Encode(&header.UDPFields{ SrcPort: h.srcAddr.Port, DstPort: h.dstAddr.Port, - Length: l, + Length: uint16(header.UDPMinimumSize + len(payload)), }) // Calculate the UDP pseudo-header checksum. @@ -439,17 +455,12 @@ func (c *testContext) injectV6Packet(payload []byte, h *header4Tuple, valid bool xsum = header.Checksum(payload, xsum) u.SetChecksum(^u.CalculateChecksum(xsum)) - // Inject packet. - c.linkEP.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ - Data: buf.ToVectorisedView(), - }) + return buf } -// injectV4Packet creates a V4 test packet with the given payload and header -// values, and injects it into the link endpoint. valid indicates if the -// caller intends to inject a packet with a valid or an invalid UDP header. -// We can invalidate the header by corrupting the UDP payload length. -func (c *testContext) injectV4Packet(payload []byte, h *header4Tuple, valid bool) { +// buildV4Packet creates a V4 test packet with the given payload and header +// values in a buffer. +func (c *testContext) buildV4Packet(payload []byte, h *header4Tuple) buffer.View { // Allocate a buffer for data and headers. buf := buffer.NewView(header.UDPMinimumSize + header.IPv4MinimumSize + len(payload)) payloadStart := len(buf) - len(payload) @@ -483,11 +494,7 @@ func (c *testContext) injectV4Packet(payload []byte, h *header4Tuple, valid bool xsum = header.Checksum(payload, xsum) u.SetChecksum(^u.CalculateChecksum(xsum)) - // Inject packet. - - c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ - Data: buf.ToVectorisedView(), - }) + return buf } func newPayload() []byte { @@ -509,7 +516,7 @@ func TestBindToDeviceOption(t *testing.T) { ep, err := s.NewEndpoint(udp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) if err != nil { - t.Fatalf("NewEndpoint failed; %v", err) + t.Fatalf("NewEndpoint failed; %s", err) } defer ep.Close() @@ -643,7 +650,7 @@ func TestBindEphemeralPort(t *testing.T) { c.createEndpoint(ipv6.ProtocolNumber) if err := c.ep.Bind(tcpip.FullAddress{}); err != nil { - t.Fatalf("ep.Bind(...) failed: %v", err) + t.Fatalf("ep.Bind(...) failed: %s", err) } } @@ -654,19 +661,19 @@ func TestBindReservedPort(t *testing.T) { c.createEndpoint(ipv6.ProtocolNumber) if err := c.ep.Connect(tcpip.FullAddress{Addr: testV6Addr, Port: testPort}); err != nil { - c.t.Fatalf("Connect failed: %v", err) + c.t.Fatalf("Connect failed: %s", err) } addr, err := c.ep.GetLocalAddress() if err != nil { - t.Fatalf("GetLocalAddress failed: %v", err) + t.Fatalf("GetLocalAddress failed: %s", err) } // We can't bind the address reserved by the connected endpoint above. { ep, err := c.s.NewEndpoint(udp.ProtocolNumber, ipv6.ProtocolNumber, &c.wq) if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) + t.Fatalf("NewEndpoint failed: %s", err) } defer ep.Close() if got, want := ep.Bind(addr), tcpip.ErrPortInUse; got != want { @@ -677,7 +684,7 @@ func TestBindReservedPort(t *testing.T) { func() { ep, err := c.s.NewEndpoint(udp.ProtocolNumber, ipv4.ProtocolNumber, &c.wq) if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) + t.Fatalf("NewEndpoint failed: %s", err) } defer ep.Close() // We can't bind ipv4-any on the port reserved by the connected endpoint @@ -687,7 +694,7 @@ func TestBindReservedPort(t *testing.T) { } // We can bind an ipv4 address on this port, though. if err := ep.Bind(tcpip.FullAddress{Addr: stackAddr, Port: addr.Port}); err != nil { - t.Fatalf("ep.Bind(...) failed: %v", err) + t.Fatalf("ep.Bind(...) failed: %s", err) } }() @@ -697,11 +704,11 @@ func TestBindReservedPort(t *testing.T) { func() { ep, err := c.s.NewEndpoint(udp.ProtocolNumber, ipv4.ProtocolNumber, &c.wq) if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) + t.Fatalf("NewEndpoint