diff options
Diffstat (limited to 'pkg/tcpip')
103 files changed, 11822 insertions, 1595 deletions
diff --git a/pkg/tcpip/BUILD b/pkg/tcpip/BUILD index 3fd9e3134..65d4d0cd8 100644 --- a/pkg/tcpip/BUILD +++ b/pkg/tcpip/BUILD @@ -1,12 +1,13 @@ load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) -load("//tools/go_stateify:defs.bzl", "go_library") - go_library( name = "tcpip", srcs = [ + "packet_buffer.go", + "packet_buffer_state.go", "tcpip.go", "time_unsafe.go", ], diff --git a/pkg/tcpip/adapters/gonet/gonet_test.go b/pkg/tcpip/adapters/gonet/gonet_test.go index 8ced960bb..ee077ae83 100644 --- a/pkg/tcpip/adapters/gonet/gonet_test.go +++ b/pkg/tcpip/adapters/gonet/gonet_test.go @@ -151,10 +151,8 @@ func TestCloseReader(t *testing.T) { buf := make([]byte, 256) n, err := c.Read(buf) - got, ok := err.(*net.OpError) - want := tcpip.ErrConnectionAborted - if n != 0 || !ok || got.Err.Error() != want.String() { - t.Errorf("c.Read() = (%d, %v), want (0, OpError(%v))", n, err, want) + if n != 0 || err != io.EOF { + t.Errorf("c.Read() = (%d, %v), want (0, EOF)", n, err) } }() sender, err := connect(s, addr) @@ -203,10 +201,8 @@ func TestCloseReaderWithForwarder(t *testing.T) { buf := make([]byte, 256) n, e := c.Read(buf) - got, ok := e.(*net.OpError) - want := tcpip.ErrConnectionAborted - if n != 0 || !ok || got.Err.Error() != want.String() { - t.Errorf("c.Read() = (%d, %v), want (0, OpError(%v))", n, e, want) + if n != 0 || e != io.EOF { + t.Errorf("c.Read() = (%d, %v), want (0, EOF)", n, e) } }) s.SetTransportProtocolHandler(tcp.ProtocolNumber, fwd.HandlePacket) diff --git a/pkg/tcpip/buffer/BUILD b/pkg/tcpip/buffer/BUILD index b4e8d6810..d6c31bfa2 100644 --- a/pkg/tcpip/buffer/BUILD +++ b/pkg/tcpip/buffer/BUILD @@ -1,9 +1,8 @@ load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) -load("//tools/go_stateify:defs.bzl", "go_library") - go_library( name = "buffer", srcs = [ diff --git a/pkg/tcpip/buffer/prependable.go b/pkg/tcpip/buffer/prependable.go index 4287464f3..48a2a2713 100644 --- a/pkg/tcpip/buffer/prependable.go +++ b/pkg/tcpip/buffer/prependable.go @@ -41,6 +41,11 @@ func NewPrependableFromView(v View) Prependable { return Prependable{buf: v, usedIdx: 0} } +// NewEmptyPrependableFromView creates a new prependable buffer from a View. +func NewEmptyPrependableFromView(v View) Prependable { + return Prependable{buf: v, usedIdx: len(v)} +} + // View returns a View of the backing buffer that contains all prepended // data so far. func (p Prependable) View() View { diff --git a/pkg/tcpip/checker/BUILD b/pkg/tcpip/checker/BUILD index 4cecfb989..b6fa6fc37 100644 --- a/pkg/tcpip/checker/BUILD +++ b/pkg/tcpip/checker/BUILD @@ -10,6 +10,7 @@ go_library( visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", + "//pkg/tcpip/buffer", "//pkg/tcpip/header", "//pkg/tcpip/seqnum", ], diff --git a/pkg/tcpip/checker/checker.go b/pkg/tcpip/checker/checker.go index 096ad71ab..2f15bf1f1 100644 --- a/pkg/tcpip/checker/checker.go +++ b/pkg/tcpip/checker/checker.go @@ -22,6 +22,7 @@ import ( "testing" "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/seqnum" ) @@ -639,6 +640,8 @@ func ICMPv4Code(want byte) TransportChecker { // ICMPv6 creates a checker that checks that the transport protocol is ICMPv6 and // potentially additional ICMPv6 header fields. +// +// ICMPv6 will validate the checksum field before calling checkers. func ICMPv6(checkers ...TransportChecker) NetworkChecker { return func(t *testing.T, h []header.Network) { t.Helper() @@ -650,6 +653,10 @@ func ICMPv6(checkers ...TransportChecker) NetworkChecker { } icmp := header.ICMPv6(last.Payload()) + if got, want := icmp.Checksum(), header.ICMPv6Checksum(icmp, last.SourceAddress(), last.DestinationAddress(), buffer.VectorisedView{}); got != want { + t.Fatalf("Bad ICMPv6 checksum; got %d, want %d", got, want) + } + for _, f := range checkers { f(t, icmp) } @@ -686,3 +693,64 @@ func ICMPv6Code(want byte) TransportChecker { } } } + +// NDP creates a checker that checks that the packet contains a valid NDP +// message for type of ty, with potentially additional checks specified by +// checkers. +// +// checkers may assume that a valid ICMPv6 is passed to it containing a valid +// NDP message as far as the size of the message (minSize) is concerned. The +// values within the message are up to checkers to validate. +func NDP(msgType header.ICMPv6Type, minSize int, checkers ...TransportChecker) NetworkChecker { + return func(t *testing.T, h []header.Network) { + t.Helper() + + // Check normal ICMPv6 first. + ICMPv6( + ICMPv6Type(msgType), + ICMPv6Code(0))(t, h) + + last := h[len(h)-1] + + icmp := header.ICMPv6(last.Payload()) + if got := len(icmp.NDPPayload()); got < minSize { + t.Fatalf("ICMPv6 NDP (type = %d) payload size of %d is less than the minimum size of %d", msgType, got, minSize) + } + + for _, f := range checkers { + f(t, icmp) + } + if t.Failed() { + t.FailNow() + } + } +} + +// NDPNS creates a checker that checks that the packet contains a valid NDP +// Neighbor Solicitation message (as per the raw wire format), with potentially +// additional checks specified by checkers. +// +// checkers may assume that a valid ICMPv6 is passed to it containing a valid +// NDPNS message as far as the size of the messages concerned. The values within +// the message are up to checkers to validate. +func NDPNS(checkers ...TransportChecker) NetworkChecker { + return NDP(header.ICMPv6NeighborSolicit, header.NDPNSMinimumSize, checkers...) +} + +// NDPNSTargetAddress creates a checker that checks the Target Address field of +// a header.NDPNeighborSolicit. +// +// The returned TransportChecker assumes that a valid ICMPv6 is passed to it +// containing a valid NDPNS message as far as the size is concerned. +func NDPNSTargetAddress(want tcpip.Address) TransportChecker { + return func(t *testing.T, h header.Transport) { + t.Helper() + + icmp := h.(header.ICMPv6) + ns := header.NDPNeighborSolicit(icmp.NDPPayload()) + + if got := ns.TargetAddress(); got != want { + t.Fatalf("got %T.TargetAddress = %s, want = %s", ns, got, want) + } + } +} diff --git a/pkg/tcpip/header/BUILD b/pkg/tcpip/header/BUILD index b558350c3..a3485b35c 100644 --- a/pkg/tcpip/header/BUILD +++ b/pkg/tcpip/header/BUILD @@ -1,9 +1,8 @@ load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) -load("//tools/go_stateify:defs.bzl", "go_library") - go_library( name = "header", srcs = [ @@ -17,6 +16,10 @@ go_library( "ipv4.go", "ipv6.go", "ipv6_fragment.go", + "ndp_neighbor_advert.go", + "ndp_neighbor_solicit.go", + "ndp_options.go", + "ndp_router_advert.go", "tcp.go", "udp.go", ], @@ -31,13 +34,26 @@ go_library( ) go_test( - name = "header_test", + name = "header_x_test", size = "small", srcs = [ + "checksum_test.go", "ipversion_test.go", "tcp_test.go", ], deps = [ ":header", + "//pkg/tcpip/buffer", + ], +) + +go_test( + name = "header_test", + size = "small", + srcs = [ + "eth_test.go", + "ndp_test.go", ], + embed = [":header"], + deps = ["//pkg/tcpip"], ) diff --git a/pkg/tcpip/header/checksum.go b/pkg/tcpip/header/checksum.go index 39a4d69be..9749c7f4d 100644 --- a/pkg/tcpip/header/checksum.go +++ b/pkg/tcpip/header/checksum.go @@ -23,11 +23,17 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/buffer" ) -func calculateChecksum(buf []byte, initial uint32) uint16 { +func calculateChecksum(buf []byte, odd bool, initial uint32) (uint16, bool) { v := initial + if odd { + v += uint32(buf[0]) + buf = buf[1:] + } + l := len(buf) - if l&1 != 0 { + odd = l&1 != 0 + if odd { l-- v += uint32(buf[l]) << 8 } @@ -36,7 +42,7 @@ func calculateChecksum(buf []byte, initial uint32) uint16 { v += (uint32(buf[i]) << 8) + uint32(buf[i+1]) } - return ChecksumCombine(uint16(v), uint16(v>>16)) + return ChecksumCombine(uint16(v), uint16(v>>16)), odd } // Checksum calculates the checksum (as defined in RFC 1071) of the bytes in the @@ -44,7 +50,8 @@ func calculateChecksum(buf []byte, initial uint32) uint16 { // // The initial checksum must have been computed on an even number of bytes. func Checksum(buf []byte, initial uint16) uint16 { - return calculateChecksum(buf, uint32(initial)) + s, _ := calculateChecksum(buf, false, uint32(initial)) + return s } // ChecksumVV calculates the checksum (as defined in RFC 1071) of the bytes in @@ -52,19 +59,40 @@ func Checksum(buf []byte, initial uint16) uint16 { // // The initial checksum must have been computed on an even number of bytes. func ChecksumVV(vv buffer.VectorisedView, initial uint16) uint16 { - var odd bool + return ChecksumVVWithOffset(vv, initial, 0, vv.Size()) +} + +// ChecksumVVWithOffset calculates the checksum (as defined in RFC 1071) of the +// bytes in the given VectorizedView. +// +// The initial checksum must have been computed on an even number of bytes. +func ChecksumVVWithOffset(vv buffer.VectorisedView, initial uint16, off int, size int) uint16 { + odd := false sum := initial for _, v := range vv.Views() { if len(v) == 0 { continue } - s := uint32(sum) - if odd { - s += uint32(v[0]) - v = v[1:] + + if off >= len(v) { + off -= len(v) + continue + } + v = v[off:] + + l := len(v) + if l > size { + l = size + } + v = v[:l] + + sum, odd = calculateChecksum(v, odd, uint32(sum)) + + size -= len(v) + if size == 0 { + break } - odd = len(v)&1 != 0 - sum = calculateChecksum(v, s) + off = 0 } return sum } diff --git a/pkg/tcpip/header/checksum_test.go b/pkg/tcpip/header/checksum_test.go new file mode 100644 index 000000000..86b466c1c --- /dev/null +++ b/pkg/tcpip/header/checksum_test.go @@ -0,0 +1,109 @@ +// 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 header provides the implementation of the encoding and decoding of +// network protocol headers. +package header_test + +import ( + "testing" + + "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" +) + +func TestChecksumVVWithOffset(t *testing.T) { + testCases := []struct { + name string + vv buffer.VectorisedView + off, size int + initial uint16 + want uint16 + }{ + { + name: "empty", + vv: buffer.NewVectorisedView(0, []buffer.View{ + buffer.NewViewFromBytes([]byte{1, 9, 0, 5, 4}), + }), + off: 0, + size: 0, + want: 0, + }, + { + name: "OneView", + vv: buffer.NewVectorisedView(0, []buffer.View{ + buffer.NewViewFromBytes([]byte{1, 9, 0, 5, 4}), + }), + off: 0, + size: 5, + want: 1294, + }, + { + name: "TwoViews", + vv: buffer.NewVectorisedView(0, []buffer.View{ + buffer.NewViewFromBytes([]byte{1, 9, 0, 5, 4}), + buffer.NewViewFromBytes([]byte{4, 3, 7, 1, 2, 123}), + }), + off: 0, + size: 11, + want: 33819, + }, + { + name: "TwoViewsWithOffset", + vv: buffer.NewVectorisedView(0, []buffer.View{ + buffer.NewViewFromBytes([]byte{98, 1, 9, 0, 5, 4}), + buffer.NewViewFromBytes([]byte{4, 3, 7, 1, 2, 123}), + }), + off: 1, + size: 11, + want: 33819, + }, + { + name: "ThreeViewsWithOffset", + vv: buffer.NewVectorisedView(0, []buffer.View{ + buffer.NewViewFromBytes([]byte{98, 1, 9, 0, 5, 4}), + buffer.NewViewFromBytes([]byte{98, 1, 9, 0, 5, 4}), + buffer.NewViewFromBytes([]byte{4, 3, 7, 1, 2, 123}), + }), + off: 7, + size: 11, + want: 33819, + }, + { + name: "ThreeViewsWithInitial", + vv: buffer.NewVectorisedView(0, []buffer.View{ + buffer.NewViewFromBytes([]byte{77, 11, 33, 0, 55, 44}), + buffer.NewViewFromBytes([]byte{98, 1, 9, 0, 5, 4}), + buffer.NewViewFromBytes([]byte{4, 3, 7, 1, 2, 123, 99}), + }), + initial: 77, + off: 7, + size: 11, + want: 33896, + }, + } + for _, tc := range testCases { + t.Run(tc.name, func(t *testing.T) { + if got, want := header.ChecksumVVWithOffset(tc.vv, tc.initial, tc.off, tc.size), tc.want; got != want { + t.Errorf("header.ChecksumVVWithOffset(%v) = %v, want: %v", tc, got, tc.want) + } + v := tc.vv.ToView() + v.TrimFront(tc.off) + v.CapLength(tc.size) + if got, want := header.Checksum(v, tc.initial), tc.want; got != want { + t.Errorf("header.Checksum(%v) = %v, want: %v", tc, got, tc.want) + } + }) + } +} diff --git a/pkg/tcpip/header/eth.go b/pkg/tcpip/header/eth.go index 4c3d3311f..f5d2c127f 100644 --- a/pkg/tcpip/header/eth.go +++ b/pkg/tcpip/header/eth.go @@ -48,8 +48,48 @@ const ( // EthernetAddressSize is the size, in bytes, of an ethernet address. EthernetAddressSize = 6 + + // unspecifiedEthernetAddress is the unspecified ethernet address + // (all bits set to 0). + unspecifiedEthernetAddress = tcpip.LinkAddress("\x00\x00\x00\x00\x00\x00") + + // 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 + // the masked bit is a 1, then the address is a multicast, unicast + // otherwise. + // + // See the IEEE Std 802-2001 document for more details. Specifically, + // section 9.2.1 of http://ieee802.org/secmail/pdfocSP2xXA6d.pdf: + // "A 48-bit universal address consists of two parts. The first 24 bits + // correspond to the OUI as assigned by the IEEE, expect that the + // assignee may set the LSB of the first octet to 1 for group addresses + // or set it to 0 for individual addresses." + unicastMulticastFlagMask = 1 + + // unicastMulticastFlagByteIdx is the byte that holds the + // unicast/multicast flag. See unicastMulticastFlagMask. + unicastMulticastFlagByteIdx = 0 +) + +const ( + // EthernetProtocolAll is a catch-all for all protocols carried inside + // an ethernet frame. It is mainly used to create packet sockets that + // capture all traffic. + EthernetProtocolAll tcpip.NetworkProtocolNumber = 0x0003 + + // EthernetProtocolPUP is the PARC Universial Packet protocol ethertype. + EthernetProtocolPUP tcpip.NetworkProtocolNumber = 0x0200 ) +// Ethertypes holds the protocol numbers describing the payload of an ethernet +// frame. These types aren't necessarily supported by netstack, but can be used +// to catch all traffic of a type via packet endpoints. +var Ethertypes = []tcpip.NetworkProtocolNumber{ + EthernetProtocolAll, + EthernetProtocolPUP, +} + // SourceAddress returns the "MAC source" field of the ethernet frame header. func (b Ethernet) SourceAddress() tcpip.LinkAddress { return tcpip.LinkAddress(b[srcMAC:][:EthernetAddressSize]) @@ -72,3 +112,25 @@ func (b Ethernet) Encode(e *EthernetFields) { copy(b[srcMAC:][:EthernetAddressSize], e.SrcAddr) copy(b[dstMAC:][:EthernetAddressSize], e.DstAddr) } + +// IsValidUnicastEthernetAddress returns true if addr is a valid unicast +// ethernet address. +func IsValidUnicastEthernetAddress(addr tcpip.LinkAddress) bool { + // Must be of the right length. + if len(addr) != EthernetAddressSize { + return false + } + + // Must not be unspecified. + if addr == unspecifiedEthernetAddress { + return false + } + + // Must not be a multicast. + if addr[unicastMulticastFlagByteIdx]&unicastMulticastFlagMask != 0 { + return false + } + + // addr is a valid unicast ethernet address. + return true +} diff --git a/pkg/tcpip/header/eth_test.go b/pkg/tcpip/header/eth_test.go new file mode 100644 index 000000000..6634c90f5 --- /dev/null +++ b/pkg/tcpip/header/eth_test.go @@ -0,0 +1,68 @@ +// Copyright 2018 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 header + +import ( + "testing" + + "gvisor.dev/gvisor/pkg/tcpip" +) + +func TestIsValidUnicastEthernetAddress(t *testing.T) { + tests := []struct { + name string + addr tcpip.LinkAddress + expected bool + }{ + { + "Nil", + tcpip.LinkAddress([]byte(nil)), + false, + }, + { + "Empty", + tcpip.LinkAddress(""), + false, + }, + { + "InvalidLength", + tcpip.LinkAddress("\x01\x02\x03"), + false, + }, + { + "Unspecified", + unspecifiedEthernetAddress, + false, + }, + { + "Multicast", + tcpip.LinkAddress("\x01\x02\x03\x04\x05\x06"), + false, + }, + { + "Valid", + tcpip.LinkAddress("\x02\x02\x03\x04\x05\x06"), + true, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + if got := IsValidUnicastEthernetAddress(test.addr); got != test.expected { + t.Fatalf("got IsValidUnicastEthernetAddress = %t, want = %t", got, test.expected) + } + }) + } +} diff --git a/pkg/tcpip/header/icmpv6.go b/pkg/tcpip/header/icmpv6.go index 1125a7d14..b4037b6c8 100644 --- a/pkg/tcpip/header/icmpv6.go +++ b/pkg/tcpip/header/icmpv6.go @@ -25,6 +25,12 @@ import ( type ICMPv6 []byte const ( + // ICMPv6HeaderSize is the size of the ICMPv6 header. That is, the + // sum of the size of the ICMPv6 Type, Code and Checksum fields, as + // per RFC 4443 section 2.1. After the ICMPv6 header, the ICMPv6 + // message body begins. + ICMPv6HeaderSize = 4 + // ICMPv6MinimumSize is the minimum size of a valid ICMP packet. ICMPv6MinimumSize = 8 @@ -37,10 +43,16 @@ const ( // ICMPv6NeighborSolicitMinimumSize is the minimum size of a // neighbor solicitation packet. - ICMPv6NeighborSolicitMinimumSize = ICMPv6MinimumSize + 16 + ICMPv6NeighborSolicitMinimumSize = ICMPv6HeaderSize + NDPNSMinimumSize + + // ICMPv6NeighborAdvertMinimumSize is the minimum size of a + // neighbor advertisement packet. + ICMPv6NeighborAdvertMinimumSize = ICMPv6HeaderSize + NDPNAMinimumSize - // ICMPv6NeighborAdvertSize is size of a neighbor advertisement. - ICMPv6NeighborAdvertSize = 32 + // ICMPv6NeighborAdvertSize is size of a neighbor advertisement + // including the NDP Target Link Layer option for an Ethernet + // address. + ICMPv6NeighborAdvertSize = ICMPv6HeaderSize + NDPNAMinimumSize + ndpTargetEthernetLinkLayerAddressSize // ICMPv6EchoMinimumSize is the minimum size of a valid ICMP echo packet. ICMPv6EchoMinimumSize = 8 @@ -68,6 +80,13 @@ const ( // icmpv6SequenceOffset is the offset of the sequence field // in a ICMPv6 Echo Request/Reply message. icmpv6SequenceOffset = 6 + + // NDPHopLimit is the expected IP hop limit value of 255 for received + // NDP packets, as per RFC 4861 sections 4.1 - 4.5, 6.1.1, 6.1.2, 7.1.1, + // 7.1.2 and 8.1. If the hop limit value is not 255, nodes MUST silently + // drop the NDP packet. All outgoing NDP packets must use this value for + // its IP hop limit field. + NDPHopLimit = 255 ) // ICMPv6Type is the ICMP type field described in RFC 4443 and friends. @@ -113,7 +132,7 @@ func (b ICMPv6) Checksum() uint16 { return binary.BigEndian.Uint16(b[icmpv6ChecksumOffset:]) } -// SetChecksum calculates and sets the ICMP checksum field. +// SetChecksum sets the ICMP checksum field. func (b ICMPv6) SetChecksum(checksum uint16) { binary.BigEndian.PutUint16(b[icmpv6ChecksumOffset:], checksum) } @@ -166,12 +185,19 @@ func (b ICMPv6) SetSequence(sequence uint16) { binary.BigEndian.PutUint16(b[icmpv6SequenceOffset:], sequence) } +// NDPPayload returns the NDP payload buffer. That is, it returns the ICMPv6 +// packet's message body as defined by RFC 4443 section 2.1; the portion of the +// ICMPv6 buffer after the first ICMPv6HeaderSize bytes. +func (b ICMPv6) NDPPayload() []byte { + return b[ICMPv6HeaderSize:] +} + // Payload implements Transport.Payload. func (b ICMPv6) Payload() []byte { return b[ICMPv6PayloadOffset:] } -// ICMPv6Checksum calculates the ICMP checksum over the provided ICMP header, +// ICMPv6Checksum calculates the ICMP checksum over the provided ICMPv6 header, // IPv6 src/dst addresses and the payload. func ICMPv6Checksum(h ICMPv6, src, dst tcpip.Address, vv buffer.VectorisedView) uint16 { // Calculate the IPv6 pseudo-header upper-layer checksum. diff --git a/pkg/tcpip/header/ipv4.go b/pkg/tcpip/header/ipv4.go index 554632a64..e5360e7c1 100644 --- a/pkg/tcpip/header/ipv4.go +++ b/pkg/tcpip/header/ipv4.go @@ -284,7 +284,7 @@ func (b IPv4) IsValid(pktSize int) bool { hlen := int(b.HeaderLength()) tlen := int(b.TotalLength()) - if hlen > tlen || tlen > pktSize { + if hlen < IPv4MinimumSize || hlen > tlen || tlen > pktSize { return false } diff --git a/pkg/tcpip/header/ipv6.go b/pkg/tcpip/header/ipv6.go index 9d3abc0e4..0caa51c1e 100644 --- a/pkg/tcpip/header/ipv6.go +++ b/pkg/tcpip/header/ipv6.go @@ -87,11 +87,14 @@ const ( // section 5. IPv6MinimumMTU = 1280 - // IPv6Any is the non-routable IPv6 "any" meta address. + // IPv6Any is the non-routable IPv6 "any" meta address. It is also + // known as the unspecified address. IPv6Any tcpip.Address = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00" ) -// IPv6EmptySubnet is the empty IPv6 subnet. +// IPv6EmptySubnet is the empty IPv6 subnet. It may also be known as the +// catch-all or wildcard subnet. That is, all IPv6 addresses are considered to +// be contained within this subnet. var IPv6EmptySubnet = func() tcpip.Subnet { subnet, err := tcpip.NewSubnet(IPv6Any, tcpip.AddressMask(IPv6Any)) if err != nil { @@ -100,6 +103,15 @@ var IPv6EmptySubnet = func() tcpip.Subnet { return subnet }() +// IPv6LinkLocalPrefix is the prefix for IPv6 link-local addresses, as defined +// by RFC 4291 section 2.5.6. +// +// The prefix is fe80::/64 +var IPv6LinkLocalPrefix = tcpip.AddressWithPrefix{ + Address: "\xfe\x80\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00", + PrefixLen: 64, +} + // PayloadLength returns the value of the "payload length" field of the ipv6 // header. func (b IPv6) PayloadLength() uint16 { diff --git a/pkg/tcpip/header/ndp_neighbor_advert.go b/pkg/tcpip/header/ndp_neighbor_advert.go new file mode 100644 index 000000000..505c92668 --- /dev/null +++ b/pkg/tcpip/header/ndp_neighbor_advert.go @@ -0,0 +1,110 @@ +// 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 header + +import "gvisor.dev/gvisor/pkg/tcpip" + +// NDPNeighborAdvert is an NDP Neighbor Advertisement message. It will +// only contain the body of an ICMPv6 packet. +// +// See RFC 4861 section 4.4 for more details. +type NDPNeighborAdvert []byte + +const ( + // NDPNAMinimumSize is the minimum size of a valid NDP Neighbor + // Advertisement message (body of an ICMPv6 packet). + NDPNAMinimumSize = 20 + + // ndpNATargetAddressOffset is the start of the Target Address + // field within an NDPNeighborAdvert. + ndpNATargetAddressOffset = 4 + + // ndpNAOptionsOffset is the start of the NDP options in an + // NDPNeighborAdvert. + ndpNAOptionsOffset = ndpNATargetAddressOffset + IPv6AddressSize + + // ndpNAFlagsOffset is the offset of the flags within an + // NDPNeighborAdvert + ndpNAFlagsOffset = 0 + + // ndpNARouterFlagMask is the mask of the Router Flag field in + // the flags byte within in an NDPNeighborAdvert. + ndpNARouterFlagMask = (1 << 7) + + // ndpNASolicitedFlagMask is the mask of the Solicited Flag field in + // the flags byte within in an NDPNeighborAdvert. + ndpNASolicitedFlagMask = (1 << 6) + + // ndpNAOverrideFlagMask is the mask of the Override Flag field in + // the flags byte within in an NDPNeighborAdvert. + ndpNAOverrideFlagMask = (1 << 5) +) + +// TargetAddress returns the value within the Target Address field. +func (b NDPNeighborAdvert) TargetAddress() tcpip.Address { + return tcpip.Address(b[ndpNATargetAddressOffset:][:IPv6AddressSize]) +} + +// SetTargetAddress sets the value within the Target Address field. +func (b NDPNeighborAdvert) SetTargetAddress(addr tcpip.Address) { + copy(b[ndpNATargetAddressOffset:][:IPv6AddressSize], addr) +} + +// RouterFlag returns the value of the Router Flag field. +func (b NDPNeighborAdvert) RouterFlag() bool { + return b[ndpNAFlagsOffset]&ndpNARouterFlagMask != 0 +} + +// SetRouterFlag sets the value in the Router Flag field. +func (b NDPNeighborAdvert) SetRouterFlag(f bool) { + if f { + b[ndpNAFlagsOffset] |= ndpNARouterFlagMask + } else { + b[ndpNAFlagsOffset] &^= ndpNARouterFlagMask + } +} + +// SolicitedFlag returns the value of the Solicited Flag field. +func (b NDPNeighborAdvert) SolicitedFlag() bool { + return b[ndpNAFlagsOffset]&ndpNASolicitedFlagMask != 0 +} + +// SetSolicitedFlag sets the value in the Solicited Flag field. +func (b NDPNeighborAdvert) SetSolicitedFlag(f bool) { + if f { + b[ndpNAFlagsOffset] |= ndpNASolicitedFlagMask + } else { + b[ndpNAFlagsOffset] &^= ndpNASolicitedFlagMask + } +} + +// OverrideFlag returns the value of the Override Flag field. +func (b NDPNeighborAdvert) OverrideFlag() bool { + return b[ndpNAFlagsOffset]&ndpNAOverrideFlagMask != 0 +} + +// SetOverrideFlag sets the value in the Override Flag field. +func (b NDPNeighborAdvert) SetOverrideFlag(f bool) { + if f { + b[ndpNAFlagsOffset] |= ndpNAOverrideFlagMask + } else { + b[ndpNAFlagsOffset] &^= ndpNAOverrideFlagMask + } +} + +// Options returns an NDPOptions of the the options body. +func (b NDPNeighborAdvert) Options() NDPOptions { + return NDPOptions(b[ndpNAOptionsOffset:]) +} diff --git a/pkg/tcpip/header/ndp_neighbor_solicit.go b/pkg/tcpip/header/ndp_neighbor_solicit.go new file mode 100644 index 000000000..3a1b8e139 --- /dev/null +++ b/pkg/tcpip/header/ndp_neighbor_solicit.go @@ -0,0 +1,52 @@ +// 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 header + +import "gvisor.dev/gvisor/pkg/tcpip" + +// NDPNeighborSolicit is an NDP Neighbor Solicitation message. It will only +// contain the body of an ICMPv6 packet. +// +// See RFC 4861 section 4.3 for more details. +type NDPNeighborSolicit []byte + +const ( + // NDPNSMinimumSize is the minimum size of a valid NDP Neighbor + // Solicitation message (body of an ICMPv6 packet). + NDPNSMinimumSize = 20 + + // ndpNSTargetAddessOffset is the start of the Target Address + // field within an NDPNeighborSolicit. + ndpNSTargetAddessOffset = 4 + + // ndpNSOptionsOffset is the start of the NDP options in an + // NDPNeighborSolicit. + ndpNSOptionsOffset = ndpNSTargetAddessOffset + IPv6AddressSize +) + +// TargetAddress returns the value within the Target Address field. +func (b NDPNeighborSolicit) TargetAddress() tcpip.Address { + return tcpip.Address(b[ndpNSTargetAddessOffset:][:IPv6AddressSize]) +} + +// SetTargetAddress sets the value within the Target Address field. +func (b NDPNeighborSolicit) SetTargetAddress(addr tcpip.Address) { + copy(b[ndpNSTargetAddessOffset:][:IPv6AddressSize], addr) +} + +// Options returns an NDPOptions of the the options body. +func (b NDPNeighborSolicit) Options() NDPOptions { + return NDPOptions(b[ndpNSOptionsOffset:]) +} diff --git a/pkg/tcpip/header/ndp_options.go b/pkg/tcpip/header/ndp_options.go new file mode 100644 index 000000000..a2b9d7435 --- /dev/null +++ b/pkg/tcpip/header/ndp_options.go @@ -0,0 +1,463 @@ +// 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 header + +import ( + "encoding/binary" + "errors" + "time" + + "gvisor.dev/gvisor/pkg/tcpip" +) + +const ( + // NDPTargetLinkLayerAddressOptionType is the type of the Target + // Link-Layer Address option, as per RFC 4861 section 4.6.1. + NDPTargetLinkLayerAddressOptionType = 2 + + // ndpTargetEthernetLinkLayerAddressSize is the size of a Target + // Link Layer Option for an Ethernet address. + ndpTargetEthernetLinkLayerAddressSize = 8 + + // NDPPrefixInformationType is the type of the Prefix Information + // option, as per RFC 4861 section 4.6.2. + NDPPrefixInformationType = 3 + + // ndpPrefixInformationLength is the expected length, in bytes, of the + // body of an NDP Prefix Information option, as per RFC 4861 section + // 4.6.2 which specifies that the Length field is 4. Given this, the + // expected length, in bytes, is 30 becuase 4 * lengthByteUnits (8) - 2 + // (Type & Length) = 30. + ndpPrefixInformationLength = 30 + + // ndpPrefixInformationPrefixLengthOffset is the offset of the Prefix + // Length field within an NDPPrefixInformation. + ndpPrefixInformationPrefixLengthOffset = 0 + + // ndpPrefixInformationFlagsOffset is the offset of the flags byte + // within an NDPPrefixInformation. + ndpPrefixInformationFlagsOffset = 1 + + // ndpPrefixInformationOnLinkFlagMask is the mask of the On-Link Flag + // field in the flags byte within an NDPPrefixInformation. + ndpPrefixInformationOnLinkFlagMask = (1 << 7) + + // ndpPrefixInformationAutoAddrConfFlagMask is the mask of the + // Autonomous Address-Configuration flag field in the flags byte within + // an NDPPrefixInformation. + ndpPrefixInformationAutoAddrConfFlagMask = (1 << 6) + + // ndpPrefixInformationReserved1FlagsMask is the mask of the Reserved1 + // field in the flags byte within an NDPPrefixInformation. + ndpPrefixInformationReserved1FlagsMask = 63 + + // ndpPrefixInformationValidLifetimeOffset is the start of the 4-byte + // Valid Lifetime field within an NDPPrefixInformation. + ndpPrefixInformationValidLifetimeOffset = 2 + + // ndpPrefixInformationPreferredLifetimeOffset is the start of the + // 4-byte Preferred Lifetime field within an NDPPrefixInformation. + ndpPrefixInformationPreferredLifetimeOffset = 6 + + // ndpPrefixInformationReserved2Offset is the start of the 4-byte + // Reserved2 field within an NDPPrefixInformation. + ndpPrefixInformationReserved2Offset = 10 + + // ndpPrefixInformationReserved2Length is the length of the Reserved2 + // field. + // + // It is 4 bytes. + ndpPrefixInformationReserved2Length = 4 + + // ndpPrefixInformationPrefixOffset is the start of the Prefix field + // within an NDPPrefixInformation. + ndpPrefixInformationPrefixOffset = 14 + + // NDPPrefixInformationInfiniteLifetime is a value that represents + // infinity for the Valid and Preferred Lifetime fields in a NDP Prefix + // Information option. Its value is (2^32 - 1)s = 4294967295s + NDPPrefixInformationInfiniteLifetime = time.Second * 4294967295 + + // lengthByteUnits is the multiplier factor for the Length field of an + // NDP option. That is, the length field for NDP options is in units of + // 8 octets, as per RFC 4861 section 4.6. + lengthByteUnits = 8 +) + +// NDPOptionIterator is an iterator of NDPOption. +// +// Note, between when an NDPOptionIterator is obtained and last used, no changes +// to the NDPOptions may happen. Doing so may cause undefined and unexpected +// behaviour. It is fine to obtain an NDPOptionIterator, iterate over the first +// few NDPOption then modify the backing NDPOptions so long as the +// NDPOptionIterator obtained before modification is no longer used. +type NDPOptionIterator struct { + // The NDPOptions this NDPOptionIterator is iterating over. + opts NDPOptions +} + +// Potential errors when iterating over an NDPOptions. +var ( + ErrNDPOptBufExhausted = errors.New("Buffer unexpectedly exhausted") + ErrNDPOptZeroLength = errors.New("NDP option has zero-valued Length field") + ErrNDPOptMalformedBody = errors.New("NDP option has a malformed body") +) + +// Next returns the next element in the backing NDPOptions, or true if we are +// done, or false if an error occured. +// +// The return can be read as option, done, error. Note, option should only be +// used if done is false and error is nil. +func (i *NDPOptionIterator) Next() (NDPOption, bool, error) { + for { + // Do we still have elements to look at? + if len(i.opts) == 0 { + return nil, true, nil + } + + // Do we have enough bytes for an NDP option that has a Length + // field of at least 1? Note, 0 in the Length field is invalid. + if len(i.opts) < lengthByteUnits { + return nil, true, ErrNDPOptBufExhausted + } + + // Get the Type field. + t := i.opts[0] + + // Get the Length field. + l := i.opts[1] + + // This would indicate an erroneous NDP option as the Length + // field should never be 0. + if l == 0 { + return nil, true, ErrNDPOptZeroLength + } + + // How many bytes are in the option body? + numBytes := int(l) * lengthByteUnits + numBodyBytes := numBytes - 2 + + potentialBody := i.opts[2:] + + // This would indicate an erroenous NDPOptions buffer as we ran + // out of the buffer in the middle of an NDP option. + if left := len(potentialBody); left < numBodyBytes { + return nil, true, ErrNDPOptBufExhausted + } + + // Get only the options body, leaving the rest of the options + // buffer alone. + body := potentialBody[:numBodyBytes] + + // Update opts with the remaining options body. + i.opts = i.opts[numBytes:] + + switch t { + case NDPTargetLinkLayerAddressOptionType: + return NDPTargetLinkLayerAddressOption(body), false, nil + + case NDPPrefixInformationType: + // Make sure the length of a Prefix Information option + // body is ndpPrefixInformationLength, as per RFC 4861 + // section 4.6.2. + if numBodyBytes != ndpPrefixInformationLength { + return nil, true, ErrNDPOptMalformedBody + } + + return NDPPrefixInformation(body), false, nil + default: + // We do not yet recognize the option, just skip for + // now. This is okay because RFC 4861 allows us to + // skip/ignore any unrecognized options. However, + // we MUST recognized all the options in RFC 4861. + // + // TODO(b/141487990): Handle all NDP options as defined + // by RFC 4861. + } + } +} + +// NDPOptions is a buffer of NDP options as defined by RFC 4861 section 4.6. +type NDPOptions []byte + +// Iter returns an iterator of NDPOption. +// +// If check is true, Iter will do an integrity check on the options by iterating +// over it and returning an error if detected. +// +// See NDPOptionIterator for more information. +func (b NDPOptions) Iter(check bool) (NDPOptionIterator, error) { + it := NDPOptionIterator{opts: b} + + if check { + for it2 := it; true; { + if _, done, err := it2.Next(); err != nil || done { + return it, err + } + } + } + + return it, nil +} + +// Serialize serializes the provided list of NDP options into o. +// +// Note, b must be of sufficient size to hold all the options in s. See +// NDPOptionsSerializer.Length for details on the getting the total size +// of a serialized NDPOptionsSerializer. +// +// Serialize may panic if b is not of sufficient size to hold all the options +// in s. +func (b NDPOptions) Serialize(s NDPOptionsSerializer) int { + done := 0 + + for _, o := range s { + l := paddedLength(o) + + if l == 0 { + continue + } + + b[0] = o.Type() + + // We know this safe because paddedLength would have returned + // 0 if o had an invalid length (> 255 * lengthByteUnits). + b[1] = uint8(l / lengthByteUnits) + + // Serialize NDP option body. + used := o.serializeInto(b[2:]) + + // Zero out remaining (padding) bytes, if any exists. + for i := used + 2; i < l; i++ { + b[i] = 0 + } + + b = b[l:] + done += l + } + + return done +} + +// NDPOption is the set of functions to be implemented by all NDP option types. +type NDPOption interface { + // Type returns the type of the receiver. + Type() uint8 + + // Length returns the length of the body of the receiver, in bytes. + Length() int + + // serializeInto serializes the receiver into the provided byte + // buffer. + // + // Note, the caller MUST provide a byte buffer with size of at least + // Length. Implementers of this function may assume that the byte buffer + // is of sufficient size. serializeInto MAY panic if the provided byte + // buffer is not of sufficient size. + // + // serializeInto will return the number of bytes that was used to + // serialize the receiver. Implementers must only use the number of + // bytes required to serialize the receiver. Callers MAY provide a + // larger buffer than required to serialize into. + serializeInto([]byte) int +} + +// paddedLength returns the length of o, in bytes, with any padding bytes, if +// required. +func paddedLength(o NDPOption) int { + l := o.Length() + + if l == 0 { + return 0 + } + + // Length excludes the 2 Type and Length bytes. + l += 2 + + // Add extra bytes if needed to make sure the option is + // lengthByteUnits-byte aligned. We do this by adding lengthByteUnits-1 + // to l and then stripping off the last few LSBits from l. This will + // make sure that l is rounded up to the nearest unit of + // lengthByteUnits. This works since lengthByteUnits is a power of 2 + // (= 8). + mask := lengthByteUnits - 1 + l += mask + l &^= mask + + if l/lengthByteUnits > 255 { + // Should never happen because an option can only have a max + // value of 255 for its Length field, so just return 0 so this + // option does not get serialized. + // + // Returning 0 here will make sure that this option does not get + // serialized when NDPOptions.Serialize is called with the + // NDPOptionsSerializer that holds this option, effectively + // skipping this option during serialization. Also note that + // a value of zero for the Length field in an NDP option is + // invalid so this is another sign to the caller that this NDP + // option is malformed, as per RFC 4861 section 4.6. + return 0 + } + + return l +} + +// NDPOptionsSerializer is a serializer for NDP options. +type NDPOptionsSerializer []NDPOption + +// Length returns the total number of bytes required to serialize. +func (b NDPOptionsSerializer) Length() int { + l := 0 + + for _, o := range b { + l += paddedLength(o) + } + + return l +} + +// NDPTargetLinkLayerAddressOption is the NDP Target Link Layer Option +// as defined by RFC 4861 section 4.6.1. +// +// It is the first X bytes following the NDP option's Type and Length field +// where X is the value in Length multiplied by lengthByteUnits - 2 bytes. +type NDPTargetLinkLayerAddressOption tcpip.LinkAddress + +// Type implements NDPOption.Type. +func (o NDPTargetLinkLayerAddressOption) Type() uint8 { + return NDPTargetLinkLayerAddressOptionType +} + +// Length implements NDPOption.Length. +func (o NDPTargetLinkLayerAddressOption) Length() int { + return len(o) +} + +// serializeInto implements NDPOption.serializeInto. +func (o NDPTargetLinkLayerAddressOption) serializeInto(b []byte) int { + return copy(b, o) +} + +// EthernetAddress will return an ethernet (MAC) address if the +// NDPTargetLinkLayerAddressOption's body has at minimum EthernetAddressSize +// bytes. If the body has more than EthernetAddressSize bytes, only the first +// EthernetAddressSize bytes are returned as that is all that is needed for an +// Ethernet address. +func (o NDPTargetLinkLayerAddressOption) EthernetAddress() tcpip.LinkAddress { + if len(o) >= EthernetAddressSize { + return tcpip.LinkAddress(o[:EthernetAddressSize]) + } + + return tcpip.LinkAddress([]byte(nil)) +} + +// NDPPrefixInformation is the NDP Prefix Information option as defined by +// RFC 4861 section 4.6.2. +// +// The length, in bytes, of a valid NDP Prefix Information option body MUST be +// ndpPrefixInformationLength bytes. +type NDPPrefixInformation []byte + +// Type implements NDPOption.Type. +func (o NDPPrefixInformation) Type() uint8 { + return NDPPrefixInformationType +} + +// Length implements NDPOption.Length. +func (o NDPPrefixInformation) Length() int { + return ndpPrefixInformationLength +} + +// serializeInto implements NDPOption.serializeInto. +func (o NDPPrefixInformation) serializeInto(b []byte) int { + used := copy(b, o) + + // Zero out the Reserved1 field. + b[ndpPrefixInformationFlagsOffset] &^= ndpPrefixInformationReserved1FlagsMask + + // Zero out the Reserved2 field. + reserved2 := b[ndpPrefixInformationReserved2Offset:][:ndpPrefixInformationReserved2Length] + for i := range reserved2 { + reserved2[i] = 0 + } + + return used +} + +// PrefixLength returns the value in the number of leading bits in the Prefix +// that are valid. +// +// Valid values are in the range [0, 128], but o may not always contain valid +// values. It is up to the caller to valdiate the Prefix Information option. +func (o NDPPrefixInformation) PrefixLength() uint8 { + return o[ndpPrefixInformationPrefixLengthOffset] +} + +// OnLinkFlag returns true of the prefix is considered on-link. On-link means +// that a forwarding node is not needed to send packets to other nodes on the +// same prefix. +// +// Note, when this function returns false, no statement is made about the +// on-link property of a prefix. That is, if OnLinkFlag returns false, the +// caller MUST NOT conclude that the prefix is off-link and MUST NOT update any +// previously stored state for this prefix about its on-link status. +func (o NDPPrefixInformation) OnLinkFlag() bool { + return o[ndpPrefixInformationFlagsOffset]&ndpPrefixInformationOnLinkFlagMask != 0 +} + +// AutonomousAddressConfigurationFlag returns true if the prefix can be used for +// Stateless Address Auto-Configuration (as specified in RFC 4862). +func (o NDPPrefixInformation) AutonomousAddressConfigurationFlag() bool { + return o[ndpPrefixInformationFlagsOffset]&ndpPrefixInformationAutoAddrConfFlagMask != 0 +} + +// ValidLifetime returns the length of time that the prefix is valid for the +// purpose of on-link determination. This value is relative to the send time of +// the packet that the Prefix Information option was present in. +// +// Note, a value of 0 implies the prefix should not be considered as on-link, +// and a value of infinity/forever is represented by +// NDPPrefixInformationInfiniteLifetime. +func (o NDPPrefixInformation) ValidLifetime() time.Duration { + // The field is the time in seconds, as per RFC 4861 section 4.6.2. + return time.Second * time.Duration(binary.BigEndian.Uint32(o[ndpPrefixInformationValidLifetimeOffset:])) +} + +// PreferredLifetime returns the length of time that an address generated from +// the prefix via Stateless Address Auto-Configuration remains preferred. This +// value is relative to the send time of the packet that the Prefix Information +// option was present in. +// +// Note, a value of 0 implies that addresses generated from the prefix should +// no longer remain preferred, and a value of infinity is represented by +// NDPPrefixInformationInfiniteLifetime. +// +// Also note that the value of this field MUST NOT exceed the Valid Lifetime +// field to avoid preferring addresses that are no longer valid, for the +// purpose of Stateless Address Auto-Configuration. +func (o NDPPrefixInformation) PreferredLifetime() time.Duration { + // The field is the time in seconds, as per RFC 4861 section 4.6.2. + return time.Second * time.Duration(binary.BigEndian.Uint32(o[ndpPrefixInformationPreferredLifetimeOffset:])) +} + +// Prefix returns an IPv6 address or a prefix of an IPv6 address. The Prefix +// Length field (see NDPPrefixInformation.PrefixLength) contains the number +// of valid leading bits in the prefix. +// +// Hosts SHOULD ignore an NDP Prefix Information option where the Prefix field +// holds the link-local prefix (fe80::). +func (o NDPPrefixInformation) Prefix() tcpip.Address { + return tcpip.Address(o[ndpPrefixInformationPrefixOffset:][:IPv6AddressSize]) +} diff --git a/pkg/tcpip/header/ndp_router_advert.go b/pkg/tcpip/header/ndp_router_advert.go new file mode 100644 index 000000000..bf7610863 --- /dev/null +++ b/pkg/tcpip/header/ndp_router_advert.go @@ -0,0 +1,112 @@ +// 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 header + +import ( + "encoding/binary" + "time" +) + +// NDPRouterAdvert is an NDP Router Advertisement message. It will only contain +// the body of an ICMPv6 packet. +// +// See RFC 4861 section 4.2 for more details. +type NDPRouterAdvert []byte + +const ( + // NDPRAMinimumSize is the minimum size of a valid NDP Router + // Advertisement message (body of an ICMPv6 packet). + NDPRAMinimumSize = 12 + + // ndpRACurrHopLimitOffset is the byte of the Curr Hop Limit field + // within an NDPRouterAdvert. + ndpRACurrHopLimitOffset = 0 + + // ndpRAFlagsOffset is the byte with the NDP RA bit-fields/flags + // within an NDPRouterAdvert. + ndpRAFlagsOffset = 1 + + // ndpRAManagedAddrConfFlagMask is the mask of the Managed Address + // Configuration flag within the bit-field/flags byte of an + // NDPRouterAdvert. + ndpRAManagedAddrConfFlagMask = (1 << 7) + + // ndpRAOtherConfFlagMask is the mask of the Other Configuration flag + // within the bit-field/flags byte of an NDPRouterAdvert. + ndpRAOtherConfFlagMask = (1 << 6) + + // ndpRARouterLifetimeOffset is the start of the 2-byte Router Lifetime + // field within an NDPRouterAdvert. + ndpRARouterLifetimeOffset = 2 + + // ndpRAReachableTimeOffset is the start of the 4-byte Reachable Time + // field within an NDPRouterAdvert. + ndpRAReachableTimeOffset = 4 + + // ndpRARetransTimerOffset is the start of the 4-byte Retrans Timer + // field within an NDPRouterAdvert. + ndpRARetransTimerOffset = 8 + + // ndpRAOptionsOffset is the start of the NDP options in an + // NDPRouterAdvert. + ndpRAOptionsOffset = 12 +) + +// CurrHopLimit returns the value of the Curr Hop Limit field. +func (b NDPRouterAdvert) CurrHopLimit() uint8 { + return b[ndpRACurrHopLimitOffset] +} + +// ManagedAddrConfFlag returns the value of the Managed Address Configuration +// flag. +func (b NDPRouterAdvert) ManagedAddrConfFlag() bool { + return b[ndpRAFlagsOffset]&ndpRAManagedAddrConfFlagMask != 0 +} + +// OtherConfFlag returns the value of the Other Configuration flag. +func (b NDPRouterAdvert) OtherConfFlag() bool { + return b[ndpRAFlagsOffset]&ndpRAOtherConfFlagMask != 0 +} + +// RouterLifetime returns the lifetime associated with the default router. A +// value of 0 means the source of the Router Advertisement is not a default +// router and SHOULD NOT appear on the default router list. Note, a value of 0 +// only means that the router should not be used as a default router, it does +// not apply to other information contained in the Router Advertisement. +func (b NDPRouterAdvert) RouterLifetime() time.Duration { + // The field is the time in seconds, as per RFC 4861 section 4.2. + return time.Second * time.Duration(binary.BigEndian.Uint16(b[ndpRARouterLifetimeOffset:])) +} + +// ReachableTime returns the time that a node assumes a neighbor is reachable +// after having received a reachability confirmation. A value of 0 means +// that it is unspecified by the source of the Router Advertisement message. +func (b NDPRouterAdvert) ReachableTime() time.Duration { + // The field is the time in milliseconds, as per RFC 4861 section 4.2. + return time.Millisecond * time.Duration(binary.BigEndian.Uint32(b[ndpRAReachableTimeOffset:])) +} + +// RetransTimer returns the time between retransmitted Neighbor Solicitation +// messages. A value of 0 means that it is unspecified by the source of the +// Router Advertisement message. +func (b NDPRouterAdvert) RetransTimer() time.Duration { + // The field is the time in milliseconds, as per RFC 4861 section 4.2. + return time.Millisecond * time.Duration(binary.BigEndian.Uint32(b[ndpRARetransTimerOffset:])) +} + +// Options returns an NDPOptions of the the options body. +func (b NDPRouterAdvert) Options() NDPOptions { + return NDPOptions(b[ndpRAOptionsOffset:]) +} diff --git a/pkg/tcpip/header/ndp_test.go b/pkg/tcpip/header/ndp_test.go new file mode 100644 index 000000000..ad6daafcd --- /dev/null +++ b/pkg/tcpip/header/ndp_test.go @@ -0,0 +1,570 @@ +// 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 header + +import ( + "bytes" + "testing" + "time" + + "gvisor.dev/gvisor/pkg/tcpip" +) + +// TestNDPNeighborSolicit tests the functions of NDPNeighborSolicit. +func TestNDPNeighborSolicit(t *testing.T) { + b := []byte{ + 0, 0, 0, 0, + 1, 2, 3, 4, + 5, 6, 7, 8, + 9, 10, 11, 12, + 13, 14, 15, 16, + } + + // Test getting the Target Address. + ns := NDPNeighborSolicit(b) + addr := tcpip.Address("\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x0f\x10") + if got := ns.TargetAddress(); got != addr { + t.Errorf("got ns.TargetAddress = %s, want %s", got, addr) + } + + // Test updating the Target Address. + addr2 := tcpip.Address("\x11\x12\x13\x14\x15\x16\x17\x18\x19\x1a\x1b\x1c\x1d\x1e\x1f\x11") + ns.SetTargetAddress(addr2) + if got := ns.TargetAddress(); got != addr2 { + t.Errorf("got ns.TargetAddress = %s, want %s", got, addr2) + } + // Make sure the address got updated in the backing buffer. + if got := tcpip.Address(b[ndpNSTargetAddessOffset:][:IPv6AddressSize]); got != addr2 { + t.Errorf("got targetaddress buffer = %s, want %s", got, addr2) + } +} + +// TestNDPNeighborAdvert tests the functions of NDPNeighborAdvert. +func TestNDPNeighborAdvert(t *testing.T) { + b := []byte{ + 160, 0, 0, 0, + 1, 2, 3, 4, + 5, 6, 7, 8, + 9, 10, 11, 12, + 13, 14, 15, 16, + } + + // Test getting the Target Address. + na := NDPNeighborAdvert(b) + addr := tcpip.Address("\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x0f\x10") + if got := na.TargetAddress(); got != addr { + t.Errorf("got TargetAddress = %s, want %s", got, addr) + } + + // Test getting the Router Flag. + if got := na.RouterFlag(); !got { + t.Errorf("got RouterFlag = false, want = true") + } + + // Test getting the Solicited Flag. + if got := na.SolicitedFlag(); got { + t.Errorf("got SolicitedFlag = true, want = false") + } + + // Test getting the Override Flag. + if got := na.OverrideFlag(); !got { + t.Errorf("got OverrideFlag = false, want = true") + } + + // Test updating the Target Address. + addr2 := tcpip.Address("\x11\x12\x13\x14\x15\x16\x17\x18\x19\x1a\x1b\x1c\x1d\x1e\x1f\x11") + na.SetTargetAddress(addr2) + if got := na.TargetAddress(); got != addr2 { + t.Errorf("got TargetAddress = %s, want %s", got, addr2) + } + // Make sure the address got updated in the backing buffer. + if got := tcpip.Address(b[ndpNATargetAddressOffset:][:IPv6AddressSize]); got != addr2 { + t.Errorf("got targetaddress buffer = %s, want %s", got, addr2) + } + + // Test updating the Router Flag. + na.SetRouterFlag(false) + if got := na.RouterFlag(); got { + t.Errorf("got RouterFlag = true, want = false") + } + + // Test updating the Solicited Flag. + na.SetSolicitedFlag(true) + if got := na.SolicitedFlag(); !got { + t.Errorf("got SolicitedFlag = false, want = true") + } + + // Test updating the Override Flag. + na.SetOverrideFlag(false) + if got := na.OverrideFlag(); got { + t.Errorf("got OverrideFlag = true, want = false") + } + + // Make sure flags got updated in the backing buffer. + if got := b[ndpNAFlagsOffset]; got != 64 { + t.Errorf("got flags byte = %d, want = 64") + } +} + +func TestNDPRouterAdvert(t *testing.T) { + b := []byte{ + 64, 128, 1, 2, + 3, 4, 5, 6, + 7, 8, 9, 10, + } + + ra := NDPRouterAdvert(b) + + if got := ra.CurrHopLimit(); got != 64 { + t.Errorf("got ra.CurrHopLimit = %d, want = 64", got) + } + + if got := ra.ManagedAddrConfFlag(); !got { + t.Errorf("got ManagedAddrConfFlag = false, want = true") + } + + if got := ra.OtherConfFlag(); got { + t.Errorf("got OtherConfFlag = true, want = false") + } + + if got, want := ra.RouterLifetime(), time.Second*258; got != want { + t.Errorf("got ra.RouterLifetime = %d, want = %d", got, want) + } + + if got, want := ra.ReachableTime(), time.Millisecond*50595078; got != want { + t.Errorf("got ra.ReachableTime = %d, want = %d", got, want) + } + + if got, want := ra.RetransTimer(), time.Millisecond*117967114; got != want { + t.Errorf("got ra.RetransTimer = %d, want = %d", got, want) + } +} + +// TestNDPTargetLinkLayerAddressOptionEthernetAddress tests getting the +// Ethernet address from an NDPTargetLinkLayerAddressOption. +func TestNDPTargetLinkLayerAddressOptionEthernetAddress(t *testing.T) { + tests := []struct { + name string + buf []byte + expected tcpip.LinkAddress + }{ + { + "ValidMAC", + []byte{1, 2, 3, 4, 5, 6}, + tcpip.LinkAddress("\x01\x02\x03\x04\x05\x06"), + }, + { + "TLLBodyTooShort", + []byte{1, 2, 3, 4, 5}, + tcpip.LinkAddress([]byte(nil)), + }, + { + "TLLBodyLargerThanNeeded", + []byte{1, 2, 3, 4, 5, 6, 7, 8}, + tcpip.LinkAddress("\x01\x02\x03\x04\x05\x06"), + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + tll := NDPTargetLinkLayerAddressOption(test.buf) + if got := tll.EthernetAddress(); got != test.expected { + t.Errorf("got tll.EthernetAddress = %s, want = %s", got, test.expected) + } + }) + } + +} + +// TestNDPTargetLinkLayerAddressOptionSerialize tests serializing a +// NDPTargetLinkLayerAddressOption. +func TestNDPTargetLinkLayerAddressOptionSerialize(t *testing.T) { + tests := []struct { + name string + buf []byte + expectedBuf []byte + addr tcpip.LinkAddress + }{ + { + "Ethernet", + make([]byte, 8), + []byte{2, 1, 1, 2, 3, 4, 5, 6}, + "\x01\x02\x03\x04\x05\x06", + }, + { + "Padding", + []byte{1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, + []byte{2, 2, 1, 2, 3, 4, 5, 6, 7, 8, 0, 0, 0, 0, 0, 0}, + "\x01\x02\x03\x04\x05\x06\x07\x08", + }, + { + "Empty", + []byte{}, + []byte{}, + "", + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + opts := NDPOptions(test.buf) + serializer := NDPOptionsSerializer{ + NDPTargetLinkLayerAddressOption(test.addr), + } + if got, want := int(serializer.Length()), len(test.expectedBuf); got != want { + t.Fatalf("got Length = %d, want = %d", got, want) + } + opts.Serialize(serializer) + if !bytes.Equal(test.buf, test.expectedBuf) { + t.Fatalf("got b = %d, want = %d", test.buf, test.expectedBuf) + } + + it, err := opts.Iter(true) + if err != nil { + t.Fatalf("got Iter = (_, %s), want = (_, nil)", err) + } + + if len(test.expectedBuf) > 0 { + next, done, err := it.Next() + if err != nil { + t.Fatalf("got Next = (_, _, %s), want = (_, _, nil)", err) + } + if done { + t.Fatal("got Next = (_, true, _), want = (_, false, _)") + } + if got := next.Type(); got != NDPTargetLinkLayerAddressOptionType { + t.Fatalf("got Type %= %d, want = %d", got, NDPTargetLinkLayerAddressOptionType) + } + tll := next.(NDPTargetLinkLayerAddressOption) + if got, want := []byte(tll), test.expectedBuf[2:]; !bytes.Equal(got, want) { + t.Fatalf("got Next = (%x, _, _), want = (%x, _, _)", got, want) + } + + if got, want := tll.EthernetAddress(), tcpip.LinkAddress(test.expectedBuf[2:][:EthernetAddressSize]); got != want { + t.Errorf("got tll.MACAddress = %s, want = %s", got, want) + } + } + + // Iterator should not return anything else. + next, done, err := it.Next() + if err != nil { + t.Errorf("got Next = (_, _, %s), want = (_, _, nil)", err) + } + if !done { + t.Error("got Next = (_, false, _), want = (_, true, _)") + } + if next != nil { + t.Errorf("got Next = (%x, _, _), want = (nil, _, _)", next) + } + }) + } +} + +// TestNDPPrefixInformationOption tests the field getters and serialization of a +// NDPPrefixInformation. +func TestNDPPrefixInformationOption(t *testing.T) { + b := []byte{ + 43, 127, + 1, 2, 3, 4, + 5, 6, 7, 8, + 5, 5, 5, 5, + 9, 10, 11, 12, + 13, 14, 15, 16, + 17, 18, 19, 20, + 21, 22, 23, 24, + } + + targetBuf := []byte{1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1} + opts := NDPOptions(targetBuf) + serializer := NDPOptionsSerializer{ + NDPPrefixInformation(b), + } + opts.Serialize(serializer) + expectedBuf := []byte{ + 3, 4, 43, 64, + 1, 2, 3, 4, + 5, 6, 7, 8, + 0, 0, 0, 0, + 9, 10, 11, 12, + 13, 14, 15, 16, + 17, 18, 19, 20, + 21, 22, 23, 24, + } + if !bytes.Equal(targetBuf, expectedBuf) { + t.Fatalf("got targetBuf = %x, want = %x", targetBuf, expectedBuf) + } + + it, err := opts.Iter(true) + if err != nil { + t.Fatalf("got Iter = (_, %s), want = (_, nil)", err) + } + + next, done, err := it.Next() + if err != nil { + t.Fatalf("got Next = (_, _, %s), want = (_, _, nil)", err) + } + if done { + t.Fatal("got Next = (_, true, _), want = (_, false, _)") + } + if got := next.Type(); got != NDPPrefixInformationType { + t.Errorf("got Type = %d, want = %d", got, NDPPrefixInformationType) + } + + pi := next.(NDPPrefixInformation) + + if got := pi.Type(); got != 3 { + t.Errorf("got Type = %d, want = 3", got) + } + + if got := pi.Length(); got != 30 { + t.Errorf("got Length = %d, want = 30", got) + } + + if got := pi.PrefixLength(); got != 43 { + t.Errorf("got PrefixLength = %d, want = 43", got) + } + + if pi.OnLinkFlag() { + t.Error("got OnLinkFlag = true, want = false") + } + + if !pi.AutonomousAddressConfigurationFlag() { + t.Error("got AutonomousAddressConfigurationFlag = false, want = true") + } + + if got, want := pi.ValidLifetime(), 16909060*time.Second; got != want { + t.Errorf("got ValidLifetime = %d, want = %d", got, want) + } + + if got, want := pi.PreferredLifetime(), 84281096*time.Second; got != want { + t.Errorf("got PreferredLifetime = %d, want = %d", got, want) + } + + if got, want := pi.Prefix(), tcpip.Address("\x09\x0a\x0b\x0c\x0d\x0e\x0f\x10\x11\x12\x13\x14\x15\x16\x17\x18"); got != want { + t.Errorf("got Prefix = %s, want = %s", got, want) + } + + // Iterator should not return anything else. + next, done, err = it.Next() + if err != nil { + t.Errorf("got Next = (_, _, %s), want = (_, _, nil)", err) + } + if !done { + t.Error("got Next = (_, false, _), want = (_, true, _)") + } + if next != nil { + t.Errorf("got Next = (%x, _, _), want = (nil, _, _)", next) + } +} + +// TestNDPOptionsIterCheck tests that Iter will return false if the NDPOptions +// the iterator was returned for is malformed. +func TestNDPOptionsIterCheck(t *testing.T) { + tests := []struct { + name string + buf []byte + expected error + }{ + { + "ZeroLengthField", + []byte{0, 0, 0, 0, 0, 0, 0, 0}, + ErrNDPOptZeroLength, + }, + { + "ValidTargetLinkLayerAddressOption", + []byte{2, 1, 1, 2, 3, 4, 5, 6}, + nil, + }, + { + "TooSmallTargetLinkLayerAddressOption", + []byte{2, 1, 1, 2, 3, 4, 5}, + ErrNDPOptBufExhausted, + }, + { + "ValidPrefixInformation", + []byte{ + 3, 4, 43, 64, + 1, 2, 3, 4, + 5, 6, 7, 8, + 0, 0, 0, 0, + 9, 10, 11, 12, + 13, 14, 15, 16, + 17, 18, 19, 20, + 21, 22, 23, 24, + }, + nil, + }, + { + "TooSmallPrefixInformation", + []byte{ + 3, 4, 43, 64, + 1, 2, 3, 4, + 5, 6, 7, 8, + 0, 0, 0, 0, + 9, 10, 11, 12, + 13, 14, 15, 16, + 17, 18, 19, 20, + 21, 22, 23, + }, + ErrNDPOptBufExhausted, + }, + { + "InvalidPrefixInformationLength", + []byte{ + 3, 3, 43, 64, + 1, 2, 3, 4, + 5, 6, 7, 8, + 0, 0, 0, 0, + 9, 10, 11, 12, + 13, 14, 15, 16, + }, + ErrNDPOptMalformedBody, + }, + { + "ValidTargetLinkLayerAddressWithPrefixInformation", + []byte{ + // Target Link-Layer Address. + 2, 1, 1, 2, 3, 4, 5, 6, + + // Prefix information. + 3, 4, 43, 64, + 1, 2, 3, 4, + 5, 6, 7, 8, + 0, 0, 0, 0, + 9, 10, 11, 12, + 13, 14, 15, 16, + 17, 18, 19, 20, + 21, 22, 23, 24, + }, + nil, + }, + { + "ValidTargetLinkLayerAddressWithPrefixInformationWithUnrecognized", + []byte{ + // Target Link-Layer Address. + 2, 1, 1, 2, 3, 4, 5, 6, + + // 255 is an unrecognized type. If 255 ends up + // being the type for some recognized type, + // update 255 to some other unrecognized value. + 255, 2, 1, 2, 3, 4, 5, 6, 1, 2, 3, 4, 5, 6, 7, 8, + + // Prefix information. + 3, 4, 43, 64, + 1, 2, 3, 4, + 5, 6, 7, 8, + 0, 0, 0, 0, + 9, 10, 11, 12, + 13, 14, 15, 16, + 17, 18, 19, 20, + 21, 22, 23, 24, + }, + nil, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + opts := NDPOptions(test.buf) + + if _, err := opts.Iter(true); err != test.expected { + t.Fatalf("got Iter(true) = (_, %v), want = (_, %v)", err, test.expected) + } + + // test.buf may be malformed but we chose not to check + // the iterator so it must return true. + if _, err := opts.Iter(false); err != nil { + t.Fatalf("got Iter(false) = (_, %s), want = (_, nil)", err) + } + }) + } +} + +// TestNDPOptionsIter tests that we can iterator over a valid NDPOptions. Note, +// this test does not actually check any of the option's getters, it simply +// checks the option Type and Body. We have other tests that tests the option +// field gettings given an option body and don't need to duplicate those tests +// here. +func TestNDPOptionsIter(t *testing.T) { + buf := []byte{ + // Target Link-Layer Address. + 2, 1, 1, 2, 3, 4, 5, 6, + + // 255 is an unrecognized type. If 255 ends up being the type + // for some recognized type, update 255 to some other + // unrecognized value. Note, this option should be skipped when + // iterating. + 255, 2, 1, 2, 3, 4, 5, 6, 1, 2, 3, 4, 5, 6, 7, 8, + + // Prefix information. + 3, 4, 43, 64, + 1, 2, 3, 4, + 5, 6, 7, 8, + 0, 0, 0, 0, + 9, 10, 11, 12, + 13, 14, 15, 16, + 17, 18, 19, 20, + 21, 22, 23, 24, + } + + opts := NDPOptions(buf) + it, err := opts.Iter(true) + if err != nil { + t.Fatalf("got Iter = (_, %s), want = (_, nil)", err) + } + + // Test the first (Taret Link-Layer) option. + next, done, err := it.Next() + if err != nil { + t.Fatalf("got Next = (_, _, %s), want = (_, _, nil)", err) + } + if done { + t.Fatal("got Next = (_, true, _), want = (_, false, _)") + } + if got, want := []byte(next.