failed: %s", err) } defer ep.Close() if err := ep.Bind(tcpip.FullAddress{Port: addr.Port}); err != nil { - t.Fatalf("ep.Bind(...) failed: %v", err) + t.Fatalf("ep.Bind(...) failed: %s", err) } }() } @@ -714,7 +721,7 @@ func TestV4ReadOnV6(t *testing.T) { // Bind to wildcard. if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } // Test acceptance. @@ -729,7 +736,7 @@ func TestV4ReadOnBoundToV4MappedWildcard(t *testing.T) { // Bind to v4 mapped wildcard. if err := c.ep.Bind(tcpip.FullAddress{Addr: v4MappedWildcardAddr, Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } // Test acceptance. @@ -744,7 +751,7 @@ func TestV4ReadOnBoundToV4Mapped(t *testing.T) { // Bind to local address. if err := c.ep.Bind(tcpip.FullAddress{Addr: stackV4MappedAddr, Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } // Test acceptance. @@ -759,7 +766,7 @@ func TestV6ReadOnV6(t *testing.T) { // Bind to wildcard. if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } // Test acceptance. @@ -796,7 +803,10 @@ func TestV4ReadSelfSource(t *testing.T) { h := unicastV4.header4Tuple(incoming) h.srcAddr = h.dstAddr - c.injectV4Packet(payload, &h, true /* valid */) + buf := c.buildV4Packet(payload, &h) + c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) if got := c.s.Stats().IP.InvalidSourceAddressesReceived.Value(); got != tt.wantInvalidSource { t.Errorf("c.s.Stats().IP.InvalidSourceAddressesReceived got %d, want %d", got, tt.wantInvalidSource) @@ -817,7 +827,7 @@ func TestV4ReadOnV4(t *testing.T) { // Bind to wildcard. if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } // Test acceptance. @@ -880,6 +890,60 @@ func TestV4ReadOnBoundToBroadcast(t *testing.T) { } } +// TestReadFromMulticast checks that an endpoint will NOT receive a packet +// that was sent with multicast SOURCE address. +func TestReadFromMulticast(t *testing.T) { + for _, flow := range []testFlow{reverseMulticast4, reverseMulticast6} { + t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpointForFlow(flow) + + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + t.Fatalf("Bind failed: %s", err) + } + testFailingRead(c, flow, false /* expectReadError */) + }) + } +} + +// TestReadFromMulticaststats checks that a discarded packet +// that that was sent with multicast SOURCE address increments +// the correct counters and that a regular packet does not. +func TestReadFromMulticastStats(t *testing.T) { + t.Helper() + for _, flow := range []testFlow{reverseMulticast4, reverseMulticast6, unicastV4} { + t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpointForFlow(flow) + + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + t.Fatalf("Bind failed: %s", err) + } + + payload := newPayload() + c.injectPacket(flow, payload) + + var want uint64 = 0 + if flow.isReverseMulticast() { + want = 1 + } + if got := c.s.Stats().IP.InvalidSourceAddressesReceived.Value(); got != want { + t.Errorf("got stats.IP.InvalidSourceAddressesReceived.Value() = %d, want = %d", got, want) + } + if got := c.s.Stats().UDP.InvalidSourceAddress.Value(); got != want { + t.Errorf("got stats.UDP.InvalidSourceAddress.Value() = %d, want = %d", got, want) + } + if got := c.ep.Stats().