(NDPTargetLinkLayerAddressOption)), buf[2:][:6]; !bytes.Equal(got, want) { + t.Errorf("got Next = (%x, _, _), want = (%x, _, _)", got, want) + } + if got := next.Type(); got != NDPTargetLinkLayerAddressOptionType { + t.Errorf("got Type = %d, want = %d", got, NDPTargetLinkLayerAddressOptionType) + } + + // Test the next (Prefix Information) option. + // Note, the unrecognized option should be skipped. + next, done, err = it.Next() + if err != nil { + t.Fatalf("got Next = (_, _, %s), want = (_, _, nil)", err) + } + if done { + t.Fatal("got Next = (_, true, _), want = (_, false, _)") + } + if got, want := next.(NDPPrefixInformation), buf[26:][:30]; !bytes.Equal(got, want) { + t.Errorf("got Next = (%x, _, _), want = (%x, _, _)", got, want) + } + if got := next.Type(); got != NDPPrefixInformationType { + t.Errorf("got Type = %d, want = %d", got, NDPPrefixInformationType) + } + + // Iterator should not return anything else. + next, done, err = it.Next() + if err != nil { + t.Errorf("got Next = (_, _, %s), want = (_, _, nil)", err) + } + if !done { + t.Error("got Next = (_, false, _), want = (_, true, _)") + } + if next != nil { + t.Errorf("got Next = (%x, _, _), want = (nil, _, _)", next) + } +} diff --git a/pkg/tcpip/iptables/BUILD b/pkg/tcpip/iptables/BUILD index 3fc14bacd..cc5f531e2 100644 --- a/pkg/tcpip/iptables/BUILD +++ b/pkg/tcpip/iptables/BUILD @@ -1,7 +1,7 @@ -package(licenses = ["notice"]) - load("//tools/go_stateify:defs.bzl", "go_library") +package(licenses = ["notice"]) + go_library( name = "iptables", srcs = [ diff --git a/pkg/tcpip/link/channel/channel.go b/pkg/tcpip/link/channel/channel.go index 18adb2085..22eefb564 100644 --- a/pkg/tcpip/link/channel/channel.go +++ b/pkg/tcpip/link/channel/channel.go @@ -65,14 +65,14 @@ func (e *Endpoint) Drain() int { } } -// Inject injects an inbound packet. -func (e *Endpoint) Inject(protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { - e.InjectLinkAddr(protocol, "", vv) +// InjectInbound injects an inbound packet. +func (e *Endpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { + e.InjectLinkAddr(protocol, "", pkt) } // InjectLinkAddr injects an inbound packet with a remote link address. -func (e *Endpoint) InjectLinkAddr(protocol tcpip.NetworkProtocolNumber, remote tcpip.LinkAddress, vv buffer.VectorisedView) { - e.dispatcher.DeliverNetworkPacket(e, remote, "" /* local */, protocol, vv.Clone(nil)) +func (e *Endpoint) InjectLinkAddr(protocol tcpip.NetworkProtocolNumber, remote tcpip.LinkAddress, pkt tcpip.PacketBuffer) { + e.dispatcher.DeliverNetworkPacket(e, remote, "" /* local */, protocol, pkt) } // Attach saves the stack network-layer dispatcher for use later when packets @@ -96,7 +96,7 @@ func (e *Endpoint) MTU() uint32 { func (e *Endpoint) Capabilities() stack.LinkEndpointCapabilities { caps := stack.LinkEndpointCapabilities(0) if e.GSO { - caps |= stack.CapabilityGSO + caps |= stack.CapabilityHardwareGSO } return caps } @@ -134,5 +134,49 @@ func (e *Endpoint) WritePacket(_ *stack.Route, gso *stack.GSO, hdr buffer.Prepen return nil } +// WritePackets stores outbound packets into the channel. +func (e *Endpoint) WritePackets(_ *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + payloadView := payload.ToView() + n := 0 +packetLoop: + for i := range hdrs { + hdr := &hdrs[i].Hdr + off := hdrs[i].Off + size := hdrs[i].Size + p := PacketInfo{ + Header: hdr.View(), + Proto: protocol, + Payload: buffer.NewViewFromBytes(payloadView[off : off+size]), + GSO: gso, + } + + select { + case e.C <- p: + n++ + default: + break packetLoop + } + } + + return n, nil +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *Endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { + p := PacketInfo{ + Header: packet.ToView(), + Proto: 0, + Payload: buffer.View{}, + GSO: nil, + } + + select { + case e.C <- p: + default: + } + + return nil +} + // Wait implements stack.LinkEndpoint.Wait. func (*Endpoint) Wait() {} diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go index 584db710e..edef7db26 100644 --- a/pkg/tcpip/link/fdbased/endpoint.go +++ b/pkg/tcpip/link/fdbased/endpoint.go @@ -82,7 +82,19 @@ const ( PacketMMap ) -// An endpoint implements the link-layer using a message-oriented file descriptor. +func (p PacketDispatchMode) String() string { + switch p { + case Readv: + return "Readv" + case RecvMMsg: + return "RecvMMsg" + case PacketMMap: + return "PacketMMap" + default: + return fmt.Sprintf("unknown packet dispatch mode %v", p) + } +} + type endpoint struct { // fds is the set of file descriptors each identifying one inbound/outbound // channel. The endpoint will dispatch from all inbound channels as well as @@ -153,6 +165,9 @@ type Options struct { // disabled. GSOMaxSize uint32 + // SoftwareGSOEnabled indicates whether software GSO is enabled or not. + SoftwareGSOEnabled bool + // PacketDispatchMode specifies the type of inbound dispatcher to be // used for this endpoint. PacketDispatchMode PacketDispatchMode @@ -166,6 +181,14 @@ type Options struct { RXChecksumOffload bool } +// fanoutID is used for AF_PACKET based endpoints to enable PACKET_FANOUT +// support in the host kernel. This allows us to use multiple FD's to receive +// from the same underlying NIC. The fanoutID needs to be the same for a given +// set of FD's that point to the same NIC. Trying to set the PACKET_FANOUT +// option for an FD with a fanoutID already in use by another FD for a different +// NIC will return an EINVAL. +var fanoutID = 1 + // New creates a new fd-based endpoint. // // Makes fd non-blocking, but does not take ownership of fd, which must remain @@ -222,7 +245,11 @@ func New(opts *Options) (stack.LinkEndpoint, error) { } if isSocket { if opts.GSOMaxSize != 0 { - e.caps |= stack.CapabilityGSO + if opts.SoftwareGSOEnabled { + e.caps |= stack.CapabilitySoftwareGSO + } else { + e.caps |= stack.CapabilityHardwareGSO + } e.gsoMaxSize = opts.GSOMaxSize } } @@ -233,6 +260,10 @@ func New(opts *Options) (stack.LinkEndpoint, error) { e.inboundDispatchers = append(e.inboundDispatchers, inboundDispatcher) } + // Increment fanoutID to ensure that we don't re-use the same fanoutID for + // the next endpoint. + fanoutID++ + return e, nil } @@ -253,7 +284,6 @@ func createInboundDispatcher(e *endpoint, fd int, isSocket bool) (linkDispatcher case *unix.SockaddrLinklayer: // enable PACKET_FANOUT mode is the underlying socket is // of type AF_PACKET. - const fanoutID = 1 const fanoutType = 0x8000 // PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG fanoutArg := fanoutID | fanoutType<<16 if err := syscall.SetsockoptInt(fd, syscall.SOL_PACKET, unix.PACKET_FANOUT, fanoutArg); err != nil { @@ -374,7 +404,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen eth.Encode(ethHdr) } - if e.Capabilities()&stack.CapabilityGSO != 0 { + if e.Capabilities()&stack.CapabilityHardwareGSO != 0 { vnetHdr := virtioNetHdr{} vnetHdrBuf := vnetHdrToByteSlice(&vnetHdr) if gso != nil { @@ -407,8 +437,130 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen return rawfile.NonBlockingWrite3(e.fds[0], hdr.View(), payload.ToView(), nil) } -// WriteRawPacket writes a raw packet directly to the file descriptor. -func (e *endpoint) WriteRawPacket(dest tcpip.Address, packet []byte) *tcpip.Error { +// WritePackets writes outbound packets to the file descriptor. If it is not +// currently writable, the packet is dropped. +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + var ethHdrBuf []byte + // hdr + data + iovLen := 2 + if e.hdrSize > 0 { + // Add ethernet header if needed. + ethHdrBuf = make([]byte, header.EthernetMinimumSize) + eth := header.Ethernet(ethHdrBuf) + ethHdr := &header.EthernetFields{ + DstAddr: r.RemoteLinkAddress, + Type: protocol, + } + + // Preserve the src address if it's set in the route. + if r.LocalLinkAddress != "" { + ethHdr.SrcAddr = r.LocalLinkAddress + } else { + ethHdr.SrcAddr = e.addr + } + eth.Encode(ethHdr) + iovLen++ + } + + n := len(hdrs) + + views := payload.Views() + /* + * Each bondary in views can add one more iovec. + * + * payload | | | | + * ----------------------------- + * packets | | | | | | | + * ----------------------------- + * iovecs | | | | | | | | | + */ + iovec := make([]syscall.Iovec, n*iovLen+len(views)-1) + mmsgHdrs := make([]rawfile.MMsgHdr, n) + + iovecIdx := 0 + viewIdx := 0 + viewOff := 0 + off := 0 + nextOff := 0 + for i := range hdrs { + prevIovecIdx := iovecIdx + mmsgHdr := &mmsgHdrs[i] + mmsgHdr.Msg.Iov = &iovec[iovecIdx] + packetSize := hdrs[i].Size + hdr := &hdrs[i].Hdr + + off = hdrs[i].Off + if off != nextOff { + // We stop in a different point last time. + size := packetSize + viewIdx = 0 + viewOff = 0 + for size > 0 { + if size >= len(views[viewIdx]) { + viewIdx++ + viewOff = 0 + size -= len(views[viewIdx]) + } else { + viewOff = size + size = 0 + } + } + } + nextOff = off + packetSize + + if ethHdrBuf != nil { + v := &iovec[iovecIdx] + v.Base = ðHdrBuf[0] + v.Len = uint64(len(ethHdrBuf)) + iovecIdx++ + } + + v := &iovec[iovecIdx] + hdrView := hdr.View() + v.Base = &hdrView[0] + v.Len = uint64(len(hdrView)) + iovecIdx++ + + for packetSize > 0 { + vec := &iovec[iovecIdx] + iovecIdx++ + + v := views[viewIdx] + vec.Base = &v[viewOff] + s := len(v) - viewOff + if s <= packetSize { + viewIdx++ + viewOff = 0 + } else { + s = packetSize + viewOff += s + } + vec.Len = uint64(s) + packetSize -= s + } + + mmsgHdr.Msg.Iovlen = uint64(iovecIdx - prevIovecIdx) + } + + packets := 0 + for packets < n { + sent, err := rawfile.NonBlockingSendMMsg(e.fds[0], mmsgHdrs) + if err != nil { + return packets, err + } + packets += sent + mmsgHdrs = mmsgHdrs[sent:] + } + return packets, nil +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { + return rawfile.NonBlockingWrite(e.fds[0], packet.ToView()) +} + +// InjectOutobund implements stack.InjectableEndpoint.InjectOutbound. +func (e *endpoint) InjectOutbound(dest tcpip.Address, packet []byte) *tcpip.Error { return rawfile.NonBlockingWrite(e.fds[0], packet) } @@ -445,9 +597,9 @@ func (e *InjectableEndpoint) Attach(dispatcher stack.NetworkDispatcher) { e.dispatcher = dispatcher } -// Inject injects an inbound packet. -func (e *InjectableEndpoint) Inject(protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { - e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, vv) +// InjectInbound injects an inbound packet. +func (e *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { + e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, pkt) } // NewInjectable creates a new fd-based InjectableEndpoint. diff --git a/pkg/tcpip/link/fdbased/endpoint_test.go b/pkg/tcpip/link/fdbased/endpoint_test.go index 04406bc9a..7e08e033b 100644 --- a/pkg/tcpip/link/fdbased/endpoint_test.go +++ b/pkg/tcpip/link/fdbased/endpoint_test.go @@ -45,7 +45,7 @@ const ( type packetInfo struct { raddr tcpip.LinkAddress proto tcpip.NetworkProtocolNumber - contents buffer.View + contents tcpip.PacketBuffer } type context struct { @@ -92,8 +92,8 @@ func (c *context) cleanup() { syscall.Close(c.fds[1]) } -func (c *context) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote tcpip.LinkAddress, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { - c.ch <- packetInfo{remote, protocol, vv.ToView()} +func (c *context) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote tcpip.LinkAddress, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { + c.ch <- packetInfo{remote, protocol, pkt} } func TestNoEthernetProperties(t *testing.T) { @@ -293,11 +293,12 @@ func TestDeliverPacket(t *testing.T) { b[i] = uint8(rand.Intn(256)) } + var hdr header.Ethernet if !eth { // So that it looks like an IPv4 packet. b[0] = 0x40 } else { - hdr := make(header.Ethernet, header.EthernetMinimumSize) + hdr = make(header.Ethernet, header.EthernetMinimumSize) hdr.Encode(&header.EthernetFields{ SrcAddr: raddr, DstAddr: laddr, @@ -315,14 +316,21 @@ func TestDeliverPacket(t *testing.T) { select { case pi := <-c.ch: want := packetInfo{ - raddr: raddr, - proto: proto, - contents: b, + raddr: raddr, + proto: proto, + contents: tcpip.PacketBuffer{ + Data: buffer.View(b).ToVectorisedView(), + LinkHeader: buffer.View(hdr), + }, } if !eth { want.proto = header.IPv4ProtocolNumber want.raddr = "" } + // want.contents.Data will be a single + // view, so make pi do the same for the + // DeepEqual check. + pi.contents.Data = pi.contents.Data.ToView().ToVectorisedView() if !reflect.DeepEqual(want, pi) { t.Fatalf("Unexpected received packet: %+v, want %+v", pi, want) } diff --git a/pkg/tcpip/link/fdbased/mmap.go b/pkg/tcpip/link/fdbased/mmap.go index 8bfeb97e4..62ed1e569 100644 --- a/pkg/tcpip/link/fdbased/mmap.go +++ b/pkg/tcpip/link/fdbased/mmap.go @@ -169,9 +169,10 @@ func (d *packetMMapDispatcher) dispatch() (bool, *tcpip.Error) { var ( p tcpip.NetworkProtocolNumber remote, local tcpip.LinkAddress + eth header.Ethernet ) if d.e.hdrSize > 0 { - eth := header.Ethernet(pkt) + eth = header.Ethernet(pkt) p = eth.Type() remote = eth.SourceAddress() local = eth.DestinationAddress() @@ -189,6 +190,9 @@ func (d *packetMMapDispatcher) dispatch() (bool, *tcpip.Error) { } pkt = pkt[d.e.hdrSize:] - d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, buffer.NewVectorisedView(len(pkt), []buffer.View{buffer.View(pkt)})) + d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, tcpip.PacketBuffer{ + Data: buffer.View(pkt).ToVectorisedView(), + LinkHeader: buffer.View(eth), + }) return true, nil } diff --git a/pkg/tcpip/link/fdbased/packet_dispatchers.go b/pkg/tcpip/link/fdbased/packet_dispatchers.go index 7ca217e5b..c67d684ce 100644 --- a/pkg/tcpip/link/fdbased/packet_dispatchers.go +++ b/pkg/tcpip/link/fdbased/packet_dispatchers.go @@ -53,7 +53,7 @@ func newReadVDispatcher(fd int, e *endpoint) (linkDispatcher, error) { d := &readVDispatcher{fd: fd, e: e} d.views = make([]buffer.View, len(BufConfig)) iovLen := len(BufConfig) - if d.e.Capabilities()&stack.CapabilityGSO != 0 { + if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 { iovLen++ } d.iovecs = make([]syscall.Iovec, iovLen) @@ -63,7 +63,7 @@ func newReadVDispatcher(fd int, e *endpoint) (linkDispatcher, error) { func (d *readVDispatcher) allocateViews(bufConfig []int) { var vnetHdr [virtioNetHdrSize]byte vnetHdrOff := 0 - if d.e.Capabilities()&stack.CapabilityGSO != 0 { + if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 { // The kernel adds virtioNetHdr before each packet, but // we don't use it, so so we allocate a buffer for it, // add it in iovecs but don't add it in a view. @@ -106,7 +106,7 @@ func (d *readVDispatcher) dispatch() (bool, *tcpip.Error) { if err != nil { return false, err } - if d.e.Capabilities()&stack.CapabilityGSO != 0 { + if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 { // Skip virtioNetHdr which is added before each packet, it // isn't used and it isn't in a view. n -= virtioNetHdrSize @@ -118,9 +118,10 @@ func (d *readVDispatcher) dispatch() (bool, *tcpip.Error) { var ( p tcpip.NetworkProtocolNumber remote, local tcpip.LinkAddress + eth header.Ethernet ) if d.e.hdrSize > 0 { - eth := header.Ethernet(d.views[0]) + eth = header.Ethernet(d.views[0][:header.EthernetMinimumSize]) p = eth.Type() remote = eth.SourceAddress() local = eth.DestinationAddress() @@ -138,10 +139,13 @@ func (d *readVDispatcher) dispatch() (bool, *tcpip.Error) { } used := d.capViews(n, BufConfig) - vv := buffer.NewVectorisedView(n, d.views[:used]) - vv.TrimFront(d.e.hdrSize) + pkt := tcpip.PacketBuffer{ + Data: buffer.NewVectorisedView(n, append([]buffer.View(nil), d.views[:used]...)), + LinkHeader: buffer.View(eth), + } + pkt.Data.TrimFront(d.e.hdrSize) - d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, vv) + d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, pkt) // Prepare e.views for another packet: release used views. for i := 0; i < used; i++ { @@ -194,7 +198,7 @@ func newRecvMMsgDispatcher(fd int, e *endpoint) (linkDispatcher, error) { } d.iovecs = make([][]syscall.Iovec, MaxMsgsPerRecv) iovLen := len(BufConfig) - if d.e.Capabilities()&stack.CapabilityGSO != 0 { + if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 { // virtioNetHdr is prepended before each packet. iovLen++ } @@ -225,7 +229,7 @@ func (d *recvMMsgDispatcher) allocateViews(bufConfig []int) { for k := 0; k < len(d.views); k++ { var vnetHdr [virtioNetHdrSize]byte vnetHdrOff := 0 - if d.e.Capabilities()&stack.CapabilityGSO != 0 { + if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 { // The kernel adds virtioNetHdr before each packet, but // we don't use it, so so we allocate a buffer for it, // add it in iovecs but don't add it in a view. @@ -261,7 +265,7 @@ func (d *recvMMsgDispatcher) dispatch() (bool, *tcpip.Error) { // Process each of received packets. for k := 0; k < nMsgs; k++ { n := int(d.msgHdrs[k].Len) - if d.e.Capabilities()&stack.CapabilityGSO != 0 { + if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 { n -= virtioNetHdrSize } if n <= d.e.hdrSize { @@ -271,9 +275,10 @@ func (d *recvMMsgDispatcher) dispatch() (bool, *tcpip.Error) { var ( p tcpip.NetworkProtocolNumber remote, local tcpip.LinkAddress + eth header.Ethernet ) if d.e.hdrSize > 0 { - eth := header.Ethernet(d.views[k][0]) + eth = header.Ethernet(d.views[k][0]) p = eth.Type() remote = eth.SourceAddress() local = eth.DestinationAddress() @@ -291,9 +296,12 @@ func (d *recvMMsgDispatcher) dispatch() (bool, *tcpip.Error) { } used := d.capViews(k, int(n), BufConfig) - vv := buffer.NewVectorisedView(int(n), d.views[k][:used]) - vv.TrimFront(d.e.hdrSize) - d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, vv) + pkt := tcpip.PacketBuffer{ + Data: buffer.NewVectorisedView(int(n), append([]buffer.View(nil), d.views[k][:used]...)), + LinkHeader: buffer.View(eth), + } + pkt.Data.TrimFront(d.e.hdrSize) + d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, pkt) // Prepare e.views for another packet: release used views. for i := 0; i < used; i++ { diff --git a/pkg/tcpip/link/loopback/BUILD b/pkg/tcpip/link/loopback/BUILD index 47a54845c..23e4d1418 100644 --- a/pkg/tcpip/link/loopback/BUILD +++ b/pkg/tcpip/link/loopback/BUILD @@ -10,6 +10,7 @@ go_library( deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/header", "//pkg/tcpip/stack", ], ) diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go index b36629d2c..bc5d8a2f3 100644 --- a/pkg/tcpip/link/loopback/loopback.go +++ b/pkg/tcpip/link/loopback/loopback.go @@ -23,6 +23,7 @@ package loopback import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -70,21 +71,45 @@ func (*endpoint) LinkAddress() tcpip.LinkAddress { return "" } +// Wait implements stack.LinkEndpoint.Wait. +func (*endpoint) Wait() {} + // WritePacket implements stack.LinkEndpoint.WritePacket. It delivers outbound // packets to the network-layer dispatcher. func (e *endpoint) WritePacket(_ *stack.Route, _ *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { views := make([]buffer.View, 1, 1+len(payload.Views())) views[0] = hdr.View() views = append(views, payload.Views()...) - vv := buffer.NewVectorisedView(len(views[0])+payload.Size(), views) // Because we're immediately turning around and writing the packet back to the // rx path, we intentionally don't preserve the remote and local link // addresses from the stack.Route we're passed. - e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, vv) + e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, tcpip.PacketBuffer{ + Data: buffer.NewVectorisedView(len(views[0])+payload.Size(), views), + }) return nil } -// Wait implements stack.LinkEndpoint.Wait. -func (*endpoint) Wait() {} +// WritePackets implements stack.LinkEndpoint.WritePackets. +func (e *endpoint) WritePackets(_ *stack.Route, _ *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + panic("not implemented") +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { + // Reject the packet if it's shorter than an ethernet header. + if packet.Size() < header.EthernetMinimumSize { + return tcpip.ErrBadAddress + } + + // There should be an ethernet header at the beginning of packet. + linkHeader := header.Ethernet(packet.First()[:header.EthernetMinimumSize]) + packet.TrimFront(len(linkHeader)) + e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, linkHeader.Type(), tcpip.PacketBuffer{ + Data: packet, + LinkHeader: buffer.View(linkHeader), + }) + + return nil +} diff --git a/pkg/tcpip/link/muxed/injectable.go b/pkg/tcpip/link/muxed/injectable.go index 7c946101d..9a8e8ebfe 100644 --- a/pkg/tcpip/link/muxed/injectable.go +++ b/pkg/tcpip/link/muxed/injectable.go @@ -79,29 +79,47 @@ func (m *InjectableEndpoint) IsAttached() bool { return m.dispatcher != nil } -// Inject implements stack.InjectableLinkEndpoint. -func (m *InjectableEndpoint) Inject(protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { - m.dispatcher.DeliverNetworkPacket(m, "" /* remote */, "" /* local */, protocol, vv) +// InjectInbound implements stack.InjectableLinkEndpoint. +func (m *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { + m.dispatcher.DeliverNetworkPacket(m, "" /* remote */, "" /* local */, protocol, pkt) +} + +// WritePackets writes outbound packets to the appropriate +// LinkInjectableEndpoint based on the RemoteAddress. HandleLocal only works if +// r.RemoteAddress has a route registered in this endpoint. +func (m *InjectableEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + endpoint, ok := m.routes[r.RemoteAddress] + if !ok { + return 0, tcpip.ErrNoRoute + } + return endpoint.WritePackets(r, gso, hdrs, payload, protocol) } // WritePacket writes outbound packets to the appropriate LinkInjectableEndpoint // based on the RemoteAddress. HandleLocal only works if r.RemoteAddress has a // route registered in this endpoint. -func (m *InjectableEndpoint) WritePacket(r *stack.Route, _ *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (m *InjectableEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { if endpoint, ok := m.routes[r.RemoteAddress]; ok { - return endpoint.WritePacket(r, nil /* gso */, hdr, payload, protocol) + return endpoint.WritePacket(r, gso, hdr, payload, protocol) } return tcpip.ErrNoRoute } -// WriteRawPacket writes outbound packets to the appropriate +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (m *InjectableEndpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { + // WriteRawPacket doesn't get a route or network address, so there's + // nowhere to write this. + return tcpip.ErrNoRoute +} + +// InjectOutbound writes outbound packets to the appropriate // LinkInjectableEndpoint based on the dest address. -func (m *InjectableEndpoint) WriteRawPacket(dest tcpip.Address, packet []byte) *tcpip.Error { +func (m *InjectableEndpoint) InjectOutbound(dest tcpip.Address, packet []byte) *tcpip.Error { endpoint, ok := m.routes[dest] if !ok { return tcpip.ErrNoRoute } - return endpoint.WriteRawPacket(dest, packet) + return endpoint.InjectOutbound(dest, packet) } // Wait implements stack.LinkEndpoint.Wait. diff --git a/pkg/tcpip/link/muxed/injectable_test.go b/pkg/tcpip/link/muxed/injectable_test.go index 3086fec00..9cd300af8 100644 --- a/pkg/tcpip/link/muxed/injectable_test.go +++ b/pkg/tcpip/link/muxed/injectable_test.go @@ -31,7 +31,7 @@ import ( func TestInjectableEndpointRawDispatch(t *testing.T) { endpoint, sock, dstIP := makeTestInjectableEndpoint(t) - endpoint.WriteRawPacket(dstIP, []byte{0xFA}) + endpoint.InjectOutbound(dstIP, []byte{0xFA}) buf := make([]byte, ipv4.MaxTotalSize) bytesRead, err := sock.Read(buf) diff --git a/pkg/tcpip/link/rawfile/BUILD b/pkg/tcpip/link/rawfile/BUILD index 2e8bc772a..05c7b8024 100644 --- a/pkg/tcpip/link/rawfile/BUILD +++ b/pkg/tcpip/link/rawfile/BUILD @@ -16,5 +16,8 @@ go_library( visibility = [ "//visibility:public", ], - deps = ["//pkg/tcpip"], + deps = [ + "//pkg/tcpip", + "@org_golang_x_sys//unix:go_default_library", + ], ) diff --git a/pkg/tcpip/link/rawfile/blockingpoll_yield_unsafe.go b/pkg/tcpip/link/rawfile/blockingpoll_yield_unsafe.go index dda3b10a6..0b5a6cf49 100644 --- a/pkg/tcpip/link/rawfile/blockingpoll_yield_unsafe.go +++ b/pkg/tcpip/link/rawfile/blockingpoll_yield_unsafe.go @@ -14,7 +14,7 @@ // +build linux,amd64 linux,arm64 // +build go1.12 -// +build !go1.14 +// +build !go1.15 // Check go:linkname function signatures when updating Go version. diff --git a/pkg/tcpip/link/rawfile/rawfile_unsafe.go b/pkg/tcpip/link/rawfile/rawfile_unsafe.go index 7e286a3a6..44e25d475 100644 --- a/pkg/tcpip/link/rawfile/rawfile_unsafe.go +++ b/pkg/tcpip/link/rawfile/rawfile_unsafe.go @@ -22,6 +22,7 @@ import ( "syscall" "unsafe" + "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/tcpip" ) @@ -101,6 +102,16 @@ func NonBlockingWrite3(fd int, b1, b2, b3 []byte) *tcpip.Error { return nil } +// NonBlockingSendMMsg sends multiple messages on a socket. +func NonBlockingSendMMsg(fd int, msgHdrs []MMsgHdr) (int, *tcpip.Error) { + n, _, e := syscall.RawSyscall6(unix.SYS_SENDMMSG, uintptr(fd), uintptr(unsafe.Pointer(&msgHdrs[0])), uintptr(len(msgHdrs)), syscall.MSG_DONTWAIT, 0, 0) + if e != 0 { + return 0, TranslateErrno(e) + } + + return int(n), nil +} + // PollEvent represents the pollfd structure passed to a poll() system call. type PollEvent struct { FD int32 diff --git a/pkg/tcpip/link/sharedmem/sharedmem.go b/pkg/tcpip/link/sharedmem/sharedmem.go index 9e71d4edf..2bace5298 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem.go +++ b/pkg/tcpip/link/sharedmem/sharedmem.go @@ -212,6 +212,26 @@ func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, hdr buffer.Prependa return nil } +// WritePackets implements stack.LinkEndpoint.WritePackets. +func (e *endpoint) WritePackets(r *stack.Route, _ *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + panic("not implemented") +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { + v := packet.ToView() + // Transmit the packet. + e.mu.Lock() + ok := e.tx.transmit(v, buffer.View{}) + e.mu.Unlock() + + if !ok { + return tcpip.ErrWouldBlock + } + + return nil +} + // dispatchLoop reads packets from the rx queue in a loop and dispatches them // to the network stack. func (e *endpoint) dispatchLoop(d stack.NetworkDispatcher) { @@ -253,8 +273,11 @@ func (e *endpoint) dispatchLoop(d stack.NetworkDispatcher) { } // Send packet up the stack. - eth := header.Ethernet(b) - d.DeliverNetworkPacket(e, eth.SourceAddress(), eth.DestinationAddress(), eth.Type(), buffer.View(b[header.EthernetMinimumSize:]).ToVectorisedView()) + eth := header.Ethernet(b[:header.EthernetMinimumSize]) + d.DeliverNetworkPacket(e, eth.SourceAddress(), eth.DestinationAddress(), eth.Type(), tcpip.PacketBuffer{ + Data: buffer.View(b[header.EthernetMinimumSize:]).ToVectorisedView(), + LinkHeader: buffer.View(eth), + }) } // Clean state. diff --git a/pkg/tcpip/link/sharedmem/sharedmem_test.go b/pkg/tcpip/link/sharedmem/sharedmem_test.go index 0e9ba0846..199406886 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem_test.go +++ b/pkg/tcpip/link/sharedmem/sharedmem_test.go @@ -78,9 +78,10 @@ func (q *queueBuffers) cleanup() { } type packetInfo struct { - addr tcpip.LinkAddress - proto tcpip.NetworkProtocolNumber - vv buffer.VectorisedView + addr tcpip.LinkAddress + proto tcpip.NetworkProtocolNumber + vv buffer.VectorisedView + linkHeader buffer.View } type testContext struct { @@ -130,12 +131,12 @@ func newTestContext(t *testing.T, mtu, bufferSize uint32, addr tcpip.LinkAddress return c } -func (c *testContext) DeliverNetworkPacket(_ stack.LinkEndpoint, remoteLinkAddr, localLinkAddr tcpip.LinkAddress, proto tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { +func (c *testContext) DeliverNetworkPacket(_ stack.LinkEndpoint, remoteLinkAddr, localLinkAddr tcpip.LinkAddress, proto tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { c.mu.Lock() c.packets = append(c.packets, packetInfo{ addr: remoteLinkAddr, proto: proto, - vv: vv.Clone(nil), + vv: pkt.Data.Clone(nil), }) c.mu.Unlock() diff --git a/pkg/tcpip/link/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go index e401dce44..d71a03cd2 100644 --- a/pkg/tcpip/link/sniffer/sniffer.go +++ b/pkg/tcpip/link/sniffer/sniffer.go @@ -116,19 +116,19 @@ func NewWithFile(lower stack.LinkEndpoint, file *os.File, snapLen uint32) (stack // DeliverNetworkPacket implements the stack.NetworkDispatcher interface. It is // called by the link-layer endpoint being wrapped when a packet arrives, and // logs the packet before forwarding to the actual dispatcher. -func (e *endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { +func (e *endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { if atomic.LoadUint32(&LogPackets) == 1 && e.file == nil { - logPacket("recv", protocol, vv.First(), nil) + logPacket("recv", protocol, pkt.Data.First(), nil) } if e.file != nil && atomic.LoadUint32(&LogPacketsToFile) == 1 { - vs := vv.Views() - length := vv.Size() + vs := pkt.Data.Views() + length := pkt.Data.Size() if length > int(e.maxPCAPLen) { length = int(e.maxPCAPLen) } buf := bytes.NewBuffer(make([]byte, 0, pcapPacketHeaderLen+length)) - if err := binary.Write(buf, binary.BigEndian, newPCAPPacketHeader(uint32(length), uint32(vv.Size()))); err != nil { + if err := binary.Write(buf, binary.BigEndian, newPCAPPacketHeader(uint32(length), uint32(pkt.Data.Size()))); err != nil { panic(err) } for _, v := range vs { @@ -147,7 +147,7 @@ func (e *endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local panic(err) } } - e.dispatcher.DeliverNetworkPacket(e, remote, local, protocol, vv) + e.dispatcher.DeliverNetworkPacket(e, remote, local, protocol, pkt) } // Attach implements the stack.LinkEndpoint interface. It saves the dispatcher @@ -193,10 +193,7 @@ func (e *endpoint) GSOMaxSize() uint32 { return 0 } -// WritePacket implements the stack.LinkEndpoint interface. It is called by -// higher-level protocols to write packets; it just logs the packet and forwards -// the request to the lower endpoint. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (e *endpoint) dumpPacket(gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) { if atomic.LoadUint32(&LogPackets) == 1 && e.file == nil { logPacket("send", protocol, hdr.View(), gso) } @@ -218,28 +215,74 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen panic(err) } length -= len(hdrBuf) - if length > 0 { - for _, v := range payload.Views() { - if len(v) > length { - v = v[:length] - } - n, err := buf.Write(v) - if err != nil { - panic(err) - } - length -= n - if length == 0 { - break - } - } - } + logVectorisedView(payload, length, buf) if _, err := e.file.Write(buf.Bytes()); err != nil { panic(err) } } +} + +// WritePacket implements the stack.LinkEndpoint interface. It is called by +// higher-level protocols to write packets; it just logs the packet and +// forwards the request to the lower endpoint. +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { + e.dumpPacket(gso, hdr, payload, protocol) return e.lower.WritePacket(r, gso, hdr, payload, protocol) } +// WritePackets implements the stack.LinkEndpoint interface. It is called by +// higher-level protocols to write packets; it just logs the packet and +// forwards the request to the lower endpoint. +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + view := payload.ToView() + for _, d := range hdrs { + e.dumpPacket(gso, d.Hdr, buffer.NewVectorisedView(d.Size, []buffer.View{view[d.Off:][:d.Size]}), protocol) + } + return e.lower.WritePackets(r, gso, hdrs, payload, protocol) +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { + if atomic.LoadUint32(&LogPackets) == 1 && e.file == nil { + logPacket("send", 0, buffer.View("[raw packet, no header available]"), nil /* gso */) + } + if e.file != nil && atomic.LoadUint32(&LogPacketsToFile) == 1 { + length := packet.Size() + if length > int(e.maxPCAPLen) { + length = int(e.maxPCAPLen) + } + + buf := bytes.NewBuffer(make([]byte, 0, pcapPacketHeaderLen+length)) + if err := binary.Write(buf, binary.BigEndian, newPCAPPacketHeader(uint32(length), uint32(packet.Size()))); err != nil { + panic(err) + } + logVectorisedView(packet, length, buf) + if _, err := e.file.Write(buf.Bytes()); err != nil { + panic(err) + } + } + return e.lower.WriteRawPacket(packet) +} + +func logVectorisedView(vv buffer.VectorisedView, length int, buf *bytes.Buffer) { + if length <= 0 { + return + } + for _, v := range vv.Views() { + if len(v) > length { + v = v[:length] + } + n, err := buf.Write(v) + if err != nil { + panic(err) + } + length -= n + if length == 0 { + return + } + } +} + // Wait implements stack.LinkEndpoint.Wait. func (*endpoint) Wait() {} diff --git a/pkg/tcpip/link/waitable/waitable.go b/pkg/tcpip/link/waitable/waitable.go index 5a1791cb5..b440970e0 100644 --- a/pkg/tcpip/link/waitable/waitable.go +++ b/pkg/tcpip/link/waitable/waitable.go @@ -50,12 +50,12 @@ func New(lower stack.LinkEndpoint) *Endpoint { // It is called by the link-layer endpoint being wrapped when a packet arrives, // and only forwards to the actual dispatcher if Wait or WaitDispatch haven't // been called. -func (e *Endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { +func (e *Endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { if !e.dispatchGate.Enter() { return } - e.dispatcher.DeliverNetworkPacket(e, remote, local, protocol, vv) + e.dispatcher.DeliverNetworkPacket(e, remote, local, protocol, pkt) e.dispatchGate.Leave() } @@ -109,6 +109,30 @@ func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen return err } +// WritePackets implements stack.LinkEndpoint.WritePackets. It is called by +// higher-level protocols to write packets. It only forwards packets to the +// lower endpoint if Wait or WaitWrite haven't been called. +func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + if !e.writeGate.Enter() { + return len(hdrs), nil + } + + n, err := e.lower.WritePackets(r, gso, hdrs, payload, protocol) + e.writeGate.Leave() + return n, err +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *Endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { + if !e.writeGate.Enter() { + return nil + } + + err := e.lower.WriteRawPacket(packet) + e.writeGate.Leave() + return err +} + // WaitWrite prevents new calls to WritePacket from reaching the lower endpoint, // and waits for inflight ones to finish before returning. func (e *Endpoint) WaitWrite() { diff --git a/pkg/tcpip/link/waitable/waitable_test.go b/pkg/tcpip/link/waitable/waitable_test.go index ae23c96b7..df2e70e54 100644 --- a/pkg/tcpip/link/waitable/waitable_test.go +++ b/pkg/tcpip/link/waitable/waitable_test.go @@ -35,7 +35,7 @@ type countedEndpoint struct { dispatcher stack.NetworkDispatcher } -func (e *countedEndpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { +func (e *countedEndpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { e.dispatchCount++ } @@ -70,6 +70,17 @@ func (e *countedEndpoint) WritePacket(r *stack.Route, _ *stack.GSO, hdr buffer.P return nil } +// WritePackets implements stack.LinkEndpoint.WritePackets. +func (e *countedEndpoint) WritePackets(r *stack.Route, _ *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + e.writeCount += len(hdrs) + return len(hdrs), nil +} + +func (e *countedEndpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { + e.writeCount++ + return nil +} + // Wait implements stack.LinkEndpoint.Wait. func (*countedEndpoint) Wait() {} @@ -109,21 +120,21 @@ func TestWaitDispatch(t *testing.T) { } // Dispatch and check that it goes through. - ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, buffer.VectorisedView{}) + ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, tcpip.PacketBuffer{}) if want := 1; ep.dispatchCount != want { t.Fatalf("Unexpected dispatchCount: got=%v, want=%v", ep.dispatchCount, want) } // Wait on writes, then try to dispatch. It must go through. wep.WaitWrite() - ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, buffer.VectorisedView{}) + ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, tcpip.PacketBuffer{}) if want := 2; ep.dispatchCount != want { t.Fatalf("Unexpected dispatchCount: got=%v, want=%v", ep.dispatchCount, want) } // Wait on dispatches, then try to dispatch. It must not go through. wep.WaitDispatch() - ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, buffer.VectorisedView{}) + ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, tcpip.PacketBuffer{}) if want := 2; ep.dispatchCount != want { t.Fatalf("Unexpected dispatchCount: got=%v, want=%v", ep.dispatchCount, want) } diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index 26cf1c528..0ee509ebe 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -42,7 +42,7 @@ const ( // endpoint implements stack.NetworkEndpoint. type endpoint struct { - nicid tcpip.NICID + nicID tcpip.NICID linkEP stack.LinkEndpoint linkAddrCache stack.LinkAddressCache } @@ -58,7 +58,7 @@ func (e *endpoint) MTU() uint32 { } func (e *endpoint) NICID() tcpip.NICID { - return e.nicid + return e.nicID } func (e *endpoint) Capabilities() stack.LinkEndpointCapabilities { @@ -79,16 +79,21 @@ func (e *endpoint) MaxHeaderLength() uint16 { func (e *endpoint) Close() {} -func (e *endpoint) WritePacket(*stack.Route, *stack.GSO, buffer.Prependable, buffer.VectorisedView, tcpip.TransportProtocolNumber, uint8, stack.PacketLooping) *tcpip.Error { +func (e *endpoint) WritePacket(*stack.Route, *stack.GSO, buffer.Prependable, buffer.VectorisedView, stack.NetworkHeaderParams, stack.PacketLooping) *tcpip.Error { return tcpip.ErrNotSupported } +// WritePackets implements stack.NetworkEndpoint.WritePackets. +func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, []stack.PacketDescriptor, buffer.VectorisedView, stack.NetworkHeaderParams, stack.PacketLooping) (int, *tcpip.Error) { + return 0, tcpip.ErrNotSupported +} + func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error { return tcpip.ErrNotSupported } -func (e *endpoint) HandlePacket(r *stack.Route, vv buffer.VectorisedView) { - v := vv.First() +func (e *endpoint) HandlePacket(r *stack.Route, pkt tcpip.PacketBuffer) { + v := pkt.Data.First() h := header.ARP(v) if !h.IsValid() { return @@ -97,7 +102,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, vv buffer.VectorisedView) { switch h.Op() { case header.ARPRequest: localAddr := tcpip.Address(h.ProtocolAddressTarget()) - if e.linkAddrCache.CheckLocalAddress(e.nicid, header.IPv4ProtocolNumber, localAddr) == 0 { + if e.linkAddrCache.CheckLocalAddress(e.nicID, header.IPv4ProtocolNumber, localAddr) == 0 { return // we have no useful answer, ignore the request } hdr := buffer.NewPrependable(int(e.linkEP.MaxHeaderLength()) + header.ARPSize) @@ -109,7 +114,11 @@ func (e *endpoint) HandlePacket(r *stack.Route, vv buffer.VectorisedView) { copy(pkt.HardwareAddressTarget(), h.HardwareAddressSender()) copy(pkt.ProtocolAddressTarget(), h.ProtocolAddressSender()) e.linkEP.WritePacket(r, nil /* gso */, hdr, buffer.VectorisedView{}, ProtocolNumber) + fallthrough // also fill the cache from requests case header.ARPReply: + addr := tcpip.Address(h.ProtocolAddressSender()) + linkAddr := tcpip.LinkAddress(h.HardwareAddressSender()) + e.linkAddrCache.AddLinkAddress(e.nicID, addr, linkAddr) } } @@ -126,12 +135,12 @@ func (*protocol) ParseAddresses(v buffer.View) (src, dst tcpip.Address) { return tcpip.Address(h.ProtocolAddressSender()), ProtocolAddress } -func (p *protocol) NewEndpoint(nicid tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, sender stack.LinkEndpoint) (stack.NetworkEndpoint, *tcpip.Error) { +func (p *protocol) NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, sender stack.LinkEndpoint) (stack.NetworkEndpoint, *tcpip.Error) { if addrWithPrefix.Address != ProtocolAddress { return nil, tcpip.ErrBadLocalAddress } return &endpoint{ - nicid: nicid, + nicID: nicID, linkEP: sender, linkAddrCache: linkAddrCache, }, nil diff --git a/pkg/tcpip/network/arp/arp_test.go b/pkg/tcpip/network/arp/arp_test.go index 88b57ec03..47098bfdc 100644 --- a/pkg/tcpip/network/arp/arp_test.go +++ b/pkg/tcpip/network/arp/arp_test.go @@ -102,7 +102,9 @@ func TestDirectRequest(t *testing.T) { inject := func(addr tcpip.Address) { copy(h.ProtocolAddressTarget(), addr) - c.linkEP.Inject(arp.ProtocolNumber, v.ToVectorisedView()) + c.linkEP.InjectInbound(arp.ProtocolNumber, tcpip.PacketBuffer{ + Data: v.ToVectorisedView(), + }) } for i, address := range []tcpip.Address{stackAddr1, stackAddr2} { diff --git a/pkg/tcpip/network/fragmentation/BUILD b/pkg/tcpip/network/fragmentation/BUILD index c5c7aad86..2cad0a0b6 100644 --- a/pkg/tcpip/network/fragmentation/BUILD +++ b/pkg/tcpip/network/fragmentation/BUILD @@ -1,10 +1,9 @@ load("@io_bazel_rules_go//go:def.bzl", "go_test") - -package(licenses = ["notice"]) - load("//tools/go_generics:defs.bzl", "go_template_instance") load("//tools/go_stateify:defs.bzl", "go_library") +package(licenses = ["notice"]) + go_template_instance( name = "reassembler_list", out = "reassembler_list.go", @@ -29,6 +28,7 @@ go_library( visibility = ["//:sandbox"], deps = [ "//pkg/log", + "//pkg/tcpip", "//pkg/tcpip/buffer", ], ) diff --git a/pkg/tcpip/network/fragmentation/fragmentation.go b/pkg/tcpip/network/fragmentation/fragmentation.go index 1628a82be..6da5238ec 100644 --- a/pkg/tcpip/network/fragmentation/fragmentation.go +++ b/pkg/tcpip/network/fragmentation/fragmentation.go @@ -17,6 +17,7 @@ package fragmentation import ( + "fmt" "log" "sync" "time" @@ -82,7 +83,7 @@ func NewFragmentation(highMemoryLimit, lowMemoryLimit int, reassemblingTimeout t // 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. -func (f *Fragmentation) Process(id uint32, first, last uint16, more bool, vv buffer.VectorisedView) (buffer.VectorisedView, bool) { +func (f *Fragmentation) Process(id uint32, first, last uint16, more bool, vv buffer.VectorisedView) (buffer.VectorisedView, bool, error) { f.mu.Lock() r, ok := f.reassemblers[id] if ok && r.tooOld(f.timeout) { @@ -97,8 +98,15 @@ func (f *Fragmentation) Process(id uint32, first, last uint16, more bool, vv buf } f.mu.Unlock() - res, done, consumed := r.process(first, last, more, vv) - + res, done, consumed, err := r.process(first, last, more, vv) + if err != nil { + // We probably got an invalid sequence of fragments. Just + // discard the reassembler and move on. + f.mu.Lock() + f.release(r) + f.mu.Unlock() + return buffer.VectorisedView{}, false, fmt.Errorf("fragmentation processing error: %v", err) + } f.mu.Lock() f.size += consumed if done { @@ -114,7 +122,7 @@ func (f *Fragmentation) Process(id uint32, first, last uint16, more bool, vv buf } } f.mu.Unlock() - return res, done + return res, done, nil } func (f *Fragmentation) release(r *reassembler) { diff --git a/pkg/tcpip/network/fragmentation/fragmentation_test.go b/pkg/tcpip/network/fragmentation/fragmentation_test.go index 799798544..72c0f53be 100644 --- a/pkg/tcpip/network/fragmentation/fragmentation_test.go +++ b/pkg/tcpip/network/fragmentation/fragmentation_test.go @@ -83,7 +83,10 @@ func TestFragmentationProcess(t *testing.T) { t.Run(c.comment, func(t *testing.T) { f := NewFragmentation(1024, 512, DefaultReassembleTimeout) for i, in := range c.in { - vv, done := f.Process(in.id, in.first, in.last, in.more, in.vv) + vv, done, err := f.Process(in.id, in.first, in.last, in.more, in.vv) + if err != nil { + t.Fatalf("f.Process(%+v, %+d, %+d, %t, %+v) failed: %v", in.id, in.first, in.last, in.more, in.vv, err) + } if !reflect.DeepEqual(vv, c.out[i].vv) { t.Errorf("got Process(%d) = %+v, want = %+v", i, vv, c.out[i].vv) } @@ -114,7 +117,10 @@ func TestReassemblingTimeout(t *testing.T) { time.Sleep(2 * timeout) // Send another fragment that completes a packet. // However, no packet should be reassembled because the fragment arrived after the timeout. - _, done := f.Process(0, 1, 1, false, vv(1, "1")) + _, done, err := f.Process(0, 1, 1, false, vv(1, "1")) + if err != nil { + t.Fatalf("f.Process(0, 1, 1, false, vv(1, \"1\")) failed: %v", err) + } if done { t.Errorf("Fragmentation does not respect the reassembling timeout.") } diff --git a/pkg/tcpip/network/fragmentation/reassembler.go b/pkg/tcpip/network/fragmentation/reassembler.go index 8037f734b..9e002e396 100644 --- a/pkg/tcpip/network/fragmentation/reassembler.go +++ b/pkg/tcpip/network/fragmentation/reassembler.go @@ -78,7 +78,7 @@ func (r *reassembler) updateHoles(first, last uint16, more bool) bool { return used } -func (r *reassembler) process(first, last uint16, more bool, vv buffer.VectorisedView) (buffer.VectorisedView, bool, int) { +func (r *reassembler) process(first, last uint16, more bool, vv buffer.VectorisedView) (buffer.VectorisedView, bool, int, error) { r.mu.Lock() defer r.mu.Unlock() consumed := 0 @@ -86,7 +86,7 @@ func (r *reassembler) process(first, last uint16, more bool, vv buffer.Vectorise // A concurrent goroutine might have already reassembled // the packet and emptied the heap while this goroutine // was waiting on the mutex. We don't have to do anything in this case. - return buffer.VectorisedView{}, false, consumed + return buffer.VectorisedView{}, false, consumed, nil } if r.updateHoles(first, last, more) { // We store the incoming packet only if it filled some holes. @@ -96,13 +96,13 @@ func (r *reassembler) process(first, last uint16, more bool, vv buffer.Vectorise } // Check if all the holes have been deleted and we are ready to reassamble. if r.deleted < len(r.holes) { - return buffer.VectorisedView{}, false, consumed + return buffer.VectorisedView{}, false, consumed, nil } res, err := r.heap.reassemble() if err != nil { - panic(fmt.Sprintf("reassemble failed with: %v. There is probably a bug in the code handling the holes.", err)) + return buffer.VectorisedView{}, false, consumed, fmt.Errorf("fragment reassembly failed: %v", err) } - return res, true, consumed + return res, true, consumed, nil } func (r *reassembler) tooOld(timeout time.Duration) bool { diff --git a/pkg/tcpip/network/ip_test.go b/pkg/tcpip/network/ip_test.go index a9741622e..fe499d47e 100644 --- a/pkg/tcpip/network/ip_test.go +++ b/pkg/tcpip/network/ip_test.go @@ -96,16 +96,16 @@ func (t *testObject) checkValues(protocol tcpip.TransportProtocolNumber, vv buff // DeliverTransportPacket is called by network endpoints after parsing incoming // packets. This is used by the test object to verify that the results of the // parsing are expected. -func (t *testObject) DeliverTransportPacket(r *stack.Route, protocol tcpip.TransportProtocolNumber, netHeader buffer.View, vv buffer.VectorisedView) { - t.checkValues(protocol, vv, r.RemoteAddress, r.LocalAddress) +func (t *testObject) DeliverTransportPacket(r *stack.Route, protocol tcpip.TransportProtocolNumber, pkt tcpip.PacketBuffer) { + t.checkValues(protocol, pkt.Data, r.RemoteAddress, r.LocalAddress) t.dataCalls++ } // DeliverTransportControlPacket is called by network endpoints after parsing // incoming control (ICMP) packets. This is used by the test object to verify // that the results of the parsing are expected. -func (t *testObject) DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ stack.ControlType, extra uint32, vv buffer.VectorisedView) { - t.checkValues(trans, vv, remote, local) +func (t *testObject) DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ stack.ControlType, extra uint32, pkt tcpip.PacketBuffer) { + t.checkValues(trans, pkt.Data, remote, local) if typ != t.typ { t.t.Errorf("typ = %v, want %v", typ, t.typ) } @@ -171,6 +171,15 @@ func (t *testObject) WritePacket(_ *stack.Route, _ *stack.GSO, hdr buffer.Prepen return nil } +// WritePackets implements stack.LinkEndpoint.WritePackets. +func (t *testObject) WritePackets(_ *stack.Route, _ *stack.GSO, hdr []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + panic("not implemented") +} + +func (t *testObject) WriteRawPacket(_ buffer.VectorisedView) *tcpip.Error { + return tcpip.ErrNotSupported +} + func buildIPv4Route(local, remote tcpip.Address) (stack.Route, *tcpip.Error) { s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, @@ -230,7 +239,7 @@ func TestIPv4Send(t *testing.T) { if err != nil { t.Fatalf("could not find route: %v", err) } - if err := ep.WritePacket(&r, nil /* gso */, hdr, payload.ToVectorisedView(), 123, 123, stack.PacketOut); err != nil { + if err := ep.WritePacket(&r, nil /* gso */, hdr, payload.ToVectorisedView(), stack.NetworkHeaderParams{Protocol: 123, TTL: 123, TOS: stack.DefaultTOS}, stack.PacketOut); err != nil { t.Fatalf("WritePacket failed: %v", err) } } @@ -270,7 +279,9 @@ func TestIPv4Receive(t *testing.T) { if err != nil { t.Fatalf("could not find route: %v", err) } - ep.HandlePacket(&r, view.ToVectorisedView()) + ep.HandlePacket(&r, tcpip.PacketBuffer{ + Data: view.ToVectorisedView(), + }) if o.dataCalls != 1 { t.Fatalf("Bad number of data calls: got %x, want 1", o.dataCalls) } @@ -358,7 +369,9 @@ func TestIPv4ReceiveControl(t *testing.T) { o.extra = c.expectedExtra vv := view[:len(view)-c.trunc].ToVectorisedView() - ep.HandlePacket(&r, vv) + ep.HandlePacket(&r, tcpip.PacketBuffer{ + Data: vv, + }) if want := c.expectedCount; o.controlCalls != want { t.Fatalf("Bad number of control calls for %q case: got %v, want %v", c.name, o.controlCalls, want) } @@ -421,13 +434,17 @@ func TestIPv4FragmentationReceive(t *testing.T) { } // Send first segment. - ep.HandlePacket(&r, frag1.ToVectorisedView()) + ep.HandlePacket(&r, tcpip.PacketBuffer{ + Data: frag1.ToVectorisedView(), + }) if o.dataCalls != 0 { t.Fatalf("Bad number of data calls: got %x, want 0", o.dataCalls) } // Send second segment. - ep.HandlePacket(&r, frag2.ToVectorisedView()) + ep.HandlePacket(&r, tcpip.PacketBuffer{ + Data: frag2.ToVectorisedView(), + }) if o.dataCalls != 1 { t.Fatalf("Bad number of data calls: got %x, want 1", o.dataCalls) } @@ -460,7 +477,7 @@ func TestIPv6Send(t *testing.T) { if err != nil { t.Fatalf("could not find route: %v", err) } - if err := ep.WritePacket(&r, nil /* gso */, hdr, payload.ToVectorisedView(), 123, 123, stack.PacketOut); err != nil { + if err := ep.WritePacket(&r, nil /* gso */, hdr, payload.ToVectorisedView(), stack.NetworkHeaderParams{Protocol: 123, TTL: 123, TOS: stack.DefaultTOS}, stack.PacketOut); err != nil { t.Fatalf("WritePacket failed: %v", err) } } @@ -500,7 +517,9 @@ func TestIPv6Receive(t *testing.T) { t.Fatalf("could not find route: %v", err) } - ep.HandlePacket(&r, view.ToVectorisedView()) + ep.HandlePacket(&r, tcpip.PacketBuffer{ + Data: view.ToVectorisedView(), + }) if o.dataCalls != 1 { t.Fatalf("Bad number of data calls: got %x, want 1", o.dataCalls) } @@ -510,6 +529,7 @@ func TestIPv6ReceiveControl(t *testing.T) { newUint16 := func(v uint16) *uint16 { return &v } const mtu = 0xffff + const outerSrcAddr = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xaa" cases := []struct { name string expectedCount int @@ -561,7 +581,7 @@ func TestIPv6ReceiveControl(t *testing.T) { PayloadLength: uint16(len(view) - header.IPv6MinimumSize - c.trunc), NextHeader: uint8(header.ICMPv6ProtocolNumber), HopLimit: 20, - SrcAddr: "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xaa", + SrcAddr: outerSrcAddr, DstAddr: localIpv6Addr, }) @@ -608,8 +628,12 @@ func TestIPv6ReceiveControl(t *testing.T) { o.typ = c.expectedTyp o.extra = c.expectedExtra - vv := view[:len(view)-c.trunc].ToVectorisedView() - ep.HandlePacket(&r, vv) + // Set ICMPv6 checksum. + icmp.SetChecksum(header.ICMPv6Checksum(icmp, outerSrcAddr, localIpv6Addr, buffer.VectorisedView{})) + + ep.HandlePacket(&r, tcpip.PacketBuffer{ + Data: view[:len(view)-c.trunc].ToVectorisedView(), + }) if want := c.expectedCount; o.controlCalls != want { t.Fatalf("Bad number of control calls for %q case: got %v, want %v", c.name, o.controlCalls, want) } diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go index a25756443..ce771631c 100644 --- a/pkg/tcpip/network/ipv4/icmp.go +++ b/pkg/tcpip/network/ipv4/icmp.go @@ -15,6 +15,7 @@ package ipv4 import ( + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" @@ -24,8 +25,8 @@ import ( // the original packet that caused the ICMP one to be sent. This information is // used to find out which transport endpoint must be notified about the ICMP // packet. -func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, vv buffer.VectorisedView) { - h := header.IPv4(vv.First()) +func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt tcpip.PacketBuffer) { + h := header.IPv4(pkt.Data.First()) // We don't use IsValid() here because ICMP only requires that the IP // header plus 8 bytes of the transport header be included. So it's @@ -39,7 +40,7 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, vv buffer. } hlen := int(h.HeaderLength()) - if vv.Size() < hlen || h.FragmentOffset() != 0 { + if pkt.Data.Size() < hlen || h.FragmentOffset() != 0 { // We won't be able to handle this if it doesn't contain the // full IPv4 header, or if it's a fragment not at offset 0 // (because it won't have the transport header). @@ -47,15 +48,15 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, vv buffer. } // Skip the ip header, then deliver control message. - vv.TrimFront(hlen) + pkt.Data.TrimFront(hlen) p := h.TransportProtocol() - e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, h.DestinationAddress(), ProtocolNumber, p, typ, extra, vv) + e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, h.