(*tcpip.TransportEndpointStats).ReceiveErrors.MalformedPacketsReceived.Value(); got != want { + t.Errorf("got EP Stats.ReceiveErrors.MalformedPacketsReceived stats = %d, want = %d", got, want) + } + }) + } +} + // TestV4ReadBroadcastOnBoundToWildcard checks that an endpoint can bind to ANY // and receive broadcast and unicast data. func TestV4ReadBroadcastOnBoundToWildcard(t *testing.T) { @@ -955,7 +1019,7 @@ func testWriteInternal(c *testContext, flow testFlow, setDest bool, checkers ... payload := buffer.View(newPayload()) n, _, err := c.ep.Write(tcpip.SlicePayload(payload), writeOpts) if err != nil { - c.t.Fatalf("Write failed: %v", err) + c.t.Fatalf("Write failed: %s", err) } if n != int64(len(payload)) { c.t.Fatalf("Bad number of bytes written: got %v, want %v", n, len(payload)) @@ -1005,7 +1069,7 @@ func TestDualWriteBoundToWildcard(t *testing.T) { // Bind to wildcard. if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } p := testDualWrite(c) @@ -1022,7 +1086,7 @@ func TestDualWriteConnectedToV6(t *testing.T) { // Connect to v6 address. if err := c.ep.Connect(tcpip.FullAddress{Addr: testV6Addr, Port: testPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } testWrite(c, unicastV6) @@ -1043,7 +1107,7 @@ func TestDualWriteConnectedToV4Mapped(t *testing.T) { // Connect to v4 mapped address. if err := c.ep.Connect(tcpip.FullAddress{Addr: testV4MappedAddr, Port: testPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } testWrite(c, unicastV4in6) @@ -1070,7 +1134,7 @@ func TestV6WriteOnBoundToV4Mapped(t *testing.T) { // Bind to v4 mapped address. if err := c.ep.Bind(tcpip.FullAddress{Addr: stackV4MappedAddr, Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } // Write to v6 address. @@ -1085,7 +1149,7 @@ func TestV6WriteOnConnected(t *testing.T) { // Connect to v6 address. if err := c.ep.Connect(tcpip.FullAddress{Addr: testV6Addr, Port: testPort}); err != nil { - c.t.Fatalf("Connect failed: %v", err) + c.t.Fatalf("Connect failed: %s", err) } testWriteWithoutDestination(c, unicastV6) @@ -1099,7 +1163,7 @@ func TestV4WriteOnConnected(t *testing.T) { // Connect to v4 mapped address. if err := c.ep.Connect(tcpip.FullAddress{Addr: testV4MappedAddr, Port: testPort}); err != nil { - c.t.Fatalf("Connect failed: %v", err) + c.t.Fatalf("Connect failed: %s", err) } testWriteWithoutDestination(c, unicastV4) @@ -1234,7 +1298,7 @@ func TestReadIncrementsPacketsReceived(t *testing.T) { // Bind to wildcard. if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } testRead(c, unicastV4) @@ -1259,6 +1323,30 @@ func TestWriteIncrementsPacketsSent(t *testing.T) { } } +func TestNoChecksum(t *testing.T) { + for _, flow := range []testFlow{unicastV4, unicastV6} { + t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpointForFlow(flow) + + // Disable the checksum generation. + if err := c.ep.SetSockOptBool(tcpip.NoChecksumOption, true); err != nil { + t.Fatalf("SetSockOptBool failed: %s", err) + } + // This option is effective on IPv4 only. + testWrite(c, flow, checker.UDP(checker.NoChecksum(flow.isV4()))) + + // Enable the checksum generation. + if err := c.ep.SetSockOptBool(tcpip.NoChecksumOption, false); err != nil { + t.Fatalf("SetSockOptBool failed: %s", err) + } + testWrite(c, flow, checker.UDP(checker.NoChecksum(false))) + }) + } +} + func TestTTL(t *testing.T) { for _, flow := range []testFlow{unicastV4, unicastV4in6, unicastV6, unicastV6Only, multicastV4, multicastV4in6, multicastV6, broadcast, broadcastIn6} { t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { @@ -1506,12 +1594,12 @@ func TestMulticastInterfaceOption(t *testing.T) { Port: stackPort, } if err := c.ep.Connect(addr); err != nil { - c.t.Fatalf("Connect failed: %v", err) + c.t.Fatalf("Connect failed: %s", err) } } if err := c.ep.SetSockOpt(ifoptSet); err != nil { - c.t.Fatalf("SetSockOpt failed: %v", err) + c.t.Fatalf("SetSockOpt failed: %s", err) } // Verify multicast interface addr and NIC were set correctly. @@ -1519,7 +1607,7 @@ func TestMulticastInterfaceOption(t *testing.T) { ifoptWant := tcpip.MulticastInterfaceOption{NIC: 1, InterfaceAddr: ifoptSet.InterfaceAddr} var ifoptGot tcpip.MulticastInterfaceOption if err := c.ep.GetSockOpt(&ifoptGot); err != nil { - c.t.Fatalf("GetSockOpt failed: %v", err) + c.t.Fatalf("GetSockOpt failed: %s", err) } if ifoptGot != ifoptWant { c.t.Errorf("got GetSockOpt() = %#v, want = %#v", ifoptGot, ifoptWant) @@ -1691,7 +1779,7 @@ func TestV6UnknownDestination(t *testing.T) { } // TestIncrementMalformedPacketsReceived verifies if the malformed received -// global and endpoint stats get incremented. +// global and endpoint stats are incremented. func TestIncrementMalformedPacketsReceived(t *testing.T) { c := newDualTestContext(t, defaultMTU) defer c.cleanup() @@ -1699,20 +1787,27 @@ func TestIncrementMalformedPacketsReceived(t *testing.T) { c.createEndpoint(ipv6.ProtocolNumber) // Bind to wildcard. if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } payload := newPayload() - c.t.Helper() h := unicastV6.header4Tuple(incoming) - c.injectV6Packet(payload, &h, false /* !valid */) + buf := c.buildV6Packet(payload, &h) - var want uint64 = 1 + // Invalidate the UDP header length field. + u := header.UDP(buf[header.IPv6MinimumSize:]) + u.SetLength(u.Length() + 1) + + c.linkEP.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) + + const want = 1 if got := c.s.Stats().UDP.MalformedPacketsReceived.Value(); got != want { - t.Errorf("got stats.UDP.MalformedPacketsReceived.Value() = %v, want = %v", got, want) + t.Errorf("got stats.UDP.MalformedPacketsReceived.Value() = %d, want = %d", got, want) } if got := c.ep.Stats().(*tcpip.TransportEndpointStats).ReceiveErrors.MalformedPacketsReceived.Value(); got != want { - t.Errorf("got EP Stats.ReceiveErrors.MalformedPacketsReceived stats = %v, want = %v", got, want) + t.Errorf("got EP Stats.ReceiveErrors.MalformedPacketsReceived stats = %d, want = %d", got, want) } } @@ -1728,7 +1823,6 @@ func TestShortHeader(t *testing.T) { c.t.Fatalf("Bind failed: %s", err) } - c.t.Helper() h := unicastV6.header4Tuple(incoming) // Allocate a buffer for an IPv6 and too-short UDP header. @@ -1768,6 +1862,199 @@ func TestShortHeader(t *testing.T) { } } +// TestIncrementChecksumErrorsV4 verifies if a checksum error is detected, +// global and endpoint stats are incremented. +func TestIncrementChecksumErrorsV4(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpoint(ipv4.ProtocolNumber) + // Bind to wildcard. + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + c.t.Fatalf("Bind failed: %s", err) + } + + payload := newPayload() + h := unicastV4.header4Tuple(incoming) + buf := c.buildV4Packet(payload, &h) + + // Invalidate the UDP header checksum field, taking care to avoid + // overflow to zero, which would disable checksum validation. + for u := header.UDP(buf[header.IPv4MinimumSize:]); ; { + u.SetChecksum(u.Checksum() + 1) + if u.Checksum() != 0 { + break + } + } + + c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) + + const want = 1 + if got := c.s.Stats().UDP.ChecksumErrors.Value(); got != want { + t.Errorf("got stats.UDP.ChecksumErrors.Value() = %d, want = %d", got, want) + } + if got := c.ep.Stats().(*tcpip.TransportEndpointStats).ReceiveErrors.ChecksumErrors.Value(); got != want { + t.Errorf("got EP Stats.ReceiveErrors.ChecksumErrors stats = %d, want = %d", got, want) + } +} + +// TestIncrementChecksumErrorsV6 verifies if a checksum error is detected, +// global and endpoint stats are incremented. +func TestIncrementChecksumErrorsV6(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpoint(ipv6.ProtocolNumber) + // Bind to wildcard. + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + c.t.Fatalf("Bind failed: %s", err) + } + + payload := newPayload() + h := unicastV6.header4Tuple(incoming) + buf := c.buildV6Packet(payload, &h) + + // Invalidate the UDP header checksum field. + u := header.UDP(buf[header.IPv6MinimumSize:]) + u.SetChecksum(u.Checksum() + 1) + + c.linkEP.