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) } -func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, vv buffer.VectorisedView) { +func (e *endpoint) handleICMP(r *stack.Route, pkt tcpip.PacketBuffer) { stats := r.Stats() received := stats.ICMP.V4PacketsReceived - v := vv.First() + v := pkt.Data.First() if len(v) < header.ICMPv4MinimumSize { received.Invalid.Increment() return @@ -73,20 +74,23 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, vv buffer.V // checksum. We'll have to reset this before we hand the packet // off. h.SetChecksum(0) - gotChecksum := ^header.ChecksumVV(vv, 0 /* initial */) + gotChecksum := ^header.ChecksumVV(pkt.Data, 0 /* initial */) if gotChecksum != wantChecksum { // It's possible that a raw socket expects to receive this. h.SetChecksum(wantChecksum) - e.dispatcher.DeliverTransportPacket(r, header.ICMPv4ProtocolNumber, netHeader, vv) + e.dispatcher.DeliverTransportPacket(r, header.ICMPv4ProtocolNumber, pkt) received.Invalid.Increment() return } // It's possible that a raw socket expects to receive this. h.SetChecksum(wantChecksum) - e.dispatcher.DeliverTransportPacket(r, header.ICMPv4ProtocolNumber, netHeader, vv) + e.dispatcher.DeliverTransportPacket(r, header.ICMPv4ProtocolNumber, tcpip.PacketBuffer{ + Data: pkt.Data.Clone(nil), + NetworkHeader: append(buffer.View(nil), pkt.NetworkHeader...), + }) - vv := vv.Clone(nil) + vv := pkt.Data.Clone(nil) vv.TrimFront(header.ICMPv4MinimumSize) hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv4MinimumSize) pkt := header.ICMPv4(hdr.Prepend(header.ICMPv4MinimumSize)) @@ -95,7 +99,7 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, vv buffer.V pkt.SetChecksum(0) pkt.SetChecksum(^header.Checksum(pkt, header.ChecksumVV(vv, 0))) sent := stats.ICMP.V4PacketsSent - if err := r.WritePacket(nil /* gso */, hdr, vv, header.ICMPv4ProtocolNumber, r.DefaultTTL()); err != nil { + if err := r.WritePacket(nil /* gso */, hdr, vv, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}); err != nil { sent.Dropped.Increment() return } @@ -104,19 +108,19 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, vv buffer.V case header.ICMPv4EchoReply: received.EchoReply.Increment() - e.dispatcher.DeliverTransportPacket(r, header.ICMPv4ProtocolNumber, netHeader, vv) + e.dispatcher.DeliverTransportPacket(r, header.ICMPv4ProtocolNumber, pkt) case header.ICMPv4DstUnreachable: received.DstUnreachable.Increment() - vv.TrimFront(header.ICMPv4MinimumSize) + pkt.Data.TrimFront(header.ICMPv4MinimumSize) switch h.Code() { case header.ICMPv4PortUnreachable: - e.handleControl(stack.ControlPortUnreachable, 0, vv) + e.handleControl(stack.ControlPortUnreachable, 0, pkt) case header.ICMPv4FragmentationNeeded: mtu := uint32(h.MTU()) - e.handleControl(stack.ControlPacketTooBig, calculateMTU(mtu), vv) + e.handleControl(stack.ControlPacketTooBig, calculateMTU(mtu), pkt) } case header.ICMPv4SrcQuench: diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index b7b07a6c1..ac16c8add 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -39,12 +39,15 @@ const ( // TotalLength field of the ipv4 header. MaxTotalSize = 0xffff + // DefaultTTL is the default time-to-live value for this endpoint. + DefaultTTL = 64 + // buckets is the number of identifier buckets. buckets = 2048 ) type endpoint struct { - nicid tcpip.NICID + nicID tcpip.NICID id stack.NetworkEndpointID prefixLen int linkEP stack.LinkEndpoint @@ -54,9 +57,9 @@ type endpoint struct { } // NewEndpoint creates a new ipv4 endpoint. -func (p *protocol) NewEndpoint(nicid tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, linkEP stack.LinkEndpoint) (stack.NetworkEndpoint, *tcpip.Error) { +func (p *protocol) NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, linkEP stack.LinkEndpoint) (stack.NetworkEndpoint, *tcpip.Error) { e := &endpoint{ - nicid: nicid, + nicID: nicID, id: stack.NetworkEndpointID{LocalAddress: addrWithPrefix.Address}, prefixLen: addrWithPrefix.PrefixLen, linkEP: linkEP, @@ -70,7 +73,7 @@ func (p *protocol) NewEndpoint(nicid tcpip.NICID, addrWithPrefix tcpip.AddressWi // DefaultTTL is the default time-to-live value for this endpoint. func (e *endpoint) DefaultTTL() uint8 { - return 255 + return e.protocol.DefaultTTL() } // MTU implements stack.NetworkEndpoint.MTU. It returns the link-layer MTU minus @@ -86,7 +89,7 @@ func (e *endpoint) Capabilities() stack.LinkEndpointCapabilities { // NICID returns the ID of the NIC this endpoint belongs to. func (e *endpoint) NICID() tcpip.NICID { - return e.nicid + return e.nicID } // ID returns the ipv4 endpoint ID. @@ -195,34 +198,44 @@ func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, hdr buff return nil } -// WritePacket writes a packet to the given destination address and protocol. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.TransportProtocolNumber, ttl uint8, loop stack.PacketLooping) *tcpip.Error { +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() + payload.Size()) + 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, protocol, e.protocol.hashIV)%buckets], 1) + id = atomic.AddUint32(&e.protocol.ids[hashRoute(r, params.Protocol, e.protocol.hashIV)%buckets], 1) } ip.Encode(&header.IPv4Fields{ IHL: header.IPv4MinimumSize, TotalLength: length, ID: uint16(id), - TTL: ttl, - Protocol: uint8(protocol), + TTL: params.TTL, + TOS: params.TOS, + Protocol: uint8(params.Protocol), SrcAddr: r.LocalAddress, DstAddr: r.RemoteAddress, }) ip.SetChecksum(^ip.CalculateChecksum()) + return ip +} + +// WritePacket writes a packet to the given destination address and protocol. +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) *tcpip.Error { + ip := e.addIPHeader(r, &hdr, payload.Size(), params) if loop&stack.PacketLoop != 0 { views := make([]buffer.View, 1, 1+len(payload.Views())) views[0] = hdr.View() views = append(views, payload.Views()...) - vv := buffer.NewVectorisedView(len(views[0])+payload.Size(), views) loopedR := r.MakeLoopedRoute() - e.HandlePacket(&loopedR, vv) + + e.HandlePacket(&loopedR, tcpip.PacketBuffer{ + Data: buffer.NewVectorisedView(len(views[0])+payload.Size(), views), + NetworkHeader: buffer.View(ip), + }) + loopedR.Release() } if loop&stack.PacketOut == 0 { @@ -238,6 +251,23 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen return nil } +// WritePackets implements stack.NetworkEndpoint.WritePackets. +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) { + if loop&stack.PacketLoop != 0 { + panic("multiple packets in local loop") + } + if loop&stack.PacketOut == 0 { + return len(hdrs), nil + } + + for i := range hdrs { + e.addIPHeader(r, &hdrs[i].Hdr, hdrs[i].Size, params) + } + n, err := e.linkEP.WritePackets(r, gso, hdrs, payload, ProtocolNumber) + r.Stats().IP.PacketsSent.IncrementBy(uint64(n)) + return n, err +} + // WriteHeaderIncludedPacket writes a packet already containing a network // header through the given route. func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error { @@ -276,7 +306,10 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.Vect ip.SetChecksum(^ip.CalculateChecksum()) if loop&stack.PacketLoop != 0 { - e.HandlePacket(r, payload) + e.HandlePacket(r, tcpip.PacketBuffer{ + Data: payload, + NetworkHeader: buffer.View(ip), + }) } if loop&stack.PacketOut == 0 { return nil @@ -289,24 +322,47 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.Vect // HandlePacket is called by the link layer when new ipv4 packets arrive for // this endpoint. -func (e *endpoint) HandlePacket(r *stack.Route, vv buffer.VectorisedView) { - headerView := vv.First() +func (e *endpoint) HandlePacket(r *stack.Route, pkt tcpip.PacketBuffer) { + headerView := pkt.Data.First() h := header.IPv4(headerView) - if !h.IsValid(vv.Size()) { + if !h.IsValid(pkt.Data.Size()) { + r.Stats().IP.MalformedPacketsReceived.Increment() return } + pkt.NetworkHeader = headerView[:h.HeaderLength()] hlen := int(h.HeaderLength()) tlen := int(h.TotalLength()) - vv.TrimFront(hlen) - vv.CapLength(tlen - hlen) + pkt.Data.TrimFront(hlen) + pkt.Data.CapLength(tlen - hlen) more := (h.Flags() & header.IPv4FlagMoreFragments) != 0 if more || h.FragmentOffset() != 0 { + if pkt.Data.Size() == 0 { + // Drop the packet as it's marked as a fragment but has + // no payload. + r.Stats().IP.MalformedPacketsReceived.Increment() + r.Stats().IP.MalformedFragmentsReceived.Increment() + return + } // The packet is a fragment, let's try to reassemble it. - last := h.FragmentOffset() + uint16(vv.Size()) - 1 + last := h.FragmentOffset() + uint16(pkt.Data.Size()) - 1 + // Drop the packet if the fragmentOffset is incorrect. i.e the + // combination of fragmentOffset and pkt.Data.size() causes a + // wrap around resulting in last being less than the offset. + if last < h.FragmentOffset() { + r.Stats().IP.MalformedPacketsReceived.Increment() + r.Stats().IP.MalformedFragmentsReceived.Increment() + return + } var ready bool - vv, ready = e.fragmentation.Process(hash.IPv4FragmentHash(h), h.FragmentOffset(), last, more, vv) + var err error + pkt.Data, ready, err = e.fragmentation.Process(hash.IPv4FragmentHash(h), h.FragmentOffset(), last, more, pkt.Data) + if err != nil { + r.Stats().IP.MalformedPacketsReceived.Increment() + r.Stats().IP.MalformedFragmentsReceived.Increment() + return + } if !ready { return } @@ -314,11 +370,11 @@ func (e *endpoint) HandlePacket(r *stack.Route, vv buffer.VectorisedView) { p := h.TransportProtocol() if p == header.ICMPv4ProtocolNumber { headerView.CapLength(hlen) - e.handleICMP(r, headerView, vv) + e.handleICMP(r, pkt) return } r.Stats().IP.PacketsDelivered.Increment() - e.dispatcher.DeliverTransportPacket(r, p, headerView, vv) + e.dispatcher.DeliverTransportPacket(r, p, pkt) } // Close cleans up resources associated with the endpoint. @@ -327,6 +383,11 @@ func (e *endpoint) Close() {} type protocol struct { ids []uint32 hashIV uint32 + + // defaultTTL is the current default TTL for the protocol. Only the + // uint8 portion of it is meaningful and it must be accessed + // atomically. + defaultTTL uint32 } // Number returns the ipv4 protocol number. @@ -352,12 +413,34 @@ func (*protocol) ParseAddresses(v buffer.View) (src, dst tcpip.Address) { // SetOption implements NetworkProtocol.SetOption. func (p *protocol) SetOption(option interface{}) *tcpip.Error { - return tcpip.ErrUnknownProtocolOption + switch v := option.(type) { + case tcpip.DefaultTTLOption: + p.SetDefaultTTL(uint8(v)) + return nil + default: + return tcpip.ErrUnknownProtocolOption + } } // Option implements NetworkProtocol.Option. func (p *protocol) Option(option interface{}) *tcpip.Error { - return tcpip.ErrUnknownProtocolOption + switch v := option.(type) { + case *tcpip.DefaultTTLOption: + *v = tcpip.DefaultTTLOption(p.DefaultTTL()) + return nil + default: + return tcpip.ErrUnknownProtocolOption + } +} + +// SetDefaultTTL sets the default TTL for endpoints created with this protocol. +func (p *protocol) SetDefaultTTL(ttl uint8) { + atomic.StoreUint32(&p.defaultTTL, uint32(ttl)) +} + +// DefaultTTL returns the default TTL for endpoints created with this protocol. +func (p *protocol) DefaultTTL() uint8 { + return uint8(atomic.LoadUint32(&p.defaultTTL)) } // calculateMTU calculates the network-layer payload MTU based on the link-layer @@ -391,5 +474,5 @@ func NewProtocol() stack.NetworkProtocol { } hashIV := r[buckets] - return &protocol{ids: ids, hashIV: hashIV} + return &protocol{ids: ids, hashIV: hashIV, defaultTTL: DefaultTTL} } diff --git a/pkg/tcpip/network/ipv4/ipv4_test.go b/pkg/tcpip/network/ipv4/ipv4_test.go index b6641ccc3..01dfb5f20 100644 --- a/pkg/tcpip/network/ipv4/ipv4_test.go +++ b/pkg/tcpip/network/ipv4/ipv4_test.go @@ -47,13 +47,6 @@ func TestExcludeBroadcast(t *testing.T) { t.Fatalf("CreateNIC failed: %v", err) } - if err := s.AddAddress(1, ipv4.ProtocolNumber, header.IPv4Broadcast); err != nil { - t.Fatalf("AddAddress failed: %v", err) - } - if err := s.AddAddress(1, ipv4.ProtocolNumber, header.IPv4Any); err != nil { - t.Fatalf("AddAddress failed: %v", err) - } - s.SetRouteTable([]tcpip.Route{{ Destination: header.IPv4EmptySubnet, NIC: 1, @@ -305,7 +298,7 @@ func TestFragmentation(t *testing.T) { Payload: payload.Clone([]buffer.View{}), } c := buildContext(t, nil, ft.mtu) - err := c.Route.WritePacket(ft.gso, hdr, payload, tcp.ProtocolNumber, 42) + err := c.Route.WritePacket(ft.gso, hdr, payload, stack.NetworkHeaderParams{Protocol: tcp.ProtocolNumber, TTL: 42, TOS: stack.DefaultTOS}) if err != nil { t.Errorf("err got %v, want %v", err, nil) } @@ -352,7 +345,7 @@ func TestFragmentationErrors(t *testing.T) { t.Run(ft.description, func(t *testing.T) { hdr, payload := makeHdrAndPayload(ft.hdrLength, header.IPv4MinimumSize, ft.payloadViewsSizes) c := buildContext(t, ft.packetCollectorErrors, ft.mtu) - err := c.Route.WritePacket(&stack.GSO{}, hdr, payload, tcp.ProtocolNumber, 42) + err := c.Route.WritePacket(&stack.GSO{}, hdr, payload, stack.NetworkHeaderParams{Protocol: tcp.ProtocolNumber, TTL: 42, TOS: stack.DefaultTOS}) for i := 0; i < len(ft.packetCollectorErrors)-1; i++ { if got, want := ft.packetCollectorErrors[i], (*tcpip.Error)(nil); got != want { t.Errorf("ft.packetCollectorErrors[%d] got %v, want %v", i, got, want) @@ -369,3 +362,119 @@ func TestFragmentationErrors(t *testing.T) { }) } } + +func TestInvalidFragments(t *testing.T) { + // These packets have both IHL and TotalLength set to 0. + testCases := []struct { + name string + packets [][]byte + wantMalformedIPPackets uint64 + wantMalformedFragments uint64 + }{ + { + "ihl_totallen_zero_valid_frag_offset", + [][]byte{ + {0x40, 0x30, 0x00, 0x00, 0x6c, 0x74, 0x7d, 0x30, 0x30, 0x30, 0x30, 0x30, 0x39, 0x32, 0x39, 0x33, 0xff, 0xff, 0xff, 0xff, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30}, + }, + 1, + 0, + }, + { + "ihl_totallen_zero_invalid_frag_offset", + [][]byte{ + {0x40, 0x30, 0x00, 0x00, 0x6c, 0x74, 0x20, 0x00, 0x30, 0x30, 0x30, 0x30, 0x39, 0x32, 0x39, 0x33, 0xff, 0xff, 0xff, 0xff, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30}, + }, + 1, + 0, + }, + { + // Total Length of 37(20 bytes IP header + 17 bytes of + // payload) + // Frag Offset of 0x1ffe = 8190*8 = 65520 + // Leading to the fragment end to be past 65535. + "ihl_totallen_valid_invalid_frag_offset_1", + [][]byte{ + {0x45, 0x30, 0x00, 0x25, 0x6c, 0x74, 0x1f, 0xfe, 0x30, 0x30, 0x30, 0x30, 0x39, 0x32, 0x39, 0x33, 0xff, 0xff, 0xff, 0xff, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30}, + }, + 1, + 1, + }, + // The following 3 tests were found by running a fuzzer and were + // triggering a panic in the IPv4 reassembler code. + { + "ihl_less_than_ipv4_minimum_size_1", + [][]byte{ + {0x42, 0x30, 0x0, 0x30, 0x30, 0x40, 0x0, 0xf3, 0x30, 0x1, 0x30, 0x30, 0x73, 0x73, 0x69, 0x6e, 0xff, 0xff, 0xff, 0xff, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30}, + {0x42, 0x30, 0x0, 0x8, 0x30, 0x40, 0x20, 0x0, 0x30, 0x1, 0x30, 0x30, 0x73, 0x73, 0x69, 0x6e, 0xff, 0xff, 0xff, 0xff, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30}, + }, + 2, + 0, + }, + { + "ihl_less_than_ipv4_minimum_size_2", + [][]byte{ + {0x42, 0x30, 0x0, 0x30, 0x30, 0x40, 0xb3, 0x12, 0x30, 0x6, 0x30, 0x30, 0x73, 0x73, 0x69, 0x6e, 0xff, 0xff, 0xff, 0xff, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30}, + {0x42, 0x30, 0x0, 0x8, 0x30, 0x40, 0x20, 0x0, 0x30, 0x6, 0x30, 0x30, 0x73, 0x73, 0x69, 0x6e, 0xff, 0xff, 0xff, 0xff, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30}, + }, + 2, + 0, + }, + { + "ihl_less_than_ipv4_minimum_size_3", + [][]byte{ + {0x42, 0x30, 0x0, 0x30, 0x30, 0x40, 0xb3, 0x30, 0x30, 0x6, 0x30, 0x30, 0x73, 0x73, 0x69, 0x6e, 0xff, 0xff, 0xff, 0xff, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30}, + {0x42, 0x30, 0x0, 0x8, 0x30, 0x40, 0x20, 0x0, 0x30, 0x6, 0x30, 0x30, 0x73, 0x73, 0x69, 0x6e, 0xff, 0xff, 0xff, 0xff, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30}, + }, + 2, + 0, + }, + { + "fragment_with_short_total_len_extra_payload", + [][]byte{ + {0x46, 0x30, 0x00, 0x30, 0x30, 0x40, 0x0e, 0x12, 0x30, 0x06, 0x30, 0x30, 0x73, 0x73, 0x69, 0x6e, 0xff, 0xff, 0xff, 0xff, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30}, + {0x46, 0x30, 0x00, 0x18, 0x30, 0x40, 0x20, 0x00, 0x30, 0x06, 0x30, 0x30, 0x73, 0x73, 0x69, 0x6e, 0xff, 0xff, 0xff, 0xff, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30}, + }, + 1, + 1, + }, + { + "multiple_fragments_with_more_fragments_set_to_false", + [][]byte{ + {0x45, 0x00, 0x00, 0x1c, 0x30, 0x40, 0x00, 0x10, 0x00, 0x06, 0x34, 0x69, 0x73, 0x73, 0x69, 0x6e, 0xff, 0xff, 0xff, 0xff, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00}, + {0x45, 0x00, 0x00, 0x1c, 0x30, 0x40, 0x00, 0x01, 0x61, 0x06, 0x34, 0x69, 0x73, 0x73, 0x69, 0x6e, 0xff, 0xff, 0xff, 0xff, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00}, + {0x45, 0x00, 0x00, 0x1c, 0x30, 0x40, 0x20, 0x00, 0x00, 0x06, 0x34, 0x1e, 0x73, 0x73, 0x69, 0x6e, 0xff, 0xff, 0xff, 0xff, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00}, + }, + 1, + 1, + }, + } + + for _, tc := range testCases { + t.Run(tc.name, func(t *testing.T) { + const nicID tcpip.NICID = 42 + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ + ipv4.NewProtocol(), + }, + }) + + var linkAddr = tcpip.LinkAddress([]byte{0x30, 0x30, 0x30, 0x30, 0x30, 0x30}) + var remoteLinkAddr = tcpip.LinkAddress([]byte{0x30, 0x30, 0x30, 0x30, 0x30, 0x31}) + ep := channel.New(10, 1500, linkAddr) + s.CreateNIC(nicID, sniffer.New(ep)) + + for _, pkt := range tc.packets { + ep.InjectLinkAddr(header.IPv4ProtocolNumber, remoteLinkAddr, tcpip.PacketBuffer{ + Data: buffer.NewVectorisedView(len(pkt), []buffer.View{pkt}), + }) + } + + if got, want := s.Stats().IP.MalformedPacketsReceived.Value(), tc.wantMalformedIPPackets; got != want { + t.Errorf("incorrect Stats.IP.MalformedPacketsReceived, got: %d, want: %d", got, want) + } + if got, want := s.Stats().IP.MalformedFragmentsReceived.Value(), tc.wantMalformedFragments; got != want { + t.Errorf("incorrect Stats.IP.MalformedFragmentsReceived, got: %d, want: %d", got, want) + } + }) + } +} diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index b4d0295bf..6629951c6 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -21,21 +21,12 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/stack" ) -const ( - // ndpHopLimit is the expected IP hop limit value of 255 for received - // NDP packets, as per RFC 4861 sections 4.1 - 4.5, 6.1.1, 6.1.2, 7.1.1, - // 7.1.2 and 8.1. If the hop limit value is not 255, nodes MUST silently - // drop the NDP packet. All outgoing NDP packets must use this value for - // its IP hop limit field. - ndpHopLimit = 255 -) - // handleControl handles the case when an ICMP packet contains the headers of // the original packet that caused the ICMP one to be sent. This information is // used to find out which transport endpoint must be notified about the ICMP // packet. -func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, vv buffer.VectorisedView) { - h := header.IPv6(vv.First()) +func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt tcpip.PacketBuffer) { + h := header.IPv6(pkt.Data.First()) // We don't use IsValid() here because ICMP only requires that up to // 1280 bytes of the original packet be included. So it's likely that it @@ -49,10 +40,10 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, vv buffer. // Skip the IP header, then handle the fragmentation header if there // is one. - vv.TrimFront(header.IPv6MinimumSize) + pkt.Data.TrimFront(header.IPv6MinimumSize) p := h.TransportProtocol() if p == header.IPv6FragmentHeader { - f := header.IPv6Fragment(vv.First()) + f := header.IPv6Fragment(pkt.Data.First()) if !f.IsValid() || f.FragmentOffset() != 0 { // We can't handle fragments that aren't at offset 0 // because they don't have the transport headers. @@ -61,35 +52,54 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, vv buffer. // Skip fragmentation header and find out the actual protocol // number. - vv.TrimFront(header.IPv6FragmentHeaderSize) + pkt.Data.TrimFront(header.IPv6FragmentHeaderSize) p = f.TransportProtocol() } // Deliver the control packet to the transport endpoint. - e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, h.DestinationAddress(), ProtocolNumber, p, typ, extra, vv) + e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, h.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) } -func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, vv buffer.VectorisedView) { +func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt tcpip.PacketBuffer) { stats := r.Stats().ICMP sent := stats.V6PacketsSent received := stats.V6PacketsReceived - v := vv.First() + v := pkt.Data.First() if len(v) < header.ICMPv6MinimumSize { received.Invalid.Increment() return } h := header.ICMPv6(v) + iph := header.IPv6(netHeader) + + // Validate ICMPv6 checksum before processing the packet. + // + // Only the first view in vv is accounted for by h. To account for the + // rest of vv, a shallow copy is made and the first view is removed. + // This copy is used as extra payload during the checksum calculation. + payload := pkt.Data + payload.RemoveFirst() + if got, want := h.Checksum(), header.ICMPv6Checksum(h, iph.SourceAddress(), iph.DestinationAddress(), payload); got != want { + received.Invalid.Increment() + return + } // As per RFC 4861 sections 4.1 - 4.5, 6.1.1, 6.1.2, 7.1.1, 7.1.2 and // 8.1, nodes MUST silently drop NDP packets where the Hop Limit field - // in the IPv6 header is not set to 255. + // in the IPv6 header is not set to 255, or the ICMPv6 Code field is not + // set to 0. switch h.Type() { case header.ICMPv6NeighborSolicit, header.ICMPv6NeighborAdvert, header.ICMPv6RouterSolicit, header.ICMPv6RouterAdvert, header.ICMPv6RedirectMsg: - if header.IPv6(netHeader).HopLimit() != ndpHopLimit { + if iph.HopLimit() != header.NDPHopLimit { + received.Invalid.Increment() + return + } + + if h.Code() != 0 { received.Invalid.Increment() return } @@ -103,9 +113,9 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, vv buffer.V received.Invalid.Increment() return } - vv.TrimFront(header.ICMPv6PacketTooBigMinimumSize) + pkt.Data.TrimFront(header.ICMPv6PacketTooBigMinimumSize) mtu := h.MTU() - e.handleControl(stack.ControlPacketTooBig, calculateMTU(mtu), vv) + e.handleControl(stack.ControlPacketTooBig, calculateMTU(mtu), pkt) case header.ICMPv6DstUnreachable: received.DstUnreachable.Increment() @@ -113,33 +123,80 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, vv buffer.V received.Invalid.Increment() return } - vv.TrimFront(header.ICMPv6DstUnreachableMinimumSize) + pkt.Data.TrimFront(header.ICMPv6DstUnreachableMinimumSize) switch h.Code() { case header.ICMPv6PortUnreachable: - e.handleControl(stack.ControlPortUnreachable, 0, vv) + e.handleControl(stack.ControlPortUnreachable, 0, pkt) } case header.ICMPv6NeighborSolicit: received.NeighborSolicit.Increment() - if len(v) < header.ICMPv6NeighborSolicitMinimumSize { received.Invalid.Increment() return } - targetAddr := tcpip.Address(v[8:][:header.IPv6AddressSize]) - if e.linkAddrCache.CheckLocalAddress(e.nicid, ProtocolNumber, targetAddr) == 0 { + + ns := header.NDPNeighborSolicit(h.NDPPayload()) + targetAddr := ns.TargetAddress() + s := r.Stack() + rxNICID := r.NICID() + + isTentative, err := s.IsAddrTentative(rxNICID, targetAddr) + if err != nil { + // We will only get an error if rxNICID is unrecognized, + // which should not happen. For now short-circuit this + // packet. + // + // TODO(b/141002840): Handle this better? + return + } + + if isTentative { + // If the target address is tentative and the source + // of the packet is a unicast (specified) address, then + // the source of the packet is attempting to perform + // address resolution on the target. In this case, the + // solicitation is silently ignored, as per RFC 4862 + // section 5.4.3. + // + // If the target address is tentative and the source of + // the packet is the unspecified address (::), then we + // know another node is also performing DAD for the + // same address (since targetAddr is tentative for us, + // we know we are also performing DAD on it). In this + // case we let the stack know so it can handle such a + // scenario and do nothing further with the NDP NS. + if iph.SourceAddress() == header.IPv6Any { + s.DupTentativeAddrDetected(rxNICID, targetAddr) + } + + // Do not handle neighbor solicitations targeted + // to an address that is tentative on the received + // NIC any further. + return + } + + // At this point we know that targetAddr is not tentative on + // rxNICID so the packet is processed as defined in RFC 4861, + // as per RFC 4862 section 5.4.3. + + if e.linkAddrCache.CheckLocalAddress(e.nicID, ProtocolNumber, targetAddr) == 0 { // We don't have a useful answer; the best we can do is ignore the request. return } - hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv6NeighborAdvertSize) - pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborAdvertSize)) - pkt.SetType(header.ICMPv6NeighborAdvert) - pkt[icmpV6FlagOffset] = ndpSolicitedFlag | ndpOverrideFlag - copy(pkt[icmpV6OptOffset-len(targetAddr):], targetAddr) - pkt[icmpV6OptOffset] = ndpOptDstLinkAddr - pkt[icmpV6LengthOffset] = 1 - copy(pkt[icmpV6LengthOffset+1:], r.LocalLinkAddress[:]) + optsSerializer := header.NDPOptionsSerializer{ + header.NDPTargetLinkLayerAddressOption(r.LocalLinkAddress[:]), + } + hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv6NeighborAdvertMinimumSize + int(optsSerializer.Length())) + packet := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborAdvertSize)) + packet.SetType(header.ICMPv6NeighborAdvert) + na := header.NDPNeighborAdvert(packet.NDPPayload()) + na.SetSolicitedFlag(true) + na.SetOverrideFlag(true) + na.SetTargetAddress(targetAddr) + opts := na.Options() + opts.Serialize(optsSerializer) // ICMPv6 Neighbor Solicit messages are always sent to // specially crafted IPv6 multicast addresses. As a result, the @@ -152,9 +209,24 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, vv buffer.V r := r.Clone() defer r.Release() r.LocalAddress = targetAddr - pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) + packet.SetChecksum(header.ICMPv6Checksum(packet, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) - if err := r.WritePacket(nil /* gso */, hdr, buffer.VectorisedView{}, header.ICMPv6ProtocolNumber, r.DefaultTTL()); err != nil { + // TODO(tamird/ghanan): there exists an explicit NDP option that is + // used to update the neighbor table with link addresses for a + // neighbor from an NS (see the Source Link Layer option RFC + // 4861 section 4.6.1 and section 7.2.3). + // + // Furthermore, the entirety of NDP handling here seems to be + // contradicted by RFC 4861. + e.linkAddrCache.AddLinkAddress(e.nicID, r.RemoteAddress, r.RemoteLinkAddress) + + // RFC 4861 Neighbor Discovery for IP version 6 (IPv6) + // + // 7.1.2. Validation of Neighbor Advertisements + // + // The IP Hop Limit field has a value of 255, i.e., the packet + // could not possibly have been forwarded by a router. + if err := r.WritePacket(nil /* gso */, hdr, buffer.VectorisedView{}, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: stack.DefaultTOS}); err != nil { sent.Dropped.Increment() return } @@ -166,10 +238,45 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, vv buffer.V received.Invalid.Increment() return } - targetAddr := tcpip.Address(v[8:][:header.IPv6AddressSize]) - e.linkAddrCache.AddLinkAddress(e.nicid, targetAddr, r.RemoteLinkAddress) + + na := header.NDPNeighborAdvert(h.NDPPayload()) + targetAddr := na.TargetAddress() + stack := r.Stack() + rxNICID := r.NICID() + + isTentative, err := stack.IsAddrTentative(rxNICID, targetAddr) + if err != nil { + // We will only get an error if rxNICID is unrecognized, + // which should not happen. For now short-circuit this + // packet. + // + // TODO(b/141002840): Handle this better? + return + } + + if isTentative { + // We just got an NA from a node that owns an address we + // are performing DAD on, implying the address is not + // unique. In this case we let the stack know so it can + // handle such a scenario and do nothing furthur with + // the NDP NA. + stack.DupTentativeAddrDetected(rxNICID, targetAddr) + return + } + + // At this point we know that the targetAddress is not tentative + // on rxNICID. However, targetAddr may still be assigned to + // rxNICID but not tentative (it could be permanent). Such a + // scenario is beyond the scope of RFC 4862. As such, we simply + // ignore such a scenario for now and proceed as normal. + // + // TODO(b/143147598): Handle the scenario described above. Also + // inform the netstack integration that a duplicate address was + // detected outside of DAD. + + e.linkAddrCache.AddLinkAddress(e.nicID, targetAddr, r.RemoteLinkAddress) if targetAddr != r.RemoteAddress { - e.linkAddrCache.AddLinkAddress(e.nicid, r.RemoteAddress, r.RemoteLinkAddress) + e.linkAddrCache.AddLinkAddress(e.nicID, r.RemoteAddress, r.RemoteLinkAddress) } case header.ICMPv6EchoRequest: @@ -178,14 +285,13 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, vv buffer.V received.Invalid.Increment() return } - - vv.TrimFront(header.ICMPv6EchoMinimumSize) + pkt.Data.TrimFront(header.ICMPv6EchoMinimumSize) hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv6EchoMinimumSize) - pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6EchoMinimumSize)) - copy(pkt, h) - pkt.SetType(header.ICMPv6EchoReply) - pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, vv)) - if err := r.WritePacket(nil /* gso */, hdr, vv, header.ICMPv6ProtocolNumber, r.DefaultTTL()); err != nil { + packet := header.ICMPv6(hdr.Prepend(header.ICMPv6EchoMinimumSize)) + copy(packet, h) + packet.SetType(header.ICMPv6EchoReply) + packet.SetChecksum(header.ICMPv6Checksum(packet, r.LocalAddress, r.RemoteAddress, pkt.Data)) + if err := r.WritePacket(nil /* gso */, hdr, pkt.Data, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}); err != nil { sent.Dropped.Increment() return } @@ -197,7 +303,7 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, vv buffer.V received.Invalid.Increment() return } - e.dispatcher.DeliverTransportPacket(r, header.ICMPv6ProtocolNumber, netHeader, vv) + e.dispatcher.DeliverTransportPacket(r, header.ICMPv6ProtocolNumber, pkt) case header.ICMPv6TimeExceeded: received.TimeExceeded.Increment() @@ -209,8 +315,51 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, vv buffer.V received.RouterSolicit.Increment() case header.ICMPv6RouterAdvert: + routerAddr := iph.SourceAddress() + + // + // Validate the RA as per RFC 4861 section 6.1.2. + // + + // Is the IP Source Address a link-local address? + if !header.IsV6LinkLocalAddress(routerAddr) { + // ...No, silently drop the packet. + received.Invalid.Increment() + return + } + + p := h.NDPPayload() + + // Is the NDP payload of sufficient size to hold a Router + // Advertisement? + if len(p) < header.NDPRAMinimumSize { + // ...No, silently drop the packet. + received.Invalid.Increment() + return + } + + ra := header.NDPRouterAdvert(p) + opts := ra.Options() + + // Are options valid as per the wire format? + if _, err := opts.Iter(true); err != nil { + // ...No, silently drop the packet. + received.Invalid.Increment() + return + } + + // + // At this point, we have a valid Router Advertisement, as far + // as RFC 4861 section 6.1.2 is concerned. + // + received.RouterAdvert.Increment() + // Tell the NIC to handle the RA. + stack := r.Stack() + rxNICID := r.NICID() + stack.HandleNDPRA(rxNICID, routerAddr, ra) + case header.ICMPv6RedirectMsg: received.RedirectMsg.Increment() @@ -262,7 +411,7 @@ func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, linkEP stack. ip.Encode(&header.IPv6Fields{ PayloadLength: length, NextHeader: uint8(header.ICMPv6ProtocolNumber), - HopLimit: defaultIPv6HopLimit, + HopLimit: header.NDPHopLimit, SrcAddr: r.LocalAddress, DstAddr: r.RemoteAddress, }) diff --git a/pkg/tcpip/network/ipv6/icmp_test.go b/pkg/tcpip/network/ipv6/icmp_test.go index 01f5a17ec..6037a1ef8 100644 --- a/pkg/tcpip/network/ipv6/icmp_test.go +++ b/pkg/tcpip/network/ipv6/icmp_test.go @@ -15,7 +15,6 @@ package ipv6 import ( - "fmt" "reflect" "strings" "testing" @@ -31,7 +30,7 @@ import ( ) const ( - linkAddr0 = tcpip.LinkAddress("\x01\x02\x03\x04\x05\x06") + linkAddr0 = tcpip.LinkAddress("\x02\x02\x03\x04\x05\x06") linkAddr1 = tcpip.LinkAddress("\x0a\x0b\x0c\x0d\x0e\x0f") ) @@ -66,7 +65,7 @@ type stubDispatcher struct { stack.TransportDispatcher } -func (*stubDispatcher) DeliverTransportPacket(*stack.Route, tcpip.TransportProtocolNumber, buffer.View, buffer.VectorisedView) { +func (*stubDispatcher) DeliverTransportPacket(*stack.Route, tcpip.TransportProtocolNumber, tcpip.PacketBuffer) { } type stubLinkAddressCache struct { @@ -132,7 +131,7 @@ func TestICMPCounts(t *testing.T) { {header.ICMPv6EchoRequest, header.ICMPv6EchoMinimumSize}, {header.ICMPv6EchoReply, header.ICMPv6EchoMinimumSize}, {header.ICMPv6RouterSolicit, header.ICMPv6MinimumSize}, - {header.ICMPv6RouterAdvert, header.ICMPv6MinimumSize}, + {header.ICMPv6RouterAdvert, header.ICMPv6HeaderSize + header.NDPRAMinimumSize}, {header.ICMPv6NeighborSolicit, header.ICMPv6NeighborSolicitMinimumSize}, {header.ICMPv6NeighborAdvert, header.ICMPv6NeighborAdvertSize}, {header.ICMPv6RedirectMsg, header.ICMPv6MinimumSize}, @@ -144,11 +143,13 @@ func TestICMPCounts(t *testing.T) { ip.Encode(&header.IPv6Fields{ PayloadLength: uint16(payloadLength), NextHeader: uint8(header.ICMPv6ProtocolNumber), - HopLimit: r.DefaultTTL(), + HopLimit: header.NDPHopLimit, SrcAddr: r.LocalAddress, DstAddr: r.RemoteAddress, }) - ep.HandlePacket(&r, hdr.View().ToVectorisedView()) + ep.HandlePacket(&r, tcpip.PacketBuffer{ + Data: hdr.View().ToVectorisedView(), + }) } for _, typ := range types { @@ -179,13 +180,10 @@ func visitStats(v reflect.Value, f func(string, *tcpip.StatCounter)) { t := v.Type() for i := 0; i < v.NumField(); i++ { v := v.Field(i) - switch v.Kind() { - case reflect.Ptr: - f(t.Field(i).Name, v.Interface().(*tcpip.StatCounter)) - case reflect.Struct: + if s, ok := v.Interface().(*tcpip.StatCounter); ok { + f(t.Field(i).Name, s) + } else { visitStats(v, f) - default: - panic(fmt.Sprintf("unexpected type %s", v.Type())) } } } @@ -284,7 +282,9 @@ func routeICMPv6Packet(t *testing.T, args routeArgs, fn func(*testing.T, header. views := []buffer.View{pkt.Header, pkt.Payload} size := len(pkt.Header) + len(pkt.Payload) vv := buffer.NewVectorisedView(size, views) - args.dst.InjectLinkAddr(pkt.Proto, args.dst.LinkAddress(), vv) + args.dst.InjectLinkAddr(pkt.Proto, args.dst.LinkAddress(), tcpip.PacketBuffer{ + Data: vv, + }) } if pkt.Proto != ProtocolNumber { @@ -363,3 +363,537 @@ func TestLinkResolution(t *testing.T) { routeICMPv6Packet(t, args, nil) } } + +func TestICMPChecksumValidationSimple(t *testing.T) { + types := []struct { + name string + typ header.ICMPv6Type + size int + statCounter func(tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter + }{ + { + "DstUnreachable", + header.ICMPv6DstUnreachable, + header.ICMPv6DstUnreachableMinimumSize, + func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return stats.DstUnreachable + }, + }, + { + "PacketTooBig", + header.ICMPv6PacketTooBig, + header.ICMPv6PacketTooBigMinimumSize, + func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return stats.PacketTooBig + }, + }, + { + "TimeExceeded", + header.ICMPv6TimeExceeded, + header.ICMPv6MinimumSize, + func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return stats.TimeExceeded + }, + }, + { + "ParamProblem", + header.ICMPv6ParamProblem, + header.ICMPv6MinimumSize, + func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return stats.ParamProblem + }, + }, + { + "EchoRequest", + header.ICMPv6EchoRequest, + header.ICMPv6EchoMinimumSize, + func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return stats.EchoRequest + }, + }, + { + "EchoReply", + header.ICMPv6EchoReply, + header.ICMPv6EchoMinimumSize, + func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return stats.EchoReply + }, + }, + { + "RouterSolicit", + header.ICMPv6RouterSolicit, + header.ICMPv6MinimumSize, + func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return stats.RouterSolicit + }, + }, + { + "RouterAdvert", + header.ICMPv6RouterAdvert, + header.ICMPv6HeaderSize + header.NDPRAMinimumSize, + func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return stats.RouterAdvert + }, + }, + { + "NeighborSolicit", + header.ICMPv6NeighborSolicit, + header.ICMPv6NeighborSolicitMinimumSize, + func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return stats.NeighborSolicit + }, + }, + { + "NeighborAdvert", + header.ICMPv6NeighborAdvert, + header.ICMPv6NeighborAdvertSize, + func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return stats.NeighborAdvert + }, + }, + { + "RedirectMsg", + header.ICMPv6RedirectMsg, + header.ICMPv6MinimumSize, + func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return stats.RedirectMsg + }, + }, + } + + for _, typ := range types { + t.Run(typ.name, func(t *testing.T) { + e := channel.New(10, 1280, linkAddr0) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + }) + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(_) = %s", err) + } + + if err := s.AddAddress(1, ProtocolNumber, lladdr0); err != nil { + t.Fatalf("AddAddress(_, %d, %s) = %s", ProtocolNumber, lladdr0, err) + } + { + subnet, err := tcpip.NewSubnet(lladdr1, tcpip.AddressMask(strings.Repeat("\xff", len(lladdr1)))) + if err != nil { + t.Fatal(err) + } + s.SetRouteTable( + []tcpip.Route{{ + Destination: subnet, + NIC: 1, + }}, + ) + } + + handleIPv6Payload := func(typ header.ICMPv6Type, size int, checksum bool) { + hdr := buffer.NewPrependable(header.IPv6MinimumSize + size) + pkt := header.ICMPv6(hdr.Prepend(size)) + pkt.SetType(typ) + if checksum { + pkt.SetChecksum(header.ICMPv6Checksum(pkt, lladdr1, lladdr0, buffer.VectorisedView{})) + } + ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) + ip.Encode(&header.IPv6Fields{ + PayloadLength: uint16(size), + NextHeader: uint8(header.ICMPv6ProtocolNumber), + HopLimit: header.NDPHopLimit, + SrcAddr: lladdr1, + DstAddr: lladdr0, + }) + e.InjectInbound(ProtocolNumber, tcpip.PacketBuffer{ + Data: hdr.View().ToVectorisedView(), + }) + } + + stats := s.Stats().ICMP.V6PacketsReceived + invalid := stats.Invalid + typStat := typ.statCounter(stats) + + // Initial stat counts should be 0. + if got := invalid.Value(); got != 0 { + t.Fatalf("got invalid = %d, want = 0", got) + } + if got := typStat.Value(); got != 0 { + t.Fatalf("got %s = %d, want = 0", typ.name, got) + } + + // Without setting checksum, the incoming packet should + // be invalid. + handleIPv6Payload(typ.typ, typ.size, false) + if got := invalid.Value(); got != 1 { + t.Fatalf("got invalid = %d, want = 1", got) + } + // Rx count of type typ.typ should not have increased. + if got := typStat.Value(); got != 0 { + t.Fatalf("got %s = %d, want = 0", typ.name, got) + } + + // When checksum is set, it should be received. + handleIPv6Payload(typ.typ, typ.size, true) + if got := typStat.Value(); got != 1 { + t.Fatalf("got %s = %d, want = 1", typ.name, got) + } + // Invalid count should not have increased again. + if got := invalid.Value(); got != 1 { + t.Fatalf("got invalid = %d, want = 1", got) + } + }) + } +} + +func TestICMPChecksumValidationWithPayload(t *testing.T) { + const simpleBodySize = 64 + simpleBody := func(view buffer.View) { + for i := 0; i < simpleBodySize; i++ { + view[i] = uint8(i) + } + } + + const errorICMPBodySize = header.IPv6MinimumSize + simpleBodySize + errorICMPBody := func(view buffer.View) { + ip := header.IPv6(view) + ip.Encode(&header.IPv6Fields{ + PayloadLength: simpleBodySize, + NextHeader: 10, + HopLimit: 20, + SrcAddr: lladdr0, + DstAddr: lladdr1, + }) + simpleBody(view[header.IPv6MinimumSize:]) + } + + types := []struct { + name string + typ header.ICMPv6Type + size int + statCounter func(tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter + payloadSize int + payload func(buffer.View) + }{ + { + "DstUnreachable", + header.ICMPv6DstUnreachable, + header.ICMPv6DstUnreachableMinimumSize, + func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return stats.DstUnreachable + }, + errorICMPBodySize, + errorICMPBody, + }, + { + "PacketTooBig", + header.ICMPv6PacketTooBig, + header.ICMPv6PacketTooBigMinimumSize, + func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return stats.PacketTooBig + }, + errorICMPBodySize, + errorICMPBody, + }, + { + "TimeExceeded", + header.ICMPv6TimeExceeded, + header.ICMPv6MinimumSize, + func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return stats.TimeExceeded + }, + errorICMPBodySize, + errorICMPBody, + }, + { + "ParamProblem", + header.ICMPv6ParamProblem, + header.ICMPv6MinimumSize, + func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return stats.ParamProblem + }, + errorICMPBodySize, + errorICMPBody, + }, + { + "EchoRequest", + header.ICMPv6EchoRequest, + header.ICMPv6EchoMinimumSize, + func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return stats.EchoRequest + }, + simpleBodySize, + simpleBody, + }, + { + "EchoReply", + header.ICMPv6EchoReply, + header.ICMPv6EchoMinimumSize, + func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return stats.EchoReply + }, + simpleBodySize, + simpleBody, + }, + } + + for _, typ := range types { + t.Run(typ.name, func(t *testing.T) { + e := channel.New(10, 1280, linkAddr0) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + }) + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(_) = %s", err) + } + + if err := s.AddAddress(1, ProtocolNumber, lladdr0); err != nil { + t.Fatalf("AddAddress(_, %d, %s) = %s", ProtocolNumber, lladdr0, err) + } + { + subnet, err := tcpip.NewSubnet(lladdr1, tcpip.AddressMask(strings.Repeat("\xff", len(lladdr1)))) + if err != nil { + t.Fatal(err) + } + s.SetRouteTable( + []tcpip.Route{{ + Destination: subnet, + NIC: 1, + }}, + ) + } + + handleIPv6Payload := func(typ header.ICMPv6Type, size, payloadSize int, payloadFn func(buffer.View), checksum bool) { + icmpSize := size + payloadSize + hdr := buffer.NewPrependable(header.IPv6MinimumSize + icmpSize) + pkt := header.ICMPv6(hdr.Prepend(icmpSize)) + pkt.SetType(typ) + payloadFn(pkt.Payload()) + + if checksum { + pkt.SetChecksum(header.ICMPv6Checksum(pkt, lladdr1, lladdr0, buffer.VectorisedView{})) + } + + ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) + ip.Encode(&header.IPv6Fields{ + PayloadLength: uint16(icmpSize), + NextHeader: uint8(header.ICMPv6ProtocolNumber), + HopLimit: header.NDPHopLimit, + SrcAddr: lladdr1, + DstAddr: lladdr0, + }) + e.InjectInbound(ProtocolNumber, tcpip.PacketBuffer{ + Data: hdr.View().ToVectorisedView(), + }) + } + + stats := s.Stats().ICMP.V6PacketsReceived + invalid := stats.Invalid + typStat := typ.statCounter(stats) + + // Initial stat counts should be 0. + if got := invalid.Value(); got != 0 { + t.Fatalf("got invalid = %d, want = 0", got) + } + if got := typStat.Value(); got != 0 { + t.Fatalf("got %s = %d, want = 0", typ.name, got) + } + + // Without setting checksum, the incoming packet should + // be invalid. + handleIPv6Payload(typ.typ, typ.size, typ.payloadSize, typ.payload, false) + if got := invalid.Value(); got != 1 { + t.Fatalf("got invalid = %d, want = 1", got) + } + // Rx count of type typ.typ should not have increased. + if got := typStat.Value(); got != 0 { + t.Fatalf("got %s = %d, want = 0", typ.name, got) + } + + // When checksum is set, it should be received. + handleIPv6Payload(typ.typ, typ.size, typ.payloadSize, typ.payload, true) + if got := typStat.Value(); got != 1 { + t.Fatalf("got %s = %d, want = 1", typ.name, got) + } + // Invalid count should not have increased again. + if got := invalid.Value(); got != 1 { + t.Fatalf("got invalid = %d, want = 1", got) + } + }) + } +} + +func TestICMPChecksumValidationWithPayloadMultipleViews(t *testing.T) { + const simpleBodySize = 64 + simpleBody := func(view buffer.View) { + for i := 0; i < simpleBodySize; i++ { + view[i] = uint8(i) + } + } + + const errorICMPBodySize = header.IPv6MinimumSize + simpleBodySize + errorICMPBody := func(view buffer.View) { + ip := header.IPv6(view) + ip.Encode(&header.IPv6Fields{ + PayloadLength: simpleBodySize, + NextHeader: 10, + HopLimit: 20, + SrcAddr: lladdr0, + DstAddr: lladdr1, + }) + simpleBody(view[header.IPv6MinimumSize:]) + } + + types := []struct { + name string + typ header.ICMPv6Type + size int + statCounter func(tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter + payloadSize int + payload func(buffer.View) + }{ + { + "DstUnreachable", + header.ICMPv6DstUnreachable, + header.ICMPv6DstUnreachableMinimumSize, + func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return stats.DstUnreachable + }, + errorICMPBodySize, + errorICMPBody, + }, + { + "PacketTooBig", + header.ICMPv6PacketTooBig, + header.ICMPv6PacketTooBigMinimumSize, + func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return stats.PacketTooBig + }, + errorICMPBodySize, + errorICMPBody, + }, + { + "TimeExceeded", + header.ICMPv6TimeExceeded, + header.ICMPv6MinimumSize, + func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return stats.TimeExceeded + }, + errorICMPBodySize, + errorICMPBody, + }, + { + "ParamProblem", + header.ICMPv6ParamProblem, + header.ICMPv6MinimumSize, + func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return stats.ParamProblem + }, + errorICMPBodySize, + errorICMPBody, + }, + { + "EchoRequest", + header.ICMPv6EchoRequest, + header.ICMPv6EchoMinimumSize, + func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return stats.EchoRequest + }, + simpleBodySize, + simpleBody, + }, + { + "EchoReply", + header.ICMPv6EchoReply, + header.ICMPv6EchoMinimumSize, + func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return stats.EchoReply + }, + simpleBodySize, + simpleBody, + }, + } + + for _, typ := range types { + t.Run(typ.name, func(t *testing.T) { + e := channel.New(10, 1280, linkAddr0) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + }) + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(_) = %s", err) + } + + if err := s.AddAddress(1, ProtocolNumber, lladdr0); err != nil { + t.Fatalf("AddAddress(_, %d, %s) = %s", ProtocolNumber, lladdr0, err) + } + { + subnet, err := tcpip.NewSubnet(lladdr1, tcpip.AddressMask(strings.Repeat("\xff", len(lladdr1)))) + if err != nil { + t.Fatal(err) + } + s.SetRouteTable( + []tcpip.Route{{ + Destination: subnet, + NIC: 1, + }}, + ) + } + + handleIPv6Payload := func(typ header.ICMPv6Type, size, payloadSize int, payloadFn func(buffer.View), checksum bool) { + hdr := buffer.NewPrependable(header.IPv6MinimumSize + size) + pkt := header.ICMPv6(hdr.Prepend(size)) + pkt.SetType(typ) + + payload := buffer.NewView(payloadSize) + payloadFn(payload) + + if checksum { + pkt.SetChecksum(header.ICMPv6Checksum(pkt, lladdr1, lladdr0, payload.ToVectorisedView())) + } + + ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) + ip.Encode(&header.IPv6Fields{ + PayloadLength: uint16(size + payloadSize), + NextHeader: uint8(header.ICMPv6ProtocolNumber), + HopLimit: header.NDPHopLimit, + SrcAddr: lladdr1, + DstAddr: lladdr0, + }) + e.InjectInbound(ProtocolNumber, tcpip.PacketBuffer{ + Data: buffer.NewVectorisedView(header.IPv6MinimumSize+size+payloadSize, []buffer.View{hdr.View(), payload}), + }) + } + + stats := s.Stats().ICMP.V6PacketsReceived + invalid := stats.Invalid + typStat := typ.statCounter(stats) + + // Initial stat counts should be 0. + if got := invalid.Value(); got != 0 { + t.Fatalf("got invalid = %d, want = 0", got) + } + if got := typStat.Value(); got != 0 { + t.Fatalf("got %s = %d, want = 0", typ.name, got) + } + + // Without setting checksum, the incoming packet should + // be invalid. + handleIPv6Payload(typ.typ, typ.size, typ.payloadSize, typ.payload, false) + if got := invalid.Value(); got != 1 { + t.Fatalf("got invalid = %d, want = 1", got) + } + // Rx count of type typ.typ should not have increased. + if got := typStat.Value(); got != 0 { + t.Fatalf("got %s = %d, want = 0", typ.name, got) + } + + // When checksum is set, it should be received. + handleIPv6Payload(typ.typ, typ.size, typ.payloadSize, typ.payload, true) + if got := typStat.Value(); got != 1 { + t.Fatalf("got %s = %d, want = 1", typ.name, got) + } + // Invalid count should not have increased again. + if got := invalid.Value(); got != 1 { + t.Fatalf("got invalid = %d, want = 1", got) + } + }) + } +} diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index 7de6a4546..4cee848a1 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -21,6 +21,8 @@ package ipv6 import ( + "sync/atomic" + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -35,23 +37,24 @@ const ( // PayloadLength field of the ipv6 header. maxPayloadSize = 0xffff - // defaultIPv6HopLimit is the default hop limit for IPv6 Packets - // egressed by Netstack. - defaultIPv6HopLimit = 255 + // DefaultTTL is the default hop limit for IPv6 Packets egressed by + // Netstack. + DefaultTTL = 64 ) type endpoint struct { - nicid tcpip.NICID + nicID tcpip.NICID id stack.NetworkEndpointID prefixLen int linkEP stack.LinkEndpoint linkAddrCache stack.LinkAddressCache dispatcher stack.TransportDispatcher + protocol *protocol } // DefaultTTL is the default hop limit for this endpoint. func (e *endpoint) DefaultTTL() uint8 { - return 255 + return e.protocol.DefaultTTL() } // MTU implements stack.NetworkEndpoint.MTU. It returns the link-layer MTU minus @@ -62,7 +65,7 @@ func (e *endpoint) MTU() uint32 { // NICID returns the ID of the NIC this endpoint belongs to. func (e *endpoint) NICID() tcpip.NICID { - return e.nicid + return e.nicID } // ID returns the ipv6 endpoint ID. @@ -94,25 +97,35 @@ func (e *endpoint) GSOMaxSize() uint32 { return 0 } -// WritePacket writes a packet to the given destination address and protocol. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.TransportProtocolNumber, ttl uint8, loop stack.PacketLooping) *tcpip.Error { - length := uint16(hdr.UsedLength() + payload.Size()) +func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadSize int, params stack.NetworkHeaderParams) header.IPv6 { + length := uint16(hdr.UsedLength() + payloadSize) ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) ip.Encode(&header.IPv6Fields{ PayloadLength: length, - NextHeader: uint8(protocol), - HopLimit: ttl, + NextHeader: uint8(params.Protocol), + HopLimit: params.TTL, + TrafficClass: params.TOS, SrcAddr: r.LocalAddress, DstAddr: r.RemoteAddress, }) + return ip +} + +// WritePacket writes a packet to the given destination address and protocol. +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) *tcpip.Error { + ip := e.addIPHeader(r, &hdr, payload.Size(), params) if loop&stack.PacketLoop != 0 { views := make([]buffer.View, 1, 1+len(payload.Views())) views[0] = hdr.View() views = append(views, payload.Views()...) - vv := buffer.NewVectorisedView(len(views[0])+payload.Size(), views) loopedR := r.MakeLoopedRoute() - e.HandlePacket(&loopedR, vv) + + e.HandlePacket(&loopedR, tcpip.PacketBuffer{ + Data: buffer.NewVectorisedView(len(views[0])+payload.Size(), views), + NetworkHeader: buffer.View(ip), + }) + loopedR.Release() } if loop&stack.PacketOut == 0 { @@ -123,6 +136,26 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen return e.linkEP.WritePacket(r, gso, hdr, payload, ProtocolNumber) } +// WritePackets implements stack.LinkEndpoint.WritePackets. +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) { + if loop&stack.PacketLoop != 0 { + panic("not implemented") + } + if loop&stack.PacketOut == 0 { + return len(hdrs), nil + } + + for i := range hdrs { + hdr := &hdrs[i].Hdr + size := hdrs[i].Size + e.addIPHeader(r, hdr, size, params) + } + + n, err := e.linkEP.WritePackets(r, gso, hdrs, payload, ProtocolNumber) + r.Stats().IP.PacketsSent.IncrementBy(uint64(n)) + return n, err +} + // WriteHeaderIncludedPacker implements stack.NetworkEndpoint. It is not yet // supported by IPv6. func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error { @@ -132,30 +165,36 @@ func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.Vector // HandlePacket is called by the link layer when new ipv6 packets arrive for // this endpoint. -func (e *endpoint) HandlePacket(r *stack.Route, vv buffer.VectorisedView) { - headerView := vv.First() +func (e *endpoint) HandlePacket(r *stack.Route, pkt tcpip.PacketBuffer) { + headerView := pkt.Data.First() h := header.IPv6(headerView) - if !h.IsValid(vv.Size()) { + if !h.IsValid(pkt.Data.Size()) { return } - vv.TrimFront(header.IPv6MinimumSize) - vv.CapLength(int(h.PayloadLength())) + pkt.NetworkHeader = headerView[:header.IPv6MinimumSize] + pkt.Data.TrimFront(header.IPv6MinimumSize) + pkt.Data.CapLength(int(h.PayloadLength())) p := h.TransportProtocol() if p == header.ICMPv6ProtocolNumber { - e.handleICMP(r, headerView, vv) + e.handleICMP(r, headerView, pkt) return } r.Stats().IP.PacketsDelivered.Increment() - e.dispatcher.DeliverTransportPacket(r, p, headerView, vv) + e.dispatcher.DeliverTransportPacket(r, p, pkt) } // Close cleans up resources associated with the endpoint. func (*endpoint) Close() {} -type protocol struct{} +type protocol struct { + // defaultTTL is the current default TTL for the protocol. Only the + // uint8 portion of it is meaningful and it must be accessed + // atomically. + defaultTTL uint32 +} // Number returns the ipv6 protocol number. func (p *protocol) Number() tcpip.NetworkProtocolNumber { @@ -179,25 +218,48 @@ func (*protocol) ParseAddresses(v buffer.View) (src, dst tcpip.Address) { } // NewEndpoint creates a new ipv6 endpoint. -func (p *protocol) NewEndpoint(nicid tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, linkEP stack.LinkEndpoint) (stack.NetworkEndpoint, *tcpip.Error) { +func (p *protocol) NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, linkEP stack.LinkEndpoint) (stack.NetworkEndpoint, *tcpip.Error) { return &endpoint{ - nicid: nicid, + nicID: nicID, id: stack.NetworkEndpointID{LocalAddress: addrWithPrefix.Address}, prefixLen: addrWithPrefix.PrefixLen, linkEP: linkEP, linkAddrCache: linkAddrCache, dispatcher: dispatcher, + protocol: p, }, nil } // SetOption implements NetworkProtocol.SetOption. func (p *protocol) SetOption(option interface{}) *tcpip.Error { - return tcpip.ErrUnknownProtocolOption + switch v := option.(type) { + case tcpip.DefaultTTLOption: + p.SetDefaultTTL(uint8(v)) + return nil + default: + return tcpip.ErrUnknownProtocolOption + } } // Option implements NetworkProtocol.Option. func (p *protocol) Option(option interface{}) *tcpip.Error { - return tcpip.ErrUnknownProtocolOption + switch v := option.(type) { + case *tcpip.DefaultTTLOption: + *v = tcpip.DefaultTTLOption(p.DefaultTTL()) + return nil + default: + return tcpip.ErrUnknownProtocolOption + } +} + +// SetDefaultTTL sets the default TTL for endpoints created with this protocol. +func (p *protocol) SetDefaultTTL(ttl uint8) { + atomic.StoreUint32(&p.defaultTTL, uint32(ttl)) +} + +// DefaultTTL returns the default TTL for endpoints created with this protocol. +func (p *protocol) DefaultTTL() uint8 { + return uint8(atomic.LoadUint32(&p.defaultTTL)) } // calculateMTU calculates the network-layer payload MTU based on the link-layer @@ -212,5 +274,5 @@ func calculateMTU(mtu uint32) uint32 { // NewProtocol returns an IPv6 network protocol. func NewProtocol() stack.NetworkProtocol { - return &protocol{} + return &protocol{defaultTTL: DefaultTTL} } diff --git a/pkg/tcpip/network/ipv6/ipv6_test.go b/pkg/tcpip/network/ipv6/ipv6_test.go index 78c674c2c..1cbfa7278 100644 --- a/pkg/tcpip/network/ipv6/ipv6_test.go +++ b/pkg/tcpip/network/ipv6/ipv6_test.go @@ -55,7 +55,9 @@ func testReceiveICMP(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst DstAddr: dst, }) - e.Inject(ProtocolNumber, hdr.View().ToVectorisedView()) + e.InjectInbound(ProtocolNumber, tcpip.PacketBuffer{ + Data: hdr.View().ToVectorisedView(), + }) stats := s.Stats().ICMP.V6PacketsReceived @@ -64,7 +66,7 @@ func testReceiveICMP(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst } } -// testReceiveICMP tests receiving a UDP packet from src to dst. want is the +// testReceiveUDP tests receiving a UDP packet from src to dst. want is the // expected UDP received count after receiving the packet. func testReceiveUDP(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst tcpip.Address, want uint64) { t.Helper() @@ -111,7 +113,9 @@ func testReceiveUDP(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst DstAddr: dst, }) - e.Inject(ProtocolNumber, hdr.View().ToVectorisedView()) + e.InjectInbound(ProtocolNumber, tcpip.PacketBuffer{ + Data: hdr.View().ToVectorisedView(), + }) stat := s.Stats().UDP.PacketsReceived diff --git a/pkg/tcpip/network/ipv6/ndp_test.go b/pkg/tcpip/network/ipv6/ndp_test.go index e30791fe3..0dbce14a0 100644 --- a/pkg/tcpip/network/ipv6/ndp_test.go +++ b/pkg/tcpip/network/ipv6/ndp_test.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/link/channel" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/tcpip/transport/icmp" ) @@ -97,7 +98,9 @@ func TestHopLimitValidation(t *testing.T) { SrcAddr: r.LocalAddress, DstAddr: r.RemoteAddress, }) - ep.HandlePacket(r, hdr.View().ToVectorisedView()) + ep.HandlePacket(r, tcpip.PacketBuffer{ + Data: hdr.View().ToVectorisedView(), + }) } types := []struct { @@ -109,7 +112,7 @@ func TestHopLimitValidation(t *testing.T) { {"RouterSolicit", header.ICMPv6RouterSolicit, header.ICMPv6MinimumSize, func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { return stats.RouterSolicit }}, - {"RouterAdvert", header.ICMPv6RouterAdvert, header.ICMPv6MinimumSize, func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + {"RouterAdvert", header.ICMPv6RouterAdvert, header.ICMPv6HeaderSize + header.NDPRAMinimumSize, func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { return stats.RouterAdvert }}, {"NeighborSolicit", header.ICMPv6NeighborSolicit, header.ICMPv6NeighborSolicitMinimumSize, func(stats tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { @@ -150,7 +153,7 @@ func TestHopLimitValidation(t *testing.T) { // Receive the NDP packet with an invalid hop limit // value. - handleIPv6Payload(hdr, ndpHopLimit-1, ep, &r) + handleIPv6Payload(hdr, header.NDPHopLimit-1, ep, &r) // Invalid count should have increased. if got := invalid.Value(); got != 1 { @@ -164,7 +167,7 @@ func TestHopLimitValidation(t *testing.T) { } // Receive the NDP packet with a valid hop limit value. - handleIPv6Payload(hdr, ndpHopLimit, ep, &r) + handleIPv6Payload(hdr, header.NDPHopLimit, ep, &r) // Rx count of NDP packet of type typ.typ should have // increased. @@ -179,3 +182,191 @@ func TestHopLimitValidation(t *testing.T) { }) } } + +// TestRouterAdvertValidation tests that when the NIC is configured to handle +// NDP Router Advertisement packets, it validates the Router Advertisement +// properly before handling them. +func TestRouterAdvertValidation(t *testing.T) { + tests := []struct { + name string + src tcpip.Address + hopLimit uint8 + code uint8 + ndpPayload []byte + expectedSuccess bool + }{ + { + "OK", + lladdr0, + 255, + 0, + []byte{ + 0, 0, 0, 0, + 0, 0, 0, 0, + 0, 0, 0, 0, + }, + true, + }, + { + "NonLinkLocalSourceAddr", + addr1, + 255, + 0, + []byte{ + 0, 0, 0, 0, + 0, 0, 0, 0, + 0, 0, 0, 0, + }, + false, + }, + { + "HopLimitNot255", + lladdr0, + 254, + 0, + []byte{ + 0, 0, 0, 0, + 0, 0, 0, 0, + 0, 0, 0, 0, + }, + false, + }, + { + "NonZeroCode", + lladdr0, + 255, + 1, + []byte{ + 0, 0, 0, 0, + 0, 0, 0, 0, + 0, 0, 0, 0, + }, + false, + }, + { + "NDPPayloadTooSmall", + lladdr0, + 255, + 0, + []byte{ + 0, 0, 0, 0, + 0, 0, 0, 0, + 0, 0, 0, + }, + false, + }, + { + "OKWithOptions", + lladdr0, + 255, + 0, + []byte{ + // RA payload + 0, 0, 0, 0, + 0, 0, 0, 0, + 0, 0, 0, 0, + + // Option #1 (TargetLinkLayerAddress) + 2, 1, 0, 0, 0, 0, 0, 0, + + // Option #2 (unrecognized) + 255, 1, 0, 0, 0, 0, 0, 0, + + // Option #3 (PrefixInformation) + 3, 4, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + }, + true, + }, + { + "OptionWithZeroLength", + lladdr0, + 255, + 0, + []byte{ + // RA payload + 0, 0, 0, 0, + 0, 0, 0, 0, + 0, 0, 0, 0, + + // Option #1 (TargetLinkLayerAddress) + // Invalid as it has 0 length. + 2, 0, 0, 0, 0, 0, 0, 0, + + // Option #2 (unrecognized) + 255, 1, 0, 0, 0, 0, 0, 0, + + // Option #3 (PrefixInformation) + 3, 4, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + }, + false, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + e := channel.New(10, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + }) + + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(_) = %s", err) + } + + icmpSize := header.ICMPv6HeaderSize + len(test.ndpPayload) + hdr := buffer.NewPrependable(header.IPv6MinimumSize + icmpSize) + pkt := header.ICMPv6(hdr.Prepend(icmpSize)) + pkt.SetType(header.ICMPv6RouterAdvert) + pkt.SetCode(test.code) + copy(pkt.NDPPayload(), test.ndpPayload) + payloadLength := hdr.UsedLength() + pkt.SetChecksum(header.ICMPv6Checksum(pkt, test.src, header.IPv6AllNodesMulticastAddress, buffer.VectorisedView{})) + ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) + ip.Encode(&header.IPv6Fields{ + PayloadLength: uint16(payloadLength), + NextHeader: uint8(icmp.ProtocolNumber6), + HopLimit: test.hopLimit, + SrcAddr: test.src, + DstAddr: header.IPv6AllNodesMulticastAddress, + }) + + stats := s.Stats().ICMP.V6PacketsReceived + invalid := stats.Invalid + rxRA := stats.RouterAdvert + + if got := invalid.Value(); got != 0 { + t.Fatalf("got invalid = %d, want = 0", got) + } + if got := rxRA.Value(); got != 0 { + t.Fatalf("got rxRA = %d, want = 0", got) + } + + e.InjectInbound(header.IPv6ProtocolNumber, tcpip.PacketBuffer{ + Data: hdr.View().ToVectorisedView(), + }) + + if test.expectedSuccess { + if got := invalid.Value(); got != 0 { + t.Fatalf("got invalid = %d, want = 0", got) + } + if got := rxRA.Value(); got != 1 { + t.Fatalf("got rxRA = %d, want = 1", got) + } + + } else { + if got := invalid.Value(); got != 1 { + t.Fatalf("got invalid = %d, want = 1", got) + } + if got := rxRA.Value(); got != 0 { + t.Fatalf("got rxRA = %d, want = 0", got) + } + } + }) + } +} diff --git a/pkg/tcpip/packet_buffer.go b/pkg/tcpip/packet_buffer.go new file mode 100644 index 000000000..10b04239d --- /dev/null +++ b/pkg/tcpip/packet_buffer.go @@ -0,0 +1,54 @@ +// 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 tcpip + +import "gvisor.dev/gvisor/pkg/tcpip/buffer" + +// A PacketBuffer contains all the data of a network packet. +// +// As a PacketBuffer traverses up the stack, it may be necessary to pass it to +// multiple endpoints. Clone() should be called in such cases so that +// modifications to the Data field do not affect other copies. +// +// +stateify savable +type PacketBuffer struct { + // Data holds the payload of the packet. For inbound packets, it also + // holds the headers, which are consumed as the packet moves up the + // stack. Headers are guaranteed not to be split across views. + // + // The bytes backing Data are immutable, but Data itself may be trimmed + // or otherwise modified. + Data buffer.VectorisedView + + // The bytes backing these views are immutable. Each field may be nil + // if either it has not been set yet or no such header exists (e.g. + // packets sent via loopback may not have a link header). + // + // These fields may be Views into other Views. SR dosen't support this, + // so deep copies are necessary in some cases. + LinkHeader buffer.View + NetworkHeader buffer.View + TransportHeader buffer.View +} + +// Clone makes a copy of pk. It clones the Data field, which creates a new +// VectorisedView but does not deep copy the underlying bytes. +func (pk PacketBuffer) Clone() PacketBuffer { + return PacketBuffer{ + Data: pk.Data.Clone(nil), + LinkHeader: pk.LinkHeader, + NetworkHeader: pk.NetworkHeader, + TransportHeader: pk.TransportHeader, + } +} diff --git a/pkg/tcpip/packet_buffer_state.go b/pkg/tcpip/packet_buffer_state.go new file mode 100644 index 000000000..04c4cf136 --- /dev/null +++ b/pkg/tcpip/packet_buffer_state.go @@ -0,0 +1,26 @@ +// 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 tcpip + +import "gvisor.dev/gvisor/pkg/tcpip/buffer" + +// beforeSave is invoked by stateify. +func (pk *PacketBuffer) beforeSave() { + // Non-Data fields may be slices of the Data field. This causes + // problems for SR, so during save we make each header independent. + pk.LinkHeader = append(buffer.View(nil), pk.LinkHeader...) + pk.NetworkHeader = append(buffer.View(nil), pk.NetworkHeader...) + pk.TransportHeader = append(buffer.View(nil), pk.TransportHeader...) +} diff --git a/pkg/tcpip/ports/ports.go b/pkg/tcpip/ports/ports.go index 315780c0c..30cea8996 100644 --- a/pkg/tcpip/ports/ports.go +++ b/pkg/tcpip/ports/ports.go @@ -19,6 +19,7 @@ import ( "math" "math/rand" "sync" + "sync/atomic" "gvisor.dev/gvisor/pkg/tcpip" ) @@ -27,6 +28,10 @@ const ( // FirstEphemeral is the first ephemeral port. FirstEphemeral = 16000 + // numEphemeralPorts it the mnumber of available ephemeral ports to + // Netstack. + numEphemeralPorts = math.MaxUint16 - FirstEphemeral + 1 + anyIPAddress tcpip.Address = "" ) @@ -40,6 +45,13 @@ type portDescriptor struct { type PortManager struct { mu sync.RWMutex allocatedPorts map[portDescriptor]bindAddresses + + // hint is used to pick ports ephemeral ports in a stable order for + // a given port offset. + // + // hint must be accessed using the portHint/incPortHint helpers. + // TODO(gvisor.dev/issue/940): S/R this field. + hint uint32 } type portNode struct { @@ -47,43 +59,76 @@ type portNode struct { refs int } -// bindAddresses is a set of IP addresses. -type bindAddresses map[tcpip.Address]portNode +// deviceNode is never empty. When it has no elements, it is removed from the +// map that references it. +type deviceNode map[tcpip.NICID]portNode -// isAvailable checks whether an IP address is available to bind to. -func (b bindAddresses) isAvailable(addr tcpip.Address, reuse bool) bool { - if addr == anyIPAddress { - if len(b) == 0 { - return true - } +// isAvailable checks whether binding is possible by device. If not binding to a +// device, check against all portNodes. If binding to a specific device, check +// against the unspecified device and the provided device. +func (d deviceNode) isAvailable(reuse bool, bindToDevice tcpip.NICID) bool { + if bindToDevice == 0 { + // Trying to binding all devices. if !reuse { + // Can't bind because the (addr,port) is already bound. return false } - for _, n := range b { - if !n.reuse { + for _, p := range d { + if !p.reuse { + // Can't bind because the (addr,port) was previously bound without reuse. return false } } return true } - // If all addresses for this portDescriptor are already bound, no - // address is available. - if n, ok := b[anyIPAddress]; ok { - if !reuse { + if p, ok := d[0]; ok { + if !reuse || !p.reuse { return false } - if !n.reuse { + } + + if p, ok := d[bindToDevice]; ok { + if !reuse || !p.reuse { return false } } - if n, ok := b[addr]; ok { - if !reuse { + return true +} + +// bindAddresses is a set of IP addresses. +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, reuse bool, bindToDevice tcpip.NICID) 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(reuse, bindToDevice) { + return false + } + } + return true + } + + // Check that there is no conflict with the "any" address. + if d, ok := b[anyIPAddress]; ok { + if !d.isAvailable(reuse, bindToDevice) { return false } - return n.reuse } + + // Check that this is no conflict with the provided address. + if d, ok := b[addr]; ok { + if !d.isAvailable(reuse, bindToDevice) { + return false + } + } + return true } @@ -97,11 +142,40 @@ func NewPortManager() *PortManager { // is suitable for its needs, and stopping when a port is found or an error // occurs. func (s *PortManager) PickEphemeralPort(testPort func(p uint16) (bool, *tcpip.Error)) (port uint16, err *tcpip.Error) { - count := uint16(math.MaxUint16 - FirstEphemeral + 1) - offset := uint16(rand.Int31n(int32(count))) + offset := uint32(rand.Int31n(numEphemeralPorts)) + return s.pickEphemeralPort(offset, numEphemeralPorts, testPort) +} + +// portHint atomically reads and returns the s.hint value. +func (s *PortManager) portHint() uint32 { + return atomic.LoadUint32(&s.hint) +} + +// incPortHint atomically increments s.hint by 1. +func (s *PortManager) incPortHint() { + atomic.AddUint32(&s.hint, 1) +} - for i := uint16(0); i < count; i++ { - port = FirstEphemeral + (offset+i)%count +// PickEphemeralPortStable starts at the specified offset + s.portHint and +// iterates over all ephemeral ports, allowing the caller to decide whether a +// given port is suitable for its needs and stopping when a port is found or an +// error occurs. +func (s *PortManager) PickEphemeralPortStable(offset uint32, testPort func(p uint16) (bool, *tcpip.Error)) (port uint16, err *tcpip.Error) { + p, err := s.pickEphemeralPort(s.portHint()+offset, numEphemeralPorts, testPort) + if err == nil { + s.incPortHint() + } + return p, err + +} + +// pickEphemeralPort starts at the offset specified from the FirstEphemeral port +// and iterates over the number of ports specified by count and allows the +// caller to decide whether a given port is suitable for its needs, and stopping +// when a port is found or an error occurs. +func (s *PortManager) pickEphemeralPort(offset, count uint32, testPort func(p uint16) (bool, *tcpip.Error)) (port uint16, err *tcpip.Error) { + for i := uint32(0); i < count; i++ { + port = uint16(FirstEphemeral + (offset+i)%count) ok, err := testPort(port) if err != nil { return 0, err @@ -116,17 +190,17 @@ func (s *PortManager) PickEphemeralPort(testPort func(p uint16) (bool, *tcpip.Er } // 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, reuse bool) bool { +func (s *PortManager) IsPortAvailable(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool, bindToDevice tcpip.NICID) bool { s.mu.Lock() defer s.mu.Unlock() - return s.isPortAvailableLocked(networks, transport, addr, port, reuse) + return s.isPortAvailableLocked(networks, transport, addr, port, reuse, bindToDevice) } -func (s *PortManager) isPortAvailableLocked(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool) bool { +func (s *PortManager) isPortAvailableLocked(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool, bindToDevice tcpip.NICID) bool { for _, network := range networks { desc := portDescriptor{network, transport, port} if addrs, ok := s.allocatedPorts[desc]; ok { - if !addrs.isAvailable(addr, reuse) { + if !addrs.isAvailable(addr, reuse, bindToDevice) { return false } } @@ -138,14 +212,14 @@ 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, reuse bool) (reservedPort uint16, err *tcpip.Error) { +func (s *PortManager) ReservePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool, bindToDevice tcpip.NICID) (reservedPort uint16, err *tcpip.Error) { s.mu.Lock() defer s.mu.Unlock() // If a port is specified, just try to reserve it for all network // protocols. if port != 0 { - if !s.reserveSpecificPort(networks, transport, addr, port, reuse) { + if !s.reserveSpecificPort(networks, transport, addr, port, reuse, bindToDevice) { return 0, tcpip.ErrPortInUse } return port, nil @@ -153,13 +227,13 @@ 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, reuse), nil + return s.reserveSpecificPort(networks, transport, addr, p, reuse, bindToDevice), 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, reuse bool) bool { - if !s.isPortAvailableLocked(networks, transport, addr, port, reuse) { +func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool, bindToDevice tcpip.NICID) bool { + if !s.isPortAvailableLocked(networks, transport, addr, port, reuse, bindToDevice) { return false } @@ -171,11 +245,16 @@ func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber m = make(bindAddresses) s.allocatedPorts[desc] = m } - if n, ok := m[addr]; ok { + d, ok := m[addr] + if !ok { + d = make(deviceNode) + m[addr] = d + } + if n, ok := d[bindToDevice]; ok { n.refs++ - m[addr] = n + d[bindToDevice] = n } else { - m[addr] = portNode{reuse: reuse, refs: 1} + d[bindToDevice] = portNode{reuse: reuse, refs: 1} } } @@ -184,22 +263,28 @@ 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) { +func (s *PortManager) ReleasePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, bindToDevice tcpip.NICID) { s.mu.Lock() defer s.mu.Unlock() for _, network := range networks { desc := portDescriptor{network, transport, port} if m, ok := s.allocatedPorts[desc]; ok { - n, ok := m[addr] + d, ok := m[addr] + if !ok { + continue + } + n, ok := d[bindToDevice] if !ok { continue } n.refs-- + d[bindToDevice] = n if n.refs == 0 { + delete(d, bindToDevice) + } + if len(d) == 0 { delete(m, addr) - } else { - m[addr] = n } if len(m) == 0 { delete(s.allocatedPorts, desc) diff --git a/pkg/tcpip/ports/ports_test.go b/pkg/tcpip/ports/ports_test.go index 689401661..19f4833fc 100644 --- a/pkg/tcpip/ports/ports_test.go +++ b/pkg/tcpip/ports/ports_test.go @@ -15,6 +15,7 @@ package ports import ( + "math/rand" "testing" "gvisor.dev/gvisor/pkg/tcpip" @@ -34,6 +35,7 @@ type portReserveTestAction struct { want *tcpip.Error reuse bool release bool + device tcpip.NICID } func TestPortReservation(t *testing.T) { @@ -100,6 +102,112 @@ func TestPortReservation(t *testing.T) { {port: 24, ip: anyIPAddress, release: true}, {port: 24, ip: anyIPAddress, reuse: false, want: nil}, }, + }, { + tname: "bind twice with device fails", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 3, want: nil}, + {port: 24, ip: fakeIPAddress, device: 3, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind to device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 1, want: nil}, + {port: 24, ip: fakeIPAddress, device: 2, want: nil}, + }, + }, { + tname: "bind to device and then without device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind without device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, reuse: true, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind with device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 0, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 456, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 789, want: nil}, + {port: 24, ip: fakeIPAddress, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, reuse: true, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind with reuse", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: nil}, + }, + }, { + tname: "binding with reuse and device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 456, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 789, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 999, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "mixing reuse and not reuse by binding to device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 456, want: nil}, + {port: 24, ip: fakeIPAddress, device: 789, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 999, want: nil}, + }, + }, { + tname: "can't bind to 0 after mixing reuse and not reuse", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 456, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind and release", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 345, reuse: false, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 789, reuse: true, want: nil}, + + // Release the bind to device 0 and try again. + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: nil, release: true}, + {port: 24, ip: fakeIPAddress, device: 345, reuse: false, want: nil}, + }, + }, { + tname: "bind twice with reuse once", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: false, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "release an unreserved device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: false, want: nil}, + {port: 24, ip: fakeIPAddress, device: 456, reuse: false, want: nil}, + // The below don't exist. + {port: 24, ip: fakeIPAddress, device: 345, reuse: false, want: nil, release: true}, + {port: 9999, ip: fakeIPAddress, device: 123, reuse: false, want: nil, release: true}, + // Release all. + {port: 24, ip: fakeIPAddress, device: 123, reuse: false, want: nil, release: true}, + {port: 24, ip: fakeIPAddress, device: 456, reuse: false, want: nil, release: true}, + }, }, } { t.Run(test.tname, func(t *testing.T) { @@ -108,12 +216,12 @@ func TestPortReservation(t *testing.T) { for _, test := range test.actions { if test.release { - pm.ReleasePort(net, fakeTransNumber, test.ip, test.port) + pm.ReleasePort(net, fakeTransNumber, test.ip, test.port, test.device) continue } - gotPort, err := pm.ReservePort(net, fakeTransNumber, test.ip, test.port, test.reuse) + gotPort, err := pm.ReservePort(net, fakeTransNumber, test.ip, test.port, test.reuse, test.device) if err != test.want { - t.Fatalf("ReservePort(.., .., %s, %d, %t) = %v, want %v", test.ip, test.port, test.release, err, test.want) + t.Fatalf("ReservePort(.., .., %s, %d, %t, %d) = %v, want %v", test.ip, test.port, test.reuse, test.device, 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) @@ -125,7 +233,6 @@ func TestPortReservation(t *testing.T) { } func TestPickEphemeralPort(t *testing.T) { - pm := NewPortManager() customErr := &tcpip.Error{} for _, test := range []struct { name string @@ -169,9 +276,63 @@ func TestPickEphemeralPort(t *testing.T) { }, } { t.Run(test.name, func(t *testing.T) { + pm := NewPortManager() if port, err := pm.PickEphemeralPort(test.f); port != test.wantPort || err != test.wantErr { t.Errorf("PickEphemeralPort(..) = (port %d, err %v); want (port %d, err %v)", port, err, test.wantPort, test.wantErr) } }) } } + +func TestPickEphemeralPortStable(t *testing.T) { + customErr := &tcpip.Error{} + for _, test := range []struct { + name string + f func(port uint16) (bool, *tcpip.Error) + wantErr *tcpip.Error + wantPort uint16 + }{ + { + name: "no-port-available", + f: func(port uint16) (bool, *tcpip.Error) { + return false, nil + }, + wantErr: tcpip.ErrNoPortAvailable, + }, + { + name: "port-tester-error", + f: func(port uint16) (bool, *tcpip.Error) { + return false, customErr + }, + wantErr: customErr, + }, + { + name: "only-port-16042-available", + f: func(port uint16) (bool, *tcpip.Error) { + if port == FirstEphemeral+42 { + return true, nil + } + return false, nil + }, + wantPort: FirstEphemeral + 42, + }, + { + name: "only-port-under-16000-available", + f: func(port uint16) (bool, *tcpip.Error) { + if port < FirstEphemeral { + return true, nil + } + return false, nil + }, + wantErr: tcpip.ErrNoPortAvailable, + }, + } { + t.Run(test.name, func(t *testing.T) { + pm := NewPortManager() + portOffset := uint32(rand.Int31n(int32(numEphemeralPorts))) + if port, err := pm.PickEphemeralPortStable(portOffset, test.f); port != test.wantPort || err != test.wantErr { + t.Errorf("PickEphemeralPort(..) = (port %d, err %v); want (port %d, err %v)", port, err, test.wantPort, test.wantErr) + } + }) + } +} diff --git a/pkg/tcpip/seqnum/BUILD b/pkg/tcpip/seqnum/BUILD index 76b5f4ffa..29b7d761c 100644 --- a/pkg/tcpip/seqnum/BUILD +++ b/pkg/tcpip/seqnum/BUILD @@ -1,7 +1,7 @@ -package(licenses = ["notice"]) - load("//tools/go_stateify:defs.bzl", "go_library") +package(licenses = ["notice"]) + go_library( name = "seqnum", srcs = ["seqnum.go"], diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD index 28c49e8ff..460db3cf8 100644 --- a/pkg/tcpip/stack/BUILD +++ b/pkg/tcpip/stack/BUILD @@ -1,10 +1,9 @@ load("@io_bazel_rules_go//go:def.bzl", "go_test") - -package(licenses = ["notice"]) - load("//tools/go_generics:defs.bzl", "go_template_instance") load("//tools/go_stateify:defs.bzl", "go_library") +package(licenses = ["notice"]) + go_template_instance( name = "linkaddrentry_list", out = "linkaddrentry_list.go", @@ -23,6 +22,7 @@ go_library( "icmp_rate_limit.go", "linkaddrcache.go", "linkaddrentry_list.go", + "ndp.go", "nic.go", "registration.go", "route.go", @@ -36,6 +36,7 @@ go_library( ], deps = [ "//pkg/ilist", + "//pkg/rand", "//pkg/sleep", "//pkg/tcpip", "//pkg/tcpip/buffer", @@ -53,18 +54,26 @@ go_test( name = "stack_x_test", size = "small", srcs = [ + "ndp_test.go", "stack_test.go", + "transport_demuxer_test.go", "transport_test.go", ], deps = [ ":stack", "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/checker", "//pkg/tcpip/header", "//pkg/tcpip/iptables", "//pkg/tcpip/link/channel", "//pkg/tcpip/link/loopback", + "//pkg/tcpip/network/ipv4", + "//pkg/tcpip/network/ipv6", + "//pkg/tcpip/transport/icmp", + "//pkg/tcpip/transport/udp", "//pkg/waiter", + "@com_github_google_go-cmp//cmp:go_default_library", ], ) diff --git a/pkg/tcpip/stack/ndp.go b/pkg/tcpip/stack/ndp.go new file mode 100644 index 000000000..8e49f7a56 --- /dev/null +++ b/pkg/tcpip/stack/ndp.go @@ -0,0 +1,534 @@ +// 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 ( + "fmt" + "log" + "time" + + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" +) + +const ( + // defaultDupAddrDetectTransmits is the default number of NDP Neighbor + // Solicitation messages to send when doing Duplicate Address Detection + // for a tentative address. + // + // 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 + + // defaultHandleRAs is the default configuration for whether or not to + // handle incoming Router Advertisements as a host. + // + // Default = true. + defaultHandleRAs = true + + // defaultDiscoverDefaultRouters is the default configuration for + // whether or not to discover default routers from incoming Router + // Advertisements as a host. + // + // Default = true. + defaultDiscoverDefaultRouters = 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. + // + // Min = 1ms. + minimumRetransmitTimer = time.Millisecond + + // MaxDiscoveredDefaultRouters is the maximum number of discovered + // default routers. The stack should stop discovering new routers after + // discovering MaxDiscoveredDefaultRouters routers. + // + // This value MUST be at minimum 2 as per RFC 4861 section 6.3.4, and + // SHOULD be more. + // + // Max = 10. + MaxDiscoveredDefaultRouters = 10 +) + +// NDPDispatcher is the interface integrators of netstack must implement to +// receive and handle NDP related events. +type NDPDispatcher interface { + // OnDuplicateAddressDetectionStatus will be called when the DAD process + // for an address (addr) on a NIC (with ID nicID) completes. resolved + // will be set to true if DAD completed successfully (no duplicate addr + // detected); false otherwise (addr was detected to be a duplicate on + // the link the NIC is a part of, or it was stopped for some other + // reason, such as the address being removed). If an error occured + // during DAD, err will be set and resolved must be ignored. + // + // This function is permitted to block indefinitely without interfering + // with the stack's operation. + OnDuplicateAddressDetectionStatus(nicID tcpip.NICID, addr tcpip.Address, resolved bool, err *tcpip.Error) + + // OnDefaultRouterDiscovered will be called when a new default router is + // discovered. Implementations must return true along with a new valid + // route table if the newly discovered router should be remembered. If + // an implementation returns false, the second return value will be + // ignored. + // + // This function is not permitted to block indefinitely. This function + // is also not permitted to call into the stack. + OnDefaultRouterDiscovered(nicID tcpip.NICID, addr tcpip.Address) (bool, []tcpip.Route) + + // OnDefaultRouterInvalidated will be called when a discovered default + // router is invalidated. Implementers must return a new valid route + // table. + // + // This function is not permitted to block indefinitely. This function + // is also not permitted to call into the stack. + OnDefaultRouterInvalidated(nicID tcpip.NICID, addr tcpip.Address) []tcpip.Route +} + +// NDPConfigurations is the NDP configurations for the netstack. +type NDPConfigurations struct { + // The number of Neighbor Solicitation messages to send when doing + // Duplicate Address Detection for a tentative address. + // + // Note, a value of zero effectively disables DAD. + DupAddrDetectTransmits uint8 + + // The amount of time to wait between sending Neighbor solicitation + // messages. + // + // Must be greater than 0.5s. + RetransmitTimer time.Duration + + // HandleRAs determines whether or not Router Advertisements will be + // processed. + HandleRAs bool + + // DiscoverDefaultRouters determines whether or not default routers will + // be discovered from Router Advertisements. This configuration is + // ignored if HandleRAs is false. + DiscoverDefaultRouters bool +} + +// DefaultNDPConfigurations returns an NDPConfigurations populated with +// default values. +func DefaultNDPConfigurations() NDPConfigurations { + return NDPConfigurations{ + DupAddrDetectTransmits: defaultDupAddrDetectTransmits, + RetransmitTimer: defaultRetransmitTimer, + HandleRAs: defaultHandleRAs, + DiscoverDefaultRouters: defaultDiscoverDefaultRouters, + } +} + +// validate modifies an NDPConfigurations with valid values. If invalid values +// are present in c, the corresponding default values will be used instead. +// +// If RetransmitTimer is less than minimumRetransmitTimer, then a value of +// defaultRetransmitTimer will be used. +func (c *NDPConfigurations) validate() { + if c.RetransmitTimer < minimumRetransmitTimer { + c.RetransmitTimer = defaultRetransmitTimer + } +} + +// ndpState is the per-interface NDP state. +type ndpState struct { + // The NIC this ndpState is for. + nic *NIC + + // configs is the per-interface NDP configurations. + configs NDPConfigurations + + // The DAD state to send the next NS message, or resolve the address. + dad map[tcpip.Address]dadState + + // The default routers discovered through Router Advertisements. + defaultRouters map[tcpip.Address]defaultRouterState +} + +// dadState holds the Duplicate Address Detection timer and channel to signal +// 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 + + // Used to let the DAD timer know that it has been stopped. + // + // Must only be read from or written to while protected by the lock of + // the NIC this dadState is associated with. + done *bool +} + +// defaultRouterState holds data associated with a default router discovered by +// a Router Advertisement. +type defaultRouterState struct { + invalidationTimer *time.Timer + + // Used to signal the timer not to invalidate the default router (R) in + // a race condition (T1 is a goroutine that handles an RA from R and T2 + // is the goroutine that handles R's invalidation timer firing): + // T1: Receive a new RA from R + // T1: Obtain the NIC's lock before processing the RA + // T2: R's invalidation timer fires, and gets blocked on obtaining the + // NIC's lock + // T1: Refreshes/extends R's lifetime & releases NIC's lock + // T2: Obtains NIC's lock & invalidates R immediately + // + // To resolve this, T1 will check to see if the timer already fired, and + // signal the timer using this channel to not invalidate R, so that once + // T2 obtains the lock, it will see that there is an event on this + // channel and do nothing further. + doNotInvalidateC chan struct{} +} + +// startDuplicateAddressDetection performs Duplicate Address Detection. +// +// This function must only be called by IPv6 addresses that are currently +// tentative. +// +// The NIC that ndp belongs to MUST be locked. +func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, ref *referencedNetworkEndpoint) *tcpip.Error { + // addr must be a valid unicast IPv6 address. + if !header.IsV6UnicastAddress(addr) { + return tcpip.ErrAddressFamilyNotSupported + } + + // Should not attempt to perform DAD on an address that is currently in + // the DAD process. + if _, ok := ndp.dad[addr]; ok { + // Should never happen because we should only ever call this + // function for newly created addresses. If we attemped to + // "add" an address that already existed, we would returned an + // error since we attempted to add a duplicate address, or its + // reference count would have been increased without doing the + // work that would have been done for an address that was brand + // new. See NIC.addPermanentAddressLocked. + panic(fmt.Sprintf("ndpdad: already performing DAD for addr %s on NIC(%d)", addr, ndp.nic.ID())) + } + + remaining := ndp.configs.DupAddrDetectTransmits + + { + done, err := ndp.doDuplicateAddressDetection(addr, remaining, ref) + if err != nil { + return err + } + if done { + return nil + } + } + + remaining-- + + var done bool + var timer *time.Timer + timer = time.AfterFunc(ndp.configs.RetransmitTimer, func() { + var d bool + var err *tcpip.Error + + // doDadIteration does a single iteration of the DAD loop. + // + // Returns true if the integrator needs to be informed of DAD + // completing. + doDadIteration := func() bool { + ndp.nic.mu.Lock() + defer ndp.nic.mu.Unlock() + + if done { + // If we reach this point, it means that the DAD + // timer fired after another goroutine already + // obtained the NIC lock and stopped DAD before + // this function obtained the NIC lock. Simply + // return here and do nothing further. + return false + } + + ref, ok := ndp.nic.endpoints[NetworkEndpointID{addr}] + if !ok { + // This should never happen. + // We should have an endpoint for addr since we + // are still performing DAD on it. If the + // endpoint does not exist, but we are doing DAD + // on it, then we started DAD at some point, but + // forgot to stop it when the endpoint was + // deleted. + panic(fmt.Sprintf("ndpdad: unrecognized addr %s for NIC(%d)", addr, ndp.nic.ID())) + } + + d, err = ndp.doDuplicateAddressDetection(addr, remaining, ref) + if err != nil || d { + delete(ndp.dad, addr) + + if err != nil { + log.Printf("ndpdad: Error occured during DAD iteration for addr (%s) on NIC(%d); err = %s", addr, ndp.nic.ID(), err) + } + + // Let the integrator know DAD has completed. + return true + } + + remaining-- + timer.Reset(ndp.nic.stack.ndpConfigs.RetransmitTimer) + return false + } + + if doDadIteration() && ndp.nic.stack.ndpDisp != nil { + ndp.nic.stack.ndpDisp.OnDuplicateAddressDetectionStatus(ndp.nic.ID(), addr, d, err) + } + }) + + ndp.dad[addr] = dadState{ + timer: timer, + done: &done, + } + + return nil +} + +// doDuplicateAddressDetection is called on every iteration of the timer, and +// when DAD starts. +// +// It handles resolving the address (if there are no more NS to send), or +// sending the next NS if there are more NS to send. +// +// This function must only be called by IPv6 addresses that are currently +// tentative. +// +// The NIC that ndp belongs to (n) MUST be locked. +// +// Returns true if DAD has resolved; false if DAD is still ongoing. +func (ndp *ndpState) doDuplicateAddressDetection(addr tcpip.Address, remaining uint8, ref *referencedNetworkEndpoint) (bool, *tcpip.Error) { + if ref.getKind() != permanentTentative { + // The endpoint should still be marked as tentative + // since we are still performing DAD on it. + panic(fmt.Sprintf("ndpdad: addr %s is not tentative on NIC(%d)", addr, ndp.nic.ID())) + } + + if remaining == 0 { + // DAD has resolved. + ref.setKind(permanent) + return true, nil + } + + // Send a new NS. + snmc := header.SolicitedNodeAddr(addr) + snmcRef, ok := ndp.nic.endpoints[NetworkEndpointID{snmc}] + if !ok { + // This should never happen as if we have the + // address, we should have the solicited-node + // address. + panic(fmt.Sprintf("ndpdad: NIC(%d) is not in the solicited-node multicast group (%s) but it has addr %s", ndp.nic.ID(), snmc, addr)) + } + + // Use the unspecified address as the source address when performing + // DAD. + r := makeRoute(header.IPv6ProtocolNumber, header.IPv6Any, snmc, ndp.nic.linkEP.LinkAddress(), snmcRef, false, false) + + hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv6NeighborSolicitMinimumSize) + pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborSolicitMinimumSize)) + pkt.SetType(header.ICMPv6NeighborSolicit) + ns := header.NDPNeighborSolicit(pkt.NDPPayload()) + ns.SetTargetAddress(addr) + pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) + + sent := r.Stats().ICMP.V6PacketsSent + if err := r.WritePacket(nil, hdr, buffer.VectorisedView{}, NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: DefaultTOS}); err != nil { + sent.Dropped.Increment() + return false, err + } + sent.NeighborSolicit.Increment() + + return false, nil +} + +// stopDuplicateAddressDetection ends a running Duplicate Address Detection +// process. Note, this may leave the DAD process for a tentative address in +// such a state forever, unless some other external event resolves the DAD +// process (receiving an NA from the true owner of addr, or an NS for addr +// (implying another node is attempting to use addr)). It is up to the caller +// of this function to handle such a scenario. Normally, addr will be removed +// from n right after this function returns or the address successfully +// resolved. +// +// The NIC that ndp belongs to MUST be locked. +func (ndp *ndpState) stopDuplicateAddressDetection(addr tcpip.Address) { + dad, ok := ndp.dad[addr] + if !ok { + // Not currently performing DAD on addr, just return. + return + } + + if dad.timer != nil { + dad.timer.Stop() + dad.timer = nil + + *dad.done = true + dad.done = nil + } + + delete(ndp.dad, addr) + + // Let the integrator know DAD did not resolve. + if ndp.nic.stack.ndpDisp != nil { + go ndp.nic.stack.ndpDisp.OnDuplicateAddressDetectionStatus(ndp.nic.ID(), addr, false, nil) + } +} + +// handleRA handles a Router Advertisement message that arrived on the NIC +// this ndp is for. Does nothing if the NIC is configured to not handle RAs. +// +// The NIC that ndp belongs to and its associated stack MUST be locked. +func (ndp *ndpState) handleRA(ip tcpip.Address, ra header.NDPRouterAdvert) { + // Is the NIC configured to handle RAs at all? + // + // Currently, the stack does not determine router interface status on a + // per-interface basis; it is a stack-wide configuration, so we check + // stack's forwarding flag to determine if the NIC is a routing + // interface. + if !ndp.configs.HandleRAs || ndp.nic.stack.forwarding { + return + } + + // Is the NIC configured to discover default routers? + if ndp.configs.DiscoverDefaultRouters { + rtr, ok := ndp.defaultRouters[ip] + rl := ra.RouterLifetime() + switch { + case !ok && rl != 0: + // This is a new default router we are discovering. + // + // Only remember it if we currently know about less than + // MaxDiscoveredDefaultRouters routers. + if len(ndp.defaultRouters) < MaxDiscoveredDefaultRouters { + ndp.rememberDefaultRouter(ip, rl) + } + + case ok && rl != 0: + // This is an already discovered default router. Update + // the invalidation timer. + timer := rtr.invalidationTimer + + // We should ALWAYS have an invalidation timer for a + // discovered router. + if timer == nil { + panic("ndphandlera: RA invalidation timer should not be nil") + } + + if !timer.Stop() { + // If we reach this point, then we know the + // timer fired after we already took the NIC + // lock. Signal the timer so that once it + // obtains the lock, it doesn't actually + // invalidate the router as we just got a new + // RA that refreshes its lifetime to a non-zero + // value. See + // defaultRouterState.doNotInvalidateC for more + // details. + rtr.doNotInvalidateC <- struct{}{} + } + + timer.Reset(rl) + + case ok && rl == 0: + // We know about the router but it is no longer to be + // used as a default router so invalidate it. + ndp.invalidateDefaultRouter(ip) + } + } + + // TODO(b/140948104): Do Prefix Discovery. + // TODO(b/141556115): Do Parameter Discovery. +} + +// invalidateDefaultRouter invalidates a discovered default router. +// +// The NIC that ndp belongs to and its associated stack MUST be locked. +func (ndp *ndpState) invalidateDefaultRouter(ip tcpip.Address) { + rtr, ok := ndp.defaultRouters[ip] + + // Is the router still discovered? + if !ok { + // ...Nope, do nothing further. + return + } + + rtr.invalidationTimer.Stop() + rtr.invalidationTimer = nil + close(rtr.doNotInvalidateC) + rtr.doNotInvalidateC = nil + + delete(ndp.defaultRouters, ip) + + // Let the integrator know a discovered default router is invalidated. + if ndp.nic.stack.ndpDisp != nil { + ndp.nic.stack.routeTable = ndp.nic.stack.ndpDisp.OnDefaultRouterInvalidated(ndp.nic.ID(), ip) + } +} + +// rememberDefaultRouter remembers a newly discovered default router with IPv6 +// link-local address ip with lifetime rl. +// +// The router identified by ip MUST NOT already be known by the NIC. +// +// The NIC that ndp belongs to and its associated stack MUST be locked. +func (ndp *ndpState) rememberDefaultRouter(ip tcpip.Address, rl time.Duration) { + if ndp.nic.stack.ndpDisp == nil { + return + } + + // Inform the integrator when we discovered a default router. + remember, routeTable := ndp.nic.stack.ndpDisp.OnDefaultRouterDiscovered(ndp.nic.ID(), ip) + if !remember { + // Informed by the integrator to not remember the router, do + // nothing further. + return + } + + // Used to signal the timer not to invalidate the default router (R) in + // a race condition. See defaultRouterState.doNotInvalidateC for more + // details. + doNotInvalidateC := make(chan struct{}, 1) + + ndp.defaultRouters[ip] = defaultRouterState{ + invalidationTimer: time.AfterFunc(rl, func() { + ndp.nic.stack.mu.Lock() + defer ndp.nic.stack.mu.Unlock() + ndp.nic.mu.Lock() + defer ndp.nic.mu.Unlock() + + select { + case <-doNotInvalidateC: + return + default: + } + + ndp.invalidateDefaultRouter(ip) + }), + doNotInvalidateC: doNotInvalidateC, + } + + ndp.nic.stack.routeTable = routeTable +} diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go new file mode 100644 index 000000000..50ce1bbfa --- /dev/null +++ b/pkg/tcpip/stack/ndp_test.go @@ -0,0 +1,1013 @@ +// 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_test + +import ( + "encoding/binary" + "fmt" + "testing" + "time" + + "github.com/google/go-cmp/cmp" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/checker" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/link/channel" + "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" + "gvisor.dev/gvisor/pkg/tcpip/stack" + "gvisor.dev/gvisor/pkg/tcpip/transport/icmp" +) + +const ( + addr1 = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01" + addr2 = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02" + addr3 = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x03" + linkAddr1 = "\x02\x02\x03\x04\x05\x06" + linkAddr2 = "\x02\x02\x03\x04\x05\x07" + linkAddr3 = "\x02\x02\x03\x04\x05\x08" + defaultTimeout = 250 * time.Millisecond +) + +var ( + llAddr1 = header.LinkLocalAddr(linkAddr1) + llAddr2 = header.LinkLocalAddr(linkAddr2) + llAddr3 = header.LinkLocalAddr(linkAddr3) +) + +// TestDADDisabled tests that an address successfully resolves immediately +// when DAD is not enabled (the default for an empty stack.Options). +func TestDADDisabled(t *testing.T) { + opts := stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + } + + e := channel.New(10, 1280, linkAddr1) + s := stack.New(opts) + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(_) = %s", err) + } + + if err := s.AddAddress(1, header.IPv6ProtocolNumber, addr1); err != nil { + t.Fatalf("AddAddress(_, %d, %s) = %s", header.IPv6ProtocolNumber, addr1, err) + } + + // Should get the address immediately since we should not have performed + // DAD on it. + addr, err := s.GetMainNICAddress(1, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("stack.GetMainNICAddress(_, _) err = %s", err) + } + if addr.Address != addr1 { + t.Fatalf("got stack.GetMainNICAddress(_, _) = %s, want = %s", addr, addr1) + } + + // We should not have sent any NDP NS messages. + if got := s.Stats().ICMP.V6PacketsSent.NeighborSolicit.Value(); got != 0 { + t.Fatalf("got NeighborSolicit = %d, want = 0", got) + } +} + +// ndpDADEvent is a set of parameters that was passed to +// ndpDispatcher.OnDuplicateAddressDetectionStatus. +type ndpDADEvent struct { + nicID tcpip.NICID + addr tcpip.Address + resolved bool + err *tcpip.Error +} + +type ndpRouterEvent struct { + nicID tcpip.NICID + addr tcpip.Address + // true if router was discovered, false if invalidated. + discovered bool +} + +var _ stack.NDPDispatcher = (*ndpDispatcher)(nil) + +// ndpDispatcher implements NDPDispatcher so tests can know when various NDP +// related events happen for test purposes. +type ndpDispatcher struct { + dadC chan ndpDADEvent + routerC chan ndpRouterEvent + rememberRouter bool + routeTable []tcpip.Route +} + +// Implements stack.NDPDispatcher.OnDuplicateAddressDetectionStatus. +func (n *ndpDispatcher) OnDuplicateAddressDetectionStatus(nicID tcpip.NICID, addr tcpip.Address, resolved bool, err *tcpip.Error) { + if n.dadC != nil { + n.dadC <- ndpDADEvent{ + nicID, + addr, + resolved, + err, + } + } +} + +// Implements stack.NDPDispatcher.OnDefaultRouterDiscovered. +func (n *ndpDispatcher) OnDefaultRouterDiscovered(nicID tcpip.NICID, addr tcpip.Address) (bool, []tcpip.Route) { + if n.routerC != nil { + n.routerC <- ndpRouterEvent{ + nicID, + addr, + true, + } + } + + if !n.rememberRouter { + return false, nil + } + + rt := append([]tcpip.Route(nil), n.routeTable...) + rt = append(rt, tcpip.Route{ + Destination: header.IPv6EmptySubnet, + Gateway: addr, + NIC: nicID, + }) + n.routeTable = rt + return true, rt +} + +// Implements stack.NDPDispatcher.OnDefaultRouterInvalidated. +func (n *ndpDispatcher) OnDefaultRouterInvalidated(nicID tcpip.NICID, addr tcpip.Address) []tcpip.Route { + if n.routerC != nil { + n.routerC <- ndpRouterEvent{ + nicID, + addr, + false, + } + } + + var rt []tcpip.Route + exclude := tcpip.Route{ + Destination: header.IPv6EmptySubnet, + Gateway: addr, + NIC: nicID, + } + + for _, r := range n.routeTable { + if r != exclude { + rt = append(rt, r) + } + } + n.routeTable = rt + return rt +} + +// TestDADResolve tests that an address successfully resolves after performing +// DAD for various values of DupAddrDetectTransmits and RetransmitTimer. +// Included in the subtests is a test to make sure that an invalid +// RetransmitTimer (<1ms) values get fixed to the default RetransmitTimer of 1s. +func TestDADResolve(t *testing.T) { + tests := []struct { + name string + dupAddrDetectTransmits uint8 + retransTimer time.Duration + expectedRetransmitTimer time.Duration + }{ + {"1:1s:1s", 1, time.Second, time.Second}, + {"2:1s:1s", 2, time.Second, time.Second}, + {"1:2s:2s", 1, 2 * time.Second, 2 * time.Second}, + // 0s is an invalid RetransmitTimer timer and will be fixed to + // the default RetransmitTimer value of 1s. + {"1:0s:1s", 1, 0, time.Second}, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + ndpDisp := ndpDispatcher{ + dadC: make(chan ndpDADEvent), + } + opts := stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPDisp: &ndpDisp, + } + opts.NDPConfigs.RetransmitTimer = test.retransTimer + opts.NDPConfigs.DupAddrDetectTransmits = test.dupAddrDetectTransmits + + e := channel.New(10, 1280, linkAddr1) + s := stack.New(opts) + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(_) = %s", err) + } + + if err := s.AddAddress(1, header.IPv6ProtocolNumber, addr1); err != nil { + t.Fatalf("AddAddress(_, %d, %s) = %s", header.IPv6ProtocolNumber, addr1, err) + } + + stat := s.Stats().ICMP.V6PacketsSent.NeighborSolicit + + // Should have sent an NDP NS immediately. + if got := stat.Value(); got != 1 { + t.Fatalf("got NeighborSolicit = %d, want = 1", got) + + } + + // Address should not be considered bound to the NIC yet + // (DAD ongoing). + addr, err := s.GetMainNICAddress(1, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (_, %v), want = (_, nil)", err) + } + if want := (tcpip.AddressWithPrefix{}); addr != want { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want) + } + + // Wait for the remaining time - some delta (500ms), to + // make sure the address is still not resolved. + const delta = 500 * time.Millisecond + time.Sleep(test.expectedRetransmitTimer*time.Duration(test.dupAddrDetectTransmits) - delta) + addr, err = s.GetMainNICAddress(1, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (_, %v), want = (_, nil)", err) + } + if want := (tcpip.AddressWithPrefix{}); addr != want { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want) + } + + // Wait for DAD to resolve. + select { + case <-time.After(2 * delta): + // We should get a resolution event after 500ms + // (delta) since we wait for 500ms less than the + // expected resolution time above to make sure + // that the address did not yet resolve. Waiting + // for 1s (2x delta) without a resolution event + // means something is wrong. + t.Fatal("timed out waiting for DAD resolution") + case e := <-ndpDisp.dadC: + if e.err != nil { + t.Fatal("got DAD error: ", e.err) + } + if e.nicID != 1 { + t.Fatalf("got DAD event w/ nicID = %d, want = 1", e.nicID) + } + if e.addr != addr1 { + t.Fatalf("got DAD event w/ addr = %s, want = %s", addr, addr1) + } + if !e.resolved { + t.Fatal("got DAD event w/ resolved = false, want = true") + } + } + addr, err = s.GetMainNICAddress(1, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("stack.GetMainNICAddress(_, _) err = %s", err) + } + if addr.Address != addr1 { + t.Fatalf("got stack.GetMainNICAddress(_, _) = %s, want = %s", addr, addr1) + } + + // Should not have sent any more NS messages. + if got := stat.Value(); got != uint64(test.dupAddrDetectTransmits) { + t.Fatalf("got NeighborSolicit = %d, want = %d", got, test.dupAddrDetectTransmits) + } + + // Validate the sent Neighbor Solicitation messages. + for i := uint8(0); i < test.dupAddrDetectTransmits; i++ { + p := <-e.C + + // Make sure its an IPv6 packet. + if p.Proto != header.IPv6ProtocolNumber { + t.Fatalf("got Proto = %d, want = %d", p.Proto, header.IPv6ProtocolNumber) + } + + // Check NDP packet. + checker.IPv6(t, p.Header.ToVectorisedView().First(), + checker.TTL(header.NDPHopLimit), + checker.NDPNS( + checker.NDPNSTargetAddress(addr1))) + } + }) + } + +} + +// TestDADFail tests to make sure that the DAD process fails if another node is +// detected to be performing DAD on the same address (receive an NS message from +// a node doing DAD for the same address), or if another node is detected to own +// the address already (receive an NA message for the tentative address). +func TestDADFail(t *testing.T) { + tests := []struct { + name string + makeBuf func(tgt tcpip.Address) buffer.Prependable + getStat func(s tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter + }{ + { + "RxSolicit", + func(tgt tcpip.Address) buffer.Prependable { + hdr := buffer.NewPrependable(header.IPv6MinimumSize + header.ICMPv6NeighborSolicitMinimumSize) + pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborSolicitMinimumSize)) + pkt.SetType(header.ICMPv6NeighborSolicit) + ns := header.NDPNeighborSolicit(pkt.NDPPayload()) + ns.SetTargetAddress(tgt) + snmc := header.SolicitedNodeAddr(tgt) + pkt.SetChecksum(header.ICMPv6Checksum(pkt, header.IPv6Any, snmc, buffer.VectorisedView{})) + payloadLength := hdr.UsedLength() + ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) + ip.Encode(&header.IPv6Fields{ + PayloadLength: uint16(payloadLength), + NextHeader: uint8(icmp.ProtocolNumber6), + HopLimit: 255, + SrcAddr: header.IPv6Any, + DstAddr: snmc, + }) + + return hdr + + }, + func(s tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return s.NeighborSolicit + }, + }, + { + "RxAdvert", + func(tgt tcpip.Address) buffer.Prependable { + hdr := buffer.NewPrependable(header.IPv6MinimumSize + header.ICMPv6NeighborAdvertSize) + pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborAdvertSize)) + pkt.SetType(header.ICMPv6NeighborAdvert) + na := header.NDPNeighborAdvert(pkt.NDPPayload()) + na.SetSolicitedFlag(true) + na.SetOverrideFlag(true) + na.SetTargetAddress(tgt) + pkt.SetChecksum(header.ICMPv6Checksum(pkt, tgt, header.IPv6AllNodesMulticastAddress, buffer.VectorisedView{})) + payloadLength := hdr.UsedLength() + ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) + ip.Encode(&header.IPv6Fields{ + PayloadLength: uint16(payloadLength), + NextHeader: uint8(icmp.ProtocolNumber6), + HopLimit: 255, + SrcAddr: tgt, + DstAddr: header.IPv6AllNodesMulticastAddress, + }) + + return hdr + + }, + func(s tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + return s.NeighborAdvert + }, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + ndpDisp := ndpDispatcher{ + dadC: make(chan ndpDADEvent), + } + ndpConfigs := stack.DefaultNDPConfigurations() + opts := stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: ndpConfigs, + NDPDisp: &ndpDisp, + } + opts.NDPConfigs.RetransmitTimer = time.Second * 2 + + e := channel.New(10, 1280, linkAddr1) + s := stack.New(opts) + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(_) = %s", err) + } + + if err := s.AddAddress(1, header.IPv6ProtocolNumber, addr1); err != nil { + t.Fatalf("AddAddress(_, %d, %s) = %s", header.IPv6ProtocolNumber, addr1, err) + } + + // Address should not be considered bound to the NIC yet + // (DAD ongoing). + addr, err := s.GetMainNICAddress(1, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (_, %v), want = (_, nil)", err) + } + if want := (tcpip.AddressWithPrefix{}); addr != want { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want) + } + + // Receive a packet to simulate multiple nodes owning or + // attempting to own the same address. + hdr := test.makeBuf(addr1) + e.InjectInbound(header.IPv6ProtocolNumber, tcpip.PacketBuffer{ + Data: hdr.View().ToVectorisedView(), + }) + + stat := test.getStat(s.Stats().ICMP.V6PacketsReceived) + if got := stat.Value(); got != 1 { + t.Fatalf("got stat = %d, want = 1", got) + } + + // Wait for DAD to fail and make sure the address did + // not get resolved. + select { + case <-time.After(time.Duration(ndpConfigs.DupAddrDetectTransmits)*ndpConfigs.RetransmitTimer + time.Second): + // If we don't get a failure event after the + // expected resolution time + extra 1s buffer, + // something is wrong. + t.Fatal("timed out waiting for DAD failure") + case e := <-ndpDisp.dadC: + if e.err != nil { + t.Fatal("got DAD error: ", e.err) + } + if e.nicID != 1 { + t.Fatalf("got DAD event w/ nicID = %d, want = 1", e.nicID) + } + if e.addr != addr1 { + t.Fatalf("got DAD event w/ addr = %s, want = %s", addr, addr1) + } + if e.resolved { + t.Fatal("got DAD event w/ resolved = true, want = false") + } + } + addr, err = s.GetMainNICAddress(1, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (_, %v), want = (_, nil)", err) + } + if want := (tcpip.AddressWithPrefix{}); addr != want { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want) + } + }) + } +} + +// TestDADStop tests to make sure that the DAD process stops when an address is +// removed. +func TestDADStop(t *testing.T) { + ndpDisp := ndpDispatcher{ + dadC: make(chan ndpDADEvent), + } + ndpConfigs := stack.NDPConfigurations{ + RetransmitTimer: time.Second, + DupAddrDetectTransmits: 2, + } + opts := stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPDisp: &ndpDisp, + NDPConfigs: ndpConfigs, + } + + e := channel.New(10, 1280, linkAddr1) + s := stack.New(opts) + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(_) = %s", err) + } + + if err := s.AddAddress(1, header.IPv6ProtocolNumber, addr1); err != nil { + t.Fatalf("AddAddress(_, %d, %s) = %s", header.IPv6ProtocolNumber, addr1, err) + } + + // Address should not be considered bound to the NIC yet (DAD ongoing). + addr, err := s.GetMainNICAddress(1, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (_, %v), want = (_, nil)", err) + } + if want := (tcpip.AddressWithPrefix{}); addr != want { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want) + } + + // Remove the address. This should stop DAD. + if err := s.RemoveAddress(1, addr1); err != nil { + t.Fatalf("RemoveAddress(_, %s) = %s", addr1, err) + } + + // Wait for DAD to fail (since the address was removed during DAD). + select { + case <-time.After(time.Duration(ndpConfigs.DupAddrDetectTransmits)*ndpConfigs.RetransmitTimer + time.Second): + // If we don't get a failure event after the expected resolution + // time + extra 1s buffer, something is wrong. + t.Fatal("timed out waiting for DAD failure") + case e := <-ndpDisp.dadC: + if e.err != nil { + t.Fatal("got DAD error: ", e.err) + } + if e.nicID != 1 { + t.Fatalf("got DAD event w/ nicID = %d, want = 1", e.nicID) + } + if e.addr != addr1 { + t.Fatalf("got DAD event w/ addr = %s, want = %s", addr, addr1) + } + if e.resolved { + t.Fatal("got DAD event w/ resolved = true, want = false") + } + + } + addr, err = s.GetMainNICAddress(1, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (_, %v), want = (_, nil)", err) + } + if want := (tcpip.AddressWithPrefix{}); addr != want { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want) + } + + // Should not have sent more than 1 NS message. + if got := s.Stats().ICMP.V6PacketsSent.NeighborSolicit.Value(); got > 1 { + t.Fatalf("got NeighborSolicit = %d, want <= 1", got) + } +} + +// TestSetNDPConfigurationFailsForBadNICID tests to make sure we get an error if +// we attempt to update NDP configurations using an invalid NICID. +func TestSetNDPConfigurationFailsForBadNICID(t *testing.T) { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + }) + + // No NIC with ID 1 yet. + if got := s.SetNDPConfigurations(1, stack.NDPConfigurations{}); got != tcpip.ErrUnknownNICID { + t.Fatalf("got s.SetNDPConfigurations = %v, want = %s", got, tcpip.ErrUnknownNICID) + } +} + +// TestSetNDPConfigurations tests that we can update and use per-interface NDP +// configurations without affecting the default NDP configurations or other +// interfaces' configurations. +func TestSetNDPConfigurations(t *testing.T) { + tests := []struct { + name string + dupAddrDetectTransmits uint8 + retransmitTimer time.Duration + expectedRetransmitTimer time.Duration + }{ + { + "OK", + 1, + time.Second, + time.Second, + }, + { + "Invalid Retransmit Timer", + 1, + 0, + time.Second, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + ndpDisp := ndpDispatcher{ + dadC: make(chan ndpDADEvent), + } + e := channel.New(10, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPDisp: &ndpDisp, + }) + + // This NIC(1)'s NDP configurations will be updated to + // be different from the default. + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + // Created before updating NIC(1)'s NDP configurations + // but updating NIC(1)'s NDP configurations should not + // affect other existing NICs. + if err := s.CreateNIC(2, e); err != nil { + t.Fatalf("CreateNIC(2) = %s", err) + } + + // Update the NDP configurations on NIC(1) to use DAD. + configs := stack.NDPConfigurations{ + DupAddrDetectTransmits: test.dupAddrDetectTransmits, + RetransmitTimer: test.retransmitTimer, + } + if err := s.SetNDPConfigurations(1, configs); err != nil { + t.Fatalf("got SetNDPConfigurations(1, _) = %s", err) + } + + // Created after updating NIC(1)'s NDP configurations + // but the stack's default NDP configurations should not + // have been updated. + if err := s.CreateNIC(3, e); err != nil { + t.Fatalf("CreateNIC(3) = %s", err) + } + + // Add addresses for each NIC. + if err := s.AddAddress(1, header.IPv6ProtocolNumber, addr1); err != nil { + t.Fatalf("AddAddress(1, %d, %s) = %s", header.IPv6ProtocolNumber, addr1, err) + } + if err := s.AddAddress(2, header.IPv6ProtocolNumber, addr2); err != nil { + t.Fatalf("AddAddress(2, %d, %s) = %s", header.IPv6ProtocolNumber, addr2, err) + } + if err := s.AddAddress(3, header.IPv6ProtocolNumber, addr3); err != nil { + t.Fatalf("AddAddress(3, %d, %s) = %s", header.IPv6ProtocolNumber, addr3, err) + } + + // Address should not be considered bound to NIC(1) yet + // (DAD ongoing). + addr, err := s.GetMainNICAddress(1, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (_, %v), want = (_, nil)", err) + } + if want := (tcpip.AddressWithPrefix{}); addr != want { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want) + } + + // Should get the address on NIC(2) and NIC(3) + // immediately since we should not have performed DAD on + // it as the stack was configured to not do DAD by + // default and we only updated the NDP configurations on + // NIC(1). + addr, err = s.GetMainNICAddress(2, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("stack.GetMainNICAddress(2, _) err = %s", err) + } + if addr.Address != addr2 { + t.Fatalf("got stack.GetMainNICAddress(2, _) = %s, want = %s", addr, addr2) + } + addr, err = s.GetMainNICAddress(3, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("stack.GetMainNICAddress(3, _) err = %s", err) + } + if addr.Address != addr3 { + t.Fatalf("got stack.GetMainNICAddress(3, _) = %s, want = %s", addr, addr3) + } + + // Sleep until right (500ms before) before resolution to + // make sure the address didn't resolve on NIC(1) yet. + const delta = 500 * time.Millisecond + time.Sleep(time.Duration(test.dupAddrDetectTransmits)*test.expectedRetransmitTimer - delta) + addr, err = s.GetMainNICAddress(1, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (_, %v), want = (_, nil)", err) + } + if want := (tcpip.AddressWithPrefix{}); addr != want { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want) + } + + // Wait for DAD to resolve. + select { + case <-time.After(2 * delta): + // We should get a resolution event after 500ms + // (delta) since we wait for 500ms less than the + // expected resolution time above to make sure + // that the address did not yet resolve. Waiting + // for 1s (2x delta) without a resolution event + // means something is wrong. + t.Fatal("timed out waiting for DAD resolution") + case e := <-ndpDisp.dadC: + if e.err != nil { + t.Fatal("got DAD error: ", e.err) + } + if e.nicID != 1 { + t.Fatalf("got DAD event w/ nicID = %d, want = 1", e.nicID) + } + if e.addr != addr1 { + t.Fatalf("got DAD event w/ addr = %s, want = %s", addr, addr1) + } + if !e.resolved { + t.Fatal("got DAD event w/ resolved = false, want = true") + } + } + addr, err = s.GetMainNICAddress(1, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("stack.GetMainNICAddress(1, _) err = %s", err) + } + if addr.Address != addr1 { + t.Fatalf("got stack.GetMainNICAddress(1, _) = %s, want = %s", addr, addr1) + } + }) + } +} + +// raBuf returns a valid NDP Router Advertisement. +// +// Note, raBuf does not populate any of the RA fields other than the +// Router Lifetime. +func raBuf(ip tcpip.Address, rl uint16) tcpip.PacketBuffer { + icmpSize := header.ICMPv6HeaderSize + header.NDPRAMinimumSize + hdr := buffer.NewPrependable(header.IPv6MinimumSize + icmpSize) + pkt := header.ICMPv6(hdr.Prepend(icmpSize)) + pkt.SetType(header.ICMPv6RouterAdvert) + pkt.SetCode(0) + // Populate the Router Lifetime. + binary.BigEndian.PutUint16(pkt.NDPPayload()[2:], rl) + pkt.SetChecksum(header.ICMPv6Checksum(pkt, ip, header.IPv6AllNodesMulticastAddress, buffer.VectorisedView{})) + payloadLength := hdr.UsedLength() + iph := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) + iph.Encode(&header.IPv6Fields{ + PayloadLength: uint16(payloadLength), + NextHeader: uint8(icmp.ProtocolNumber6), + HopLimit: header.NDPHopLimit, + SrcAddr: ip, + DstAddr: header.IPv6AllNodesMulticastAddress, + }) + + return tcpip.PacketBuffer{Data: hdr.View().ToVectorisedView()} +} + +// TestNoRouterDiscovery tests that router discovery will not be performed if +// configured not to. +func TestNoRouterDiscovery(t *testing.T) { + // Being configured to discover routers means handle and + // discover are set to true and forwarding is set to false. + // This tests all possible combinations of the configurations, + // except for the configuration where handle = true, discover = + // true and forwarding = false (the required configuration to do + // router discovery) - that will done in other tests. + for i := 0; i < 7; i++ { + handle := i&1 != 0 + discover := i&2 != 0 + forwarding := i&4 == 0 + + t.Run(fmt.Sprintf("HandleRAs(%t), DiscoverDefaultRouters(%t), Forwarding(%t)", handle, discover, forwarding), func(t *testing.T) { + ndpDisp := ndpDispatcher{ + routerC: make(chan ndpRouterEvent, 10), + } + e := channel.New(10, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + HandleRAs: handle, + DiscoverDefaultRouters: discover, + }, + NDPDisp: &ndpDisp, + }) + s.SetForwarding(forwarding) + + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + // Rx an RA with non-zero lifetime. + e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, 1000)) + select { + case <-ndpDisp.routerC: + t.Fatal("unexpectedly discovered a router when configured not to") + case <-time.After(defaultTimeout): + } + }) + } +} + +// TestRouterDiscoveryDispatcherNoRemember tests that the stack does not +// remember a discovered router when the dispatcher asks it not to. +func TestRouterDiscoveryDispatcherNoRemember(t *testing.T) { + ndpDisp := ndpDispatcher{ + routerC: make(chan ndpRouterEvent, 10), + } + e := channel.New(10, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + HandleRAs: true, + DiscoverDefaultRouters: true, + }, + NDPDisp: &ndpDisp, + }) + + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + routeTable := []tcpip.Route{ + { + header.IPv6EmptySubnet, + llAddr3, + 1, + }, + } + s.SetRouteTable(routeTable) + + // Rx an RA with short lifetime. + lifetime := time.Duration(1) + e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, uint16(lifetime))) + select { + case r := <-ndpDisp.routerC: + if r.nicID != 1 { + t.Fatalf("got r.nicID = %d, want = 1", r.nicID) + } + if r.addr != llAddr2 { + t.Fatalf("got r.addr = %s, want = %s", r.addr, llAddr2) + } + if !r.discovered { + t.Fatal("got r.discovered = false, want = true") + } + case <-time.After(defaultTimeout): + t.Fatal("timeout waiting for router discovery event") + } + + // Original route table should not have been modified. + if got := s.GetRouteTable(); !cmp.Equal(got, routeTable) { + t.Fatalf("got GetRouteTable = %v, want = %v", got, routeTable) + } + + // Wait for the normal invalidation time plus an extra second to + // make sure we do not actually receive any invalidation events as + // we should not have remembered the router in the first place. + select { + case <-ndpDisp.routerC: + t.Fatal("should not have received any router events") + case <-time.After(lifetime*time.Second + defaultTimeout): + } + + // Original route table should not have been modified. + if got := s.GetRouteTable(); !cmp.Equal(got, routeTable) { + t.Fatalf("got GetRouteTable = %v, want = %v", got, routeTable) + } +} + +func TestRouterDiscovery(t *testing.T) { + ndpDisp := ndpDispatcher{ + routerC: make(chan ndpRouterEvent, 10), + rememberRouter: true, + } + e := channel.New(10, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + HandleRAs: true, + DiscoverDefaultRouters: true, + }, + NDPDisp: &ndpDisp, + }) + + waitForEvent := func(addr tcpip.Address, discovered bool, timeout time.Duration) { + t.Helper() + + select { + case r := <-ndpDisp.routerC: + if r.nicID != 1 { + t.Fatalf("got r.nicID = %d, want = 1", r.nicID) + } + if r.addr != addr { + t.Fatalf("got r.addr = %s, want = %s", r.addr, addr) + } + if r.discovered != discovered { + t.Fatalf("got r.discovered = %t, want = %t", r.discovered, discovered) + } + case <-time.After(timeout): + t.Fatal("timeout waiting for router discovery event") + } + } + + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + // Rx an RA from lladdr2 with zero lifetime. It should not be + // remembered. + e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, 0)) + select { + case <-ndpDisp.routerC: + t.Fatal("unexpectedly discovered a router with 0 lifetime") + case <-time.After(defaultTimeout): + } + + // Rx an RA from lladdr2 with a huge lifetime. + e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, 1000)) + waitForEvent(llAddr2, true, defaultTimeout) + + // Should have a default route through the discovered router. + if got, want := s.GetRouteTable(), []tcpip.Route{{header.IPv6EmptySubnet, llAddr2, 1}}; !cmp.Equal(got, want) { + t.Fatalf("got GetRouteTable = %v, want = %v", got, want) + } + + // Rx an RA from another router (lladdr3) with non-zero lifetime. + l3Lifetime := time.Duration(6) + e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr3, uint16(l3Lifetime))) + waitForEvent(llAddr3, true, defaultTimeout) + + // Should have default routes through the discovered routers. + if got, want := s.GetRouteTable(), []tcpip.Route{{header.IPv6EmptySubnet, llAddr2, 1}, {header.IPv6EmptySubnet, llAddr3, 1}}; !cmp.Equal(got, want) { + t.Fatalf("got GetRouteTable = %v, want = %v", got, want) + } + + // Rx an RA from lladdr2 with lesser lifetime. + l2Lifetime := time.Duration(2) + e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, uint16(l2Lifetime))) + select { + case <-ndpDisp.routerC: + t.Fatal("Should not receive a router event when updating lifetimes for known routers") + case <-time.After(defaultTimeout): + } + + // Should still have a default route through the discovered routers. + if got, want := s.GetRouteTable(), []tcpip.Route{{header.IPv6EmptySubnet, llAddr2, 1}, {header.IPv6EmptySubnet, llAddr3, 1}}; !cmp.Equal(got, want) { + t.Fatalf("got GetRouteTable = %v, want = %v", got, want) + } + + // Wait for lladdr2's router invalidation timer to fire. The lifetime + // of the router should have been updated to the most recent (smaller) + // lifetime. + // + // Wait for the normal lifetime plus an extra bit for the + // router to get invalidated. If we don't get an invalidation + // event after this time, then something is wrong. + waitForEvent(llAddr2, false, l2Lifetime*time.Second+defaultTimeout) + + // Should no longer have the default route through lladdr2. + if got, want := s.GetRouteTable(), []tcpip.Route{{header.IPv6EmptySubnet, llAddr3, 1}}; !cmp.Equal(got, want) { + t.Fatalf("got GetRouteTable = %v, want = %v", got, want) + } + + // Rx an RA from lladdr2 with huge lifetime. + e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, 1000)) + waitForEvent(llAddr2, true, defaultTimeout) + + // Should have a default route through the discovered routers. + if got, want := s.GetRouteTable(), []tcpip.Route{{header.IPv6EmptySubnet, llAddr3, 1}, {header.IPv6EmptySubnet, llAddr2, 1}}; !cmp.Equal(got, want) { + t.Fatalf("got GetRouteTable = %v, want = %v", got, want) + } + + // Rx an RA from lladdr2 with zero lifetime. It should be invalidated. + e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, 0)) + waitForEvent(llAddr2, false, defaultTimeout) + + // Should have deleted the default route through the router that just + // got invalidated. + if got, want := s.GetRouteTable(), []tcpip.Route{{header.IPv6EmptySubnet, llAddr3, 1}}; !cmp.Equal(got, want) { + t.Fatalf("got GetRouteTable = %v, want = %v", got, want) + } + + // Wait for lladdr3's router invalidation timer to fire. The lifetime + // of the router should have been updated to the most recent (smaller) + // lifetime. + // + // Wait for the normal lifetime plus an extra bit for the + // router to get invalidated. If we don't get an invalidation + // event after this time, then something is wrong. + waitForEvent(llAddr3, false, l3Lifetime*time.Second+defaultTimeout) + + // Should not have any routes now that all discovered routers have been + // invalidated. + if got := len(s.GetRouteTable()); got != 0 { + t.Fatalf("got len(s.GetRouteTable()) = %d, want = 0", got) + } +} + +// TestRouterDiscoveryMaxRouters tests that only +// stack.MaxDiscoveredDefaultRouters discovered routers are remembered. +func TestRouterDiscoveryMaxRouters(t *testing.T) { + ndpDisp := ndpDispatcher{ + routerC: make(chan ndpRouterEvent, 10), + rememberRouter: true, + } + e := channel.New(10, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + HandleRAs: true, + DiscoverDefaultRouters: true, + }, + NDPDisp: &ndpDisp, + }) + + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + expectedRt := [stack.MaxDiscoveredDefaultRouters]tcpip.Route{} + + // Receive an RA from 2 more than the max number of discovered routers. + for i := 1; i <= stack.MaxDiscoveredDefaultRouters+2; i++ { + linkAddr := []byte{2, 2, 3, 4, 5, 0} + linkAddr[5] = byte(i) + llAddr := header.LinkLocalAddr(tcpip.LinkAddress(linkAddr)) + + e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr, 5)) + + if i <= stack.MaxDiscoveredDefaultRouters { + expectedRt[i-1] = tcpip.Route{header.IPv6EmptySubnet, llAddr, 1} + select { + case r := <-ndpDisp.routerC: + if r.nicID != 1 { + t.Fatalf("got r.nicID = %d, want = 1", r.nicID) + } + if r.addr != llAddr { + t.Fatalf("got r.addr = %s, want = %s", r.addr, llAddr) + } + if !r.discovered { + t.Fatal("got r.discovered = false, want = true") + } + case <-time.After(defaultTimeout): + t.Fatal("timeout waiting for router discovery event") + } + + } else { + select { + case <-ndpDisp.routerC: + t.Fatal("should not have discovered a new router after we already discovered the max number of routers") + case <-time.After(defaultTimeout): + } + } + } + + // Should only have default routes for the first + // stack.MaxDiscoveredDefaultRouters discovered routers. + if got := s.GetRouteTable(); !cmp.Equal(got, expectedRt[:]) { + t.Fatalf("got GetRouteTable = %v, want = %v", got, expectedRt) + } +} diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 0e8a23f00..28a28ae6e 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -19,7 +19,6 @@ import ( "sync" "sync/atomic" - "gvisor.dev/gvisor/pkg/ilist" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -34,17 +33,24 @@ type NIC struct { linkEP LinkEndpoint loopback bool - demux *transportDemuxer - mu sync.RWMutex spoofing bool promiscuous bool - primary map[tcpip.NetworkProtocolNumber]*ilist.List + primary map[tcpip.NetworkProtocolNumber][]*referencedNetworkEndpoint endpoints map[NetworkEndpointID]*referencedNetworkEndpoint addressRanges []tcpip.Subnet mcastJoins map[NetworkEndpointID]int32 + // packetEPs is protected by mu, but the contained PacketEndpoint + // values are not. + packetEPs map[tcpip.NetworkProtocolNumber][]PacketEndpoint stats NICStats + + // ndp is the NDP related state for NIC. + // + // Note, read and write operations on ndp require that the NIC is + // appropriately locked. + ndp ndpState } // NICStats includes transmitted and received stats. @@ -78,17 +84,26 @@ const ( NeverPrimaryEndpoint ) +// newNIC returns a new NIC using the default NDP configurations from stack. func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, loopback bool) *NIC { - return &NIC{ + // TODO(b/141011931): Validate a LinkEndpoint (ep) is valid. For + // example, make sure that the link address it provides is a valid + // unicast ethernet address. + + // TODO(b/143357959): RFC 8200 section 5 requires that IPv6 endpoints + // observe an MTU of at least 1280 bytes. Ensure that this requirement + // of IPv6 is supported on this endpoint's LinkEndpoint. + + nic := &NIC{ stack: stack, id: id, name: name, linkEP: ep, loopback: loopback, - demux: newTransportDemuxer(stack), - primary: make(map[tcpip.NetworkProtocolNumber]*ilist.List), + primary: make(map[tcpip.NetworkProtocolNumber][]*referencedNetworkEndpoint), endpoints: make(map[NetworkEndpointID]*referencedNetworkEndpoint), mcastJoins: make(map[NetworkEndpointID]int32), + packetEPs: make(map[tcpip.NetworkProtocolNumber][]PacketEndpoint), stats: NICStats{ Tx: DirectionStats{ Packets: &tcpip.StatCounter{}, @@ -99,7 +114,23 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, loopback Bytes: &tcpip.StatCounter{}, }, }, + ndp: ndpState{ + configs: stack.ndpConfigs, + dad: make(map[tcpip.Address]dadState), + defaultRouters: make(map[tcpip.Address]defaultRouterState), + }, + } + nic.ndp.nic = nic + + // Register supported packet endpoint protocols. + for _, netProto := range header.Ethertypes { + nic.packetEPs[netProto] = []PacketEndpoint{} } + for _, netProto := range stack.networkProtocols { + nic.packetEPs[netProto.Number()] = []PacketEndpoint{} + } + + return nic } // enable enables the NIC. enable will attach the link to its LinkEndpoint and @@ -107,6 +138,16 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, loopback func (n *NIC) enable() *tcpip.Error { n.attachLinkEndpoint() + // Create an endpoint to receive broadcast packets on this interface. + if _, ok := n.stack.networkProtocols[header.IPv4ProtocolNumber]; ok { + if err := n.AddAddress(tcpip.ProtocolAddress{ + Protocol: header.IPv4ProtocolNumber, + AddressWithPrefix: tcpip.AddressWithPrefix{header.IPv4Broadcast, 8 * header.IPv4AddressSize}, + }, NeverPrimaryEndpoint); err != nil { + return err + } + } + // Join the IPv6 All-Nodes Multicast group if the stack is configured to // use IPv6. This is required to ensure that this node properly receives // and responds to the various NDP messages that are destined to the @@ -114,11 +155,50 @@ func (n *NIC) enable() *tcpip.Error { // when we perform Duplicate Address Detection, or Router Advertisement // when we do Router Discovery. See RFC 4862, section 5.4.2 and RFC 4861 // section 4.2 for more information. - if _, ok := n.stack.networkProtocols[header.IPv6ProtocolNumber]; ok { - return n.joinGroup(header.IPv6ProtocolNumber, header.IPv6AllNodesMulticastAddress) + // + // Also auto-generate an IPv6 link-local address based on the NIC's + // link address if it is configured to do so. Note, each interface is + // required to have IPv6 link-local unicast address, as per RFC 4291 + // section 2.1. + _, ok := n.stack.networkProtocols[header.IPv6ProtocolNumber] + if !ok { + return nil } - return nil + n.mu.Lock() + defer n.mu.Unlock() + + if err := n.joinGroupLocked(header.IPv6ProtocolNumber, header.IPv6AllNodesMulticastAddress); err != nil { + return err + } + + if !n.stack.autoGenIPv6LinkLocal { + return nil + } + + l2addr := n.linkEP.LinkAddress() + + // Only attempt to generate the link-local address if we have a + // valid MAC address. + // + // TODO(b/141011931): Validate a LinkEndpoint's link address + // (provided by LinkEndpoint.LinkAddress) before reaching this + // point. + if !header.IsValidUnicastEthernetAddress(l2addr) { + return nil + } + + addr := header.LinkLocalAddr(l2addr) + + _, err := n.addPermanentAddressLocked(tcpip.ProtocolAddress{ + Protocol: header.IPv6ProtocolNumber, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: addr, + PrefixLen: header.IPv6LinkLocalPrefix.PrefixLen, + }, + }, CanBePrimaryEndpoint) + + return err } // attachLinkEndpoint attaches the NIC to the endpoint, which will enable it @@ -154,18 +234,7 @@ func (n *NIC) primaryEndpoint(protocol tcpip.NetworkProtocolNumber) *referencedN n.mu.RLock() defer n.mu.RUnlock() - list := n.primary[protocol] - if list == nil { - return nil - } - - for e := list.Front(); e != nil; e = e.Next() { - r := e.(*referencedNetworkEndpoint) - // TODO(crawshaw): allow broadcast address when SO_BROADCAST is set. - switch r.ep.ID().LocalAddress { - case header.IPv4Broadcast, header.IPv4Any: - continue - } + for _, r := range n.primary[protocol] { if r.isValidForOutgoing() && r.tryIncRef() { return r } @@ -275,13 +344,35 @@ func (n *NIC) addPermanentAddressLocked(protocolAddress tcpip.ProtocolAddress, p id := NetworkEndpointID{protocolAddress.AddressWithPrefix.Address} if ref, ok := n.endpoints[id]; ok { switch ref.getKind() { - case permanent: + case permanentTentative, permanent: // The NIC already have a permanent endpoint with that address. return nil, tcpip.ErrDuplicateAddress case permanentExpired, temporary: - // Promote the endpoint to become permanent. + // Promote the endpoint to become permanent and respect + // the new peb. if ref.tryIncRef() { ref.setKind(permanent) + + refs := n.primary[ref.protocol] + for i, r := range refs { + if r == ref { + switch peb { + case CanBePrimaryEndpoint: + return ref, nil + case FirstPrimaryEndpoint: + if i == 0 { + return ref, nil + } + n.primary[r.protocol] = append(refs[:i], refs[i+1:]...) + case NeverPrimaryEndpoint: + n.primary[r.protocol] = append(refs[:i], refs[i+1:]...) + return ref, nil + } + } + } + + n.insertPrimaryEndpointLocked(ref, peb) + return ref, nil } // tryIncRef failing means the endpoint is scheduled to be removed once @@ -291,6 +382,7 @@ func (n *NIC) addPermanentAddressLocked(protocolAddress tcpip.ProtocolAddress, p n.removeEndpointLocked(ref) } } + return n.addAddressLocked(protocolAddress, peb, permanent) } @@ -314,6 +406,15 @@ func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb Primar if err != nil { return nil, err } + + isIPv6Unicast := protocolAddress.Protocol == header.IPv6ProtocolNumber && header.IsV6UnicastAddress(protocolAddress.AddressWithPrefix.Address) + + // If the address is an IPv6 address and it is a permanent address, + // mark it as tentative so it goes through the DAD process. + if isIPv6Unicast && kind == permanent { + kind = permanentTentative + } + ref := &referencedNetworkEndpoint{ refs: 1, ep: ep, @@ -331,7 +432,7 @@ func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb Primar // If we are adding an IPv6 unicast address, join the solicited-node // multicast address. - if protocolAddress.Protocol == header.IPv6ProtocolNumber && header.IsV6UnicastAddress(protocolAddress.AddressWithPrefix.Address) { + if isIPv6Unicast { snmc := header.SolicitedNodeAddr(protocolAddress.AddressWithPrefix.Address) if err := n.joinGroupLocked(protocolAddress.Protocol, snmc); err != nil { return nil, err @@ -340,17 +441,13 @@ func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb Primar n.endpoints[id] = ref - l, ok := n.primary[protocolAddress.Protocol] - if !ok { - l = &ilist.List{} - n.primary[protocolAddress.Protocol] = l - } + n.insertPrimaryEndpointLocked(ref, peb) - switch peb { - case CanBePrimaryEndpoint: - l.PushBack(ref) - case FirstPrimaryEndpoint: - l.PushFront(ref) + // If we are adding a tentative IPv6 address, start DAD. + if isIPv6Unicast && kind == permanentTentative { + if err := n.ndp.startDuplicateAddressDetection(protocolAddress.AddressWithPrefix.Address, ref); err != nil { + return nil, err + } } return ref, nil @@ -375,10 +472,12 @@ func (n *NIC) AllAddresses() []tcpip.ProtocolAddress { addrs := make([]tcpip.ProtocolAddress, 0, len(n.endpoints)) for nid, ref := range n.endpoints { - // Don't include expired or tempory endpoints to avoid confusion and - // prevent the caller from using those. + // Don't include tentative, expired or temporary endpoints to + // avoid confusion and prevent the caller from using those. switch ref.getKind() { - case permanentExpired, temporary: + case permanentTentative, permanentExpired, temporary: + // TODO(b/140898488): Should tentative addresses be + // returned? continue } addrs = append(addrs, tcpip.ProtocolAddress{ @@ -399,12 +498,12 @@ func (n *NIC) PrimaryAddresses() []tcpip.ProtocolAddress { var addrs []tcpip.ProtocolAddress for proto, list := range n.primary { - for e := list.Front(); e != nil; e = e.Next() { - ref := e.(*referencedNetworkEndpoint) - // Don't include expired or tempory endpoints to avoid confusion and - // prevent the caller from using those. + for _, ref := range list { + // Don't include tentative, expired or tempory endpoints + // to avoid confusion and prevent the caller from using + // those. switch ref.getKind() { - case permanentExpired, temporary: + case permanentTentative, permanentExpired, temporary: continue } @@ -464,6 +563,19 @@ func (n *NIC) AddressRanges() []tcpip.Subnet { return append(sns, n.addressRanges...) } +// insertPrimaryEndpointLocked adds r to n's primary endpoint list as required +// by peb. +// +// n MUST be locked. +func (n *NIC) insertPrimaryEndpointLocked(r *referencedNetworkEndpoint, peb PrimaryEndpointBehavior) { + switch peb { + case CanBePrimaryEndpoint: + n.primary[r.protocol] = append(n.primary[r.protocol], r) + case FirstPrimaryEndpoint: + n.primary[r.protocol] = append([]*referencedNetworkEndpoint{r}, n.primary[r.protocol]...) + } +} + func (n *NIC) removeEndpointLocked(r *referencedNetworkEndpoint) { id := *r.ep.ID() @@ -481,9 +593,12 @@ func (n *NIC) removeEndpointLocked(r *referencedNetworkEndpoint) { } delete(n.endpoints, id) - wasInList := r.Next() != nil || r.Prev() != nil || r == n.primary[r.protocol].Front() - if wasInList { - n.primary[r.protocol].Remove(r) + refs := n.primary[r.protocol] + for i, ref := range refs { + if ref == r { + n.primary[r.protocol] = append(refs[:i], refs[i+1:]...) + break + } } r.ep.Close() @@ -497,10 +612,22 @@ func (n *NIC) removeEndpoint(r *referencedNetworkEndpoint) { func (n *NIC) removePermanentAddressLocked(addr tcpip.Address) *tcpip.Error { r, ok := n.endpoints[NetworkEndpointID{addr}] - if !ok || r.getKind() != permanent { + if !ok { + return tcpip.ErrBadLocalAddress + } + + kind := r.getKind() + if kind != permanent && kind != permanentTentative { return tcpip.ErrBadLocalAddress } + isIPv6Unicast := r.protocol == header.IPv6ProtocolNumber && header.IsV6UnicastAddress(addr) + + // If we are removing a tentative IPv6 unicast address, stop DAD. + if isIPv6Unicast && kind == permanentTentative { + n.ndp.stopDuplicateAddressDetection(addr) + } + r.setKind(permanentExpired) if !r.decRefLocked() { // The endpoint still has references to it. @@ -511,7 +638,7 @@ func (n *NIC) removePermanentAddressLocked(addr tcpip.Address) *tcpip.Error { // If we are removing an IPv6 unicast address, leave the solicited-node // multicast address. - if r.protocol == header.IPv6ProtocolNumber && header.IsV6UnicastAddress(addr) { + if isIPv6Unicast { snmc := header.SolicitedNodeAddr(addr) if err := n.leaveGroupLocked(snmc); err != nil { return err @@ -541,6 +668,11 @@ func (n *NIC) joinGroup(protocol tcpip.NetworkProtocolNumber, addr tcpip.Address // exists yet. Otherwise it just increments its count. n MUST be locked before // joinGroupLocked is called. func (n *NIC) joinGroupLocked(protocol tcpip.NetworkProtocolNumber, addr tcpip.Address) *tcpip.Error { + // TODO(b/143102137): When implementing MLD, make sure MLD packets are + // not sent unless a valid link-local address is available for use on n + // as an MLD packet's source address must be a link-local address as + // outlined in RFC 3810 section 5. + id := NetworkEndpointID{addr} joins := n.mcastJoins[id] if joins == 0 { @@ -591,10 +723,10 @@ func (n *NIC) leaveGroupLocked(addr tcpip.Address) *tcpip.Error { return nil } -func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, localLinkAddr, remotelinkAddr tcpip.LinkAddress, ref *referencedNetworkEndpoint, vv buffer.VectorisedView) { +func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, localLinkAddr, remotelinkAddr tcpip.LinkAddress, ref *referencedNetworkEndpoint, pkt tcpip.PacketBuffer) { r := makeRoute(protocol, dst, src, localLinkAddr, ref, false /* handleLocal */, false /* multicastLoop */) r.RemoteLinkAddress = remotelinkAddr - ref.ep.HandlePacket(&r, vv) + ref.ep.HandlePacket(&r, pkt) ref.decRef() } @@ -604,9 +736,9 @@ func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, // Note that the ownership of the slice backing vv is retained by the caller. // This rule applies only to the slice itself, not to the items of the slice; // the ownership of the items is not retained by the caller. -func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, _ tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { +func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { n.stats.Rx.Packets.Increment() - n.stats.Rx.Bytes.IncrementBy(uint64(vv.Size())) + n.stats.Rx.Bytes.IncrementBy(uint64(pkt.Data.Size())) netProto, ok := n.stack.networkProtocols[protocol] if !ok { @@ -614,36 +746,39 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, _ tcpip.LinkAddr return } + // If no local link layer address is provided, assume it was sent + // directly to this NIC. + if local == "" { + local = n.linkEP.LinkAddress() + } + + // Are any packet sockets listening for this network protocol? + n.mu.RLock() + packetEPs := n.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.packetEPs[header.EthernetProtocolAll]...) + } + n.mu.RUnlock() + for _, ep := range packetEPs { + ep.HandlePacket(n.id, local, protocol, pkt.Clone()) + } + if netProto.Number() == header.IPv4ProtocolNumber || netProto.Number() == header.IPv6ProtocolNumber { n.stack.stats.IP.PacketsReceived.Increment() } - if len(vv.First()) < netProto.MinimumPacketSize() { + if len(pkt.Data.First()) < netProto.MinimumPacketSize() { n.stack.stats.MalformedRcvdPackets.Increment() return } - src, dst := netProto.ParseAddresses(vv.First()) - - n.stack.AddLinkAddress(n.id, src, remote) - - // If the packet is destined to the IPv4 Broadcast address, then make a - // route to each IPv4 network endpoint and let each endpoint handle the - // packet. - if dst == header.IPv4Broadcast { - // n.endpoints is mutex protected so acquire lock. - n.mu.RLock() - for _, ref := range n.endpoints { - if ref.isValidForIncoming() && ref.protocol == header.IPv4ProtocolNumber && ref.tryIncRef() { - handlePacket(protocol, dst, src, linkEP.LinkAddress(), remote, ref, vv) - } - } - n.mu.RUnlock() - return - } + src, dst := netProto.ParseAddresses(pkt.Data.First()) if ref := n.getRef(protocol, dst); ref != nil { - handlePacket(protocol, dst, src, linkEP.LinkAddress(), remote, ref, vv) + handlePacket(protocol, dst, src, linkEP.LinkAddress(), remote, ref, pkt) return } @@ -671,31 +806,34 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, _ tcpip.LinkAddr if ok { r.RemoteAddress = src // TODO(b/123449044): Update the source NIC as well. - ref.ep.HandlePacket(&r, vv) + ref.ep.HandlePacket(&r, pkt) ref.decRef() } else { // n doesn't have a destination endpoint. // Send the packet out of n. - hdr := buffer.NewPrependableFromView(vv.First()) - vv.RemoveFirst() + hdr := buffer.NewPrependableFromView(pkt.Data.First()) + pkt.Data.RemoveFirst() // TODO(b/128629022): use route.WritePacket. - if err := n.linkEP.WritePacket(&r, nil /* gso */, hdr, vv, protocol); err != nil { + if err := n.linkEP.WritePacket(&r, nil /* gso */, hdr, pkt.Data, protocol); err != nil { r.Stats().IP.OutgoingPacketErrors.Increment() } else { n.stats.Tx.Packets.Increment() - n.stats.Tx.Bytes.IncrementBy(uint64(hdr.UsedLength() + vv.Size())) + n.stats.Tx.Bytes.IncrementBy(uint64(hdr.UsedLength() + pkt.Data.Size())) } } return } - n.stack.stats.IP.InvalidAddressesReceived.Increment() + // If a packet socket handled the packet, don't treat it as invalid. + if len(packetEPs) == 0 { + n.stack.stats.IP.InvalidAddressesReceived.Increment() + } } // DeliverTransportPacket delivers the packets to the appropriate transport // protocol endpoint. -func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, netHeader buffer.View, vv buffer.VectorisedView) { +func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt tcpip.PacketBuffer) { state, ok := n.stack.transportProtocols[protocol] if !ok { n.stack.stats.UnknownProtocolRcvdPackets.Increment() @@ -707,46 +845,41 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN // Raw socket packets are delivered based solely on the transport // protocol number. We do not inspect the payload to ensure it's // validly formed. - if !n.demux.deliverRawPacket(r, protocol, netHeader, vv) { - n.stack.demux.deliverRawPacket(r, protocol, netHeader, vv) - } + n.stack.demux.deliverRawPacket(r, protocol, pkt) - if len(vv.First()) < transProto.MinimumPacketSize() { + if len(pkt.Data.First()) < transProto.MinimumPacketSize() { n.stack.stats.MalformedRcvdPackets.Increment() return } - srcPort, dstPort, err := transProto.ParsePorts(vv.First()) + srcPort, dstPort, err := transProto.ParsePorts(pkt.Data.First()) if err != nil { n.stack.stats.MalformedRcvdPackets.Increment() return } id := TransportEndpointID{dstPort, r.LocalAddress, srcPort, r.RemoteAddress} - if n.demux.deliverPacket(r, protocol, netHeader, vv, id) { - return - } - if n.stack.demux.deliverPacket(r, protocol, netHeader, vv, id) { + if n.stack.demux.deliverPacket(r, protocol, pkt, id) { return } // Try to deliver to per-stack default handler. if state.defaultHandler != nil { - if state.defaultHandler(r, id, netHeader, vv) { + if state.defaultHandler(r, id, pkt) { return } } // We could not find an appropriate destination for this packet, so // deliver it to the global handler. - if !transProto.HandleUnknownDestinationPacket(r, id, netHeader, vv) { + if !transProto.HandleUnknownDestinationPacket(r, id, pkt) { n.stack.stats.MalformedRcvdPackets.Increment() } } // DeliverTransportControlPacket delivers control packets to the appropriate // transport protocol endpoint. -func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, vv buffer.VectorisedView) { +func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, pkt tcpip.PacketBuffer) { state, ok := n.stack.transportProtocols[trans] if !ok { return @@ -757,20 +890,17 @@ func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcp // ICMPv4 only guarantees that 8 bytes of the transport protocol will // be present in the payload. We know that the ports are within the // first 8 bytes for all known transport protocols. - if len(vv.First()) < 8 { + if len(pkt.Data.First()) < 8 { return } - srcPort, dstPort, err := transProto.ParsePorts(vv.First()) + srcPort, dstPort, err := transProto.ParsePorts(pkt.Data.First()) if err != nil { return } id := TransportEndpointID{srcPort, local, dstPort, remote} - if n.demux.deliverControlPacket(net, trans, typ, extra, vv, id) { - return - } - if n.stack.demux.deliverControlPacket(net, trans, typ, extra, vv, id) { + if n.stack.demux.deliverControlPacket(n, net, trans, typ, extra, pkt, id) { return } } @@ -785,14 +915,78 @@ func (n *NIC) Stack() *Stack { return n.stack } +// isAddrTentative returns true if addr is tentative on n. +// +// Note that if addr is not associated with n, then this function will return +// false. It will only return true if the address is associated with the NIC +// AND it is tentative. +func (n *NIC) isAddrTentative(addr tcpip.Address) bool { + ref, ok := n.endpoints[NetworkEndpointID{addr}] + if !ok { + return false + } + + return ref.getKind() == permanentTentative +} + +// dupTentativeAddrDetected attempts to inform n that a tentative addr +// is a duplicate on a link. +// +// dupTentativeAddrDetected will delete the tentative address if it exists. +func (n *NIC) dupTentativeAddrDetected(addr tcpip.Address) *tcpip.Error { + n.mu.Lock() + defer n.mu.Unlock() + + ref, ok := n.endpoints[NetworkEndpointID{addr}] + if !ok { + return tcpip.ErrBadAddress + } + + if ref.getKind() != permanentTentative { + return tcpip.ErrInvalidEndpointState + } + + return n.removePermanentAddressLocked(addr) +} + +// setNDPConfigs sets the NDP configurations for n. +// +// Note, if c contains invalid NDP configuration values, it will be fixed to +// use default values for the erroneous values. +func (n *NIC) setNDPConfigs(c NDPConfigurations) { + c.validate() + + n.mu.Lock() + n.ndp.configs = c + n.mu.Unlock() +} + +// handleNDPRA handles an NDP Router Advertisement message that arrived on n. +func (n *NIC) handleNDPRA(ip tcpip.Address, ra header.NDPRouterAdvert) { + n.mu.Lock() + defer n.mu.Unlock() + + n.ndp.handleRA(ip, ra) +} + type networkEndpointKind int32 const ( + // A permanentTentative endpoint is a permanent address that is not yet + // considered to be fully bound to an interface in the traditional + // sense. That is, the address is associated with a NIC, but packets + // destined to the address MUST NOT be accepted and MUST be silently + // dropped, and the address MUST NOT be used as a source address for + // outgoing packets. For IPv6, addresses will be of this kind until + // NDP's Duplicate Address Detection has resolved, or be deleted if + // the process results in detecting a duplicate address. + permanentTentative networkEndpointKind = iota + // A permanent endpoint is created by adding a permanent address (vs. a // temporary one) to the NIC. Its reference count is biased by 1 to avoid // removal when no route holds a reference to it. It is removed by explicitly // removing the permanent address from the NIC. - permanent networkEndpointKind = iota + permanent // An expired permanent endoint is a permanent endoint that had its address // removed from the NIC, and it is waiting to be removed once no more routes @@ -810,8 +1004,37 @@ const ( temporary ) +func (n *NIC) registerPacketEndpoint(netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) *tcpip.Error { + n.mu.Lock() + defer n.mu.Unlock() + + eps, ok := n.packetEPs[netProto] + if !ok { + return tcpip.ErrNotSupported + } + n.packetEPs[netProto] = append(eps, ep) + + return nil +} + +func (n *NIC) unregisterPacketEndpoint(netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) { + n.mu.Lock() + defer n.mu.Unlock() + + eps, ok := n.packetEPs[netProto] + if !ok { + return + } + + for i, epOther := range eps { + if epOther == ep { + n.packetEPs[netProto] = append(eps[:i], eps[i+1:]...) + return + } + } +} + type referencedNetworkEndpoint struct { - ilist.Entry ep NetworkEndpoint nic *NIC protocol tcpip.NetworkProtocolNumber diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go index 80101d4bb..c0026f5a3 100644 --- a/pkg/tcpip/stack/registration.go +++ b/pkg/tcpip/stack/registration.go @@ -60,24 +60,64 @@ const ( // TransportEndpoint is the interface that needs to be implemented by transport // protocol (e.g., tcp, udp) endpoints that can handle packets. type TransportEndpoint interface { + // UniqueID returns an unique ID for this transport endpoint. + UniqueID() uint64 + // HandlePacket is called by the stack when new packets arrive to - // this transport endpoint. - HandlePacket(r *Route, id TransportEndpointID, vv buffer.VectorisedView) + // this transport endpoint. It sets pkt.TransportHeader. + // + // HandlePacket takes ownership of pkt. + HandlePacket(r *Route, id TransportEndpointID, pkt tcpip.PacketBuffer) - // HandleControlPacket is called by the stack when new control (e.g., + // HandleControlPacket is called by the stack when new control (e.g. // ICMP) packets arrive to this transport endpoint. - HandleControlPacket(id TransportEndpointID, typ ControlType, extra uint32, vv buffer.VectorisedView) + // HandleControlPacket takes ownership of pkt. + HandleControlPacket(id TransportEndpointID, typ ControlType, extra uint32, pkt tcpip.PacketBuffer) + + // Close puts the endpoint in a closed state and frees all resources + // associated with it. This cleanup may happen asynchronously. Wait can + // be used to block on this asynchronous cleanup. + Close() + + // Wait waits for any worker goroutines owned by the endpoint to stop. + // + // An endpoint can be requested to stop its worker goroutines by calling + // its Close method. + // + // Wait will not block if the endpoint hasn't started any goroutines + // yet, even if it might later. + Wait() } // RawTransportEndpoint is the interface that needs to be implemented by raw // transport protocol endpoints. RawTransportEndpoints receive the entire -// packet - including the link, network, and transport headers - as delivered -// to netstack. +// packet - including the network and transport headers - as delivered to +// netstack. type RawTransportEndpoint interface { // HandlePacket is called by the stack when new packets arrive to // this transport endpoint. The packet contains all data from the link // layer up. - HandlePacket(r *Route, netHeader buffer.View, packet buffer.VectorisedView) + // + // HandlePacket takes ownership of pkt. + HandlePacket(r *Route, pkt tcpip.PacketBuffer) +} + +// PacketEndpoint is the interface that needs to be implemented by packet +// transport protocol endpoints. These endpoints receive link layer headers in +// addition to whatever they contain (usually network and transport layer +// headers and a payload). +type PacketEndpoint interface { + // HandlePacket is called by the stack when new packets arrive that + // match the endpoint. + // + // Implementers should treat packet as immutable and should copy it + // before before modification. + // + // linkHeader may have a length of 0, in which case the PacketEndpoint + // should construct its own ethernet header for applications. + // + // HandlePacket takes ownership of pkt. + HandlePacket(nicID tcpip.NICID, addr tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) } // TransportProtocol is the interface that needs to be implemented by transport @@ -107,7 +147,9 @@ type TransportProtocol interface { // // The return value indicates whether the packet was well-formed (for // stats purposes only). - HandleUnknownDestinationPacket(r *Route, id TransportEndpointID, netHeader buffer.View, vv buffer.VectorisedView) bool + // + // HandleUnknownDestinationPacket takes ownership of pkt. + HandleUnknownDestinationPacket(r *Route, id TransportEndpointID, pkt tcpip.PacketBuffer) bool // SetOption allows enabling/disabling protocol specific features. // SetOption returns an error if the option is not supported or the @@ -125,13 +167,21 @@ type TransportProtocol interface { // the network layer. type TransportDispatcher interface { // DeliverTransportPacket delivers packets to the appropriate - // transport protocol endpoint. It also returns the network layer - // header for the enpoint to inspect or pass up the stack. - DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, netHeader buffer.View, vv buffer.VectorisedView) + // transport protocol endpoint. + // + // pkt.NetworkHeader must be set before calling DeliverTransportPacket. + // + // DeliverTransportPacket takes ownership of pkt. + DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt tcpip.PacketBuffer) // DeliverTransportControlPacket delivers control packets to the // appropriate transport protocol endpoint. - DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, vv buffer.VectorisedView) + // + // pkt.NetworkHeader must be set before calling + // DeliverTransportControlPacket. + // + // DeliverTransportControlPacket takes ownership of pkt. + DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, pkt tcpip.PacketBuffer) } // PacketLooping specifies where an outbound packet should be sent. @@ -146,6 +196,19 @@ const ( PacketLoop ) +// NetworkHeaderParams are the header parameters given as input by the +// transport endpoint to the network. +type NetworkHeaderParams struct { + // Protocol refers to the transport protocol number. + Protocol tcpip.TransportProtocolNumber + + // TTL refers to Time To Live field of the IP-header. + TTL uint8 + + // TOS refers to TypeOfService or TrafficClass field of the IP-header. + TOS uint8 +} + // NetworkEndpoint is the interface that needs to be implemented by endpoints // of network layer protocols (e.g., ipv4, ipv6). type NetworkEndpoint interface { @@ -170,7 +233,11 @@ type NetworkEndpoint interface { // WritePacket writes a packet to the given destination address and // protocol. - WritePacket(r *Route, gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.TransportProtocolNumber, ttl uint8, loop PacketLooping) *tcpip.Error + WritePacket(r *Route, gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params NetworkHeaderParams, loop PacketLooping) *tcpip.Error + + // WritePackets writes packets to the given destination address and + // protocol. + WritePackets(r *Route, gso *GSO, hdrs []PacketDescriptor, payload buffer.VectorisedView, params NetworkHeaderParams, loop PacketLooping) (int, *tcpip.Error) // WriteHeaderIncludedPacket writes a packet that includes a network // header to the given destination address. @@ -186,8 +253,10 @@ type NetworkEndpoint interface { NICID() tcpip.NICID // HandlePacket is called by the link layer when new packets arrive to - // this network endpoint. - HandlePacket(r *Route, vv buffer.VectorisedView) + // this network endpoint. It sets pkt.NetworkHeader. + // + // HandlePacket takes ownership of pkt. + HandlePacket(r *Route, pkt tcpip.PacketBuffer) // Close is called when the endpoint is reomved from a stack. Close() @@ -212,7 +281,7 @@ type NetworkProtocol interface { ParseAddresses(v buffer.View) (src, dst tcpip.Address) // NewEndpoint creates a new endpoint of this protocol. - NewEndpoint(nicid tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache LinkAddressCache, dispatcher TransportDispatcher, sender LinkEndpoint) (NetworkEndpoint, *tcpip.Error) + NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache LinkAddressCache, dispatcher TransportDispatcher, sender LinkEndpoint) (NetworkEndpoint, *tcpip.Error) // SetOption allows enabling/disabling protocol specific features. // SetOption returns an error if the option is not supported or the @@ -229,9 +298,15 @@ type NetworkProtocol interface { // packets to the appropriate network endpoint after it has been handled by // the data link layer. type NetworkDispatcher interface { - // DeliverNetworkPacket finds the appropriate network protocol - // endpoint and hands the packet over for further processing. - DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) + // DeliverNetworkPacket finds the appropriate network protocol endpoint + // and hands the packet over for further processing. + // + // pkt.LinkHeader may or may not be set before calling + // DeliverNetworkPacket. Some packets do not have link headers (e.g. + // packets sent via loopback), and won't have the field set. + // + // DeliverNetworkPacket takes ownership of pkt. + DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) } // LinkEndpointCapabilities is the type associated with the capabilities @@ -253,12 +328,18 @@ const ( CapabilitySaveRestore CapabilityDisconnectOk CapabilityLoopback - CapabilityGSO + CapabilityHardwareGSO + + // CapabilitySoftwareGSO indicates the link endpoint supports of sending + // multiple packets using a single call (LinkEndpoint.WritePackets). + CapabilitySoftwareGSO ) // LinkEndpoint is the interface implemented by data link layer protocols (e.g., // ethernet, loopback, raw) and used by network layer protocols to send packets -// out through the implementer's data link endpoint. +// out through the implementer's data link endpoint. When a link header exists, +// it sets each tcpip.PacketBuffer's LinkHeader field before passing it up the +// stack. type LinkEndpoint interface { // MTU is the maximum transmission unit for this endpoint. This is // usually dictated by the backing physical network; when such a @@ -288,6 +369,18 @@ type LinkEndpoint interface { // r.LocalLinkAddress if it is provided. WritePacket(r *Route, gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error + // WritePackets writes packets with the given protocol through the + // given route. + // + // Right now, WritePackets is used only when the software segmentation + // offload is enabled. If it will be used for something else, it may + // require to change syscall filters. + WritePackets(r *Route, gso *GSO, hdrs []PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) + + // WriteRawPacket writes a packet directly to the link. The packet + // should already have an ethernet header. + WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error + // Attach attaches the data link layer endpoint to the network-layer // dispatcher of the stack. Attach(dispatcher NetworkDispatcher) @@ -311,13 +404,14 @@ type LinkEndpoint interface { type InjectableLinkEndpoint interface { LinkEndpoint - // Inject injects an inbound packet. - Inject(protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) + // InjectInbound injects an inbound packet. + InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) - // WriteRawPacket writes a fully formed outbound packet directly to the link. + // InjectOutbound writes a fully formed outbound packet directly to the + // link. // // dest is used by endpoints with multiple raw destinations. - WriteRawPacket(dest tcpip.Address, packet []byte) *tcpip.Error + InjectOutbound(dest tcpip.Address, packet []byte) *tcpip.Error } // A LinkAddressResolver is an extension to a NetworkProtocol that @@ -346,10 +440,10 @@ type LinkAddressResolver interface { type LinkAddressCache interface { // CheckLocalAddress determines if the given local address exists, and if it // does not exist. - CheckLocalAddress(nicid tcpip.NICID, protocol tcpip.NetworkProtocolNumber, addr tcpip.Address) tcpip.NICID + CheckLocalAddress(nicID tcpip.NICID, protocol tcpip.NetworkProtocolNumber, addr tcpip.Address) tcpip.NICID // AddLinkAddress adds a link address to the cache. - AddLinkAddress(nicid tcpip.NICID, addr tcpip.Address, linkAddr tcpip.LinkAddress) + AddLinkAddress(nicID tcpip.NICID, addr tcpip.Address, linkAddr tcpip.LinkAddress) // GetLinkAddress looks up the cache to translate address to link address (e.g. IP -> MAC). // If the LinkEndpoint requests address resolution and there is a LinkAddressResolver @@ -360,17 +454,22 @@ type LinkAddressCache interface { // 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). - GetLinkAddress(nicid tcpip.NICID, addr, localAddr tcpip.Address, protocol tcpip.NetworkProtocolNumber, w *sleep.Waker) (tcpip.LinkAddress, <-chan struct{}, *tcpip.Error) + GetLinkAddress(nicID tcpip.NICID, addr, localAddr tcpip.Address, protocol tcpip.NetworkProtocolNumber, w *sleep.Waker) (tcpip.LinkAddress, <-chan struct{}, *tcpip.Error) // RemoveWaker removes a waker that has been added in GetLinkAddress(). - RemoveWaker(nicid tcpip.NICID, addr tcpip.Address, waker *sleep.Waker) + RemoveWaker(nicID tcpip.NICID, addr tcpip.Address, waker *sleep.Waker) } -// UnassociatedEndpointFactory produces endpoints for writing packets not -// associated with a particular transport protocol. Such endpoints can be used -// to write arbitrary packets that include the IP header. -type UnassociatedEndpointFactory interface { - NewUnassociatedRawEndpoint(stack *Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) +// RawFactory produces endpoints for writing various types of raw packets. +type RawFactory interface { + // NewUnassociatedEndpoint produces endpoints for writing packets not + // associated with a particular transport protocol. Such endpoints can + // be used to write arbitrary packets that include the network header. + NewUnassociatedEndpoint(stack *Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) + + // NewPacketEndpoint produces endpoints for reading and writing packets + // that include network and (when cooked is false) link layer headers. + NewPacketEndpoint(stack *Stack, cooked bool, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) } // GSOType is the type of GSO segments. @@ -381,8 +480,14 @@ type GSOType int // Types of gso segments. const ( GSONone GSOType = iota + + // Hardware GSO types: GSOTCPv4 GSOTCPv6 + + // GSOSW is used for software GSO segments which have to be sent by + // endpoint.WritePackets. + GSOSW ) // GSO contains generic segmentation offload properties. @@ -410,3 +515,7 @@ type GSOEndpoint interface { // GSOMaxSize returns the maximum GSO packet size. GSOMaxSize() uint32 } + +// SoftwareGSOMaxSize is a maximum allowed size of a software GSO segment. +// This isn't a hard limit, because it is never set into packet headers. +const SoftwareGSOMaxSize = (1 << 16) diff --git a/pkg/tcpip/stack/route.go b/pkg/tcpip/stack/route.go index 5c8b7977a..1a0a51b57 100644 --- a/pkg/tcpip/stack/route.go +++ b/pkg/tcpip/stack/route.go @@ -47,8 +47,8 @@ type Route struct { // starts. ref *referencedNetworkEndpoint - // loop controls where WritePacket should send packets. - loop PacketLooping + // Loop controls where WritePacket should send packets. + Loop PacketLooping } // makeRoute initializes a new route. It takes ownership of the provided @@ -59,6 +59,8 @@ func makeRoute(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr tcpip loop = PacketLoop } else if multicastLoop && (header.IsV4MulticastAddress(remoteAddr) || header.IsV6MulticastAddress(remoteAddr)) { loop |= PacketLoop + } else if remoteAddr == header.IPv4Broadcast { + loop |= PacketLoop } return Route{ @@ -67,7 +69,7 @@ func makeRoute(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr tcpip LocalLinkAddress: localLinkAddr, RemoteAddress: remoteAddr, ref: ref, - loop: loop, + Loop: loop, } } @@ -152,12 +154,12 @@ func (r *Route) IsResolutionRequired() bool { } // WritePacket writes the packet through the given route. -func (r *Route) WritePacket(gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.TransportProtocolNumber, ttl uint8) *tcpip.Error { +func (r *Route) WritePacket(gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params NetworkHeaderParams) *tcpip.Error { if !r.ref.isValidForOutgoing() { return tcpip.ErrInvalidEndpointState } - err := r.ref.ep.WritePacket(r, gso, hdr, payload, protocol, ttl, r.loop) + err := r.ref.ep.WritePacket(r, gso, hdr, payload, params, r.Loop) if err != nil { r.Stats().IP.OutgoingPacketErrors.Increment() } else { @@ -167,6 +169,44 @@ func (r *Route) WritePacket(gso *GSO, hdr buffer.Prependable, payload buffer.Vec return err } +// PacketDescriptor is a packet descriptor which contains a packet header and +// offset and size of packet data in a payload view. +type PacketDescriptor struct { + Hdr buffer.Prependable + Off int + Size int +} + +// NewPacketDescriptors allocates a set of packet descriptors. +func NewPacketDescriptors(n int, hdrSize int) []PacketDescriptor { + buf := make([]byte, n*hdrSize) + hdrs := make([]PacketDescriptor, n) + for i := range hdrs { + hdrs[i].Hdr = buffer.NewEmptyPrependableFromView(buf[i*hdrSize:][:hdrSize]) + } + return hdrs +} + +// WritePackets writes the set of packets through the given route. +func (r *Route) WritePackets(gso *GSO, hdrs []PacketDescriptor, payload buffer.VectorisedView, params NetworkHeaderParams) (int, *tcpip.Error) { + if !r.ref.isValidForOutgoing() { + return 0, tcpip.ErrInvalidEndpointState + } + + n, err := r.ref.ep.WritePackets(r, gso, hdrs, payload, params, r.Loop) + if err != nil { + r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(len(hdrs) - n)) + } + r.ref.nic.stats.Tx.Packets.IncrementBy(uint64(n)) + payloadSize := 0 + for i := 0; i < n; i++ { + r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(hdrs[i].Hdr.UsedLength())) + payloadSize += hdrs[i].Size + } + r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(payloadSize)) + return n, err +} + // WriteHeaderIncludedPacket writes a packet already containing a network // header through the given route. func (r *Route) WriteHeaderIncludedPacket(payload buffer.VectorisedView) *tcpip.Error { @@ -174,7 +214,7 @@ func (r *Route) WriteHeaderIncludedPacket(payload buffer.VectorisedView) *tcpip. return tcpip.ErrInvalidEndpointState } - if err := r.ref.ep.WriteHeaderIncludedPacket(r, payload, r.loop); err != nil { + if err := r.ref.ep.WriteHeaderIncludedPacket(r, payload, r.Loop); err != nil { r.Stats().IP.OutgoingPacketErrors.Increment() return err } @@ -208,10 +248,17 @@ func (r *Route) Clone() Route { return *r } -// MakeLoopedRoute duplicates the given route and tweaks it in case of multicast. +// MakeLoopedRoute duplicates the given route with special handling for routes +// used for sending multicast or broadcast packets. In those cases the +// multicast/broadcast address is the remote address when sending out, but for +// incoming (looped) packets it becomes the local address. Similarly, the local +// interface address that was the local address going out becomes the remote +// address coming in. This is different to unicast routes where local and +// remote addresses remain the same as they identify location (local vs remote) +// not direction (source vs destination). func (r *Route) MakeLoopedRoute() Route { l := r.Clone() - if header.IsV4MulticastAddress(r.RemoteAddress) || header.IsV6MulticastAddress(r.RemoteAddress) { + if r.RemoteAddress == header.IPv4Broadcast || header.IsV4MulticastAddress(r.RemoteAddress) || header.IsV6MulticastAddress(r.RemoteAddress) { l.RemoteAddress, l.LocalAddress = l.LocalAddress, l.RemoteAddress l.RemoteLinkAddress = l.LocalLinkAddress } diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 18d1704a5..2f8d8e822 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -20,10 +20,13 @@ package stack import ( + "encoding/binary" "sync" + "sync/atomic" "time" "golang.org/x/time/rate" + "gvisor.dev/gvisor/pkg/rand" "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" @@ -41,11 +44,14 @@ const ( resolutionTimeout = 1 * time.Second // resolutionAttempts is set to the same ARP retries used in Linux. resolutionAttempts = 3 + + // DefaultTOS is the default type of service value for network endpoints. + DefaultTOS = 0 ) type transportProtocolState struct { proto TransportProtocol - defaultHandler func(r *Route, id TransportEndpointID, netHeader buffer.View, vv buffer.VectorisedView) bool + defaultHandler func(r *Route, id TransportEndpointID, pkt tcpip.PacketBuffer) bool } // TCPProbeFunc is the expected function type for a TCP probe function to be @@ -339,6 +345,13 @@ type ResumableEndpoint interface { Resume(*Stack) } +// uniqueIDGenerator is a default unique ID generator. +type uniqueIDGenerator uint64 + +func (u *uniqueIDGenerator) UniqueID() uint64 { + return atomic.AddUint64((*uint64)(u), 1) +} + // Stack is a networking stack, with all supported protocols, NICs, and route // table. type Stack struct { @@ -346,10 +359,9 @@ type Stack struct { networkProtocols map[tcpip.NetworkProtocolNumber]NetworkProtocol linkAddrResolvers map[tcpip.NetworkProtocolNumber]LinkAddressResolver - // unassociatedFactory creates unassociated endpoints. If nil, raw - // endpoints are disabled. It is set during Stack creation and is - // immutable. - unassociatedFactory UnassociatedEndpointFactory + // rawFactory creates raw endpoints. If nil, raw endpoints are + // disabled. It is set during Stack creation and is immutable. + rawFactory RawFactory demux *transportDemuxer @@ -357,9 +369,10 @@ type Stack struct { linkAddrCache *linkAddrCache - mu sync.RWMutex - nics map[tcpip.NICID]*NIC - forwarding bool + mu sync.RWMutex + nics map[tcpip.NICID]*NIC + forwarding bool + cleanupEndpoints map[TransportEndpoint]struct{} // route is the route table passed in by the user via SetRouteTable(), // it is used by FindRoute() to build a route for a specific @@ -388,6 +401,32 @@ type Stack struct { // icmpRateLimiter is a global rate limiter for all ICMP messages generated // by the stack. icmpRateLimiter *ICMPRateLimiter + + // seed is a one-time random value initialized at stack startup + // and is used to seed the TCP port picking on active connections + // + // TODO(gvisor.dev/issue/940): S/R this field. + seed uint32 + + // ndpConfigs is the default NDP configurations used by interfaces. + ndpConfigs NDPConfigurations + + // autoGenIPv6LinkLocal determines whether or not the stack will attempt + // to auto-generate an IPv6 link-local address for newly enabled NICs. + // See the AutoGenIPv6LinkLocal field of Options for more details. + autoGenIPv6LinkLocal bool + + // ndpDisp is the NDP event dispatcher that is used to send the netstack + // integrator NDP related events. + ndpDisp NDPDispatcher + + // uniqueIDGenerator is a generator of unique identifiers. + uniqueIDGenerator UniqueID +} + +// UniqueID is an abstract generator of unique identifiers. +type UniqueID interface { + UniqueID() uint64 } // Options contains optional Stack configuration. @@ -411,14 +450,71 @@ type Options struct { // stack (false). HandleLocal bool - // UnassociatedFactory produces unassociated endpoints raw endpoints. - // Raw endpoints are enabled only if this is non-nil. - UnassociatedFactory UnassociatedEndpointFactory + // UniqueID is an optional generator of unique identifiers. + UniqueID UniqueID + + // NDPConfigs is the default NDP configurations used by interfaces. + // + // By default, NDPConfigs will have a zero value for its + // DupAddrDetectTransmits field, implying that DAD will not be performed + // before assigning an address to a NIC. + NDPConfigs NDPConfigurations + + // AutoGenIPv6LinkLocal determins whether or not the stack will attempt + // to auto-generate an IPv6 link-local address for newly enabled NICs. + // Note, setting this to true does not mean that a link-local address + // will be assigned right away, or at all. If Duplicate Address + // Detection is enabled, an address will only be assigned if it + // successfully resolves. If it fails, no further attempt will be made + // to auto-generate an IPv6 link-local address. + // + // The generated link-local address will follow RFC 4291 Appendix A + // guidelines. + AutoGenIPv6LinkLocal bool + + // NDPDisp is the NDP event dispatcher that an integrator can provide to + // receive NDP related events. + NDPDisp NDPDispatcher + + // RawFactory produces raw endpoints. Raw endpoints are enabled only if + // this is non-nil. + RawFactory RawFactory +} + +// TransportEndpointInfo holds useful information about a transport endpoint +// which can be queried by monitoring tools. +// +// +stateify savable +type TransportEndpointInfo struct { + // The following fields are initialized at creation time and are + // immutable. + + NetProto tcpip.NetworkProtocolNumber + TransProto tcpip.TransportProtocolNumber + + // The following fields are protected by endpoint mu. + + ID TransportEndpointID + // BindNICID and bindAddr are set via calls to Bind(). They are used to + // reject attempts to send data or connect via a different NIC or + // address + BindNICID tcpip.NICID + BindAddr tcpip.Address + // RegisterNICID is the default NICID registered as a side-effect of + // connect or datagram write. + RegisterNICID tcpip.NICID } +// IsEndpointInfo is an empty method to implement the tcpip.EndpointInfo +// marker interface. +func (*TransportEndpointInfo) IsEndpointInfo() {} + // New allocates a new networking stack with only the requested networking and // transport protocols configured with default options. // +// Note, NDPConfigurations will be fixed before being used by the Stack. That +// is, if an invalid value was provided, it will be reset to the default value. +// // Protocol options can be changed by calling the // SetNetworkProtocolOption/SetTransportProtocolOption methods provided by the // stack. Please refer to individual protocol implementations as to what options @@ -429,17 +525,30 @@ func New(opts Options) *Stack { clock = &tcpip.StdClock{} } + if opts.UniqueID == nil { + opts.UniqueID = new(uniqueIDGenerator) + } + + // Make sure opts.NDPConfigs contains valid values only. + opts.NDPConfigs.validate() + s := &Stack{ - transportProtocols: make(map[tcpip.TransportProtocolNumber]*transportProtocolState), - networkProtocols: make(map[tcpip.NetworkProtocolNumber]NetworkProtocol), - linkAddrResolvers: make(map[tcpip.NetworkProtocolNumber]LinkAddressResolver), - nics: make(map[tcpip.NICID]*NIC), - linkAddrCache: newLinkAddrCache(ageLimit, resolutionTimeout, resolutionAttempts), - PortManager: ports.NewPortManager(), - clock: clock, - stats: opts.Stats.FillIn(), - handleLocal: opts.HandleLocal, - icmpRateLimiter: NewICMPRateLimiter(), + transportProtocols: make(map[tcpip.TransportProtocolNumber]*transportProtocolState), + networkProtocols: make(map[tcpip.NetworkProtocolNumber]NetworkProtocol), + linkAddrResolvers: make(map[tcpip.NetworkProtocolNumber]LinkAddressResolver), + nics: make(map[tcpip.NICID]*NIC), + cleanupEndpoints: make(map[TransportEndpoint]struct{}), + linkAddrCache: newLinkAddrCache(ageLimit, resolutionTimeout, resolutionAttempts), + PortManager: ports.NewPortManager(), + clock: clock, + stats: opts.Stats.FillIn(), + handleLocal: opts.HandleLocal, + icmpRateLimiter: NewICMPRateLimiter(), + seed: generateRandUint32(), + ndpConfigs: opts.NDPConfigs, + autoGenIPv6LinkLocal: opts.AutoGenIPv6LinkLocal, + uniqueIDGenerator: opts.UniqueID, + ndpDisp: opts.NDPDisp, } // Add specified network protocols. @@ -457,8 +566,8 @@ func New(opts Options) *Stack { } } - // Add the factory for unassociated endpoints, if present. - s.unassociatedFactory = opts.UnassociatedFactory + // Add the factory for raw endpoints, if present. + s.rawFactory = opts.RawFactory // Create the global transport demuxer. s.demux = newTransportDemuxer(s) @@ -466,6 +575,11 @@ func New(opts Options) *Stack { return s } +// UniqueID returns a unique identifier. +func (s *Stack) UniqueID() uint64 { + return s.uniqueIDGenerator.UniqueID() +} + // SetNetworkProtocolOption allows configuring individual protocol level // options. This method returns an error if the protocol is not supported or // option is not supported by the protocol implementation or the provided value @@ -527,7 +641,7 @@ func (s *Stack) TransportProtocolOption(transport tcpip.TransportProtocolNumber, // // It must be called only during initialization of the stack. Changing it as the // stack is operating is not supported. -func (s *Stack) SetTransportProtocolHandler(p tcpip.TransportProtocolNumber, h func(*Route, TransportEndpointID, buffer.View, buffer.VectorisedView) bool) { +func (s *Stack) SetTransportProtocolHandler(p tcpip.TransportProtocolNumber, h func(*Route, TransportEndpointID, tcpip.PacketBuffer) bool) { state := s.transportProtocols[p] if state != nil { state.defaultHandler = h @@ -593,12 +707,12 @@ func (s *Stack) NewEndpoint(transport tcpip.TransportProtocolNumber, network tcp // protocol. Raw endpoints receive all traffic for a given protocol regardless // of address. func (s *Stack) NewRawEndpoint(transport tcpip.TransportProtocolNumber, network tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue, associated bool) (tcpip.Endpoint, *tcpip.Error) { - if s.unassociatedFactory == nil { + if s.rawFactory == nil { return nil, tcpip.ErrNotPermitted } if !associated { - return s.unassociatedFactory.NewUnassociatedRawEndpoint(s, network, transport, waiterQueue) + return s.rawFactory.NewUnassociatedEndpoint(s, network, transport, waiterQueue) } t, ok := s.transportProtocols[transport] @@ -609,6 +723,16 @@ func (s *Stack) NewRawEndpoint(transport tcpip.TransportProtocolNumber, network return t.proto.NewRawEndpoint(s, network, waiterQueue) } +// NewPacketEndpoint creates a new packet endpoint listening for the given +// netProto. +func (s *Stack) NewPacketEndpoint(cooked bool, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { + if s.rawFactory == nil { + return nil, tcpip.ErrNotPermitted + } + + return s.rawFactory.NewPacketEndpoint(s, cooked, netProto, waiterQueue) +} + // createNIC creates a NIC with the provided id and link-layer endpoint, and // optionally enable it. func (s *Stack) createNIC(id tcpip.NICID, name string, ep LinkEndpoint, enabled, loopback bool) *tcpip.Error { @@ -893,7 +1017,7 @@ func (s *Stack) FindRoute(id tcpip.NICID, localAddr, remoteAddr tcpip.Address, n } } else { for _, route := range s.routeTable { - if (id != 0 && id != route.NIC) || (len(remoteAddr) != 0 && !isBroadcast && !route.Destination.Contains(remoteAddr)) { + if (id != 0 && id != route.NIC) || (len(remoteAddr) != 0 && !route.Destination.Contains(remoteAddr)) { continue } if nic, ok := s.nics[route.NIC]; ok { @@ -931,13 +1055,13 @@ func (s *Stack) CheckNetworkProtocol(protocol tcpip.NetworkProtocolNumber) bool // CheckLocalAddress determines if the given local address exists, and if it // does, returns the id of the NIC it's bound to. Returns 0 if the address // does not exist. -func (s *Stack) CheckLocalAddress(nicid tcpip.NICID, protocol tcpip.NetworkProtocolNumber, addr tcpip.Address) tcpip.NICID { +func (s *Stack) CheckLocalAddress(nicID tcpip.NICID, protocol tcpip.NetworkProtocolNumber, addr tcpip.Address) tcpip.NICID { s.mu.RLock() defer s.mu.RUnlock() // If a NIC is specified, we try to find the address there only. - if nicid != 0 { - nic := s.nics[nicid] + if nicID != 0 { + nic := s.nics[nicID] if nic == nil { return 0 } @@ -996,35 +1120,35 @@ func (s *Stack) SetSpoofing(nicID tcpip.NICID, enable bool) *tcpip.Error { } // AddLinkAddress adds a link address to the stack link cache. -func (s *Stack) AddLinkAddress(nicid tcpip.NICID, addr tcpip.Address, linkAddr tcpip.LinkAddress) { - fullAddr := tcpip.FullAddress{NIC: nicid, Addr: addr} +func (s *Stack) AddLinkAddress(nicID tcpip.NICID, addr tcpip.Address, linkAddr tcpip.LinkAddress) { + fullAddr := tcpip.FullAddress{NIC: nicID, Addr: addr} s.linkAddrCache.add(fullAddr, linkAddr) // TODO: provide a way for a transport endpoint to receive a signal // that AddLinkAddress for a particular address has been called. } // GetLinkAddress implements LinkAddressCache.GetLinkAddress. -func (s *Stack) GetLinkAddress(nicid tcpip.NICID, addr, localAddr tcpip.Address, protocol tcpip.NetworkProtocolNumber, waker *sleep.Waker) (tcpip.LinkAddress, <-chan struct{}, *tcpip.Error) { +func (s *Stack) GetLinkAddress(nicID tcpip.NICID, addr, localAddr tcpip.Address, protocol tcpip.NetworkProtocolNumber, waker *sleep.Waker) (tcpip.LinkAddress, <-chan struct{}, *tcpip.Error) { s.mu.RLock() - nic := s.nics[nicid] + nic := s.nics[nicID] if nic == nil { s.mu.RUnlock() return "", nil, tcpip.ErrUnknownNICID } s.mu.RUnlock() - fullAddr := tcpip.FullAddress{NIC: nicid, Addr: addr} + fullAddr := tcpip.FullAddress{NIC: nicID, Addr: addr} linkRes := s.linkAddrResolvers[protocol] return s.linkAddrCache.get(fullAddr, linkRes, localAddr, nic.linkEP, waker) } // RemoveWaker implements LinkAddressCache.RemoveWaker. -func (s *Stack) RemoveWaker(nicid tcpip.NICID, addr tcpip.Address, waker *sleep.Waker) { +func (s *Stack) RemoveWaker(nicID tcpip.NICID, addr tcpip.Address, waker *sleep.Waker) { s.mu.RLock() defer s.mu.RUnlock() - if nic := s.nics[nicid]; nic == nil { - fullAddr := tcpip.FullAddress{NIC: nicid, Addr: addr} + if nic := s.nics[nicID]; nic == nil { + fullAddr := tcpip.FullAddress{NIC: nicID, Addr: addr} s.linkAddrCache.removeWaker(fullAddr, waker) } } @@ -1033,73 +1157,52 @@ func (s *Stack) RemoveWaker(nicid tcpip.NICID, addr tcpip.Address, waker *sleep. // transport dispatcher. Received packets that match the provided id will be // delivered to the given endpoint; specifying a nic is optional, but // nic-specific IDs have precedence over global ones. -func (s *Stack) RegisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool) *tcpip.Error { - if nicID == 0 { - return s.demux.registerEndpoint(netProtos, protocol, id, ep, reusePort) - } - - s.mu.RLock() - defer s.mu.RUnlock() - - nic := s.nics[nicID] - if nic == nil { - return tcpip.ErrUnknownNICID - } - - return nic.demux.registerEndpoint(netProtos, protocol, id, ep, reusePort) +func (s *Stack) RegisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error { + return s.demux.registerEndpoint(netProtos, protocol, id, ep, reusePort, 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) { - if nicID == 0 { - s.demux.unregisterEndpoint(netProtos, protocol, id, ep) - return - } +func (s *Stack) UnregisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, bindToDevice tcpip.NICID) { + s.demux.unregisterEndpoint(netProtos, protocol, id, ep, bindToDevice) +} - s.mu.RLock() - defer s.mu.RUnlock() +// StartTransportEndpointCleanup removes the endpoint with the given id from +// the stack transport dispatcher. It also transitions it to the cleanup stage. +func (s *Stack) StartTransportEndpointCleanup(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, bindToDevice tcpip.NICID) { + s.mu.Lock() + defer s.mu.Unlock() - nic := s.nics[nicID] - if nic != nil { - nic.demux.unregisterEndpoint(netProtos, protocol, id, ep) - } + s.cleanupEndpoints[ep] = struct{}{} + + s.demux.unregisterEndpoint(netProtos, protocol, id, ep, bindToDevice) +} + +// CompleteTransportEndpointCleanup removes the endpoint from the cleanup +// stage. +func (s *Stack) CompleteTransportEndpointCleanup(ep TransportEndpoint) { + s.mu.Lock() + delete(s.cleanupEndpoints, ep) + s.mu.Unlock() +} + +// FindTransportEndpoint finds an endpoint that most closely matches the provided +// id. If no endpoint is found it returns nil. +func (s *Stack) FindTransportEndpoint(netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, id TransportEndpointID, r *Route) TransportEndpoint { + return s.demux.findTransportEndpoint(netProto, transProto, id, r) } // RegisterRawTransportEndpoint registers the given endpoint with the stack // transport dispatcher. Received packets that match the provided transport // protocol will be delivered to the given endpoint. func (s *Stack) RegisterRawTransportEndpoint(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, ep RawTransportEndpoint) *tcpip.Error { - if nicID == 0 { - return s.demux.registerRawEndpoint(netProto, transProto, ep) - } - - s.mu.RLock() - defer s.mu.RUnlock() - - nic := s.nics[nicID] - if nic == nil { - return tcpip.ErrUnknownNICID - } - - return nic.demux.registerRawEndpoint(netProto, transProto, ep) + return s.demux.registerRawEndpoint(netProto, transProto, ep) } // UnregisterRawTransportEndpoint removes the endpoint for the transport // protocol from the stack transport dispatcher. func (s *Stack) UnregisterRawTransportEndpoint(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, ep RawTransportEndpoint) { - if nicID == 0 { - s.demux.unregisterRawEndpoint(netProto, transProto, ep) - return - } - - s.mu.RLock() - defer s.mu.RUnlock() - - nic := s.nics[nicID] - if nic != nil { - nic.demux.unregisterRawEndpoint(netProto, transProto, ep) - } + s.demux.unregisterRawEndpoint(netProto, transProto, ep) } // RegisterRestoredEndpoint records e as an endpoint that has been restored on @@ -1110,6 +1213,69 @@ func (s *Stack) RegisterRestoredEndpoint(e ResumableEndpoint) { s.mu.Unlock() } +// RegisteredEndpoints returns all endpoints which are currently registered. +func (s *Stack) RegisteredEndpoints() []TransportEndpoint { + s.mu.Lock() + defer s.mu.Unlock() + var es []TransportEndpoint + for _, e := range s.demux.protocol { + es = append(es, e.transportEndpoints()...) + } + return es +} + +// CleanupEndpoints returns endpoints currently in the cleanup state. +func (s *Stack) CleanupEndpoints() []TransportEndpoint { + s.mu.Lock() + es := make([]TransportEndpoint, 0, len(s.cleanupEndpoints)) + for e := range s.cleanupEndpoints { + es = append(es, e) + } + s.mu.Unlock() + return es +} + +// RestoreCleanupEndpoints adds endpoints to cleanup tracking. This is useful +// for restoring a stack after a save. +func (s *Stack) RestoreCleanupEndpoints(es []TransportEndpoint) { + s.mu.Lock() + for _, e := range es { + s.cleanupEndpoints[e] = struct{}{} + } + s.mu.Unlock() +} + +// Close closes all currently registered transport endpoints. +// +// Endpoints created or modified during this call may not get closed. +func (s *Stack) Close() { + for _, e := range s.RegisteredEndpoints() { + e.Close() + } +} + +// Wait waits for all transport and link endpoints to halt their worker +// goroutines. +// +// Endpoints created or modified during this call may not get waited on. +// +// Note that link endpoints must be stopped via an implementation specific +// mechanism. +func (s *Stack) Wait() { + for _, e := range s.RegisteredEndpoints() { + e.Wait() + } + for _, e := range s.CleanupEndpoints() { + e.Wait() + } + + s.mu.RLock() + defer s.mu.RUnlock() + for _, n := range s.nics { + n.linkEP.Wait() + } +} + // Resume restarts the stack after a restore. This must be called after the // entire system has been restored. func (s *Stack) Resume() { @@ -1124,6 +1290,109 @@ func (s *Stack) Resume() { } } +// RegisterPacketEndpoint registers ep with the stack, causing it to receive +// all traffic of the specified netProto on the given NIC. If nicID is 0, it +// receives traffic from every NIC. +func (s *Stack) RegisterPacketEndpoint(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) *tcpip.Error { + s.mu.Lock() + defer s.mu.Unlock() + + // If no NIC is specified, capture on all devices. + if nicID == 0 { + // Register with each NIC. + for _, nic := range s.nics { + if err := nic.registerPacketEndpoint(netProto, ep); err != nil { + s.unregisterPacketEndpointLocked(0, netProto, ep) + return err + } + } + return nil + } + + // Capture on a specific device. + nic, ok := s.nics[nicID] + if !ok { + return tcpip.ErrUnknownNICID + } + if err := nic.registerPacketEndpoint(netProto, ep); err != nil { + return err + } + + return nil +} + +// UnregisterPacketEndpoint unregisters ep for packets of the specified +// netProto from the specified NIC. If nicID is 0, ep is unregistered from all +// NICs. +func (s *Stack) UnregisterPacketEndpoint(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) { + s.mu.Lock() + defer s.mu.Unlock() + s.unregisterPacketEndpointLocked(nicID, netProto, ep) +} + +func (s *Stack) unregisterPacketEndpointLocked(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) { + // If no NIC is specified, unregister on all devices. + if nicID == 0 { + // Unregister with each NIC. + for _, nic := range s.nics { + nic.unregisterPacketEndpoint(netProto, ep) + } + return + } + + // Unregister in a single device. + nic, ok := s.nics[nicID] + if !ok { + return + } + nic.unregisterPacketEndpoint(netProto, ep) +} + +// WritePacket writes data directly to the specified NIC. It adds an ethernet +// header based on the arguments. +func (s *Stack) WritePacket(nicID tcpip.NICID, dst tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, payload buffer.VectorisedView) *tcpip.Error { + s.mu.Lock() + nic, ok := s.nics[nicID] + s.mu.Unlock() + if !ok { + return tcpip.ErrUnknownDevice + } + + // Add our own fake ethernet header. + ethFields := header.EthernetFields{ + SrcAddr: nic.linkEP.LinkAddress(), + DstAddr: dst, + Type: netProto, + } + fakeHeader := make(header.Ethernet, header.EthernetMinimumSize) + fakeHeader.Encode(ðFields) + ethHeader := buffer.View(fakeHeader).ToVectorisedView() + ethHeader.Append(payload) + + if err := nic.linkEP.WriteRawPacket(ethHeader); err != nil { + return err + } + + return nil +} + +// WriteRawPacket writes data directly to the specified NIC without adding any +// headers. +func (s *Stack) WriteRawPacket(nicID tcpip.NICID, payload buffer.VectorisedView) *tcpip.Error { + s.mu.Lock() + nic, ok := s.nics[nicID] + s.mu.Unlock() + if !ok { + return tcpip.ErrUnknownDevice + } + + if err := nic.linkEP.WriteRawPacket(payload); err != nil { + return err + } + + return nil +} + // NetworkProtocolInstance returns the protocol instance in the stack for the // specified network protocol. This method is public for protocol implementers // and tests to use. @@ -1243,3 +1512,85 @@ func (s *Stack) SetICMPBurst(burst int) { func (s *Stack) AllowICMPMessage() bool { return s.icmpRateLimiter.Allow() } + +// IsAddrTentative returns true if addr is tentative on the NIC with ID id. +// +// Note that if addr is not associated with a NIC with id ID, then this +// function will return false. It will only return true if the address is +// associated with the NIC AND it is tentative. +func (s *Stack) IsAddrTentative(id tcpip.NICID, addr tcpip.Address) (bool, *tcpip.Error) { + s.mu.RLock() + defer s.mu.RUnlock() + + nic, ok := s.nics[id] + if !ok { + return false, tcpip.ErrUnknownNICID + } + + return nic.isAddrTentative(addr), nil +} + +// DupTentativeAddrDetected attempts to inform the NIC with ID id that a +// tentative addr on it is a duplicate on a link. +func (s *Stack) DupTentativeAddrDetected(id tcpip.NICID, addr tcpip.Address) *tcpip.Error { + s.mu.Lock() + defer s.mu.Unlock() + + nic, ok := s.nics[id] + if !ok { + return tcpip.ErrUnknownNICID + } + + return nic.dupTentativeAddrDetected(addr) +} + +// SetNDPConfigurations sets the per-interface NDP configurations on the NIC +// with ID id to c. +// +// Note, if c contains invalid NDP configuration values, it will be fixed to +// use default values for the erroneous values. +func (s *Stack) SetNDPConfigurations(id tcpip.NICID, c NDPConfigurations) *tcpip.Error { + s.mu.Lock() + defer s.mu.Unlock() + + nic, ok := s.nics[id] + if !ok { + return tcpip.ErrUnknownNICID + } + + nic.setNDPConfigs(c) + + return nil +} + +// 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 { + s.mu.Lock() + defer s.mu.Unlock() + + nic, ok := s.nics[id] + if !ok { + return tcpip.ErrUnknownNICID + } + + nic.handleNDPRA(ip, ra) + + return nil +} + +// Seed returns a 32 bit value that can be used as a seed value for port +// picking, ISN generation etc. +// +// NOTE: The seed is generated once during stack initialization only. +func (s *Stack) Seed() uint32 { + return s.seed +} + +func generateRandUint32() uint32 { + b := make([]byte, 4) + if _, err := rand.Read(b); err != nil { + panic(err) + } + return binary.LittleEndian.Uint32(b) +} diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index d2dede8a9..bf1d6974c 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -24,11 +24,14 @@ import ( "sort" "strings" "testing" + "time" + "github.com/google/go-cmp/cmp" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/link/channel" + "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -55,7 +58,7 @@ const ( // use the first three: destination address, source address, and transport // protocol. They're all one byte fields to simplify parsing. type fakeNetworkEndpoint struct { - nicid tcpip.NICID + nicID tcpip.NICID id stack.NetworkEndpointID prefixLen int proto *fakeNetworkProtocol @@ -68,7 +71,7 @@ func (f *fakeNetworkEndpoint) MTU() uint32 { } func (f *fakeNetworkEndpoint) NICID() tcpip.NICID { - return f.nicid + return f.nicID } func (f *fakeNetworkEndpoint) PrefixLen() int { @@ -83,28 +86,28 @@ func (f *fakeNetworkEndpoint) ID() *stack.NetworkEndpointID { return &f.id } -func (f *fakeNetworkEndpoint) HandlePacket(r *stack.Route, vv buffer.VectorisedView) { +func (f *fakeNetworkEndpoint) HandlePacket(r *stack.Route, pkt tcpip.PacketBuffer) { // Increment the received packet count in the protocol descriptor. f.proto.packetCount[int(f.id.LocalAddress[0])%len(f.proto.packetCount)]++ // Consume the network header. - b := vv.First() - vv.TrimFront(fakeNetHeaderLen) + b := pkt.Data.First() + pkt.Data.TrimFront(fakeNetHeaderLen) // Handle control packets. if b[2] == uint8(fakeControlProtocol) { - nb := vv.First() + nb := pkt.Data.First() if len(nb) < fakeNetHeaderLen { return } - vv.TrimFront(fakeNetHeaderLen) - f.dispatcher.DeliverTransportControlPacket(tcpip.Address(nb[1:2]), tcpip.Address(nb[0:1]), fakeNetNumber, tcpip.TransportProtocolNumber(nb[2]), stack.ControlPortUnreachable, 0, vv) + pkt.Data.TrimFront(fakeNetHeaderLen) + f.dispatcher.DeliverTransportControlPacket(tcpip.Address(nb[1:2]), tcpip.Address(nb[0:1]), fakeNetNumber, tcpip.TransportProtocolNumber(nb[2]), stack.ControlPortUnreachable, 0, pkt) return } // Dispatch the packet to the transport protocol. - f.dispatcher.DeliverTransportPacket(r, tcpip.TransportProtocolNumber(b[2]), buffer.View([]byte{}), vv) + f.dispatcher.DeliverTransportPacket(r, tcpip.TransportProtocolNumber(b[2]), pkt) } func (f *fakeNetworkEndpoint) MaxHeaderLength() uint16 { @@ -119,7 +122,7 @@ func (f *fakeNetworkEndpoint) Capabilities() stack.LinkEndpointCapabilities { return f.ep.Capabilities() } -func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.TransportProtocolNumber, _ uint8, loop stack.PacketLooping) *tcpip.Error { +func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) *tcpip.Error { // Increment the sent packet count in the protocol descriptor. f.proto.sendPacketCount[int(r.RemoteAddress[0])%len(f.proto.sendPacketCount)]++ @@ -128,14 +131,16 @@ func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr bu b := hdr.Prepend(fakeNetHeaderLen) b[0] = r.RemoteAddress[0] b[1] = f.id.LocalAddress[0] - b[2] = byte(protocol) + b[2] = byte(params.Protocol) if loop&stack.PacketLoop != 0 { views := make([]buffer.View, 1, 1+len(payload.Views())) views[0] = hdr.View() views = append(views, payload.Views()...) vv := buffer.NewVectorisedView(len(views[0])+payload.Size(), views) - f.HandlePacket(r, vv) + f.HandlePacket(r, tcpip.PacketBuffer{ + Data: vv, + }) } if loop&stack.PacketOut == 0 { return nil @@ -144,6 +149,11 @@ func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr bu return f.ep.WritePacket(r, gso, hdr, payload, fakeNetNumber) } +// WritePackets implements stack.LinkEndpoint.WritePackets. +func (f *fakeNetworkEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) { + panic("not implemented") +} + func (*fakeNetworkEndpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error { return tcpip.ErrNotSupported } @@ -189,9 +199,9 @@ func (*fakeNetworkProtocol) ParseAddresses(v buffer.View) (src, dst tcpip.Addres return tcpip.Address(v[1:2]), tcpip.Address(v[0:1]) } -func (f *fakeNetworkProtocol) NewEndpoint(nicid tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, ep stack.LinkEndpoint) (stack.NetworkEndpoint, *tcpip.Error) { +func (f *fakeNetworkProtocol) NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, ep stack.LinkEndpoint) (stack.NetworkEndpoint, *tcpip.Error) { return &fakeNetworkEndpoint{ - nicid: nicid, + nicID: nicID, id: stack.NetworkEndpointID{LocalAddress: addrWithPrefix.Address}, prefixLen: addrWithPrefix.PrefixLen, proto: f, @@ -251,7 +261,9 @@ func TestNetworkReceive(t *testing.T) { // Make sure packet with wrong address is not delivered. buf[0] = 3 - ep.Inject(fakeNetNumber, buf.ToVectorisedView()) + ep.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) if fakeNet.packetCount[1] != 0 { t.Errorf("packetCount[1] = %d, want %d", fakeNet.packetCount[1], 0) } @@ -261,7 +273,9 @@ func TestNetworkReceive(t *testing.T) { // Make sure packet is delivered to first endpoint. buf[0] = 1 - ep.Inject(fakeNetNumber, buf.ToVectorisedView()) + ep.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) if fakeNet.packetCount[1] != 1 { t.Errorf("packetCount[1] = %d, want %d", fakeNet.packetCount[1], 1) } @@ -271,7 +285,9 @@ func TestNetworkReceive(t *testing.T) { // Make sure packet is delivered to second endpoint. buf[0] = 2 - ep.Inject(fakeNetNumber, buf.ToVectorisedView()) + ep.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) if fakeNet.packetCount[1] != 1 { t.Errorf("packetCount[1] = %d, want %d", fakeNet.packetCount[1], 1) } @@ -280,7 +296,9 @@ func TestNetworkReceive(t *testing.T) { } // Make sure packet is not delivered if protocol number is wrong. - ep.Inject(fakeNetNumber-1, buf.ToVectorisedView()) + ep.InjectInbound(fakeNetNumber-1, tcpip.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) if fakeNet.packetCount[1] != 1 { t.Errorf("packetCount[1] = %d, want %d", fakeNet.packetCount[1], 1) } @@ -290,7 +308,9 @@ func TestNetworkReceive(t *testing.T) { // Make sure packet that is too small is dropped. buf.CapLength(2) - ep.Inject(fakeNetNumber, buf.ToVectorisedView()) + ep.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) if fakeNet.packetCount[1] != 1 { t.Errorf("packetCount[1] = %d, want %d", fakeNet.packetCount[1], 1) } @@ -310,7 +330,7 @@ func sendTo(s *stack.Stack, addr tcpip.Address, payload buffer.View) *tcpip.Erro func send(r stack.Route, payload buffer.View) *tcpip.Error { hdr := buffer.NewPrependable(int(r.MaxHeaderLength())) - return r.WritePacket(nil /* gso */, hdr, payload.ToVectorisedView(), fakeTransNumber, 123) + return r.WritePacket(nil /* gso */, hdr, payload.ToVectorisedView(), stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}) } func testSendTo(t *testing.T, s *stack.Stack, addr tcpip.Address, ep *channel.Endpoint, payload buffer.View) { @@ -365,7 +385,9 @@ func testFailingRecv(t *testing.T, fakeNet *fakeNetworkProtocol, localAddrByte b func testRecvInternal(t *testing.T, fakeNet *fakeNetworkProtocol, localAddrByte byte, ep *channel.Endpoint, buf buffer.View, want int) { t.Helper() - ep.Inject(fakeNetNumber, buf.ToVectorisedView()) + ep.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) if got := fakeNet.PacketCount(localAddrByte); got != want { t.Errorf("receive packet count: got = %d, want %d", got, want) } @@ -660,11 +682,11 @@ func TestAddressRemovalWithRouteHeld(t *testing.T) { } } -func verifyAddress(t *testing.T, s *stack.Stack, nicid tcpip.NICID, addr tcpip.Address) { +func verifyAddress(t *testing.T, s *stack.Stack, nicID tcpip.NICID, addr tcpip.Address) { t.Helper() - info, ok := s.NICInfo()[nicid] + info, ok := s.NICInfo()[nicID] if !ok { - t.Fatalf("NICInfo() failed to find nicid=%d", nicid) + t.Fatalf("NICInfo() failed to find nicID=%d", nicID) } if len(addr) == 0 { // No address given, verify that there is no address assigned to the NIC. @@ -697,7 +719,7 @@ func TestEndpointExpiration(t *testing.T) { localAddrByte byte = 0x01 remoteAddr tcpip.Address = "\x03" noAddr tcpip.Address = "" - nicid tcpip.NICID = 1 + nicID tcpip.NICID = 1 ) localAddr := tcpip.Address([]byte{localAddrByte}) @@ -709,7 +731,7 @@ func TestEndpointExpiration(t *testing.T) { }) ep := channel.New(10, defaultMTU, "") - if err := s.CreateNIC(nicid, ep); err != nil { + if err := s.CreateNIC(nicID, ep); err != nil { t.Fatal("CreateNIC failed:", err) } @@ -726,13 +748,13 @@ func TestEndpointExpiration(t *testing.T) { buf[0] = localAddrByte if promiscuous { - if err := s.SetPromiscuousMode(nicid, true); err != nil { + if err := s.SetPromiscuousMode(nicID, true); err != nil { t.Fatal("SetPromiscuousMode failed:", err) } } if spoofing { - if err := s.SetSpoofing(nicid, true); err != nil { + if err := s.SetSpoofing(nicID, true); err != nil { t.Fatal("SetSpoofing failed:", err) } } @@ -740,7 +762,7 @@ func TestEndpointExpiration(t *testing.T) { // 1. No Address yet, send should only work for spoofing, receive for // promiscuous mode. //----------------------- - verifyAddress(t, s, nicid, noAddr) + verifyAddress(t, s, nicID, noAddr) if promiscuous { testRecv(t, fakeNet, localAddrByte, ep, buf) } else { @@ -755,20 +777,20 @@ func TestEndpointExpiration(t *testing.T) { // 2. Add Address, everything should work. //----------------------- - if err := s.AddAddress(nicid, fakeNetNumber, localAddr); err != nil { + if err := s.AddAddress(nicID, fakeNetNumber, localAddr); err != nil { t.Fatal("AddAddress failed:", err) } - verifyAddress(t, s, nicid, localAddr) + verifyAddress(t, s, nicID, localAddr) testRecv(t, fakeNet, localAddrByte, ep, buf) testSendTo(t, s, remoteAddr, ep, nil) // 3. Remove the address, send should only work for spoofing, receive // for promiscuous mode. //----------------------- - if err := s.RemoveAddress(nicid, localAddr); err != nil { + if err := s.RemoveAddress(nicID, localAddr); err != nil { t.Fatal("RemoveAddress failed:", err) } - verifyAddress(t, s, nicid, noAddr) + verifyAddress(t, s, nicID, noAddr) if promiscuous { testRecv(t, fakeNet, localAddrByte, ep, buf) } else { @@ -783,10 +805,10 @@ func TestEndpointExpiration(t *testing.T) { // 4. Add Address back, everything should work again. //----------------------- - if err := s.AddAddress(nicid, fakeNetNumber, localAddr); err != nil { + if err := s.AddAddress(nicID, fakeNetNumber, localAddr); err != nil { t.Fatal("AddAddress failed:", err) } - verifyAddress(t, s, nicid, localAddr) + verifyAddress(t, s, nicID, localAddr) testRecv(t, fakeNet, localAddrByte, ep, buf) testSendTo(t, s, remoteAddr, ep, nil) @@ -804,10 +826,10 @@ func TestEndpointExpiration(t *testing.T) { // 6. Remove the address. Send should only work for spoofing, receive // for promiscuous mode. //----------------------- - if err := s.RemoveAddress(nicid, localAddr); err != nil { + if err := s.RemoveAddress(nicID, localAddr); err != nil { t.Fatal("RemoveAddress failed:", err) } - verifyAddress(t, s, nicid, noAddr) + verifyAddress(t, s, nicID, noAddr) if promiscuous { testRecv(t, fakeNet, localAddrByte, ep, buf) } else { @@ -823,10 +845,10 @@ func TestEndpointExpiration(t *testing.T) { // 7. Add Address back, everything should work again. //----------------------- - if err := s.AddAddress(nicid, fakeNetNumber, localAddr); err != nil { + if err := s.AddAddress(nicID, fakeNetNumber, localAddr); err != nil { t.Fatal("AddAddress failed:", err) } - verifyAddress(t, s, nicid, localAddr) + verifyAddress(t, s, nicID, localAddr) testRecv(t, fakeNet, localAddrByte, ep, buf) testSendTo(t, s, remoteAddr, ep, nil) testSend(t, r, ep, nil) @@ -834,17 +856,17 @@ func TestEndpointExpiration(t *testing.T) { // 8. Remove the route, sendTo/recv should still work. //----------------------- r.Release() - verifyAddress(t, s, nicid, localAddr) + verifyAddress(t, s, nicID, localAddr) testRecv(t, fakeNet, localAddrByte, ep, buf) testSendTo(t, s, remoteAddr, ep, nil) // 9. Remove the address. Send should only work for spoofing, receive // for promiscuous mode. //----------------------- - if err := s.RemoveAddress(nicid, localAddr); err != nil { + if err := s.RemoveAddress(nicID, localAddr); err != nil { t.Fatal("RemoveAddress failed:", err) } - verifyAddress(t, s, nicid, noAddr) + verifyAddress(t, s, nicID, noAddr) if promiscuous { testRecv(t, fakeNet, localAddrByte, ep, buf) } else { @@ -952,10 +974,10 @@ func TestSpoofingWithAddress(t *testing.T) { t.Fatal("FindRoute failed:", err) } if r.LocalAddress != nonExistentLocalAddr { - t.Errorf("Route has wrong local address: got %s, want %s", r.LocalAddress, nonExistentLocalAddr) + t.Errorf("got Route.LocalAddress = %s, want = %s", r.LocalAddress, nonExistentLocalAddr) } if r.RemoteAddress != dstAddr { - t.Errorf("Route has wrong remote address: got %s, want %s", r.RemoteAddress, dstAddr) + t.Errorf("got Route.RemoteAddress = %s, want = %s", r.RemoteAddress, dstAddr) } // Sending a packet works. testSendTo(t, s, dstAddr, ep, nil) @@ -967,10 +989,10 @@ func TestSpoofingWithAddress(t *testing.T) { t.Fatal("FindRoute failed:", err) } if r.LocalAddress != localAddr { - t.Errorf("Route has wrong local address: got %s, want %s", r.LocalAddress, nonExistentLocalAddr) + t.Errorf("got Route.LocalAddress = %s, want = %s", r.LocalAddress, nonExistentLocalAddr) } if r.RemoteAddress != dstAddr { - t.Errorf("Route has wrong remote address: got %s, want %s", r.RemoteAddress, dstAddr) + t.Errorf("got Route.RemoteAddress = %s, want = %s", r.RemoteAddress, dstAddr) } // Sending a packet using the route works. testSend(t, r, ep, nil) @@ -1016,17 +1038,33 @@ func TestSpoofingNoAddress(t *testing.T) { t.Fatal("FindRoute failed:", err) } if r.LocalAddress != nonExistentLocalAddr { - t.Errorf("Route has wrong local address: got %s, want %s", r.LocalAddress, nonExistentLocalAddr) + t.Errorf("got Route.LocalAddress = %s, want = %s", r.LocalAddress, nonExistentLocalAddr) } if r.RemoteAddress != dstAddr { - t.Errorf("Route has wrong remote address: got %s, want %s", r.RemoteAddress, dstAddr) + t.Errorf("got Route.RemoteAddress = %s, want = %s", r.RemoteAddress, dstAddr) } // Sending a packet works. // FIXME(b/139841518):Spoofing doesn't work if there is no primary address. // testSendTo(t, s, remoteAddr, ep, nil) } -func TestBroadcastNeedsNoRoute(t *testing.T) { +func verifyRoute(gotRoute, wantRoute stack.Route) error { + if gotRoute.LocalAddress != wantRoute.LocalAddress { + return fmt.Errorf("bad local address: got %s, want = %s", gotRoute.LocalAddress, wantRoute.LocalAddress) + } + if gotRoute.RemoteAddress != wantRoute.RemoteAddress { + return fmt.Errorf("bad remote address: got %s, want = %s", gotRoute.RemoteAddress, wantRoute.RemoteAddress) + } + if gotRoute.RemoteLinkAddress != wantRoute.RemoteLinkAddress { + return fmt.Errorf("bad remote link address: got %s, want = %s", gotRoute.RemoteLinkAddress, wantRoute.RemoteLinkAddress) + } + if gotRoute.NextHop != wantRoute.NextHop { + return fmt.Errorf("bad next-hop address: got %s, want = %s", gotRoute.NextHop, wantRoute.NextHop) + } + return nil +} + +func TestOutgoingBroadcastWithEmptyRouteTable(t *testing.T) { s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, }) @@ -1039,28 +1077,99 @@ func TestBroadcastNeedsNoRoute(t *testing.T) { // If there is no endpoint, it won't work. if _, err := s.FindRoute(1, header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, false /* multicastLoop */); err != tcpip.ErrNetworkUnreachable { - t.Fatalf("got FindRoute(1, %v, %v, %v) = %v, want = %s", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err, tcpip.ErrNetworkUnreachable) + t.Fatalf("got FindRoute(1, %s, %s, %d) = %s, want = %s", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err, tcpip.ErrNetworkUnreachable) } - if err := s.AddAddress(1, fakeNetNumber, header.IPv4Any); err != nil { - t.Fatalf("AddAddress(%v, %v) failed: %s", fakeNetNumber, header.IPv4Any, err) + protoAddr := tcpip.ProtocolAddress{Protocol: fakeNetNumber, AddressWithPrefix: tcpip.AddressWithPrefix{header.IPv4Any, 0}} + if err := s.AddProtocolAddress(1, protoAddr); err != nil { + t.Fatalf("AddProtocolAddress(1, %s) failed: %s", protoAddr, err) } r, err := s.FindRoute(1, header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, false /* multicastLoop */) if err != nil { - t.Fatalf("FindRoute(1, %v, %v, %v) failed: %v", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err) + t.Fatalf("FindRoute(1, %s, %s, %d) failed: %s", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err) + } + if err := verifyRoute(r, stack.Route{LocalAddress: header.IPv4Any, RemoteAddress: header.IPv4Broadcast}); err != nil { + t.Errorf("FindRoute(1, %s, %s, %d) returned unexpected Route: %s)", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err) + } + + // If the NIC doesn't exist, it won't work. + if _, err := s.FindRoute(2, header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, false /* multicastLoop */); err != tcpip.ErrNetworkUnreachable { + t.Fatalf("got FindRoute(2, %s, %s, %d) = %s want = %s", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err, tcpip.ErrNetworkUnreachable) + } +} + +func TestOutgoingBroadcastWithRouteTable(t *testing.T) { + defaultAddr := tcpip.AddressWithPrefix{header.IPv4Any, 0} + // Local subnet on NIC1: 192.168.1.58/24, gateway 192.168.1.1. + nic1Addr := tcpip.AddressWithPrefix{"\xc0\xa8\x01\x3a", 24} + nic1Gateway := tcpip.Address("\xc0\xa8\x01\x01") + // Local subnet on NIC2: 10.10.10.5/24, gateway 10.10.10.1. + nic2Addr := tcpip.AddressWithPrefix{"\x0a\x0a\x0a\x05", 24} + nic2Gateway := tcpip.Address("\x0a\x0a\x0a\x01") + + // Create a new stack with two NICs. + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) + ep := channel.New(10, defaultMTU, "") + if err := s.CreateNIC(1, ep); err != nil { + t.Fatalf("CreateNIC failed: %s", err) + } + if err := s.CreateNIC(2, ep); err != nil { + t.Fatalf("CreateNIC failed: %s", err) + } + nic1ProtoAddr := tcpip.ProtocolAddress{fakeNetNumber, nic1Addr} + if err := s.AddProtocolAddress(1, nic1ProtoAddr); err != nil { + t.Fatalf("AddProtocolAddress(1, %s) failed: %s", nic1ProtoAddr, err) } - if r.LocalAddress != header.IPv4Any { - t.Errorf("Bad local address: got %v, want = %v", r.LocalAddress, header.IPv4Any) + nic2ProtoAddr := tcpip.ProtocolAddress{fakeNetNumber, nic2Addr} + if err := s.AddProtocolAddress(2, nic2ProtoAddr); err != nil { + t.Fatalf("AddAddress(2, %s) failed: %s", nic2ProtoAddr, err) } - if r.RemoteAddress != header.IPv4Broadcast { - t.Errorf("Bad remote address: got %v, want = %v", r.RemoteAddress, header.IPv4Broadcast) + // Set the initial route table. + rt := []tcpip.Route{ + {Destination: nic1Addr.Subnet(), NIC: 1}, + {Destination: nic2Addr.Subnet(), NIC: 2}, + {Destination: defaultAddr.Subnet(), Gateway: nic2Gateway, NIC: 2}, + {Destination: defaultAddr.Subnet(), Gateway: nic1Gateway, NIC: 1}, } + s.SetRouteTable(rt) - // If the NIC doesn't exist, it won't work. - if _, err := s.FindRoute(2, header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, false /* multicastLoop */); err != tcpip.ErrNetworkUnreachable { - t.Fatalf("got FindRoute(2, %v, %v, %v) = %v want = %v", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err, tcpip.ErrNetworkUnreachable) + // When an interface is given, the route for a broadcast goes through it. + r, err := s.FindRoute(1, nic1Addr.Address, header.IPv4Broadcast, fakeNetNumber, false /* multicastLoop */) + if err != nil { + t.Fatalf("FindRoute(1, %s, %s, %d) failed: %s", nic1Addr.Address, header.IPv4Broadcast, fakeNetNumber, err) + } + if err := verifyRoute(r, stack.Route{LocalAddress: nic1Addr.Address, RemoteAddress: header.IPv4Broadcast}); err != nil { + t.Errorf("FindRoute(1, %s, %s, %d) returned unexpected Route: %s)", nic1Addr.Address, header.IPv4Broadcast, fakeNetNumber, err) + } + + // When an interface is not given, it consults the route table. + // 1. Case: Using the default route. + r, err = s.FindRoute(0, "", header.IPv4Broadcast, fakeNetNumber, false /* multicastLoop */) + if err != nil { + t.Fatalf("FindRoute(0, \"\", %s, %d) failed: %s", header.IPv4Broadcast, fakeNetNumber, err) + } + if err := verifyRoute(r, stack.Route{LocalAddress: nic2Addr.Address, RemoteAddress: header.IPv4Broadcast}); err != nil { + t.Errorf("FindRoute(0, \"\", %s, %d) returned unexpected Route: %s)", header.IPv4Broadcast, fakeNetNumber, err) + } + + // 2. Case: Having an explicit route for broadcast will select that one. + rt = append( + []tcpip.Route{ + {Destination: tcpip.AddressWithPrefix{header.IPv4Broadcast, 8 * header.IPv4AddressSize}.Subnet(), NIC: 1}, + }, + rt..., + ) + s.SetRouteTable(rt) + r, err = s.FindRoute(0, "", header.IPv4Broadcast, fakeNetNumber, false /* multicastLoop */) + if err != nil { + t.Fatalf("FindRoute(0, \"\", %s, %d) failed: %s", header.IPv4Broadcast, fakeNetNumber, err) + } + if err := verifyRoute(r, stack.Route{LocalAddress: nic1Addr.Address, RemoteAddress: header.IPv4Broadcast}); err != nil { + t.Errorf("FindRoute(0, \"\", %s, %d) returned unexpected Route: %s)", header.IPv4Broadcast, fakeNetNumber, err) } } @@ -1550,12 +1659,12 @@ func verifyAddresses(t *testing.T, expectedAddresses, gotAddresses []tcpip.Proto } func TestAddAddress(t *testing.T) { - const nicid = 1 + const nicID = 1 s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, }) ep := channel.New(10, defaultMTU, "") - if err := s.CreateNIC(nicid, ep); err != nil { + if err := s.CreateNIC(nicID, ep); err != nil { t.Fatal("CreateNIC failed:", err) } @@ -1563,7 +1672,7 @@ func TestAddAddress(t *testing.T) { expectedAddresses := make([]tcpip.ProtocolAddress, 0, 2) for _, addrLen := range []int{4, 16} { address := addrGen.next(addrLen) - if err := s.AddAddress(nicid, fakeNetNumber, address); err != nil { + if err := s.AddAddress(nicID, fakeNetNumber, address); err != nil { t.Fatalf("AddAddress(address=%s) failed: %s", address, err) } expectedAddresses = append(expectedAddresses, tcpip.ProtocolAddress{ @@ -1572,17 +1681,17 @@ func TestAddAddress(t *testing.T) { }) } - gotAddresses := s.AllAddresses()[nicid] + gotAddresses := s.AllAddresses()[nicID] verifyAddresses(t, expectedAddresses, gotAddresses) } func TestAddProtocolAddress(t *testing.T) { - const nicid = 1 + const nicID = 1 s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, }) ep := channel.New(10, defaultMTU, "") - if err := s.CreateNIC(nicid, ep); err != nil { + if err := s.CreateNIC(nicID, ep); err != nil { t.Fatal("CreateNIC failed:", err) } @@ -1599,24 +1708,24 @@ func TestAddProtocolAddress(t *testing.T) { PrefixLen: prefixLen, }, } - if err := s.AddProtocolAddress(nicid, protocolAddress); err != nil { + if err := s.AddProtocolAddress(nicID, protocolAddress); err != nil { t.Errorf("AddProtocolAddress(%+v) failed: %s", protocolAddress, err) } expectedAddresses = append(expectedAddresses, protocolAddress) } } - gotAddresses := s.AllAddresses()[nicid] + gotAddresses := s.AllAddresses()[nicID] verifyAddresses(t, expectedAddresses, gotAddresses) } func TestAddAddressWithOptions(t *testing.T) { - const nicid = 1 + const nicID = 1 s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, }) ep := channel.New(10, defaultMTU, "") - if err := s.CreateNIC(nicid, ep); err != nil { + if err := s.CreateNIC(nicID, ep); err != nil { t.Fatal("CreateNIC failed:", err) } @@ -1627,7 +1736,7 @@ func TestAddAddressWithOptions(t *testing.T) { for _, addrLen := range addrLenRange { for _, behavior := range behaviorRange { address := addrGen.next(addrLen) - if err := s.AddAddressWithOptions(nicid, fakeNetNumber, address, behavior); err != nil { + if err := s.AddAddressWithOptions(nicID, fakeNetNumber, address, behavior); err != nil { t.Fatalf("AddAddressWithOptions(address=%s, behavior=%d) failed: %s", address, behavior, err) } expectedAddresses = append(expectedAddresses, tcpip.ProtocolAddress{ @@ -1637,17 +1746,17 @@ func TestAddAddressWithOptions(t *testing.T) { } } - gotAddresses := s.AllAddresses()[nicid] + gotAddresses := s.AllAddresses()[nicID] verifyAddresses(t, expectedAddresses, gotAddresses) } func TestAddProtocolAddressWithOptions(t *testing.T) { - const nicid = 1 + const nicID = 1 s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, }) ep := channel.New(10, defaultMTU, "") - if err := s.CreateNIC(nicid, ep); err != nil { + if err := s.CreateNIC(nicID, ep); err != nil { t.Fatal("CreateNIC failed:", err) } @@ -1666,7 +1775,7 @@ func TestAddProtocolAddressWithOptions(t *testing.T) { PrefixLen: prefixLen, }, } - if err := s.AddProtocolAddressWithOptions(nicid, protocolAddress, behavior); err != nil { + if err := s.AddProtocolAddressWithOptions(nicID, protocolAddress, behavior); err != nil { t.Fatalf("AddProtocolAddressWithOptions(%+v, %d) failed: %s", protocolAddress, behavior, err) } expectedAddresses = append(expectedAddresses, protocolAddress) @@ -1674,7 +1783,7 @@ func TestAddProtocolAddressWithOptions(t *testing.T) { } } - gotAddresses := s.AllAddresses()[nicid] + gotAddresses := s.AllAddresses()[nicID] verifyAddresses(t, expectedAddresses, gotAddresses) } @@ -1700,7 +1809,9 @@ func TestNICStats(t *testing.T) { // Send a packet to address 1. buf := buffer.NewView(30) - ep1.Inject(fakeNetNumber, buf.ToVectorisedView()) + ep1.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) if got, want := s.NICInfo()[1].Stats.Rx.Packets.Value(), uint64(1); got != want { t.Errorf("got Rx.Packets.Value() = %d, want = %d", got, want) } @@ -1760,7 +1871,9 @@ func TestNICForwarding(t *testing.T) { // Send a packet to address 3. buf := buffer.NewView(30) buf[0] = 3 - ep1.Inject(fakeNetNumber, buf.ToVectorisedView()) + ep1.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) select { case <-ep2.C: @@ -1777,3 +1890,297 @@ func TestNICForwarding(t *testing.T) { t.Errorf("got Tx.Bytes.Value() = %d, want = %d", got, want) } } + +// TestNICAutoGenAddr tests the auto-generation of IPv6 link-local addresses +// (or lack there-of if disabled (default)). Note, DAD will be disabled in +// these tests. +func TestNICAutoGenAddr(t *testing.T) { + tests := []struct { + name string + autoGen bool + linkAddr tcpip.LinkAddress + shouldGen bool + }{ + { + "Disabled", + false, + linkAddr1, + false, + }, + { + "Enabled", + true, + linkAddr1, + true, + }, + { + "Nil MAC", + true, + tcpip.LinkAddress([]byte(nil)), + false, + }, + { + "Empty MAC", + true, + tcpip.LinkAddress(""), + false, + }, + { + "Invalid MAC", + true, + tcpip.LinkAddress("\x01\x02\x03"), + false, + }, + { + "Multicast MAC", + true, + tcpip.LinkAddress("\x01\x02\x03\x04\x05\x06"), + false, + }, + { + "Unspecified MAC", + true, + tcpip.LinkAddress("\x00\x00\x00\x00\x00\x00"), + false, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + opts := stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + } + + if test.autoGen { + // Only set opts.AutoGenIPv6LinkLocal when + // test.autoGen is true because + // opts.AutoGenIPv6LinkLocal should be false by + // default. + opts.AutoGenIPv6LinkLocal = true + } + + e := channel.New(10, 1280, test.linkAddr) + s := stack.New(opts) + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(_) = %s", err) + } + + addr, err := s.GetMainNICAddress(1, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("stack.GetMainNICAddress(_, _) err = %s", err) + } + + if test.shouldGen { + // Should have auto-generated an address and + // resolved immediately (DAD is disabled). + if want := (tcpip.AddressWithPrefix{Address: header.LinkLocalAddr(test.linkAddr), PrefixLen: header.IPv6LinkLocalPrefix.PrefixLen}); addr != want { + t.Fatalf("got stack.GetMainNICAddress(_, _) = %s, want = %s", addr, want) + } + } else { + // Should not have auto-generated an address. + if want := (tcpip.AddressWithPrefix{}); addr != want { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want) + } + } + }) + } +} + +// TestNICAutoGenAddrDoesDAD tests that the successful auto-generation of IPv6 +// link-local addresses will only be assigned after the DAD process resolves. +func TestNICAutoGenAddrDoesDAD(t *testing.T) { + ndpDisp := ndpDispatcher{ + dadC: make(chan ndpDADEvent), + } + ndpConfigs := stack.DefaultNDPConfigurations() + opts := stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: ndpConfigs, + AutoGenIPv6LinkLocal: true, + NDPDisp: &ndpDisp, + } + + e := channel.New(10, 1280, linkAddr1) + s := stack.New(opts) + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(_) = %s", err) + } + + // Address should not be considered bound to the + // NIC yet (DAD ongoing). + addr, err := s.GetMainNICAddress(1, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (_, %v), want = (_, nil)", err) + } + if want := (tcpip.AddressWithPrefix{}); addr != want { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want) + } + + linkLocalAddr := header.LinkLocalAddr(linkAddr1) + + // Wait for DAD to resolve. + select { + case <-time.After(time.Duration(ndpConfigs.DupAddrDetectTransmits)*ndpConfigs.RetransmitTimer + time.Second): + // We should get a resolution event after 1s (default time to + // resolve as per default NDP configurations). Waiting for that + // resolution time + an extra 1s without a resolution event + // means something is wrong. + t.Fatal("timed out waiting for DAD resolution") + case e := <-ndpDisp.dadC: + if e.err != nil { + t.Fatal("got DAD error: ", e.err) + } + if e.nicID != 1 { + t.Fatalf("got DAD event w/ nicID = %d, want = 1", e.nicID) + } + if e.addr != linkLocalAddr { + t.Fatalf("got DAD event w/ addr = %s, want = %s", addr, linkLocalAddr) + } + if !e.resolved { + t.Fatal("got DAD event w/ resolved = false, want = true") + } + } + addr, err = s.GetMainNICAddress(1, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("stack.GetMainNICAddress(_, _) err = %s", err) + } + if want := (tcpip.AddressWithPrefix{Address: linkLocalAddr, PrefixLen: header.IPv6LinkLocalPrefix.PrefixLen}); addr != want { + t.Fatalf("got stack.GetMainNICAddress(_, _) = %s, want = %s", addr, want) + } +} + +// TestNewPEB tests that a new PrimaryEndpointBehavior value (peb) is respected +// when an address's kind gets "promoted" to permanent from permanentExpired. +func TestNewPEBOnPromotionToPermanent(t *testing.T) { + pebs := []stack.PrimaryEndpointBehavior{ + stack.NeverPrimaryEndpoint, + stack.CanBePrimaryEndpoint, + stack.FirstPrimaryEndpoint, + } + + for _, pi := range pebs { + for _, ps := range pebs { + t.Run(fmt.Sprintf("%d-to-%d", pi, ps), func(t *testing.T) { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) + ep1 := channel.New(10, defaultMTU, "") + if err := s.CreateNIC(1, ep1); err != nil { + t.Fatal("CreateNIC failed:", err) + } + + // Add a permanent address with initial + // PrimaryEndpointBehavior (peb), pi. If pi is + // NeverPrimaryEndpoint, the address should not + // be returned by a call to GetMainNICAddress; + // else, it should. + if err := s.AddAddressWithOptions(1, fakeNetNumber, "\x01", pi); err != nil { + t.Fatal("AddAddressWithOptions failed:", err) + } + addr, err := s.GetMainNICAddress(1, fakeNetNumber) + if err != nil { + t.Fatal("s.GetMainNICAddress failed:", err) + } + if pi == stack.NeverPrimaryEndpoint { + if want := (tcpip.AddressWithPrefix{}); addr != want { + t.Fatalf("got GetMainNICAddress = %s, want = %s", addr, want) + + } + } else if addr.Address != "\x01" { + t.Fatalf("got GetMainNICAddress = %s, want = 1", addr.Address) + } + + { + subnet, err := tcpip.NewSubnet("\x00", "\x00") + if err != nil { + t.Fatalf("NewSubnet failed:", err) + } + s.SetRouteTable([]tcpip.Route{{Destination: subnet, Gateway: "\x00", NIC: 1}}) + } + + // Take a route through the address so its ref + // count gets incremented and does not actually + // get deleted when RemoveAddress is called + // below. This is because we want to test that a + // new peb is respected when an address gets + // "promoted" to permanent from a + // permanentExpired kind. + r, err := s.FindRoute(1, "\x01", "\x02", fakeNetNumber, false) + if err != nil { + t.Fatal("FindRoute failed:", err) + } + defer r.Release() + if err := s.RemoveAddress(1, "\x01"); err != nil { + t.Fatalf("RemoveAddress failed:", err) + } + + // + // At this point, the address should still be + // known by the NIC, but have its + // kind = permanentExpired. + // + + // Add some other address with peb set to + // FirstPrimaryEndpoint. + if err := s.AddAddressWithOptions(1, fakeNetNumber, "\x03", stack.FirstPrimaryEndpoint); err != nil { + t.Fatal("AddAddressWithOptions failed:", err) + + } + + // Add back the address we removed earlier and + // make sure the new peb was respected. + // (The address should just be promoted now). + if err := s.AddAddressWithOptions(1, fakeNetNumber, "\x01", ps); err != nil { + t.Fatal("AddAddressWithOptions failed:", err) + } + var primaryAddrs []tcpip.Address + for _, pa := range s.NICInfo()[1].ProtocolAddresses { + primaryAddrs = append(primaryAddrs, pa.AddressWithPrefix.Address) + } + var expectedList []tcpip.Address + switch ps { + case stack.FirstPrimaryEndpoint: + expectedList = []tcpip.Address{ + "\x01", + "\x03", + } + case stack.CanBePrimaryEndpoint: + expectedList = []tcpip.Address{ + "\x03", + "\x01", + } + case stack.NeverPrimaryEndpoint: + expectedList = []tcpip.Address{ + "\x03", + } + } + if !cmp.Equal(primaryAddrs, expectedList) { + t.Fatalf("got NIC's primary addresses = %v, want = %v", primaryAddrs, expectedList) + } + + // Once we remove the other address, if the new + // peb, ps, was NeverPrimaryEndpoint, no address + // should be returned by a call to + // GetMainNICAddress; else, our original address + // should be returned. + if err := s.RemoveAddress(1, "\x03"); err != nil { + t.Fatalf("RemoveAddress failed:", err) + } + addr, err = s.GetMainNICAddress(1, fakeNetNumber) + if err != nil { + t.Fatal("s.GetMainNICAddress failed:", err) + } + if ps == stack.NeverPrimaryEndpoint { + if want := (tcpip.AddressWithPrefix{}); addr != want { + t.Fatalf("got GetMainNICAddress = %s, want = %s", addr, want) + + } + } else { + if addr.Address != "\x01" { + t.Fatalf("got GetMainNICAddress = %s, want = 1", addr.Address) + } + } + }) + } + } +} diff --git a/pkg/tcpip/stack/transport_demuxer.go b/pkg/tcpip/stack/transport_demuxer.go index cf8a6d129..cb805522b 100644 --- a/pkg/tcpip/stack/transport_demuxer.go +++ b/pkg/tcpip/stack/transport_demuxer.go @@ -17,10 +17,10 @@ package stack import ( "fmt" "math/rand" + "sort" "sync" "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/hash/jenkins" "gvisor.dev/gvisor/pkg/tcpip/header" ) @@ -35,7 +35,7 @@ type protocolIDs struct { type transportEndpoints struct { // mu protects all fields of the transportEndpoints. mu sync.RWMutex - endpoints map[TransportEndpointID]TransportEndpoint + endpoints map[TransportEndpointID]*endpointsByNic // rawEndpoints contains endpoints for raw sockets, which receive all // traffic of a given protocol regardless of port. rawEndpoints []RawTransportEndpoint @@ -43,19 +43,122 @@ type transportEndpoints struct { // unregisterEndpoint unregisters the endpoint with the given id such that it // won't receive any more packets. -func (eps *transportEndpoints) unregisterEndpoint(id TransportEndpointID, ep TransportEndpoint) { +func (eps *transportEndpoints) unregisterEndpoint(id TransportEndpointID, ep TransportEndpoint, bindToDevice tcpip.NICID) { eps.mu.Lock() defer eps.mu.Unlock() - e, ok := eps.endpoints[id] + epsByNic, ok := eps.endpoints[id] if !ok { return } - if multiPortEp, ok := e.