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) + + const want = 1 + if got := c.s.Stats().UDP.ChecksumErrors.Value(); got != want { + t.Errorf("got stats.UDP.ChecksumErrors.Value() = %d, want = %d", got, want) + } + if got := c.ep.Stats().(*tcpip.TransportEndpointStats).ReceiveErrors.ChecksumErrors.Value(); got != want { + t.Errorf("got EP Stats.ReceiveErrors.ChecksumErrors stats = %d, want = %d", got, want) + } +} + +// TestPayloadModifiedV4 verifies if a checksum error is detected, +// global and endpoint stats are incremented. +func TestPayloadModifiedV4(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpoint(ipv4.ProtocolNumber) + // Bind to wildcard. + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + c.t.Fatalf("Bind failed: %s", err) + } + + payload := newPayload() + h := unicastV4.header4Tuple(incoming) + buf := c.buildV4Packet(payload, &h) + // Modify the payload so that the checksum value in the UDP header will be incorrect. + buf[len(buf)-1]++ + c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) + + const want = 1 + if got := c.s.Stats().UDP.ChecksumErrors.Value(); got != want { + t.Errorf("got stats.UDP.ChecksumErrors.Value() = %d, want = %d", got, want) + } + if got := c.ep.Stats().(*tcpip.TransportEndpointStats).ReceiveErrors.ChecksumErrors.Value(); got != want { + t.Errorf("got EP Stats.ReceiveErrors.ChecksumErrors stats = %d, want = %d", got, want) + } +} + +// TestPayloadModifiedV6 verifies if a checksum error is detected, +// global and endpoint stats are incremented. +func TestPayloadModifiedV6(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpoint(ipv6.ProtocolNumber) + // Bind to wildcard. + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + c.t.Fatalf("Bind failed: %s", err) + } + + payload := newPayload() + h := unicastV6.header4Tuple(incoming) + buf := c.buildV6Packet(payload, &h) + // Modify the payload so that the checksum value in the UDP header will be incorrect. + buf[len(buf)-1]++ + c.linkEP.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) + + const want = 1 + if got := c.s.Stats().UDP.ChecksumErrors.Value(); got != want { + t.Errorf("got stats.UDP.ChecksumErrors.Value() = %d, want = %d", got, want) + } + if got := c.ep.Stats().(*tcpip.TransportEndpointStats).ReceiveErrors.ChecksumErrors.Value(); got != want { + t.Errorf("got EP Stats.ReceiveErrors.ChecksumErrors stats = %d, want = %d", got, want) + } +} + +// TestChecksumZeroV4 verifies if the checksum value is zero, global and +// endpoint states are *not* incremented (UDP checksum is optional on IPv4). +func TestChecksumZeroV4(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpoint(ipv4.ProtocolNumber) + // Bind to wildcard. + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + c.t.Fatalf("Bind failed: %s", err) + } + + payload := newPayload() + h := unicastV4.header4Tuple(incoming) + buf := c.buildV4Packet(payload, &h) + // Set the checksum field in the UDP header to zero. + u := header.UDP(buf[header.IPv4MinimumSize:]) + u.SetChecksum(0) + c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) + + const want = 0 + if got := c.s.Stats().UDP.ChecksumErrors.Value(); got != want { + t.Errorf("got stats.UDP.ChecksumErrors.Value() = %d, want = %d", got, want) + } + if got := c.ep.Stats().