(*multiPortEndpoint); ok { - if !multiPortEp.unregisterEndpoint(ep) { + if !epsByNic.unregisterEndpoint(bindToDevice, ep) { + return + } + delete(eps.endpoints, id) +} + +func (eps *transportEndpoints) transportEndpoints() []TransportEndpoint { + eps.mu.RLock() + defer eps.mu.RUnlock() + es := make([]TransportEndpoint, 0, len(eps.endpoints)) + for _, e := range eps.endpoints { + es = append(es, e.transportEndpoints()...) + } + return es +} + +type endpointsByNic struct { + mu sync.RWMutex + endpoints map[tcpip.NICID]*multiPortEndpoint + // seed is a random secret for a jenkins hash. + seed uint32 +} + +func (epsByNic *endpointsByNic) transportEndpoints() []TransportEndpoint { + epsByNic.mu.RLock() + defer epsByNic.mu.RUnlock() + var eps []TransportEndpoint + for _, ep := range epsByNic.endpoints { + eps = append(eps, ep.transportEndpoints()...) + } + return eps +} + +// HandlePacket is called by the stack when new packets arrive to this transport +// endpoint. +func (epsByNic *endpointsByNic) handlePacket(r *Route, id TransportEndpointID, pkt tcpip.PacketBuffer) { + epsByNic.mu.RLock() + + mpep, ok := epsByNic.endpoints[r.ref.nic.ID()] + if !ok { + if mpep, ok = epsByNic.endpoints[0]; !ok { + epsByNic.mu.RUnlock() // Don't use defer for performance reasons. return } } - delete(eps.endpoints, id) + + // If this is a broadcast or multicast datagram, deliver the datagram to all + // endpoints bound to the right device. + if isMulticastOrBroadcast(id.LocalAddress) { + mpep.handlePacketAll(r, id, pkt) + epsByNic.mu.RUnlock() // Don't use defer for performance reasons. + return + } + // multiPortEndpoints are guaranteed to have at least one element. + selectEndpoint(id, mpep, epsByNic.seed).HandlePacket(r, id, pkt) + epsByNic.mu.RUnlock() // Don't use defer for performance reasons. +} + +// HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket. +func (epsByNic *endpointsByNic) handleControlPacket(n *NIC, id TransportEndpointID, typ ControlType, extra uint32, pkt tcpip.PacketBuffer) { + epsByNic.mu.RLock() + defer epsByNic.mu.RUnlock() + + mpep, ok := epsByNic.endpoints[n.ID()] + if !ok { + mpep, ok = epsByNic.endpoints[0] + } + if !ok { + return + } + + // TODO(eyalsoha): Why don't we look at id to see if this packet needs to + // broadcast like we are doing with handlePacket above? + + // multiPortEndpoints are guaranteed to have at least one element. + selectEndpoint(id, mpep, epsByNic.seed).HandleControlPacket(id, typ, extra, pkt) +} + +// registerEndpoint returns true if it succeeds. It fails and returns +// false if ep already has an element with the same key. +func (epsByNic *endpointsByNic) registerEndpoint(t TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error { + epsByNic.mu.Lock() + defer epsByNic.mu.Unlock() + + if multiPortEp, ok := epsByNic.endpoints[bindToDevice]; ok { + // There was already a bind. + return multiPortEp.singleRegisterEndpoint(t, reusePort) + } + + // This is a new binding. + multiPortEp := &multiPortEndpoint{} + multiPortEp.endpointsMap = make(map[TransportEndpoint]int) + multiPortEp.reuse = reusePort + epsByNic.endpoints[bindToDevice] = multiPortEp + return multiPortEp.singleRegisterEndpoint(t, reusePort) +} + +// unregisterEndpoint returns true if endpointsByNic has to be unregistered. +func (epsByNic *endpointsByNic) unregisterEndpoint(bindToDevice tcpip.NICID, t TransportEndpoint) bool { + epsByNic.mu.Lock() + defer epsByNic.mu.Unlock() + multiPortEp, ok := epsByNic.endpoints[bindToDevice] + if !ok { + return false + } + if multiPortEp.unregisterEndpoint(t) { + delete(epsByNic.endpoints, bindToDevice) + } + return len(epsByNic.endpoints) == 0 } // transportDemuxer demultiplexes packets targeted at a transport endpoint @@ -75,7 +178,7 @@ func newTransportDemuxer(stack *Stack) *transportDemuxer { for netProto := range stack.networkProtocols { for proto := range stack.transportProtocols { d.protocol[protocolIDs{netProto, proto}] = &transportEndpoints{ - endpoints: make(map[TransportEndpointID]TransportEndpoint), + endpoints: make(map[TransportEndpointID]*endpointsByNic), } } } @@ -85,10 +188,10 @@ func newTransportDemuxer(stack *Stack) *transportDemuxer { // registerEndpoint registers the given endpoint with the dispatcher such that // packets that match the endpoint ID are delivered to it. -func (d *transportDemuxer) registerEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool) *tcpip.Error { +func (d *transportDemuxer) registerEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error { for i, n := range netProtos { - if err := d.singleRegisterEndpoint(n, protocol, id, ep, reusePort); err != nil { - d.unregisterEndpoint(netProtos[:i], protocol, id, ep) + if err := d.singleRegisterEndpoint(n, protocol, id, ep, reusePort, bindToDevice); err != nil { + d.unregisterEndpoint(netProtos[:i], protocol, id, ep, bindToDevice) return err } } @@ -97,13 +200,27 @@ func (d *transportDemuxer) registerEndpoint(netProtos []tcpip.NetworkProtocolNum } // multiPortEndpoint is a container for TransportEndpoints which are bound to -// the same pair of address and port. +// the same pair of address and port. endpointsArr always has at least one +// element. +// +// FIXME(gvisor.dev/issue/873): Restore this properly. Currently, we just save +// this to ensure that the underlying endpoints get saved/restored, but not not +// use the restored copy. +// +// +stateify savable type multiPortEndpoint struct { - mu sync.RWMutex + mu sync.RWMutex `state:"nosave"` endpointsArr []TransportEndpoint endpointsMap map[TransportEndpoint]int - // seed is a random secret for a jenkins hash. - seed uint32 + // reuse indicates if more than one endpoint is allowed. + reuse bool +} + +func (ep *multiPortEndpoint) transportEndpoints() []TransportEndpoint { + ep.mu.RLock() + eps := append([]TransportEndpoint(nil), ep.endpointsArr...) + ep.mu.RUnlock() + return eps } // reciprocalScale scales a value into range [0, n). @@ -117,9 +234,10 @@ func reciprocalScale(val, n uint32) uint32 { // selectEndpoint calculates a hash of destination and source addresses and // ports then uses it to select a socket. In this case, all packets from one // address will be sent to same endpoint. -func (ep *multiPortEndpoint) selectEndpoint(id TransportEndpointID) TransportEndpoint { - ep.mu.RLock() - defer ep.mu.RUnlock() +func selectEndpoint(id TransportEndpointID, mpep *multiPortEndpoint, seed uint32) TransportEndpoint { + if len(mpep.endpointsArr) == 1 { + return mpep.endpointsArr[0] + } payload := []byte{ byte(id.LocalPort), @@ -128,51 +246,77 @@ func (ep *multiPortEndpoint) selectEndpoint(id TransportEndpointID) TransportEnd byte(id.RemotePort >> 8), } - h := jenkins.Sum32(ep.seed) + h := jenkins.Sum32(seed) h.Write(payload) h.Write([]byte(id.LocalAddress)) h.Write([]byte(id.RemoteAddress)) hash := h.Sum32() - idx := reciprocalScale(hash, uint32(len(ep.endpointsArr))) - return ep.endpointsArr[idx] + idx := reciprocalScale(hash, uint32(len(mpep.endpointsArr))) + return mpep.endpointsArr[idx] } -// HandlePacket is called by the stack when new packets arrive to this transport -// endpoint. -func (ep *multiPortEndpoint) HandlePacket(r *Route, id TransportEndpointID, vv buffer.VectorisedView) { - // If this is a broadcast or multicast datagram, deliver the datagram to all - // endpoints managed by ep. - if id.LocalAddress == header.IPv4Broadcast || header.IsV4MulticastAddress(id.LocalAddress) || header.IsV6MulticastAddress(id.LocalAddress) { - for i, endpoint := range ep.endpointsArr { - // HandlePacket modifies vv, so each endpoint needs its own copy. - if i == len(ep.endpointsArr)-1 { - endpoint.HandlePacket(r, id, vv) - break - } - vvCopy := buffer.NewView(vv.Size()) - copy(vvCopy, vv.ToView()) - endpoint.HandlePacket(r, id, vvCopy.ToVectorisedView()) +func (ep *multiPortEndpoint) handlePacketAll(r *Route, id TransportEndpointID, pkt tcpip.PacketBuffer) { + ep.mu.RLock() + for i, endpoint := range ep.endpointsArr { + // HandlePacket takes ownership of pkt, so each endpoint needs + // its own copy except for the final one. + if i == len(ep.endpointsArr)-1 { + endpoint.HandlePacket(r, id, pkt) + break } - } else { - ep.selectEndpoint(id).HandlePacket(r, id, vv) + endpoint.HandlePacket(r, id, pkt.Clone()) } + ep.mu.RUnlock() // Don't use defer for performance reasons. } -// HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket. -func (ep *multiPortEndpoint) HandleControlPacket(id TransportEndpointID, typ ControlType, extra uint32, vv buffer.VectorisedView) { - ep.selectEndpoint(id).HandleControlPacket(id, typ, extra, vv) +// Close implements stack.TransportEndpoint.Close. +func (ep *multiPortEndpoint) Close() { + ep.mu.RLock() + eps := append([]TransportEndpoint(nil), ep.endpointsArr...) + ep.mu.RUnlock() + for _, e := range eps { + e.Close() + } } -func (ep *multiPortEndpoint) singleRegisterEndpoint(t TransportEndpoint) { +// Wait implements stack.TransportEndpoint.Wait. +func (ep *multiPortEndpoint) Wait() { + ep.mu.RLock() + eps := append([]TransportEndpoint(nil), ep.endpointsArr...) + ep.mu.RUnlock() + for _, e := range eps { + e.Wait() + } +} + +// singleRegisterEndpoint tries to add an endpoint to the multiPortEndpoint +// list. The list might be empty already. +func (ep *multiPortEndpoint) singleRegisterEndpoint(t TransportEndpoint, reusePort bool) *tcpip.Error { ep.mu.Lock() defer ep.mu.Unlock() - // A new endpoint is added into endpointsArr and its index there is - // saved in endpointsMap. This will allows to remove endpoint from - // the array fast. + if len(ep.endpointsArr) > 0 { + // If it was previously bound, we need to check if we can bind again. + if !ep.reuse || !reusePort { + return tcpip.ErrPortInUse + } + } + + // A new endpoint is added into endpointsArr and its index there is saved in + // endpointsMap. This will allow us to remove endpoint from the array fast. ep.endpointsMap[t] = len(ep.endpointsArr) ep.endpointsArr = append(ep.endpointsArr, t) + + // ep.endpointsArr is sorted by endpoint unique IDs, so that endpoints + // can be restored in the same order. + sort.Slice(ep.endpointsArr, func(i, j int) bool { + return ep.endpointsArr[i].UniqueID() < ep.endpointsArr[j].UniqueID() + }) + for i, e := range ep.endpointsArr { + ep.endpointsMap[e] = i + } + return nil } // unregisterEndpoint returns true if multiPortEndpoint has to be unregistered. @@ -197,53 +341,41 @@ func (ep *multiPortEndpoint) unregisterEndpoint(t TransportEndpoint) bool { return true } -func (d *transportDemuxer) singleRegisterEndpoint(netProto tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool) *tcpip.Error { +func (d *transportDemuxer) singleRegisterEndpoint(netProto tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error { if id.RemotePort != 0 { + // TODO(eyalsoha): Why? reusePort = false } eps, ok := d.protocol[protocolIDs{netProto, protocol}] if !ok { - return nil + return tcpip.ErrUnknownProtocol } eps.mu.Lock() defer eps.mu.Unlock() - var multiPortEp *multiPortEndpoint - if _, ok := eps.endpoints[id]; ok { - if !reusePort { - return tcpip.ErrPortInUse - } - multiPortEp, ok = eps.endpoints[id].(*multiPortEndpoint) - if !ok { - return tcpip.ErrPortInUse - } + if epsByNic, ok := eps.endpoints[id]; ok { + // There was already a binding. + return epsByNic.registerEndpoint(ep, reusePort, bindToDevice) } - if reusePort { - if multiPortEp == nil { - multiPortEp = &multiPortEndpoint{} - multiPortEp.endpointsMap = make(map[TransportEndpoint]int) - multiPortEp.seed = rand.Uint32() - eps.endpoints[id] = multiPortEp - } - - multiPortEp.singleRegisterEndpoint(ep) - - return nil + // This is a new binding. + epsByNic := &endpointsByNic{ + endpoints: make(map[tcpip.NICID]*multiPortEndpoint), + seed: rand.Uint32(), } - eps.endpoints[id] = ep + eps.endpoints[id] = epsByNic - return nil + return epsByNic.registerEndpoint(ep, reusePort, 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) { +func (d *transportDemuxer) unregisterEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, bindToDevice tcpip.NICID) { for _, n := range netProtos { if eps, ok := d.protocol[protocolIDs{n, protocol}]; ok { - eps.unregisterEndpoint(id, ep) + eps.unregisterEndpoint(id, ep, bindToDevice) } } } @@ -259,30 +391,21 @@ var loopbackSubnet = func() tcpip.Subnet { // deliverPacket attempts to find one or more matching transport endpoints, and // then, if matches are found, delivers the packet to them. Returns true if it // found one or more endpoints, false otherwise. -func (d *transportDemuxer) deliverPacket(r *Route, protocol tcpip.TransportProtocolNumber, netHeader buffer.View, vv buffer.VectorisedView, id TransportEndpointID) bool { +func (d *transportDemuxer) deliverPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt tcpip.PacketBuffer, id TransportEndpointID) bool { eps, ok := d.protocol[protocolIDs{r.NetProto, protocol}] if !ok { return false } - // If a sender bound to the Loopback interface sends a broadcast, - // that broadcast must not be delivered to the sender. - if loopbackSubnet.Contains(r.RemoteAddress) && r.LocalAddress == header.IPv4Broadcast && id.LocalPort == id.RemotePort { - return false - } - - // If the packet is a broadcast, then find all matching transport endpoints. - // Otherwise, try to find a single matching transport endpoint. - destEps := make([]TransportEndpoint, 0, 1) eps.mu.RLock() - if protocol == header.UDPProtocolNumber && id.LocalAddress == header.IPv4Broadcast { - for epID, endpoint := range eps.endpoints { - if epID.LocalPort == id.LocalPort { - destEps = append(destEps, endpoint) - } - } - } else if ep := d.findEndpointLocked(eps, vv, id); ep != nil { + // Determine which transport endpoint or endpoints to deliver this packet to. + // If the packet is a broadcast or multicast, then find all matching + // transport endpoints. + var destEps []*endpointsByNic + if protocol == header.UDPProtocolNumber && isMulticastOrBroadcast(id.LocalAddress) { + destEps = d.findAllEndpointsLocked(eps, id) + } else if ep := d.findEndpointLocked(eps, id); ep != nil { destEps = append(destEps, ep) } @@ -297,17 +420,19 @@ func (d *transportDemuxer) deliverPacket(r *Route, protocol tcpip.TransportProto return false } - // Deliver the packet. - for _, ep := range destEps { - ep.HandlePacket(r, id, vv) + // HandlePacket takes ownership of pkt, so each endpoint needs its own + // copy except for the final one. + for _, ep := range destEps[:len(destEps)-1] { + ep.handlePacket(r, id, pkt.Clone()) } + destEps[len(destEps)-1].handlePacket(r, id, pkt) return true } // deliverRawPacket attempts to deliver the given packet and returns whether it // was delivered successfully. -func (d *transportDemuxer) deliverRawPacket(r *Route, protocol tcpip.TransportProtocolNumber, netHeader buffer.View, vv buffer.VectorisedView) bool { +func (d *transportDemuxer) deliverRawPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt tcpip.PacketBuffer) bool { eps, ok := d.protocol[protocolIDs{r.NetProto, protocol}] if !ok { return false @@ -321,7 +446,7 @@ func (d *transportDemuxer) deliverRawPacket(r *Route, protocol tcpip.TransportPr for _, rawEP := range eps.rawEndpoints { // Each endpoint gets its own copy of the packet for the sake // of save/restore. - rawEP.HandlePacket(r, buffer.NewViewFromBytes(netHeader), vv.ToView().ToVectorisedView()) + rawEP.HandlePacket(r, pkt) foundRaw = true } eps.mu.RUnlock() @@ -331,7 +456,7 @@ func (d *transportDemuxer) deliverRawPacket(r *Route, protocol tcpip.TransportPr // deliverControlPacket attempts to deliver the given control packet. Returns // true if it found an endpoint, false otherwise. -func (d *transportDemuxer) deliverControlPacket(net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, vv buffer.VectorisedView, id TransportEndpointID) bool { +func (d *transportDemuxer) deliverControlPacket(n *NIC, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, pkt tcpip.PacketBuffer, id TransportEndpointID) bool { eps, ok := d.protocol[protocolIDs{net, trans}] if !ok { return false @@ -339,7 +464,7 @@ func (d *transportDemuxer) deliverControlPacket(net tcpip.NetworkProtocolNumber, // Try to find the endpoint. eps.mu.RLock() - ep := d.findEndpointLocked(eps, vv, id) + ep := d.findEndpointLocked(eps, id) eps.mu.RUnlock() // Fail if we didn't find one. @@ -348,15 +473,16 @@ func (d *transportDemuxer) deliverControlPacket(net tcpip.NetworkProtocolNumber, } // Deliver the packet. - ep.HandleControlPacket(id, typ, extra, vv) + ep.handleControlPacket(n, id, typ, extra, pkt) return true } -func (d *transportDemuxer) findEndpointLocked(eps *transportEndpoints, vv buffer.VectorisedView, id TransportEndpointID) TransportEndpoint { +func (d *transportDemuxer) findAllEndpointsLocked(eps *transportEndpoints, id TransportEndpointID) []*endpointsByNic { + var matchedEPs []*endpointsByNic // Try to find a match with the id as provided. if ep, ok := eps.endpoints[id]; ok { - return ep + matchedEPs = append(matchedEPs, ep) } // Try to find a match with the id minus the local address. @@ -364,7 +490,7 @@ func (d *transportDemuxer) findEndpointLocked(eps *transportEndpoints, vv buffer nid.LocalAddress = "" if ep, ok := eps.endpoints[nid]; ok { - return ep + matchedEPs = append(matchedEPs, ep) } // Try to find a match with the id minus the remote part. @@ -372,15 +498,54 @@ func (d *transportDemuxer) findEndpointLocked(eps *transportEndpoints, vv buffer nid.RemoteAddress = "" nid.RemotePort = 0 if ep, ok := eps.endpoints[nid]; ok { - return ep + matchedEPs = append(matchedEPs, ep) } // Try to find a match with only the local port. nid.LocalAddress = "" if ep, ok := eps.endpoints[nid]; ok { - return ep + matchedEPs = append(matchedEPs, ep) + } + return matchedEPs +} + +// findTransportEndpoint find a single endpoint that most closely matches the provided id. +func (d *transportDemuxer) findTransportEndpoint(netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, id TransportEndpointID, r *Route) TransportEndpoint { + eps, ok := d.protocol[protocolIDs{netProto, transProto}] + if !ok { + return nil + } + // Try to find the endpoint. + eps.mu.RLock() + epsByNic := d.findEndpointLocked(eps, id) + // Fail if we didn't find one. + if epsByNic == nil { + eps.mu.RUnlock() + return nil + } + + epsByNic.mu.RLock() + eps.mu.RUnlock() + + mpep, ok := epsByNic.endpoints[r.ref.nic.ID()] + if !ok { + if mpep, ok = epsByNic.endpoints[0]; !ok { + epsByNic.mu.RUnlock() // Don't use defer for performance reasons. + return nil + } } + ep := selectEndpoint(id, mpep, epsByNic.seed) + epsByNic.mu.RUnlock() + return ep +} + +// findEndpointLocked returns the endpoint that most closely matches the given +// id. +func (d *transportDemuxer) findEndpointLocked(eps *transportEndpoints, id TransportEndpointID) *endpointsByNic { + if matchedEPs := d.findAllEndpointsLocked(eps, id); len(matchedEPs) > 0 { + return matchedEPs[0] + } return nil } @@ -391,7 +556,7 @@ func (d *transportDemuxer) findEndpointLocked(eps *transportEndpoints, vv buffer func (d *transportDemuxer) registerRawEndpoint(netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, ep RawTransportEndpoint) *tcpip.Error { eps, ok := d.protocol[protocolIDs{netProto, transProto}] if !ok { - return nil + return tcpip.ErrNotSupported } eps.mu.Lock() @@ -418,3 +583,7 @@ func (d *transportDemuxer) unregisterRawEndpoint(netProto tcpip.NetworkProtocolN } } } + +func isMulticastOrBroadcast(addr tcpip.Address) bool { + return addr == header.IPv4Broadcast || header.IsV4MulticastAddress(addr) || header.IsV6MulticastAddress(addr) +} diff --git a/pkg/tcpip/stack/transport_demuxer_test.go b/pkg/tcpip/stack/transport_demuxer_test.go new file mode 100644 index 000000000..3b28b06d0 --- /dev/null +++ b/pkg/tcpip/stack/transport_demuxer_test.go @@ -0,0 +1,354 @@ +// Copyright 2018 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" + "math/rand" + "testing" + + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/link/channel" + "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" + "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" + "gvisor.dev/gvisor/pkg/tcpip/stack" + "gvisor.dev/gvisor/pkg/tcpip/transport/udp" + "gvisor.dev/gvisor/pkg/waiter" +) + +const ( + stackV6Addr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01" + testV6Addr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02" + + stackAddr = "\x0a\x00\x00\x01" + stackPort = 1234 + testPort = 4096 +) + +type testContext struct { + t *testing.T + linkEPs map[string]*channel.Endpoint + s *stack.Stack + + ep tcpip.Endpoint + wq waiter.Queue +} + +func (c *testContext) cleanup() { + if c.ep != nil { + c.ep.Close() + } +} + +func (c *testContext) createV6Endpoint(v6only bool) { + var err *tcpip.Error + c.ep, err = c.s.NewEndpoint(udp.ProtocolNumber, ipv6.ProtocolNumber, &c.wq) + if err != nil { + c.t.Fatalf("NewEndpoint failed: %v", err) + } + + var v tcpip.V6OnlyOption + if v6only { + v = 1 + } + if err := c.ep.SetSockOpt(v); err != nil { + c.t.Fatalf("SetSockOpt failed: %v", err) + } +} + +// newDualTestContextMultiNic creates the testing context and also linkEpNames +// named NICs. +func newDualTestContextMultiNic(t *testing.T, mtu uint32, linkEpNames []string) *testContext { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}}) + linkEPs := make(map[string]*channel.Endpoint) + for i, linkEpName := range linkEpNames { + channelEP := channel.New(256, mtu, "") + nicID := tcpip.NICID(i + 1) + if err := s.CreateNamedNIC(nicID, linkEpName, channelEP); err != nil { + t.Fatalf("CreateNIC failed: %v", err) + } + linkEPs[linkEpName] = channelEP + + if err := s.AddAddress(nicID, ipv4.ProtocolNumber, stackAddr); err != nil { + t.Fatalf("AddAddress IPv4 failed: %v", err) + } + + if err := s.AddAddress(nicID, ipv6.ProtocolNumber, stackV6Addr); err != nil { + t.Fatalf("AddAddress IPv6 failed: %v", err) + } + } + + s.SetRouteTable([]tcpip.Route{ + { + Destination: header.IPv4EmptySubnet, + NIC: 1, + }, + { + Destination: header.IPv6EmptySubnet, + NIC: 1, + }, + }) + + return &testContext{ + t: t, + s: s, + linkEPs: linkEPs, + } +} + +type headers struct { + srcPort uint16 + dstPort uint16 +} + +func newPayload() []byte { + b := make([]byte, 30+rand.Intn(100)) + for i := range b { + b[i] = byte(rand.Intn(256)) + } + return b +} + +func (c *testContext) sendV6Packet(payload []byte, h *headers, linkEpName string) { + // Allocate a buffer for data and headers. + buf := buffer.NewView(header.UDPMinimumSize + header.IPv6MinimumSize + len(payload)) + copy(buf[len(buf)-len(payload):], payload) + + // Initialize the IP header. + ip := header.IPv6(buf) + ip.Encode(&header.IPv6Fields{ + PayloadLength: uint16(header.UDPMinimumSize + len(payload)), + NextHeader: uint8(udp.ProtocolNumber), + HopLimit: 65, + SrcAddr: testV6Addr, + DstAddr: stackV6Addr, + }) + + // Initialize the UDP header. + u := header.UDP(buf[header.IPv6MinimumSize:]) + u.Encode(&header.UDPFields{ + SrcPort: h.srcPort, + DstPort: h.dstPort, + Length: uint16(header.UDPMinimumSize + len(payload)), + }) + + // Calculate the UDP pseudo-header checksum. + xsum := header.PseudoHeaderChecksum(udp.ProtocolNumber, testV6Addr, stackV6Addr, uint16(len(u))) + + // Calculate the UDP checksum and set it. + xsum = header.Checksum(payload, xsum) + u.SetChecksum(^u.CalculateChecksum(xsum)) + + // Inject packet. + c.linkEPs[linkEpName].InjectInbound(ipv6.ProtocolNumber, tcpip.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) +} + +func TestTransportDemuxerRegister(t *testing.T) { + for _, test := range []struct { + name string + proto tcpip.NetworkProtocolNumber + want *tcpip.Error + }{ + {"failure", ipv6.ProtocolNumber, tcpip.ErrUnknownProtocol}, + {"success", ipv4.ProtocolNumber, nil}, + } { + t.Run(test.name, func(t *testing.T) { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}}) + if got, want := s.RegisterTransportEndpoint(0, []tcpip.NetworkProtocolNumber{test.proto}, udp.ProtocolNumber, stack.TransportEndpointID{}, nil, false, 0), test.want; got != want { + t.Fatalf("s.RegisterTransportEndpoint(...) = %v, want %v", got, want) + } + }) + } +} + +// TestReuseBindToDevice injects varied packets on input devices and checks that +// the distribution of packets received matches expectations. +func TestDistribution(t *testing.T) { + type endpointSockopts struct { + reuse int + bindToDevice string + } + for _, test := range []struct { + name string + // endpoints will received the inject packets. + endpoints []endpointSockopts + // wantedDistribution is the wanted ratio of packets received on each + // endpoint for each NIC on which packets are injected. + wantedDistributions map[string][]float64 + }{ + { + "BindPortReuse", + // 5 endpoints that all have reuse set. + []endpointSockopts{ + endpointSockopts{1, ""}, + endpointSockopts{1, ""}, + endpointSockopts{1, ""}, + endpointSockopts{1, ""}, + endpointSockopts{1, ""}, + }, + map[string][]float64{ + // Injected packets on dev0 get distributed evenly. + "dev0": []float64{0.2, 0.2, 0.2, 0.2, 0.2}, + }, + }, + { + "BindToDevice", + // 3 endpoints with various bindings. + []endpointSockopts{ + endpointSockopts{0, "dev0"}, + endpointSockopts{0, "dev1"}, + endpointSockopts{0, "dev2"}, + }, + map[string][]float64{ + // Injected packets on dev0 go only to the endpoint bound to dev0. + "dev0": []float64{1, 0, 0}, + // Injected packets on dev1 go only to the endpoint bound to dev1. + "dev1": []float64{0, 1, 0}, + // Injected packets on dev2 go only to the endpoint bound to dev2. + "dev2": []float64{0, 0, 1}, + }, + }, + { + "ReuseAndBindToDevice", + // 6 endpoints with various bindings. + []endpointSockopts{ + endpointSockopts{1, "dev0"}, + endpointSockopts{1, "dev0"}, + endpointSockopts{1, "dev1"}, + endpointSockopts{1, "dev1"}, + endpointSockopts{1, "dev1"}, + endpointSockopts{1, ""}, + }, + map[string][]float64{ + // Injected packets on dev0 get distributed among endpoints bound to + // dev0. + "dev0": []float64{0.5, 0.5, 0, 0, 0, 0}, + // Injected packets on dev1 get distributed among endpoints bound to + // dev1 or unbound. + "dev1": []float64{0, 0, 1. / 3, 1. / 3, 1. / 3, 0}, + // Injected packets on dev999 go only to the unbound. + "dev999": []float64{0, 0, 0, 0, 0, 1}, + }, + }, + } { + t.Run(test.name, func(t *testing.T) { + for device, wantedDistribution := range test.wantedDistributions { + t.Run(device, func(t *testing.T) { + var devices []string + for d := range test.wantedDistributions { + devices = append(devices, d) + } + c := newDualTestContextMultiNic(t, defaultMTU, devices) + defer c.cleanup() + + c.createV6Endpoint(false) + + eps := make(map[tcpip.Endpoint]int) + + pollChannel := make(chan tcpip.Endpoint) + for i, endpoint := range test.endpoints { + // Try to receive the data. + wq := waiter.Queue{} + we, ch := waiter.NewChannelEntry(nil) + wq.EventRegister(&we, waiter.EventIn) + defer wq.EventUnregister(&we) + defer close(ch) + + var err *tcpip.Error + ep, err := c.s.NewEndpoint(udp.ProtocolNumber, ipv6.ProtocolNumber, &wq) + if err != nil { + c.t.Fatalf("NewEndpoint failed: %v", err) + } + eps[ep] = i + + go func(ep tcpip.Endpoint) { + for range ch { + pollChannel <- ep + } + }(ep) + + defer ep.Close() + reusePortOption := tcpip.ReusePortOption(endpoint.reuse) + if err := ep.SetSockOpt(reusePortOption); err != nil { + c.t.Fatalf("SetSockOpt(%#v) on endpoint %d failed: %v", reusePortOption, i, err) + } + bindToDeviceOption := tcpip.BindToDeviceOption(endpoint.bindToDevice) + if err := ep.SetSockOpt(bindToDeviceOption); err != nil { + c.t.Fatalf("SetSockOpt(%#v) on endpoint %d failed: %v", bindToDeviceOption, i, err) + } + if err := ep.Bind(tcpip.FullAddress{Addr: stackV6Addr, Port: stackPort}); err != nil { + t.Fatalf("ep.Bind(...) on endpoint %d failed: %v", i, err) + } + } + + npackets := 100000 + nports := 10000 + if got, want := len(test.endpoints), len(wantedDistribution); got != want { + t.Fatalf("got len(test.endpoints) = %d, want %d", got, want) + } + ports := make(map[uint16]tcpip.Endpoint) + stats := make(map[tcpip.Endpoint]int) + for i := 0; i < npackets; i++ { + // Send a packet. + port := uint16(i % nports) + payload := newPayload() + c.sendV6Packet(payload, + &headers{ + srcPort: testPort + port, + dstPort: stackPort}, + device) + + var addr tcpip.FullAddress + ep := <-pollChannel + _, _, err := ep.Read(&addr) + if err != nil { + c.t.Fatalf("Read on endpoint %d failed: %v", eps[ep], err) + } + stats[ep]++ + if i < nports { + ports[uint16(i)] = ep + } else { + // Check that all packets from one client are handled by the same + // socket. + if want, got := ports[port], ep; want != got { + t.Fatalf("Packet sent on port %d expected on endpoint %d but received on endpoint %d", port, eps[want], eps[got]) + } + } + } + + // Check that a packet distribution is as expected. + for ep, i := range eps { + wantedRatio := wantedDistribution[i] + wantedRecv := wantedRatio * float64(npackets) + actualRecv := stats[ep] + actualRatio := float64(stats[ep]) / float64(npackets) + // The deviation is less than 10%. + if math.Abs(actualRatio-wantedRatio) > 0.05 { + t.Errorf("wanted about %.0f%% (%.0f of %d) packets to arrive on endpoint %d, got %.0f%% (%d of %d)", wantedRatio*100, wantedRecv, npackets, i, actualRatio*100, actualRecv, npackets) + } + } + }) + } + }) + } +} diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index 56e8a5d9b..2cacea99a 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -38,19 +38,27 @@ const ( // Headers of this protocol are fakeTransHeaderLen bytes, but we currently don't // use it. type fakeTransportEndpoint struct { - id stack.TransportEndpointID + stack.TransportEndpointInfo stack *stack.Stack - netProto tcpip.NetworkProtocolNumber proto *fakeTransportProtocol peerAddr tcpip.Address route stack.Route + uniqueID uint64 // acceptQueue is non-nil iff bound. acceptQueue []fakeTransportEndpoint } -func newFakeTransportEndpoint(stack *stack.Stack, proto *fakeTransportProtocol, netProto tcpip.NetworkProtocolNumber) tcpip.Endpoint { - return &fakeTransportEndpoint{stack: stack, netProto: netProto, proto: proto} +func (f *fakeTransportEndpoint) Info() tcpip.EndpointInfo { + return &f.TransportEndpointInfo +} + +func (f *fakeTransportEndpoint) Stats() tcpip.EndpointStats { + return nil +} + +func newFakeTransportEndpoint(s *stack.Stack, proto *fakeTransportProtocol, netProto tcpip.NetworkProtocolNumber, uniqueID uint64) tcpip.Endpoint { + return &fakeTransportEndpoint{stack: s, TransportEndpointInfo: stack.TransportEndpointInfo{NetProto: netProto}, proto: proto, uniqueID: uniqueID} } func (f *fakeTransportEndpoint) Close() { @@ -75,7 +83,7 @@ func (f *fakeTransportEndpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions if err != nil { return 0, nil, err } - if err := f.route.WritePacket(nil /* gso */, hdr, buffer.View(v).ToVectorisedView(), fakeTransNumber, 123); err != nil { + if err := f.route.WritePacket(nil /* gso */, hdr, buffer.View(v).ToVectorisedView(), stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}); err != nil { return 0, nil, err } @@ -126,8 +134,8 @@ func (f *fakeTransportEndpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { defer r.Release() // Try to register so that we can start receiving packets. - f.id.RemoteAddress = addr.Addr - err = f.stack.RegisterTransportEndpoint(0, []tcpip.NetworkProtocolNumber{fakeNetNumber}, fakeTransNumber, f.id, f, false) + f.ID.RemoteAddress = addr.Addr + err = f.stack.RegisterTransportEndpoint(0, []tcpip.NetworkProtocolNumber{fakeNetNumber}, fakeTransNumber, f.ID, f, false /* reuse */, 0 /* bindToDevice */) if err != nil { return err } @@ -137,6 +145,10 @@ func (f *fakeTransportEndpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { return nil } +func (f *fakeTransportEndpoint) UniqueID() uint64 { + return f.uniqueID +} + func (f *fakeTransportEndpoint) ConnectEndpoint(e tcpip.Endpoint) *tcpip.Error { return nil } @@ -168,7 +180,8 @@ func (f *fakeTransportEndpoint) Bind(a tcpip.FullAddress) *tcpip.Error { fakeTransNumber, stack.TransportEndpointID{LocalAddress: a.Addr}, f, - false, + false, /* reuse */ + 0, /* bindtoDevice */ ); err != nil { return err } @@ -184,14 +197,16 @@ func (*fakeTransportEndpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Erro return tcpip.FullAddress{}, nil } -func (f *fakeTransportEndpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, _ buffer.VectorisedView) { +func (f *fakeTransportEndpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, _ tcpip.PacketBuffer) { // Increment the number of received packets. f.proto.packetCount++ if f.acceptQueue != nil { f.acceptQueue = append(f.acceptQueue, fakeTransportEndpoint{ - id: id, - stack: f.stack, - netProto: f.netProto, + stack: f.stack, + TransportEndpointInfo: stack.TransportEndpointInfo{ + ID: f.ID, + NetProto: f.NetProto, + }, proto: f.proto, peerAddr: r.RemoteAddress, route: r.Clone(), @@ -199,7 +214,7 @@ func (f *fakeTransportEndpoint) HandlePacket(r *stack.Route, id stack.TransportE } } -func (f *fakeTransportEndpoint) HandleControlPacket(stack.TransportEndpointID, stack.ControlType, uint32, buffer.VectorisedView) { +func (f *fakeTransportEndpoint) HandleControlPacket(stack.TransportEndpointID, stack.ControlType, uint32, tcpip.PacketBuffer) { // Increment the number of received control packets. f.proto.controlCount++ } @@ -208,15 +223,15 @@ func (f *fakeTransportEndpoint) State() uint32 { return 0 } -func (f *fakeTransportEndpoint) ModerateRecvBuf(copied int) { -} +func (f *fakeTransportEndpoint) ModerateRecvBuf(copied int) {} func (f *fakeTransportEndpoint) IPTables() (iptables.IPTables, error) { return iptables.IPTables{}, nil } -func (f *fakeTransportEndpoint) Resume(*stack.Stack) { -} +func (f *fakeTransportEndpoint) Resume(*stack.Stack) {} + +func (f *fakeTransportEndpoint) Wait() {} type fakeTransportGoodOption bool @@ -241,7 +256,7 @@ func (*fakeTransportProtocol) Number() tcpip.TransportProtocolNumber { } func (f *fakeTransportProtocol) NewEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, _ *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { - return newFakeTransportEndpoint(stack, f, netProto), nil + return newFakeTransportEndpoint(stack, f, netProto, stack.UniqueID()), nil } func (f *fakeTransportProtocol) NewRawEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, _ *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { @@ -256,7 +271,7 @@ func (*fakeTransportProtocol) ParsePorts(buffer.View) (src, dst uint16, err *tcp return 0, 0, nil } -func (*fakeTransportProtocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, buffer.View, buffer.VectorisedView) bool { +func (*fakeTransportProtocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, tcpip.PacketBuffer) bool { return true } @@ -327,7 +342,9 @@ func TestTransportReceive(t *testing.T) { // Make sure packet with wrong protocol is not delivered. buf[0] = 1 buf[2] = 0 - linkEP.Inject(fakeNetNumber, buf.ToVectorisedView()) + linkEP.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) if fakeTrans.packetCount != 0 { t.Errorf("packetCount = %d, want %d", fakeTrans.packetCount, 0) } @@ -336,7 +353,9 @@ func TestTransportReceive(t *testing.T) { buf[0] = 1 buf[1] = 3 buf[2] = byte(fakeTransNumber) - linkEP.Inject(fakeNetNumber, buf.ToVectorisedView()) + linkEP.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) if fakeTrans.packetCount != 0 { t.Errorf("packetCount = %d, want %d", fakeTrans.packetCount, 0) } @@ -345,7 +364,9 @@ func TestTransportReceive(t *testing.T) { buf[0] = 1 buf[1] = 2 buf[2] = byte(fakeTransNumber) - linkEP.Inject(fakeNetNumber, buf.ToVectorisedView()) + linkEP.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) if fakeTrans.packetCount != 1 { t.Errorf("packetCount = %d, want %d", fakeTrans.packetCount, 1) } @@ -398,7 +419,9 @@ func TestTransportControlReceive(t *testing.T) { buf[fakeNetHeaderLen+0] = 0 buf[fakeNetHeaderLen+1] = 1 buf[fakeNetHeaderLen+2] = 0 - linkEP.Inject(fakeNetNumber, buf.ToVectorisedView()) + linkEP.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) if fakeTrans.controlCount != 0 { t.Errorf("controlCount = %d, want %d", fakeTrans.controlCount, 0) } @@ -407,7 +430,9 @@ func TestTransportControlReceive(t *testing.T) { buf[fakeNetHeaderLen+0] = 3 buf[fakeNetHeaderLen+1] = 1 buf[fakeNetHeaderLen+2] = byte(fakeTransNumber) - linkEP.Inject(fakeNetNumber, buf.ToVectorisedView()) + linkEP.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) if fakeTrans.controlCount != 0 { t.Errorf("controlCount = %d, want %d", fakeTrans.controlCount, 0) } @@ -416,7 +441,9 @@ func TestTransportControlReceive(t *testing.T) { buf[fakeNetHeaderLen+0] = 2 buf[fakeNetHeaderLen+1] = 1 buf[fakeNetHeaderLen+2] = byte(fakeTransNumber) - linkEP.Inject(fakeNetNumber, buf.ToVectorisedView()) + linkEP.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) if fakeTrans.controlCount != 1 { t.Errorf("controlCount = %d, want %d", fakeTrans.controlCount, 1) } @@ -569,7 +596,9 @@ func TestTransportForwarding(t *testing.T) { req[0] = 1 req[1] = 3 req[2] = byte(fakeTransNumber) - ep2.Inject(fakeNetNumber, req.ToVectorisedView()) + ep2.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + Data: req.ToVectorisedView(), + }) aep, _, err := ep.Accept() if err != nil || aep == nil { diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index c021c67ac..bd5eb89ca 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -57,6 +57,9 @@ type Error struct { // String implements fmt.Stringer.String. func (e *Error) String() string { + if e == nil { + return "<nil>" + } return e.msg } @@ -228,6 +231,13 @@ func (s *Subnet) Broadcast() Address { return Address(addr) } +// Equal returns true if s equals o. +// +// Needed to use cmp.Equal on Subnet as its fields are unexported. +func (s Subnet) Equal(o Subnet) bool { + return s == o +} + // NICID is a number that uniquely identifies a NIC. type NICID int32 @@ -252,7 +262,7 @@ type FullAddress struct { // This may not be used by all endpoint types. NIC NICID - // Addr is the network address. + // Addr is the network or link layer address. Addr Address // Port is the transport port. @@ -426,6 +436,26 @@ type Endpoint interface { // IPTables returns the iptables for this endpoint's stack. IPTables() (iptables.IPTables, error) + + // Info returns a copy to the transport endpoint info. + Info() EndpointInfo + + // Stats returns a reference to the endpoint stats. + Stats() EndpointStats +} + +// EndpointInfo is the interface implemented by each endpoint info struct. +type EndpointInfo interface { + // IsEndpointInfo is an empty method to implement the tcpip.EndpointInfo + // marker interface. + IsEndpointInfo() +} + +// EndpointStats is the interface implemented by each endpoint stats struct. +type EndpointStats interface { + // IsEndpointStats is an empty method to implement the tcpip.EndpointStats + // marker interface. + IsEndpointStats() } // WriteOptions contains options for Endpoint.Write. @@ -466,6 +496,11 @@ const ( // number of unread bytes in the output buffer should be returned. SendQueueSizeOption + // 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 + // TODO(b/137664753): convert all int socket options to be handled via // GetSockOptInt. ) @@ -478,11 +513,6 @@ type ErrorOption struct{} // socket is to be restricted to sending and receiving IPv6 packets only. type V6OnlyOption int -// 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. -type DelayOption int - // CorkOption is used by SetSockOpt/GetSockOpt to specify if data should be // held until segments are full by the TCP transport protocol. type CorkOption int @@ -495,6 +525,10 @@ type ReuseAddressOption int // to be bound to an identical socket address. type ReusePortOption int +// BindToDeviceOption is used by SetSockOpt/GetSockOpt to specify that sockets +// should bind only on a specific NIC. +type BindToDeviceOption string + // QuickAckOption is stubbed out in SetSockOpt/GetSockOpt. type QuickAckOption int @@ -546,6 +580,22 @@ type ModerateReceiveBufferOption bool // Maximum Segment Size(MSS) value as specified using the TCP_MAXSEG option. type MaxSegOption int +// TTLOption is used by SetSockOpt/GetSockOpt to control the default TTL/hop +// limit value for unicast messages. The default is protocol specific. +// +// A zero value indicates the default. +type TTLOption uint8 + +// TCPLingerTimeoutOption is used by SetSockOpt/GetSockOpt to set/get the +// maximum duration for which a socket lingers in the TCP_FIN_WAIT_2 state +// before being marked closed. +type TCPLingerTimeoutOption time.Duration + +// TCPTimeWaitTimeoutOption is used by SetSockOpt/GetSockOpt to set/get the +// maximum duration for which a socket lingers in the TIME_WAIT state +// before being marked closed. +type TCPTimeWaitTimeoutOption time.Duration + // MulticastTTLOption is used by SetSockOpt/GetSockOpt to control the default // TTL value for multicast messages. The default is 1. type MulticastTTLOption uint8 @@ -587,6 +637,18 @@ type OutOfBandInlineOption int // datagram sockets are allowed to send packets to a broadcast address. type BroadcastOption int +// DefaultTTLOption is used by stack.(*Stack).NetworkProtocolOption to specify +// a default TTL. +type DefaultTTLOption uint8 + +// IPv4TOSOption is used by SetSockOpt/GetSockOpt to specify TOS +// for all subsequent outgoing IPv4 packets from the endpoint. +type IPv4TOSOption uint8 + +// IPv6TrafficClassOption is used by SetSockOpt/GetSockOpt to specify TOS +// for all subsequent outgoing IPv6 packets from the endpoint. +type IPv6TrafficClassOption uint8 + // Route is a row in the routing table. It specifies through which NIC (and // gateway) sets of packets should be routed. A row is considered viable if the // masked target address matches the destination address in the row. @@ -628,6 +690,11 @@ func (s *StatCounter) Increment() { s.IncrementBy(1) } +// Decrement minuses one to the counter. +func (s *StatCounter) Decrement() { + s.IncrementBy(^uint64(0)) +} + // Value returns the current value of the counter. func (s *StatCounter) Value() uint64 { return atomic.LoadUint64(&s.count) @@ -816,6 +883,14 @@ type IPStats struct { // OutgoingPacketErrors is the total number of IP packets which failed // to write to a link-layer endpoint. OutgoingPacketErrors *StatCounter + + // MalformedPacketsReceived is the total number of IP Packets that were + // dropped due to the IP packet header failing validation checks. + MalformedPacketsReceived *StatCounter + + // MalformedFragmentsReceived is the total number of IP Fragments that were + // dropped due to the fragment failing validation checks. + MalformedFragmentsReceived *StatCounter } // TCPStats collects TCP-specific stats. @@ -828,6 +903,15 @@ type TCPStats struct { // successfully via Listen. PassiveConnectionOpenings *StatCounter + // CurrentEstablished is the number of TCP connections for which the + // current state is either ESTABLISHED or CLOSE-WAIT. + CurrentEstablished *StatCounter + + // EstablishedResets is the number of times TCP connections have made + // a direct transition to the CLOSED state from either the + // ESTABLISHED state or the CLOSE-WAIT state. + EstablishedResets *StatCounter + // ListenOverflowSynDrop is the number of times the listen queue overflowed // and a SYN was dropped. ListenOverflowSynDrop *StatCounter @@ -862,6 +946,9 @@ type TCPStats struct { // SegmentsSent is the number of TCP segments sent. SegmentsSent *StatCounter + // SegmentSendErrors is the number of TCP segments failed to be sent. + SegmentSendErrors *StatCounter + // ResetsSent is the number of TCP resets sent. ResetsSent *StatCounter @@ -914,6 +1001,9 @@ type UDPStats struct { // PacketsSent is the number of UDP datagrams sent via sendUDP. PacketsSent *StatCounter + + // PacketSendErrors is the number of datagrams failed to be sent. + PacketSendErrors *StatCounter } // Stats holds statistics about the networking stack. @@ -924,7 +1014,7 @@ type Stats struct { // stack that were for an unknown or unsupported protocol. UnknownProtocolRcvdPackets *StatCounter - // MalformedRcvPackets is the number of packets received by the stack + // MalformedRcvdPackets is the number of packets received by the stack // that were deemed malformed. MalformedRcvdPackets *StatCounter @@ -944,18 +1034,95 @@ type Stats struct { UDP UDPStats } +// ReceiveErrors collects packet receive errors within transport endpoint. +type ReceiveErrors struct { + // ReceiveBufferOverflow is the number of received packets dropped + // due to the receive buffer being full. + ReceiveBufferOverflow StatCounter + + // MalformedPacketsReceived is the number of incoming packets + // dropped due to the packet header being in a malformed state. + MalformedPacketsReceived StatCounter + + // ClosedReceiver is the number of received packets dropped because + // of receiving endpoint state being closed. + ClosedReceiver StatCounter +} + +// SendErrors collects packet send errors within the transport layer for +// an endpoint. +type SendErrors struct { + // SendToNetworkFailed is the number of packets failed to be written to + // the network endpoint. + SendToNetworkFailed StatCounter + + // NoRoute is the number of times we failed to resolve IP route. + NoRoute StatCounter + + // NoLinkAddr is the number of times we failed to resolve ARP. + NoLinkAddr StatCounter +} + +// ReadErrors collects segment read errors from an endpoint read call. +type ReadErrors struct { + // ReadClosed is the number of received packet drops because the endpoint + // was shutdown for read. + ReadClosed StatCounter + + // InvalidEndpointState is the number of times we found the endpoint state + // to be unexpected. + InvalidEndpointState StatCounter +} + +// WriteErrors collects packet write errors from an endpoint write call. +type WriteErrors struct { + // WriteClosed is the number of packet drops because the endpoint + // was shutdown for write. + WriteClosed StatCounter + + // InvalidEndpointState is the number of times we found the endpoint state + // to be unexpected. + InvalidEndpointState StatCounter + + // InvalidArgs is the number of times invalid input arguments were + // provided for endpoint Write call. + InvalidArgs StatCounter +} + +// TransportEndpointStats collects statistics about the endpoint. +type TransportEndpointStats struct { + // PacketsReceived is the number of successful packet receives. + PacketsReceived StatCounter + + // PacketsSent is the number of successful packet sends. + PacketsSent StatCounter + + // ReceiveErrors collects packet receive errors within transport layer. + ReceiveErrors ReceiveErrors + + // ReadErrors collects packet read errors from an endpoint read call. + ReadErrors ReadErrors + + // SendErrors collects packet send errors within the transport layer. + SendErrors SendErrors + + // WriteErrors collects packet write errors from an endpoint write call. + WriteErrors WriteErrors +} + +// IsEndpointStats is an empty method to implement the tcpip.EndpointStats +// marker interface. +func (*TransportEndpointStats) IsEndpointStats() {} + func fillIn(v reflect.Value) { for i := 0; i < v.NumField(); i++ { v := v.Field(i) - switch v.Kind() { - case reflect.Ptr: - if s := v.Addr().Interface().(**StatCounter); *s == nil { - *s = &StatCounter{} + if s, ok := v.Addr().Interface().(**StatCounter); ok { + if *s == nil { + *s = new(StatCounter) } - case reflect.Struct: + } else { fillIn(v) - default: - panic(fmt.Sprintf("unexpected type %s", v.Type())) } } } @@ -966,6 +1133,26 @@ func (s Stats) FillIn() Stats { return s } +// Clone returns a copy of the TransportEndpointStats by atomically reading +// each field. +func (src *TransportEndpointStats) Clone() TransportEndpointStats { + var dst TransportEndpointStats + clone(reflect.ValueOf(&dst).Elem(), reflect.ValueOf(src).Elem()) + return dst +} + +func clone(dst reflect.Value, src reflect.Value) { + for i := 0; i < dst.NumField(); i++ { + d := dst.Field(i) + s := src.Field(i) + if c, ok := s.Addr().Interface().(*StatCounter); ok { + d.Addr().Interface().(*StatCounter).IncrementBy(c.Value()) + } else { + clone(d, s) + } + } +} + // String implements the fmt.Stringer interface. func (a Address) String() string { switch len(a) { @@ -1091,6 +1278,47 @@ func (a AddressWithPrefix) String() string { return fmt.Sprintf("%s/%d", a.Address, a.PrefixLen) } +// Subnet converts the address and prefix into a Subnet value and returns it. +func (a AddressWithPrefix) Subnet() Subnet { + addrLen := len(a.Address) + if a.PrefixLen <= 0 { + return Subnet{ + address: Address(strings.Repeat("\x00", addrLen)), + mask: AddressMask(strings.Repeat("\x00", addrLen)), + } + } + if a.PrefixLen >= addrLen*8 { + return Subnet{ + address: a.Address, + mask: AddressMask(strings.Repeat("\xff", addrLen)), + } + } + + sa := make([]byte, addrLen) + sm := make([]byte, addrLen) + n := uint(a.PrefixLen) + for i := 0; i < addrLen; i++ { + if n >= 8 { + sa[i] = a.Address[i] + sm[i] = 0xff + n -= 8 + continue + } + sm[i] = ^byte(0xff >> n) + sa[i] = a.Address[i] & sm[i] + n = 0 + } + + // For extra caution, call NewSubnet rather than directly creating the Subnet + // value. If that fails it indicates a serious bug in this code, so panic is + // in order. + s, err := NewSubnet(Address(sa), AddressMask(sm)) + if err != nil { + panic("invalid subnet: " + err.Error()) + } + return s +} + // ProtocolAddress is an address and the network protocol it is associated // with. type ProtocolAddress struct { @@ -1111,8 +1339,8 @@ var ( // GetDanglingEndpoints returns all dangling endpoints. func GetDanglingEndpoints() []Endpoint { - es := make([]Endpoint, 0, len(danglingEndpoints)) danglingEndpointsMu.Lock() + es := make([]Endpoint, 0, len(danglingEndpoints)) for e := range danglingEndpoints { es = append(es, e) } diff --git a/pkg/tcpip/tcpip_test.go b/pkg/tcpip/tcpip_test.go index fb3a0a5ee..8c0aacffa 100644 --- a/pkg/tcpip/tcpip_test.go +++ b/pkg/tcpip/tcpip_test.go @@ -195,3 +195,34 @@ func TestStatsString(t *testing.T) { t.Logf(`got = fmt.Sprintf("%%+v", Stats{}.FillIn()) = %q`, got) } } + +func TestAddressWithPrefixSubnet(t *testing.T) { + tests := []struct { + addr Address + prefixLen int + subnetAddr Address + subnetMask AddressMask + }{ + {"\xaa\x55\x33\x42", -1, "\x00\x00\x00\x00", "\x00\x00\x00\x00"}, + {"\xaa\x55\x33\x42", 0, "\x00\x00\x00\x00", "\x00\x00\x00\x00"}, + {"\xaa\x55\x33\x42", 1, "\x80\x00\x00\x00", "\x80\x00\x00\x00"}, + {"\xaa\x55\x33\x42", 7, "\xaa\x00\x00\x00", "\xfe\x00\x00\x00"}, + {"\xaa\x55\x33\x42", 8, "\xaa\x00\x00\x00", "\xff\x00\x00\x00"}, + {"\xaa\x55\x33\x42", 24, "\xaa\x55\x33\x00", "\xff\xff\xff\x00"}, + {"\xaa\x55\x33\x42", 31, "\xaa\x55\x33\x42", "\xff\xff\xff\xfe"}, + {"\xaa\x55\x33\x42", 32, "\xaa\x55\x33\x42", "\xff\xff\xff\xff"}, + {"\xaa\x55\x33\x42", 33, "\xaa\x55\x33\x42", "\xff\xff\xff\xff"}, + } + for _, tt := range tests { + ap := AddressWithPrefix{Address: tt.addr, PrefixLen: tt.prefixLen} + gotSubnet := ap.Subnet() + wantSubnet, err := NewSubnet(tt.subnetAddr, tt.subnetMask) + if err != nil { + t.Error("NewSubnet(%q, %q) failed: %s", tt.subnetAddr, tt.subnetMask, err) + continue + } + if gotSubnet != wantSubnet { + t.Errorf("got subnet = %q, want = %q", gotSubnet, wantSubnet) + } + } +} diff --git a/pkg/tcpip/time_unsafe.go b/pkg/tcpip/time_unsafe.go index a52262e87..48764b978 100644 --- a/pkg/tcpip/time_unsafe.go +++ b/pkg/tcpip/time_unsafe.go @@ -13,7 +13,7 @@ // limitations under the License. // +build go1.9 -// +build !go1.14 +// +build !go1.15 // Check go:linkname function signatures when updating Go version. diff --git a/pkg/tcpip/transport/icmp/BUILD b/pkg/tcpip/transport/icmp/BUILD index d78a162b8..9254c3dea 100644 --- a/pkg/tcpip/transport/icmp/BUILD +++ b/pkg/tcpip/transport/icmp/BUILD @@ -1,8 +1,8 @@ -package(licenses = ["notice"]) - load("//tools/go_generics:defs.bzl", "go_template_instance") load("//tools/go_stateify:defs.bzl", "go_library") +package(licenses = ["notice"]) + go_template_instance( name = "icmp_packet_list", out = "icmp_packet_list.go", diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index a111fdb2a..70e008d36 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -31,9 +31,6 @@ type icmpPacket struct { senderAddress tcpip.FullAddress data buffer.VectorisedView `state:".