(*tcpip.TransportEndpointStats).ReceiveErrors.ChecksumErrors.Value(); got != want { + t.Errorf("got EP Stats.ReceiveErrors.ChecksumErrors stats = %d, want = %d", got, want) + } +} + +// TestChecksumZeroV6 verifies if the checksum value is zero, global and +// endpoint states are incremented (UDP checksum is *not* optional on IPv6). +func TestChecksumZeroV6(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpoint(ipv6.ProtocolNumber) + // Bind to wildcard. + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + c.t.Fatalf("Bind failed: %s", err) + } + + payload := newPayload() + h := unicastV6.header4Tuple(incoming) + buf := c.buildV6Packet(payload, &h) + // Set the checksum field in the UDP header to zero. + u := header.UDP(buf[header.IPv6MinimumSize:]) + u.SetChecksum(0) + c.linkEP.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) + + const want = 1 + if got := c.s.Stats().UDP.ChecksumErrors.Value(); got != want { + t.Errorf("got stats.UDP.ChecksumErrors.Value() = %d, want = %d", got, want) + } + if got := c.ep.Stats().(*tcpip.TransportEndpointStats).ReceiveErrors.ChecksumErrors.Value(); got != want { + t.Errorf("got EP Stats.ReceiveErrors.ChecksumErrors stats = %d, want = %d", got, want) + } +} + // TestShutdownRead verifies endpoint read shutdown and error // stats increment on packet receive. func TestShutdownRead(t *testing.T) { @@ -1778,15 +2065,15 @@ func TestShutdownRead(t *testing.T) { // Bind to wildcard. if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } if err := c.ep.Connect(tcpip.FullAddress{Addr: testV6Addr, Port: testPort}); err != nil { - c.t.Fatalf("Connect failed: %v", err) + c.t.Fatalf("Connect failed: %s", err) } if err := c.ep.Shutdown(tcpip.ShutdownRead); err != nil { - t.Fatalf("Shutdown failed: %v", err) + t.Fatalf("Shutdown failed: %s", err) } testFailingRead(c, unicastV6, true /* expectReadError */) @@ -1809,11 +2096,11 @@ func TestShutdownWrite(t *testing.T) { c.createEndpoint(ipv6.ProtocolNumber) if err := c.ep.Connect(tcpip.FullAddress{Addr: testV6Addr, Port: testPort}); err != nil { - c.t.Fatalf("Connect failed: %v", err) + c.t.Fatalf("Connect failed: %s", err) } if err := c.ep.Shutdown(tcpip.ShutdownWrite); err != nil { - t.Fatalf("Shutdown failed: %v", err) + t.Fatalf("Shutdown failed: %s", err) } testFailingWrite(c, unicastV6, tcpip.ErrClosedForSend) @@ -1855,3 +2142,192 @@ func (c *testContext) checkEndpointReadStats(incr uint64, want tcpip.TransportEn c.t.Errorf("Endpoint stats not matching for error %s got %+v want %+v", err, got, want) } } + +func TestOutgoingSubnetBroadcast(t *testing.T) { + const nicID1 = 1 + + ipv4Addr := tcpip.AddressWithPrefix{ + Address: "\xc0\xa8\x01\x3a", + PrefixLen: 24, + } + ipv4Subnet := ipv4Addr.Subnet() + ipv4SubnetBcast := ipv4Subnet.Broadcast() + ipv4Gateway := tcpip.Address("\xc0\xa8\x01\x01") + ipv4AddrPrefix31 := tcpip.AddressWithPrefix{ + Address: "\xc0\xa8\x01\x3a", + PrefixLen: 31, + } + ipv4Subnet31 := ipv4AddrPrefix31.Subnet() + ipv4Subnet31Bcast := ipv4Subnet31.Broadcast() + ipv4AddrPrefix32 := tcpip.AddressWithPrefix{ + Address: "\xc0\xa8\x01\x3a", + PrefixLen: 32, + } + ipv4Subnet32 := ipv4AddrPrefix32.Subnet() + ipv4Subnet32Bcast := ipv4Subnet32.Broadcast() + ipv6Addr := tcpip.AddressWithPrefix{ + Address: "\x20\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01", + PrefixLen: 64, + } + ipv6Subnet := ipv6Addr.Subnet() + ipv6SubnetBcast := ipv6Subnet.Broadcast() + remNetAddr := tcpip.AddressWithPrefix{ + Address: "\x64\x0a\x7b\x18", + PrefixLen: 24, + } + remNetSubnet := remNetAddr.Subnet() + remNetSubnetBcast := remNetSubnet.Broadcast() + + tests := []struct { + name string + nicAddr tcpip.ProtocolAddress + routes []tcpip.Route + remoteAddr tcpip.Address + requiresBroadcastOpt bool + }{ + { + name: "IPv4 Broadcast to local subnet", + nicAddr: tcpip.ProtocolAddress{ + Protocol: header.IPv4ProtocolNumber, + AddressWithPrefix: ipv4Addr, + }, + routes: []tcpip.Route{ + { + Destination: ipv4Subnet, + NIC: nicID1, + }, + }, + remoteAddr: ipv4SubnetBcast, + requiresBroadcastOpt: true, + }, + { + name: "IPv4 Broadcast to local /31 subnet", + nicAddr: tcpip.ProtocolAddress{ + Protocol: header.IPv4ProtocolNumber, + AddressWithPrefix: ipv4AddrPrefix31, + }, + routes: []tcpip.Route{ + { + Destination: ipv4Subnet31, + NIC: nicID1, + }, + }, + remoteAddr: ipv4Subnet31Bcast, + requiresBroadcastOpt: false, + }, + { + name: "IPv4 Broadcast to local /32 subnet", + nicAddr: tcpip.ProtocolAddress{ + Protocol: header.IPv4ProtocolNumber, + AddressWithPrefix: ipv4AddrPrefix32, + }, + routes: []tcpip.Route{ + { + Destination: ipv4Subnet32, + NIC: nicID1, + }, + }, + remoteAddr: ipv4Subnet32Bcast, + requiresBroadcastOpt: false, + }, + // IPv6 has no notion of a broadcast. + { + name: "IPv6 'Broadcast' to local subnet", + nicAddr: tcpip.ProtocolAddress{ + Protocol: header.IPv6ProtocolNumber, + AddressWithPrefix: ipv6Addr, + }, + routes: []tcpip.Route{ + { + Destination: ipv6Subnet, + NIC: nicID1, + }, + }, + remoteAddr: ipv6SubnetBcast, + requiresBroadcastOpt: false, + }, + { + name: "IPv4 Broadcast to remote subnet", + nicAddr: tcpip.ProtocolAddress{ + Protocol: header.IPv4ProtocolNumber, + AddressWithPrefix: ipv4Addr, + }, + routes: []tcpip.Route{ + { + Destination: remNetSubnet, + Gateway: ipv4Gateway, + NIC: nicID1, + }, + }, + remoteAddr: remNetSubnetBcast, + requiresBroadcastOpt: true, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, + + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + }) + e := channel.New(0, defaultMTU, "") + if err := s.CreateNIC(nicID1, e); err != nil { + t.Fatalf("CreateNIC(%d, _): %s", nicID1, err) + } + if err := s.AddProtocolAddress(nicID1, test.nicAddr); err != nil { + t.Fatalf("AddProtocolAddress(%d, %+v): %s", nicID1, test.nicAddr, err) + } + + s.SetRouteTable(test.routes) + + var netProto tcpip.NetworkProtocolNumber + switch l := len(test.remoteAddr); l { + case header.IPv4AddressSize: + netProto = header.IPv4ProtocolNumber + case header.IPv6AddressSize: + netProto = header.IPv6ProtocolNumber + default: + t.Fatalf("got unexpected address length = %d bytes", l) + } + + wq := waiter.Queue{} + ep, err := s.NewEndpoint(udp.ProtocolNumber, netProto, &wq) + if err != nil { + t.Fatalf("NewEndpoint(%d, %d, _): %s", udp.ProtocolNumber, netProto, err) + } + defer ep.Close() + + data := tcpip.SlicePayload([]byte{1, 2, 3, 4}) + to := tcpip.FullAddress{ + Addr: test.remoteAddr, + Port: 80, + } + opts := tcpip.WriteOptions{To: &to} + expectedErrWithoutBcastOpt := tcpip.ErrBroadcastDisabled + if !test.requiresBroadcastOpt { + expectedErrWithoutBcastOpt = nil + } + + if n, _, err := ep.Write(data, opts); err != expectedErrWithoutBcastOpt { + t.Fatalf("got ep.Write(_, _) = (%d, _, %v), want = (_, _, %v)", n, err, expectedErrWithoutBcastOpt) + } + + if err := ep.SetSockOptBool(tcpip.BroadcastOption, true); err != nil { + t.Fatalf("got SetSockOptBool(BroadcastOption, true): %s", err) + } + + if n, _, err := ep.Write(data, opts); err != nil { + t.Fatalf("got ep.Write(_, _) = (%d, _, %s), want = (_, _, nil)", n, err) + } + + if err := ep.SetSockOptBool(tcpip.BroadcastOption, false); err != nil { + t.Fatalf("got SetSockOptBool(BroadcastOption, false): %s", err) + } + + if n, _, err := ep.Write(data, opts); err != expectedErrWithoutBcastOpt { + t.Fatalf("got ep.Write(_, _) = (%d, _, %v), want = (_, _, %v)", n, err, expectedErrWithoutBcastOpt) + } + }) + } +} |