(buffer.VectorisedView)"` timestamp int64 - // views is used as buffer for data when its length is large - // enough to store a VectorisedView. - views [8]buffer.View `state:"nosave"` } type endpointState int @@ -52,12 +49,13 @@ const ( // // +stateify savable type endpoint struct { + stack.TransportEndpointInfo + // The following fields are initialized at creation time and are // immutable. stack *stack.Stack `state:"manual"` - netProto tcpip.NetworkProtocolNumber - transProto tcpip.TransportProtocolNumber waiterQueue *waiter.Queue + uniqueID uint64 // The following fields are used to manage the receive queue, and are // protected by rcvMu. @@ -73,30 +71,32 @@ type endpoint struct { sndBufSize int // shutdownFlags represent the current shutdown state of the endpoint. shutdownFlags tcpip.ShutdownFlags - id stack.TransportEndpointID state endpointState - // bindNICID and bindAddr are set via calls to Bind(). They are used to - // reject attempts to send data or connect via a different NIC or - // address - bindNICID tcpip.NICID - bindAddr tcpip.Address - // regNICID is the default NIC to be used when callers don't specify a - // NIC. - regNICID tcpip.NICID - route stack.Route `state:"manual"` -} - -func newEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { + route stack.Route `state:"manual"` + ttl uint8 + stats tcpip.TransportEndpointStats `state:"nosave"` +} + +func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { return &endpoint{ - stack: stack, - netProto: netProto, - transProto: transProto, + stack: s, + TransportEndpointInfo: stack.TransportEndpointInfo{ + NetProto: netProto, + TransProto: transProto, + }, waiterQueue: waiterQueue, rcvBufSizeMax: 32 * 1024, sndBufSize: 32 * 1024, + state: stateInitial, + uniqueID: s.UniqueID(), }, nil } +// UniqueID implements stack.TransportEndpoint.UniqueID. +func (e *endpoint) UniqueID() uint64 { + return e.uniqueID +} + // Close puts the endpoint in a closed state and frees all resources // associated with it. func (e *endpoint) Close() { @@ -104,7 +104,7 @@ func (e *endpoint) Close() { e.shutdownFlags = tcpip.ShutdownRead | tcpip.ShutdownWrite switch e.state { case stateBound, stateConnected: - e.stack.UnregisterTransportEndpoint(e.regNICID, []tcpip.NetworkProtocolNumber{e.netProto}, e.transProto, e.id, e) + e.stack.UnregisterTransportEndpoint(e.RegisterNICID, []tcpip.NetworkProtocolNumber{e.NetProto}, e.TransProto, e.ID, e, 0 /* bindToDevice */) } // Close the receive list and drain it. @@ -143,6 +143,7 @@ func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMess if e.rcvList.Empty() { err := tcpip.ErrWouldBlock if e.rcvClosed { + e.stats.ReadErrors.ReadClosed.Increment() err = tcpip.ErrClosedForReceive } e.rcvMu.Unlock() @@ -205,6 +206,29 @@ func (e *endpoint) prepareForWrite(to *tcpip.FullAddress) (retry bool, err *tcpi // Write writes data to the endpoint's peer. This method does not block // if the data cannot be written. func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { + n, ch, err := e.write(p, opts) + switch err { + case nil: + e.stats.PacketsSent.Increment() + case tcpip.ErrMessageTooLong, tcpip.ErrInvalidOptionValue: + e.stats.WriteErrors.InvalidArgs.Increment() + case tcpip.ErrClosedForSend: + e.stats.WriteErrors.WriteClosed.Increment() + case tcpip.ErrInvalidEndpointState: + e.stats.WriteErrors.InvalidEndpointState.Increment() + case tcpip.ErrNoLinkAddress: + e.stats.SendErrors.NoLinkAddr.Increment() + case tcpip.ErrNoRoute, tcpip.ErrBroadcastDisabled, tcpip.ErrNetworkUnreachable: + // Errors indicating any problem with IP routing of the packet. + e.stats.SendErrors.NoRoute.Increment() + default: + // For all other errors when writing to the network layer. + e.stats.SendErrors.SendToNetworkFailed.Increment() + } + return n, ch, err +} + +func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { // MSG_MORE is unimplemented. (This also means that MSG_EOR is a no-op.) if opts.More { return 0, nil, tcpip.ErrInvalidOptionValue @@ -254,13 +278,13 @@ func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c } else { // Reject destination address if it goes through a different // NIC than the endpoint was bound to. - nicid := to.NIC - if e.bindNICID != 0 { - if nicid != 0 && nicid != e.bindNICID { + nicID := to.NIC + if e.BindNICID != 0 { + if nicID != 0 && nicID != e.BindNICID { return 0, nil, tcpip.ErrNoRoute } - nicid = e.bindNICID + nicID = e.BindNICID } toCopy := *to @@ -271,7 +295,7 @@ func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c } // Find the enpoint. - r, err := e.stack.FindRoute(nicid, e.bindAddr, to.Addr, netProto, false /* multicastLoop */) + r, err := e.stack.FindRoute(nicID, e.BindAddr, to.Addr, netProto, false /* multicastLoop */) if err != nil { return 0, nil, err } @@ -294,12 +318,12 @@ func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c return 0, nil, err } - switch e.netProto { + switch e.NetProto { case header.IPv4ProtocolNumber: - err = send4(route, e.id.LocalPort, v) + err = send4(route, e.ID.LocalPort, v, e.ttl) case header.IPv6ProtocolNumber: - err = send6(route, e.id.LocalPort, v) + err = send6(route, e.ID.LocalPort, v, e.ttl) } if err != nil { @@ -314,8 +338,15 @@ func (e *endpoint) Peek([][]byte) (int64, tcpip.ControlMessages, *tcpip.Error) { return 0, tcpip.ControlMessages{}, nil } -// SetSockOpt sets a socket option. Currently not supported. +// SetSockOpt sets a socket option. func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { + switch o := opt.(type) { + case tcpip.TTLOption: + e.mu.Lock() + e.ttl = uint8(o) + e.mu.Unlock() + } + return nil } @@ -362,12 +393,18 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { *o = 0 return nil + case *tcpip.TTLOption: + e.rcvMu.Lock() + *o = tcpip.TTLOption(e.ttl) + e.rcvMu.Unlock() + return nil + default: return tcpip.ErrUnknownProtocolOption } } -func send4(r *stack.Route, ident uint16, data buffer.View) *tcpip.Error { +func send4(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Error { if len(data) < header.ICMPv4MinimumSize { return tcpip.ErrInvalidEndpointState } @@ -389,10 +426,13 @@ func send4(r *stack.Route, ident uint16, data buffer.View) *tcpip.Error { icmpv4.SetChecksum(0) icmpv4.SetChecksum(^header.Checksum(icmpv4, header.Checksum(data, 0))) - return r.WritePacket(nil /* gso */, hdr, data.ToVectorisedView(), header.ICMPv4ProtocolNumber, r.DefaultTTL()) + if ttl == 0 { + ttl = r.DefaultTTL() + } + return r.WritePacket(nil /* gso */, hdr, data.ToVectorisedView(), stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}) } -func send6(r *stack.Route, ident uint16, data buffer.View) *tcpip.Error { +func send6(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Error { if len(data) < header.ICMPv6EchoMinimumSize { return tcpip.ErrInvalidEndpointState } @@ -409,21 +449,24 @@ func send6(r *stack.Route, ident uint16, data buffer.View) *tcpip.Error { return tcpip.ErrInvalidEndpointState } - icmpv6.SetChecksum(0) - icmpv6.SetChecksum(^header.Checksum(icmpv6, header.Checksum(data, 0))) + dataVV := data.ToVectorisedView() + icmpv6.SetChecksum(header.ICMPv6Checksum(icmpv6, r.LocalAddress, r.RemoteAddress, dataVV)) - return r.WritePacket(nil /* gso */, hdr, data.ToVectorisedView(), header.ICMPv6ProtocolNumber, r.DefaultTTL()) + if ttl == 0 { + ttl = r.DefaultTTL() + } + return r.WritePacket(nil /* gso */, hdr, dataVV, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}) } func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress, allowMismatch bool) (tcpip.NetworkProtocolNumber, *tcpip.Error) { - netProto := e.netProto + netProto := e.NetProto if header.IsV4MappedAddress(addr.Addr) { return 0, tcpip.ErrNoRoute } // Fail if we're bound to an address length different from the one we're // checking. - if l := len(e.id.LocalAddress); !allowMismatch && l != 0 && l != len(addr.Addr) { + if l := len(e.ID.LocalAddress); !allowMismatch && l != 0 && l != len(addr.Addr) { return 0, tcpip.ErrInvalidEndpointState } @@ -440,20 +483,20 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { e.mu.Lock() defer e.mu.Unlock() - nicid := addr.NIC + nicID := addr.NIC localPort := uint16(0) switch e.state { case stateBound, stateConnected: - localPort = e.id.LocalPort - if e.bindNICID == 0 { + localPort = e.ID.LocalPort + if e.BindNICID == 0 { break } - if nicid != 0 && nicid != e.bindNICID { + if nicID != 0 && nicID != e.BindNICID { return tcpip.ErrInvalidEndpointState } - nicid = e.bindNICID + nicID = e.BindNICID default: return tcpip.ErrInvalidEndpointState } @@ -464,7 +507,7 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { } // Find a route to the desired destination. - r, err := e.stack.FindRoute(nicid, e.bindAddr, addr.Addr, netProto, false /* multicastLoop */) + r, err := e.stack.FindRoute(nicID, e.BindAddr, addr.Addr, netProto, false /* multicastLoop */) if err != nil { return err } @@ -481,14 +524,14 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { // v6only is set to false and this is an ipv6 endpoint. netProtos := []tcpip.NetworkProtocolNumber{netProto} - id, err = e.registerWithStack(nicid, netProtos, id) + id, err = e.registerWithStack(nicID, netProtos, id) if err != nil { return err } - e.id = id + e.ID = id e.route = r.Clone() - e.regNICID = nicid + e.RegisterNICID = nicID e.state = stateConnected @@ -539,18 +582,18 @@ func (*endpoint) Accept() (tcpip.Endpoint, *waiter.Queue, *tcpip.Error) { return nil, nil, tcpip.ErrNotSupported } -func (e *endpoint) registerWithStack(nicid tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, id stack.TransportEndpointID) (stack.TransportEndpointID, *tcpip.Error) { +func (e *endpoint) registerWithStack(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, id stack.TransportEndpointID) (stack.TransportEndpointID, *tcpip.Error) { if id.LocalPort != 0 { // The endpoint already has a local port, just attempt to // register it. - err := e.stack.RegisterTransportEndpoint(nicid, netProtos, e.transProto, id, e, false) + err := e.stack.RegisterTransportEndpoint(nicID, netProtos, e.TransProto, id, e, false /* reuse */, 0 /* bindToDevice */) return id, err } // We need to find a port for the endpoint. _, err := e.stack.PickEphemeralPort(func(p uint16) (bool, *tcpip.Error) { id.LocalPort = p - err := e.stack.RegisterTransportEndpoint(nicid, netProtos, e.transProto, id, e, false) + err := e.stack.RegisterTransportEndpoint(nicID, netProtos, e.TransProto, id, e, false /* reuse */, 0 /* bindtodevice */) switch err { case nil: return true, nil @@ -597,8 +640,8 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) *tcpip.Error { return err } - e.id = id - e.regNICID = addr.NIC + e.ID = id + e.RegisterNICID = addr.NIC // Mark endpoint as bound. e.state = stateBound @@ -621,8 +664,8 @@ func (e *endpoint) Bind(addr tcpip.FullAddress) *tcpip.Error { return err } - e.bindNICID = addr.NIC - e.bindAddr = addr.Addr + e.BindNICID = addr.NIC + e.BindAddr = addr.Addr return nil } @@ -633,9 +676,9 @@ func (e *endpoint) GetLocalAddress() (tcpip.FullAddress, *tcpip.Error) { defer e.mu.RUnlock() return tcpip.FullAddress{ - NIC: e.regNICID, - Addr: e.id.LocalAddress, - Port: e.id.LocalPort, + NIC: e.RegisterNICID, + Addr: e.ID.LocalAddress, + Port: e.ID.LocalPort, }, nil } @@ -649,9 +692,9 @@ func (e *endpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Error) { } return tcpip.FullAddress{ - NIC: e.regNICID, - Addr: e.id.RemoteAddress, - Port: e.id.RemotePort, + NIC: e.RegisterNICID, + Addr: e.ID.RemoteAddress, + Port: e.ID.RemotePort, }, nil } @@ -675,19 +718,21 @@ func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { // HandlePacket is called by the stack when new packets arrive to this transport // endpoint. -func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, vv buffer.VectorisedView) { +func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt tcpip.PacketBuffer) { // Only accept echo replies. - switch e.netProto { + switch e.NetProto { case header.IPv4ProtocolNumber: - h := header.ICMPv4(vv.First()) + h := header.ICMPv4(pkt.Data.First()) if h.Type() != header.ICMPv4EchoReply { e.stack.Stats().DroppedPackets.Increment() + e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() return } case header.IPv6ProtocolNumber: - h := header.ICMPv6(vv.First()) + h := header.ICMPv6(pkt.Data.First()) if h.Type() != header.ICMPv6EchoReply { e.stack.Stats().DroppedPackets.Increment() + e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() return } } @@ -695,31 +740,39 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, vv e.rcvMu.Lock() // Drop the packet if our buffer is currently full. - if !e.rcvReady || e.rcvClosed || e.rcvBufSize >= e.rcvBufSizeMax { + if !e.rcvReady || e.rcvClosed { + e.rcvMu.Unlock() e.stack.Stats().DroppedPackets.Increment() + e.stats.ReceiveErrors.ClosedReceiver.Increment() + return + } + + if e.rcvBufSize >= e.rcvBufSizeMax { e.rcvMu.Unlock() + e.stack.Stats().DroppedPackets.Increment() + e.stats.ReceiveErrors.ReceiveBufferOverflow.Increment() return } wasEmpty := e.rcvBufSize == 0 // Push new packet into receive list and increment the buffer size. - pkt := &icmpPacket{ + packet := &icmpPacket{ senderAddress: tcpip.FullAddress{ NIC: r.NICID(), Addr: id.RemoteAddress, }, } - pkt.data = vv.Clone(pkt.views[:]) + packet.data = pkt.Data - e.rcvList.PushBack(pkt) - e.rcvBufSize += pkt.data.Size() + e.rcvList.PushBack(packet) + e.rcvBufSize += packet.data.Size() - pkt.timestamp = e.stack.NowNanoseconds() + packet.timestamp = e.stack.NowNanoseconds() e.rcvMu.Unlock() - + e.stats.PacketsReceived.Increment() // Notify any waiters that there's data to be read now. if wasEmpty { e.waiterQueue.Notify(waiter.EventIn) @@ -727,7 +780,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, vv } // HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket. -func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, vv buffer.VectorisedView) { +func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, pkt tcpip.PacketBuffer) { } // State implements tcpip.Endpoint.State. The ICMP endpoint currently doesn't @@ -735,3 +788,20 @@ func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.C func (e *endpoint) State() uint32 { return 0 } + +// Info returns a copy of the endpoint info. +func (e *endpoint) Info() tcpip.EndpointInfo { + e.mu.RLock() + // Make a copy of the endpoint info. + ret := e.TransportEndpointInfo + e.mu.RUnlock() + return &ret +} + +// Stats returns a pointer to the endpoint stats. +func (e *endpoint) Stats() tcpip.EndpointStats { + return &e.stats +} + +// Wait implements stack.TransportEndpoint.Wait. +func (*endpoint) Wait() {} diff --git a/pkg/tcpip/transport/icmp/endpoint_state.go b/pkg/tcpip/transport/icmp/endpoint_state.go index c587b96b6..9d263c0ec 100644 --- a/pkg/tcpip/transport/icmp/endpoint_state.go +++ b/pkg/tcpip/transport/icmp/endpoint_state.go @@ -76,19 +76,19 @@ func (e *endpoint) Resume(s *stack.Stack) { var err *tcpip.Error if e.state == stateConnected { - e.route, err = e.stack.FindRoute(e.regNICID, e.bindAddr, e.id.RemoteAddress, e.netProto, false /* multicastLoop */) + e.route, err = e.stack.FindRoute(e.RegisterNICID, e.BindAddr, e.ID.RemoteAddress, e.NetProto, false /* multicastLoop */) if err != nil { panic(err) } - e.id.LocalAddress = e.route.LocalAddress - } else if len(e.id.LocalAddress) != 0 { // stateBound - if e.stack.CheckLocalAddress(e.regNICID, e.netProto, e.id.LocalAddress) == 0 { + e.ID.LocalAddress = e.route.LocalAddress + } else if len(e.ID.LocalAddress) != 0 { // stateBound + if e.stack.CheckLocalAddress(e.RegisterNICID, e.NetProto, e.ID.LocalAddress) == 0 { panic(tcpip.ErrBadLocalAddress) } } - e.id, err = e.registerWithStack(e.regNICID, []tcpip.NetworkProtocolNumber{e.netProto}, e.id) + e.ID, err = e.registerWithStack(e.RegisterNICID, []tcpip.NetworkProtocolNumber{e.NetProto}, e.ID) if err != nil { panic(err) } diff --git a/pkg/tcpip/transport/icmp/protocol.go b/pkg/tcpip/transport/icmp/protocol.go index bfb16f7c3..9ce500e80 100644 --- a/pkg/tcpip/transport/icmp/protocol.go +++ b/pkg/tcpip/transport/icmp/protocol.go @@ -104,7 +104,7 @@ func (p *protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) // HandleUnknownDestinationPacket handles packets targeted at this protocol but // that don't match any existing endpoint. -func (p *protocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, buffer.View, buffer.VectorisedView) bool { +func (p *protocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, tcpip.PacketBuffer) bool { return true } diff --git a/pkg/tcpip/transport/packet/BUILD b/pkg/tcpip/transport/packet/BUILD new file mode 100644 index 000000000..8ea2e6ee5 --- /dev/null +++ b/pkg/tcpip/transport/packet/BUILD @@ -0,0 +1,46 @@ +load("//tools/go_generics:defs.bzl", "go_template_instance") +load("//tools/go_stateify:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_template_instance( + name = "packet_list", + out = "packet_list.go", + package = "packet", + prefix = "packet", + template = "//pkg/ilist:generic_list", + types = { + "Element": "*packet", + "Linker": "*packet", + }, +) + +go_library( + name = "packet", + srcs = [ + "endpoint.go", + "endpoint_state.go", + "packet_list.go", + ], + importpath = "gvisor.dev/gvisor/pkg/tcpip/transport/packet", + imports = ["gvisor.dev/gvisor/pkg/tcpip/buffer"], + visibility = ["//visibility:public"], + deps = [ + "//pkg/log", + "//pkg/sleep", + "//pkg/tcpip", + "//pkg/tcpip/buffer", + "//pkg/tcpip/header", + "//pkg/tcpip/iptables", + "//pkg/tcpip/stack", + "//pkg/waiter", + ], +) + +filegroup( + name = "autogen", + srcs = [ + "packet_list.go", + ], + visibility = ["//:sandbox"], +) diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go new file mode 100644 index 000000000..0010b5e5f --- /dev/null +++ b/pkg/tcpip/transport/packet/endpoint.go @@ -0,0 +1,362 @@ +// 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 packet provides the implementation of packet sockets (see +// packet(7)). Packet sockets allow applications to: +// +// * manually write and inspect link, network, and transport headers +// * receive all traffic of a given network protocol, or all protocols +// +// Packet sockets are similar to raw sockets, but provide even more power to +// users, letting them effectively talk directly to the network device. +// +// Packet sockets skip the input and output iptables chains. +package packet + +import ( + "sync" + + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/iptables" + "gvisor.dev/gvisor/pkg/tcpip/stack" + "gvisor.dev/gvisor/pkg/waiter" +) + +// +stateify savable +type packet struct { + packetEntry + // data holds the actual packet data, including any headers and + // payload. + data buffer.VectorisedView `state:".(buffer.VectorisedView)"` + // timestampNS is the unix time at which the packet was received. + timestampNS int64 + // senderAddr is the network address of the sender. + senderAddr tcpip.FullAddress +} + +// endpoint is the packet socket implementation of tcpip.Endpoint. It is legal +// to have goroutines make concurrent calls into the endpoint. +// +// Lock order: +// endpoint.mu +// endpoint.rcvMu +// +// +stateify savable +type endpoint struct { + stack.TransportEndpointInfo + // The following fields are initialized at creation time and are + // immutable. + stack *stack.Stack `state:"manual"` + netProto tcpip.NetworkProtocolNumber + waiterQueue *waiter.Queue + cooked bool + + // The following fields are used to manage the receive queue and are + // protected by rcvMu. + rcvMu sync.Mutex `state:"nosave"` + rcvList packetList + rcvBufSizeMax int `state:".(int)"` + rcvBufSize int + rcvClosed bool + + // The following fields are protected by mu. + mu sync.RWMutex `state:"nosave"` + sndBufSize int + closed bool + stats tcpip.TransportEndpointStats `state:"nosave"` +} + +// NewEndpoint returns a new packet endpoint. +func NewEndpoint(s *stack.Stack, cooked bool, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { + ep := &endpoint{ + stack: s, + TransportEndpointInfo: stack.TransportEndpointInfo{ + NetProto: netProto, + }, + cooked: cooked, + netProto: netProto, + waiterQueue: waiterQueue, + rcvBufSizeMax: 32 * 1024, + sndBufSize: 32 * 1024, + } + + if err := s.RegisterPacketEndpoint(0, netProto, ep); err != nil { + return nil, err + } + return ep, nil +} + +// Close implements tcpip.Endpoint.Close. +func (ep *endpoint) Close() { + ep.mu.Lock() + defer ep.mu.Unlock() + + if ep.closed { + return + } + + ep.stack.UnregisterPacketEndpoint(0, ep.netProto, ep) + + ep.rcvMu.Lock() + defer ep.rcvMu.Unlock() + + // Clear the receive list. + ep.rcvClosed = true + ep.rcvBufSize = 0 + for !ep.rcvList.Empty() { + ep.rcvList.Remove(ep.rcvList.Front()) + } + + ep.closed = true + ep.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut) +} + +// ModerateRecvBuf implements tcpip.Endpoint.ModerateRecvBuf. +func (ep *endpoint) ModerateRecvBuf(copied int) {} + +// IPTables implements tcpip.Endpoint.IPTables. +func (ep *endpoint) IPTables() (iptables.IPTables, error) { + return ep.stack.IPTables(), nil +} + +// Read implements tcpip.Endpoint.Read. +func (ep *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) { + ep.rcvMu.Lock() + + // If there's no data to read, return that read would block or that the + // endpoint is closed. + if ep.rcvList.Empty() { + err := tcpip.ErrWouldBlock + if ep.rcvClosed { + ep.stats.ReadErrors.ReadClosed.Increment() + err = tcpip.ErrClosedForReceive + } + ep.rcvMu.Unlock() + return buffer.View{}, tcpip.ControlMessages{}, err + } + + packet := ep.rcvList.Front() + ep.rcvList.Remove(packet) + ep.rcvBufSize -= packet.data.Size() + + ep.rcvMu.Unlock() + + if addr != nil { + *addr = packet.senderAddr + } + + return packet.data.ToView(), tcpip.ControlMessages{HasTimestamp: true, Timestamp: packet.timestampNS}, nil +} + +func (ep *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { + // TODO(b/129292371): Implement. + return 0, nil, tcpip.ErrInvalidOptionValue +} + +// Peek implements tcpip.Endpoint.Peek. +func (ep *endpoint) Peek([][]byte) (int64, tcpip.ControlMessages, *tcpip.Error) { + return 0, tcpip.ControlMessages{}, nil +} + +// Disconnect implements tcpip.Endpoint.Disconnect. Packet sockets cannot be +// disconnected, and this function always returns tpcip.ErrNotSupported. +func (*endpoint) Disconnect() *tcpip.Error { + return tcpip.ErrNotSupported +} + +// Connect implements tcpip.Endpoint.Connect. Packet sockets cannot be +// connected, and this function always returnes tcpip.ErrNotSupported. +func (ep *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { + return tcpip.ErrNotSupported +} + +// Shutdown implements tcpip.Endpoint.Shutdown. Packet sockets cannot be used +// with Shutdown, and this function always returns tcpip.ErrNotSupported. +func (ep *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { + return tcpip.ErrNotSupported +} + +// Listen implements tcpip.Endpoint.Listen. Packet sockets cannot be used with +// Listen, and this function always returns tcpip.ErrNotSupported. +func (ep *endpoint) Listen(backlog int) *tcpip.Error { + return tcpip.ErrNotSupported +} + +// Accept implements tcpip.Endpoint.Accept. Packet sockets cannot be used with +// Accept, and this function always returns tcpip.ErrNotSupported. +func (ep *endpoint) Accept() (tcpip.Endpoint, *waiter.Queue, *tcpip.Error) { + return nil, nil, tcpip.ErrNotSupported +} + +// Bind implements tcpip.Endpoint.Bind. +func (ep *endpoint) Bind(addr tcpip.FullAddress) *tcpip.Error { + // TODO(gvisor.dev/issue/173): Add Bind support. + + // "By default, all packets of the specified protocol type are passed + // to a packet socket. To get packets only from a specific interface + // use bind(2) specifying an address in a struct sockaddr_ll to bind + // the packet socket to an interface. Fields used for binding are + // sll_family (should be AF_PACKET), sll_protocol, and sll_ifindex." + // - packet(7). + + return tcpip.ErrNotSupported +} + +// GetLocalAddress implements tcpip.Endpoint.GetLocalAddress. +func (ep *endpoint) GetLocalAddress() (tcpip.FullAddress, *tcpip.Error) { + return tcpip.FullAddress{}, tcpip.ErrNotSupported +} + +// GetRemoteAddress implements tcpip.Endpoint.GetRemoteAddress. +func (ep *endpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Error) { + // Even a connected socket doesn't return a remote address. + return tcpip.FullAddress{}, tcpip.ErrNotConnected +} + +// Readiness implements tcpip.Endpoint.Readiness. +func (ep *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { + // The endpoint is always writable. + result := waiter.EventOut & mask + + // Determine whether the endpoint is readable. + if (mask & waiter.EventIn) != 0 { + ep.rcvMu.Lock() + if !ep.rcvList.Empty() || ep.rcvClosed { + result |= waiter.EventIn + } + ep.rcvMu.Unlock() + } + + return result +} + +// SetSockOpt implements tcpip.Endpoint.SetSockOpt. Packet sockets cannot be +// used with SetSockOpt, and this function always returns +// tcpip.ErrNotSupported. +func (ep *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { + return tcpip.ErrNotSupported +} + +// SetSockOptInt implements tcpip.Endpoint.SetSockOptInt. +func (ep *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error { + return tcpip.ErrUnknownProtocolOption +} + +// GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. +func (ep *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) { + return 0, tcpip.ErrNotSupported +} + +// GetSockOpt implements tcpip.Endpoint.GetSockOpt. +func (ep *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { + return tcpip.ErrNotSupported +} + +// HandlePacket implements stack.PacketEndpoint.HandlePacket. +func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { + ep.rcvMu.Lock() + + // Drop the packet if our buffer is currently full. + if ep.rcvClosed { + ep.rcvMu.Unlock() + ep.stack.Stats().DroppedPackets.Increment() + ep.stats.ReceiveErrors.ClosedReceiver.Increment() + return + } + + if ep.rcvBufSize >= ep.rcvBufSizeMax { + ep.rcvMu.Unlock() + ep.stack.Stats().DroppedPackets.Increment() + ep.stats.ReceiveErrors.ReceiveBufferOverflow.Increment() + return + } + + wasEmpty := ep.rcvBufSize == 0 + + // Push new packet into receive list and increment the buffer size. + var packet packet + // TODO(b/129292371): Return network protocol. + if len(pkt.LinkHeader) > 0 { + // Get info directly from the ethernet header. + hdr := header.Ethernet(pkt.LinkHeader) + packet.senderAddr = tcpip.FullAddress{ + NIC: nicID, + Addr: tcpip.Address(hdr.SourceAddress()), + } + } else { + // Guess the would-be ethernet header. + packet.senderAddr = tcpip.FullAddress{ + NIC: nicID, + Addr: tcpip.Address(localAddr), + } + } + + if ep.cooked { + // Cooked packets can simply be queued. + packet.data = pkt.Data + } 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, + } + fakeHeader := make(header.Ethernet, header.EthernetMinimumSize) + fakeHeader.Encode(ðFields) + linkHeader = buffer.View(fakeHeader) + } else { + linkHeader = append(buffer.View(nil), pkt.LinkHeader...) + } + combinedVV := linkHeader.ToVectorisedView() + combinedVV.Append(pkt.Data) + packet.data = combinedVV + } + packet.timestampNS = ep.stack.NowNanoseconds() + + ep.rcvList.PushBack(&packet) + ep.rcvBufSize += packet.data.Size() + + ep.rcvMu.Unlock() + ep.stats.PacketsReceived.Increment() + // Notify waiters that there's data to be read. + if wasEmpty { + ep.waiterQueue.Notify(waiter.EventIn) + } +} + +// State implements socket.Socket.State. +func (ep *endpoint) State() uint32 { + return 0 +} + +// Info returns a copy of the endpoint info. +func (ep *endpoint) Info() tcpip.EndpointInfo { + ep.mu.RLock() + // Make a copy of the endpoint info. + ret := ep.TransportEndpointInfo + ep.mu.RUnlock() + return &ret +} + +// Stats returns a pointer to the endpoint stats. +func (ep *endpoint) Stats() tcpip.EndpointStats { + return &ep.stats +} diff --git a/pkg/tcpip/transport/packet/endpoint_state.go b/pkg/tcpip/transport/packet/endpoint_state.go new file mode 100644 index 000000000..9b88f17e4 --- /dev/null +++ b/pkg/tcpip/transport/packet/endpoint_state.go @@ -0,0 +1,72 @@ +// Copyright 2018 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 packet + +import ( + "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/stack" +) + +// saveData saves packet.data field. +func (p *packet) saveData() buffer.VectorisedView { + // We cannot save p.data directly as p.data.views may alias to p.views, + // which is not allowed by state framework (in-struct pointer). + return p.data.Clone(nil) +} + +// loadData loads packet.data field. +func (p *packet) loadData(data buffer.VectorisedView) { + // NOTE: We cannot do the p.data = data.Clone(p.views[:]) optimization + // here because data.views is not guaranteed to be loaded by now. Plus, + // data.views will be allocated anyway so there really is little point + // of utilizing p.views for data.views. + p.data = data +} + +// beforeSave is invoked by stateify. +func (ep *endpoint) beforeSave() { + // Stop incoming packets from being handled (and mutate endpoint state). + // The lock will be released after saveRcvBufSizeMax(), which would have + // saved ep.rcvBufSizeMax and set it to 0 to continue blocking incoming + // packets. + ep.rcvMu.Lock() +} + +// saveRcvBufSizeMax is invoked by stateify. +func (ep *endpoint) saveRcvBufSizeMax() int { + max := ep.rcvBufSizeMax + // Make sure no new packets will be handled regardless of the lock. + ep.rcvBufSizeMax = 0 + // Release the lock acquired in beforeSave() so regular endpoint closing + // logic can proceed after save. + ep.rcvMu.Unlock() + return max +} + +// loadRcvBufSizeMax is invoked by stateify. +func (ep *endpoint) loadRcvBufSizeMax(max int) { + ep.rcvBufSizeMax = max +} + +// afterLoad is invoked by stateify. +func (ep *endpoint) afterLoad() { + // StackFromEnv is a stack used specifically for save/restore. + ep.stack = stack.StackFromEnv + + // TODO(gvisor.dev/173): Once bind is supported, choose the right NIC. + if err := ep.stack.RegisterPacketEndpoint(0, ep.netProto, ep); err != nil { + panic(*err) + } +} diff --git a/pkg/tcpip/transport/raw/BUILD b/pkg/tcpip/transport/raw/BUILD index 7241f6c19..4af49218c 100644 --- a/pkg/tcpip/transport/raw/BUILD +++ b/pkg/tcpip/transport/raw/BUILD @@ -1,17 +1,17 @@ -package(licenses = ["notice"]) - load("//tools/go_generics:defs.bzl", "go_template_instance") load("//tools/go_stateify:defs.bzl", "go_library") +package(licenses = ["notice"]) + go_template_instance( - name = "packet_list", - out = "packet_list.go", + name = "raw_packet_list", + out = "raw_packet_list.go", package = "raw", - prefix = "packet", + prefix = "rawPacket", template = "//pkg/ilist:generic_list", types = { - "Element": "*packet", - "Linker": "*packet", + "Element": "*rawPacket", + "Linker": "*rawPacket", }, ) @@ -20,8 +20,8 @@ go_library( srcs = [ "endpoint.go", "endpoint_state.go", - "packet_list.go", "protocol.go", + "raw_packet_list.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/transport/raw", imports = ["gvisor.dev/gvisor/pkg/tcpip/buffer"], @@ -34,6 +34,7 @@ go_library( "//pkg/tcpip/header", "//pkg/tcpip/iptables", "//pkg/tcpip/stack", + "//pkg/tcpip/transport/packet", "//pkg/waiter", ], ) @@ -41,7 +42,7 @@ go_library( filegroup( name = "autogen", srcs = [ - "packet_list.go", + "raw_packet_list.go", ], visibility = ["//:sandbox"], ) diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index a02731a5d..230a1537a 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -17,8 +17,7 @@ // // * manually write and inspect transport layer headers and payloads // * receive all traffic of a given transport protocol (e.g. ICMP or UDP) -// * optionally write and inspect network layer and link layer headers for -// packets +// * optionally write and inspect network layer headers of packets // // Raw sockets don't have any notion of ports, and incoming packets are // demultiplexed solely by protocol number. Thus, a raw UDP endpoint will @@ -38,15 +37,11 @@ import ( ) // +stateify savable -type packet struct { - packetEntry +type rawPacket struct { + rawPacketEntry // data holds the actual packet data, including any headers and // payload. data buffer.VectorisedView `state:".(buffer.VectorisedView)"` - // views is pre-allocated space to back data. As long as the packet is - // made up of fewer than 8 buffer.Views, no extra allocation is - // necessary to store packet data. - views [8]buffer.View `state:"nosave"` // timestampNS is the unix time at which the packet was received. timestampNS int64 // senderAddr is the network address of the sender. @@ -62,18 +57,17 @@ type packet struct { // // +stateify savable type endpoint struct { + stack.TransportEndpointInfo // The following fields are initialized at creation time and are // immutable. stack *stack.Stack `state:"manual"` - netProto tcpip.NetworkProtocolNumber - transProto tcpip.TransportProtocolNumber waiterQueue *waiter.Queue associated bool // The following fields are used to manage the receive queue and are // protected by rcvMu. rcvMu sync.Mutex `state:"nosave"` - rcvList packetList + rcvList rawPacketList rcvBufSizeMax int `state:".(int)"` rcvBufSize int rcvClosed bool @@ -84,35 +78,28 @@ type endpoint struct { closed bool connected bool bound bool - // registeredNIC is the NIC to which th endpoint is explicitly - // registered. Is set when Connect or Bind are used to specify a NIC. - registeredNIC tcpip.NICID - // boundNIC and boundAddr are set on calls to Bind(). When callers - // attempt actions that would invalidate the binding data (e.g. sending - // data via a NIC other than boundNIC), the endpoint will return an - // error. - boundNIC tcpip.NICID - boundAddr tcpip.Address // route is the route to a remote network endpoint. It is set via // Connect(), and is valid only when conneted is true. - route stack.Route `state:"manual"` + route stack.Route `state:"manual"` + stats tcpip.TransportEndpointStats `state:"nosave"` } // NewEndpoint returns a raw endpoint for the given protocols. -// TODO(b/129292371): IP_HDRINCL and AF_PACKET. func NewEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { return newEndpoint(stack, netProto, transProto, waiterQueue, true /* associated */) } -func newEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue, associated bool) (tcpip.Endpoint, *tcpip.Error) { +func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue, associated bool) (tcpip.Endpoint, *tcpip.Error) { if netProto != header.IPv4ProtocolNumber { return nil, tcpip.ErrUnknownProtocol } - ep := &endpoint{ - stack: stack, - netProto: netProto, - transProto: transProto, + e := &endpoint{ + stack: s, + TransportEndpointInfo: stack.TransportEndpointInfo{ + NetProto: netProto, + TransProto: transProto, + }, waiterQueue: waiterQueue, rcvBufSizeMax: 32 * 1024, sndBufSize: 32 * 1024, @@ -123,114 +110,139 @@ func newEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, trans // headers included. Because they're write-only, We don't need to // register with the stack. if !associated { - ep.rcvBufSizeMax = 0 - ep.waiterQueue = nil - return ep, nil + e.rcvBufSizeMax = 0 + e.waiterQueue = nil + return e, nil } - if err := ep.stack.RegisterRawTransportEndpoint(ep.registeredNIC, ep.netProto, ep.transProto, ep); err != nil { + if err := e.stack.RegisterRawTransportEndpoint(e.RegisterNICID, e.NetProto, e.TransProto, e); err != nil { return nil, err } - return ep, nil + return e, nil } // Close implements tcpip.Endpoint.Close. -func (ep *endpoint) Close() { - ep.mu.Lock() - defer ep.mu.Unlock() +func (e *endpoint) Close() { + e.mu.Lock() + defer e.mu.Unlock() - if ep.closed || !ep.associated { + if e.closed || !e.associated { return } - ep.stack.UnregisterRawTransportEndpoint(ep.registeredNIC, ep.netProto, ep.transProto, ep) + e.stack.UnregisterRawTransportEndpoint(e.RegisterNICID, e.NetProto, e.TransProto, e) - ep.rcvMu.Lock() - defer ep.rcvMu.Unlock() + e.rcvMu.Lock() + defer e.rcvMu.Unlock() // Clear the receive list. - ep.rcvClosed = true - ep.rcvBufSize = 0 - for !ep.rcvList.Empty() { - ep.rcvList.Remove(ep.rcvList.Front()) + e.rcvClosed = true + e.rcvBufSize = 0 + for !e.rcvList.Empty() { + e.rcvList.Remove(e.rcvList.Front()) } - if ep.connected { - ep.route.Release() - ep.connected = false + if e.connected { + e.route.Release() + e.connected = false } - ep.closed = true + e.closed = true - ep.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut) + e.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut) } // ModerateRecvBuf implements tcpip.Endpoint.ModerateRecvBuf. -func (ep *endpoint) ModerateRecvBuf(copied int) {} +func (e *endpoint) ModerateRecvBuf(copied int) {} // IPTables implements tcpip.Endpoint.IPTables. -func (ep *endpoint) IPTables() (iptables.IPTables, error) { - return ep.stack.IPTables(), nil +func (e *endpoint) IPTables() (iptables.IPTables, error) { + return e.stack.IPTables(), nil } // Read implements tcpip.Endpoint.Read. -func (ep *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) { - if !ep.associated { +func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) { + if !e.associated { return buffer.View{}, tcpip.ControlMessages{}, tcpip.ErrInvalidOptionValue } - ep.rcvMu.Lock() + e.rcvMu.Lock() // If there's no data to read, return that read would block or that the // endpoint is closed. - if ep.rcvList.Empty() { + if e.rcvList.Empty() { err := tcpip.ErrWouldBlock - if ep.rcvClosed { + if e.rcvClosed { + e.stats.ReadErrors.ReadClosed.Increment() err = tcpip.ErrClosedForReceive } - ep.rcvMu.Unlock() + e.rcvMu.Unlock() return buffer.View{}, tcpip.ControlMessages{}, err } - packet := ep.rcvList.Front() - ep.rcvList.Remove(packet) - ep.rcvBufSize -= packet.data.Size() + pkt := e.rcvList.Front() + e.rcvList.Remove(pkt) + e.rcvBufSize -= pkt.data.Size() - ep.rcvMu.Unlock() + e.rcvMu.Unlock() if addr != nil { - *addr = packet.senderAddr + *addr = pkt.senderAddr } - return packet.data.ToView(), tcpip.ControlMessages{HasTimestamp: true, Timestamp: packet.timestampNS}, nil + return pkt.data.ToView(), tcpip.ControlMessages{HasTimestamp: true, Timestamp: pkt.timestampNS}, nil } // Write implements tcpip.Endpoint.Write. -func (ep *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { +func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { + n, ch, err := e.write(p, opts) + switch err { + case nil: + e.stats.PacketsSent.Increment() + case tcpip.ErrMessageTooLong, tcpip.ErrInvalidOptionValue: + e.stats.WriteErrors.InvalidArgs.Increment() + case tcpip.ErrClosedForSend: + e.stats.WriteErrors.WriteClosed.Increment() + case tcpip.ErrInvalidEndpointState: + e.stats.WriteErrors.InvalidEndpointState.Increment() + case tcpip.ErrNoLinkAddress: + e.stats.SendErrors.NoLinkAddr.Increment() + case tcpip.ErrNoRoute, tcpip.ErrBroadcastDisabled, tcpip.ErrNetworkUnreachable: + // Errors indicating any problem with IP routing of the packet. + e.stats.SendErrors.NoRoute.Increment() + default: + // For all other errors when writing to the network layer. + e.stats.SendErrors.SendToNetworkFailed.Increment() + } + return n, ch, err +} + +func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { // MSG_MORE is unimplemented. This also means that MSG_EOR is a no-op. if opts.More { return 0, nil, tcpip.ErrInvalidOptionValue } - ep.mu.RLock() + e.mu.RLock() - if ep.closed { - ep.mu.RUnlock() + if e.closed { + e.mu.RUnlock() return 0, nil, tcpip.ErrInvalidEndpointState } payloadBytes, err := p.FullPayload() if err != nil { + e.mu.RUnlock() return 0, nil, err } // If this is an unassociated socket and callee provided a nonzero // destination address, route using that address. - if !ep.associated { + if !e.associated { ip := header.IPv4(payloadBytes) if !ip.IsValid(len(payloadBytes)) { - ep.mu.RUnlock() + e.mu.RUnlock() return 0, nil, tcpip.ErrInvalidOptionValue } dstAddr := ip.DestinationAddress() @@ -251,66 +263,66 @@ func (ep *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <- if opts.To == nil { // If the user doesn't specify a destination, they should have // connected to another address. - if !ep.connected { - ep.mu.RUnlock() + if !e.connected { + e.mu.RUnlock() return 0, nil, tcpip.ErrDestinationRequired } - if ep.route.IsResolutionRequired() { - savedRoute := &ep.route + if e.route.IsResolutionRequired() { + savedRoute := &e.route // Promote lock to exclusive if using a shared route, // given that it may need to change in finishWrite. - ep.mu.RUnlock() - ep.mu.Lock() + e.mu.RUnlock() + e.mu.Lock() // Make sure that the route didn't change during the // time we didn't hold the lock. - if !ep.connected || savedRoute != &ep.route { - ep.mu.Unlock() + if !e.connected || savedRoute != &e.route { + e.mu.Unlock() return 0, nil, tcpip.ErrInvalidEndpointState } - n, ch, err := ep.finishWrite(payloadBytes, savedRoute) - ep.mu.Unlock() + n, ch, err := e.finishWrite(payloadBytes, savedRoute) + e.mu.Unlock() return n, ch, err } - n, ch, err := ep.finishWrite(payloadBytes, &ep.route) - ep.mu.RUnlock() + n, ch, err := e.finishWrite(payloadBytes, &e.route) + e.mu.RUnlock() return n, ch, err } // The caller provided a destination. Reject destination address if it // goes through a different NIC than the endpoint was bound to. nic := opts.To.NIC - if ep.bound && nic != 0 && nic != ep.boundNIC { - ep.mu.RUnlock() + if e.bound && nic != 0 && nic != e.BindNICID { + e.mu.RUnlock() 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 { - ep.mu.RUnlock() + e.mu.RUnlock() return 0, nil, tcpip.ErrInvalidEndpointState } - // Find the route to the destination. If boundAddress is 0, + // Find the route to the destination. If BindAddress is 0, // FindRoute will choose an appropriate source address. - route, err := ep.stack.FindRoute(nic, ep.boundAddr, opts.To.Addr, ep.netProto, false) + route, err := e.stack.FindRoute(nic, e.BindAddr, opts.To.Addr, e.NetProto, false) if err != nil { - ep.mu.RUnlock() + e.mu.RUnlock() return 0, nil, err } - n, ch, err := ep.finishWrite(payloadBytes, &route) + n, ch, err := e.finishWrite(payloadBytes, &route) route.Release() - ep.mu.RUnlock() + e.mu.RUnlock() return n, ch, err } // finishWrite writes the payload to a route. It resolves the route if // necessary. It's really just a helper to make defer unnecessary in Write. -func (ep *endpoint) finishWrite(payloadBytes []byte, route *stack.Route) (int64, <-chan struct{}, *tcpip.Error) { +func (e *endpoint) finishWrite(payloadBytes []byte, route *stack.Route) (int64, <-chan struct{}, *tcpip.Error) { // We may need to resolve the route (match a link layer address to the // network address). If that requires blocking (e.g. to use ARP), // return a channel on which the caller can wait. @@ -323,16 +335,16 @@ func (ep *endpoint) finishWrite(payloadBytes []byte, route *stack.Route) (int64, } } - switch ep.netProto { + switch e.NetProto { case header.IPv4ProtocolNumber: - if !ep.associated { + if !e.associated { if err := route.WriteHeaderIncludedPacket(buffer.View(payloadBytes).ToVectorisedView()); err != nil { return 0, nil, err } break } hdr := buffer.NewPrependable(len(payloadBytes) + int(route.MaxHeaderLength())) - if err := route.WritePacket(nil /* gso */, hdr, buffer.View(payloadBytes).ToVectorisedView(), ep.transProto, route.DefaultTTL()); err != nil { + if err := route.WritePacket(nil /* gso */, hdr, buffer.View(payloadBytes).ToVectorisedView(), stack.NetworkHeaderParams{Protocol: e.TransProto, TTL: route.DefaultTTL(), TOS: stack.DefaultTOS}); err != nil { return 0, nil, err } @@ -344,7 +356,7 @@ func (ep *endpoint) finishWrite(payloadBytes []byte, route *stack.Route) (int64, } // Peek implements tcpip.Endpoint.Peek. -func (ep *endpoint) Peek([][]byte) (int64, tcpip.ControlMessages, *tcpip.Error) { +func (e *endpoint) Peek([][]byte) (int64, tcpip.ControlMessages, *tcpip.Error) { return 0, tcpip.ControlMessages{}, nil } @@ -354,11 +366,11 @@ func (*endpoint) Disconnect() *tcpip.Error { } // Connect implements tcpip.Endpoint.Connect. -func (ep *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { - ep.mu.Lock() - defer ep.mu.Unlock() +func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { + e.mu.Lock() + defer e.mu.Unlock() - if ep.closed { + if e.closed { return tcpip.ErrInvalidEndpointState } @@ -368,15 +380,15 @@ func (ep *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { } nic := addr.NIC - if ep.bound { - if ep.boundNIC == 0 { + if e.bound { + if e.BindNICID == 0 { // If we're bound, but not to a specific NIC, the NIC // in addr will be used. Nothing to do here. } else if addr.NIC == 0 { // If we're bound to a specific NIC, but addr doesn't // specify a NIC, use the bound NIC. - nic = ep.boundNIC - } else if addr.NIC != ep.boundNIC { + nic = e.BindNICID + } else if addr.NIC != e.BindNICID { // We're bound and addr specifies a NIC. They must be // the same. return tcpip.ErrInvalidEndpointState @@ -384,53 +396,53 @@ func (ep *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { } // Find a route to the destination. - route, err := ep.stack.FindRoute(nic, tcpip.Address(""), addr.Addr, ep.netProto, false) + route, err := e.stack.FindRoute(nic, tcpip.Address(""), addr.Addr, e.NetProto, false) if err != nil { return err } defer route.Release() - if ep.associated { + if e.associated { // Re-register the endpoint with the appropriate NIC. - if err := ep.stack.RegisterRawTransportEndpoint(addr.NIC, ep.netProto, ep.transProto, ep); err != nil { + if err := e.stack.RegisterRawTransportEndpoint(addr.NIC, e.NetProto, e.TransProto, e); err != nil { return err } - ep.stack.UnregisterRawTransportEndpoint(ep.registeredNIC, ep.netProto, ep.transProto, ep) - ep.registeredNIC = nic + e.stack.UnregisterRawTransportEndpoint(e.RegisterNICID, e.NetProto, e.TransProto, e) + e.RegisterNICID = nic } // Save the route we've connected via. - ep.route = route.Clone() - ep.connected = true + e.route = route.Clone() + e.connected = true return nil } // Shutdown implements tcpip.Endpoint.Shutdown. It's a noop for raw sockets. -func (ep *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { - ep.mu.Lock() - defer ep.mu.Unlock() +func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { + e.mu.Lock() + defer e.mu.Unlock() - if !ep.connected { + if !e.connected { return tcpip.ErrNotConnected } return nil } // Listen implements tcpip.Endpoint.Listen. -func (ep *endpoint) Listen(backlog int) *tcpip.Error { +func (e *endpoint) Listen(backlog int) *tcpip.Error { return tcpip.ErrNotSupported } // Accept implements tcpip.Endpoint.Accept. -func (ep *endpoint) Accept() (tcpip.Endpoint, *waiter.Queue, *tcpip.Error) { +func (e *endpoint) Accept() (tcpip.Endpoint, *waiter.Queue, *tcpip.Error) { return nil, nil, tcpip.ErrNotSupported } // Bind implements tcpip.Endpoint.Bind. -func (ep *endpoint) Bind(addr tcpip.FullAddress) *tcpip.Error { - ep.mu.Lock() - defer ep.mu.Unlock() +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). @@ -439,56 +451,56 @@ func (ep *endpoint) Bind(addr tcpip.FullAddress) *tcpip.Error { } // If a local address was specified, verify that it's valid. - if len(addr.Addr) == header.IPv4AddressSize && ep.stack.CheckLocalAddress(addr.NIC, ep.netProto, addr.Addr) == 0 { + if len(addr.Addr) == header.IPv4AddressSize && e.stack.CheckLocalAddress(addr.NIC, e.NetProto, addr.Addr) == 0 { return tcpip.ErrBadLocalAddress } - if ep.associated { + if e.associated { // Re-register the endpoint with the appropriate NIC. - if err := ep.stack.RegisterRawTransportEndpoint(addr.NIC, ep.netProto, ep.transProto, ep); err != nil { + if err := e.stack.RegisterRawTransportEndpoint(addr.NIC, e.NetProto, e.TransProto, e); err != nil { return err } - ep.stack.UnregisterRawTransportEndpoint(ep.registeredNIC, ep.netProto, ep.transProto, ep) - ep.registeredNIC = addr.NIC - ep.boundNIC = addr.NIC + e.stack.UnregisterRawTransportEndpoint(e.RegisterNICID, e.NetProto, e.TransProto, e) + e.RegisterNICID = addr.NIC + e.BindNICID = addr.NIC } - ep.boundAddr = addr.Addr - ep.bound = true + e.BindAddr = addr.Addr + e.bound = true return nil } // GetLocalAddress implements tcpip.Endpoint.GetLocalAddress. -func (ep *endpoint) GetLocalAddress() (tcpip.FullAddress, *tcpip.Error) { +func (e *endpoint) GetLocalAddress() (tcpip.FullAddress, *tcpip.Error) { return tcpip.FullAddress{}, tcpip.ErrNotSupported } // GetRemoteAddress implements tcpip.Endpoint.GetRemoteAddress. -func (ep *endpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Error) { +func (e *endpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Error) { // Even a connected socket doesn't return a remote address. return tcpip.FullAddress{}, tcpip.ErrNotConnected } // Readiness implements tcpip.Endpoint.Readiness. -func (ep *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { +func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { // The endpoint is always writable. result := waiter.EventOut & mask // Determine whether the endpoint is readable. if (mask & waiter.EventIn) != 0 { - ep.rcvMu.Lock() - if !ep.rcvList.Empty() || ep.rcvClosed { + e.rcvMu.Lock() + if !e.rcvList.Empty() || e.rcvClosed { result |= waiter.EventIn } - ep.rcvMu.Unlock() + e.rcvMu.Unlock() } return result } // SetSockOpt implements tcpip.Endpoint.SetSockOpt. -func (ep *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { +func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { return tcpip.ErrUnknownProtocolOption } @@ -498,28 +510,28 @@ func (ep *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error { } // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. -func (ep *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) { +func (e *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) { switch opt { case tcpip.ReceiveQueueSizeOption: v := 0 - ep.rcvMu.Lock() - if !ep.rcvList.Empty() { - p := ep.rcvList.Front() + e.rcvMu.Lock() + if !e.rcvList.Empty() { + p := e.rcvList.Front() v = p.data.Size() } - ep.rcvMu.Unlock() + e.rcvMu.Unlock() return v, nil case tcpip.SendBufferSizeOption: - ep.mu.Lock() - v := ep.sndBufSize - ep.mu.Unlock() + e.mu.Lock() + v := e.sndBufSize + e.mu.Unlock() return v, nil case tcpip.ReceiveBufferSizeOption: - ep.rcvMu.Lock() - v := ep.rcvBufSizeMax - ep.rcvMu.Unlock() + e.rcvMu.Lock() + v := e.rcvBufSizeMax + e.rcvMu.Unlock() return v, nil } @@ -528,7 +540,7 @@ func (ep *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) { } // GetSockOpt implements tcpip.Endpoint.GetSockOpt. -func (ep *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { +func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { switch o := opt.(type) { case tcpip.ErrorOption: return nil @@ -543,63 +555,89 @@ func (ep *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { } // HandlePacket implements stack.RawTransportEndpoint.HandlePacket. -func (ep *endpoint) HandlePacket(route *stack.Route, netHeader buffer.View, vv buffer.VectorisedView) { - ep.rcvMu.Lock() +func (e *endpoint) HandlePacket(route *stack.Route, pkt tcpip.PacketBuffer) { + e.rcvMu.Lock() // Drop the packet if our buffer is currently full. - if ep.rcvClosed || ep.rcvBufSize >= ep.rcvBufSizeMax { - ep.stack.Stats().DroppedPackets.Increment() - ep.rcvMu.Unlock() + if e.rcvClosed { + e.rcvMu.Unlock() + e.stack.Stats().DroppedPackets.Increment() + e.stats.ReceiveErrors.ClosedReceiver.Increment() return } - if ep.bound { + if e.rcvBufSize >= e.rcvBufSizeMax { + e.rcvMu.Unlock() + e.stack.Stats().DroppedPackets.Increment() + e.stats.ReceiveErrors.ReceiveBufferOverflow.Increment() + return + } + + if e.bound { // If bound to a NIC, only accept data for that NIC. - if ep.boundNIC != 0 && ep.boundNIC != route.NICID() { - ep.rcvMu.Unlock() + if e.BindNICID != 0 && e.BindNICID != route.NICID() { + e.rcvMu.Unlock() return } // If bound to an address, only accept data for that address. - if ep.boundAddr != "" && ep.boundAddr != route.RemoteAddress { - ep.rcvMu.Unlock() + if e.BindAddr != "" && e.BindAddr != route.RemoteAddress { + e.rcvMu.Unlock() return } } // If connected, only accept packets from the remote address we // connected to. - if ep.connected && ep.route.RemoteAddress != route.RemoteAddress { - ep.rcvMu.Unlock() + if e.connected && e.route.RemoteAddress != route.RemoteAddress { + e.rcvMu.Unlock() return } - wasEmpty := ep.rcvBufSize == 0 + wasEmpty := e.rcvBufSize == 0 // Push new packet into receive list and increment the buffer size. - packet := &packet{ + packet := &rawPacket{ senderAddr: tcpip.FullAddress{ NIC: route.NICID(), Addr: route.RemoteAddress, }, } - combinedVV := netHeader.ToVectorisedView() - combinedVV.Append(vv) - packet.data = combinedVV.Clone(packet.views[:]) - packet.timestampNS = ep.stack.NowNanoseconds() - - ep.rcvList.PushBack(packet) - ep.rcvBufSize += packet.data.Size() + networkHeader := append(buffer.View(nil), pkt.NetworkHeader...) + combinedVV := networkHeader.ToVectorisedView() + combinedVV.Append(pkt.Data) + packet.data = combinedVV + packet.timestampNS = e.stack.NowNanoseconds() - ep.rcvMu.Unlock() + e.rcvList.PushBack(packet) + e.rcvBufSize += packet.data.Size() + e.rcvMu.Unlock() + e.stats.PacketsReceived.Increment() // Notify waiters that there's data to be read. if wasEmpty { - ep.waiterQueue.Notify(waiter.EventIn) + e.waiterQueue.Notify(waiter.EventIn) } } // State implements socket.Socket.State. -func (ep *endpoint) State() uint32 { +func (e *endpoint) State() uint32 { return 0 } + +// Info returns a copy of the endpoint info. +func (e *endpoint) Info() tcpip.EndpointInfo { + e.mu.RLock() + // Make a copy of the endpoint info. + ret := e.TransportEndpointInfo + e.mu.RUnlock() + return &ret +} + +// Stats returns a pointer to the endpoint stats. +func (e *endpoint) Stats() tcpip.EndpointStats { + return &e.stats +} + +// Wait implements stack.TransportEndpoint.Wait. +func (*endpoint) Wait() {} diff --git a/pkg/tcpip/transport/raw/endpoint_state.go b/pkg/tcpip/transport/raw/endpoint_state.go index 168953dec..33bfb56cd 100644 --- a/pkg/tcpip/transport/raw/endpoint_state.go +++ b/pkg/tcpip/transport/raw/endpoint_state.go @@ -20,15 +20,15 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/stack" ) -// saveData saves packet.data field. -func (p *packet) saveData() buffer.VectorisedView { +// saveData saves rawPacket.data field. +func (p *rawPacket) saveData() buffer.VectorisedView { // We cannot save p.data directly as p.data.views may alias to p.views, // which is not allowed by state framework (in-struct pointer). return p.data.Clone(nil) } -// loadData loads packet.data field. -func (p *packet) loadData(data buffer.VectorisedView) { +// loadData loads rawPacket.data field. +func (p *rawPacket) loadData(data buffer.VectorisedView) { // NOTE: We cannot do the p.data = data.Clone(p.views[:]) optimization // here because data.views is not guaranteed to be loaded by now. Plus, // data.views will be allocated anyway so there really is little point @@ -73,7 +73,7 @@ func (ep *endpoint) Resume(s *stack.Stack) { // If the endpoint is connected, re-connect. if ep.connected { var err *tcpip.Error - ep.route, err = ep.stack.FindRoute(ep.registeredNIC, ep.boundAddr, ep.route.RemoteAddress, ep.netProto, false) + ep.route, err = ep.stack.FindRoute(ep.RegisterNICID, ep.BindAddr, ep.route.RemoteAddress, ep.NetProto, false) if err != nil { panic(err) } @@ -81,12 +81,14 @@ func (ep *endpoint) Resume(s *stack.Stack) { // If the endpoint is bound, re-bind. if ep.bound { - if ep.stack.CheckLocalAddress(ep.registeredNIC, ep.netProto, ep.boundAddr) == 0 { + if ep.stack.CheckLocalAddress(ep.RegisterNICID, ep.NetProto, ep.BindAddr) == 0 { panic(tcpip.ErrBadLocalAddress) } } - if err := ep.stack.RegisterRawTransportEndpoint(ep.registeredNIC, ep.netProto, ep.transProto, ep); err != nil { - panic(err) + if ep.associated { + if err := ep.stack.RegisterRawTransportEndpoint(ep.RegisterNICID, ep.NetProto, ep.TransProto, ep); err != nil { + panic(err) + } } } diff --git a/pkg/tcpip/transport/raw/protocol.go b/pkg/tcpip/transport/raw/protocol.go index a2512d666..f30aa2a4a 100644 --- a/pkg/tcpip/transport/raw/protocol.go +++ b/pkg/tcpip/transport/raw/protocol.go @@ -17,13 +17,19 @@ package raw import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/stack" + "gvisor.dev/gvisor/pkg/tcpip/transport/packet" "gvisor.dev/gvisor/pkg/waiter" ) -// EndpointFactory implements stack.UnassociatedEndpointFactory. +// EndpointFactory implements stack.RawFactory. type EndpointFactory struct{} -// NewUnassociatedRawEndpoint implements stack.UnassociatedEndpointFactory. -func (EndpointFactory) NewUnassociatedRawEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { +// NewUnassociatedEndpoint implements stack.RawFactory.NewUnassociatedEndpoint. +func (EndpointFactory) NewUnassociatedEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { return newEndpoint(stack, netProto, transProto, waiterQueue, false /* associated */) } + +// NewPacketEndpoint implements stack.RawFactory.NewPacketEndpoint. +func (EndpointFactory) NewPacketEndpoint(stack *stack.Stack, cooked bool, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { + return packet.NewEndpoint(stack, cooked, netProto, waiterQueue) +} diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD index 39a839ab7..3f47b328d 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -1,10 +1,9 @@ load("@io_bazel_rules_go//go:def.bzl", "go_test") - -package(licenses = ["notice"]) - load("//tools/go_generics:defs.bzl", "go_template_instance") load("//tools/go_stateify:defs.bzl", "go_library") +package(licenses = ["notice"]) + go_template_instance( name = "tcp_segment_list", out = "tcp_segment_list.go", @@ -45,10 +44,12 @@ go_library( imports = ["gvisor.dev/gvisor/pkg/tcpip/buffer"], visibility = ["//visibility:public"], deps = [ + "//pkg/log", "//pkg/rand", "//pkg/sleep", "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/hash/jenkins", "//pkg/tcpip/header", "//pkg/tcpip/iptables", "//pkg/tcpip/seqnum", @@ -70,7 +71,7 @@ filegroup( go_test( name = "tcp_test", - size = "small", + size = "medium", srcs = [ "dual_stack_test.go", "sack_scoreboard_test.go", diff --git a/pkg/tcpip/transport/tcp/accept.go b/pkg/tcpip/transport/tcp/accept.go index 0802e984e..f24b51b91 100644 --- a/pkg/tcpip/transport/tcp/accept.go +++ b/pkg/tcpip/transport/tcp/accept.go @@ -25,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/rand" "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/tcpip/seqnum" "gvisor.dev/gvisor/pkg/tcpip/stack" @@ -229,7 +230,7 @@ func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, i } n := newEndpoint(l.stack, netProto, nil) n.v6only = l.v6only - n.id = s.id + n.ID = s.id n.boundNICID = s.route.NICID() n.route = s.route.Clone() n.effectiveNetProtos = []tcpip.NetworkProtocolNumber{s.route.NetProto} @@ -242,7 +243,7 @@ func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, i n.initGSO() // Register new endpoint so that packets are routed to it. - if err := n.stack.RegisterTransportEndpoint(n.boundNICID, n.effectiveNetProtos, ProtocolNumber, n.id, n, n.reusePort); err != nil { + if err := n.stack.RegisterTransportEndpoint(n.boundNICID, n.effectiveNetProtos, ProtocolNumber, n.ID, n, n.reusePort, n.bindToDevice); err != nil { n.Close() return nil, err } @@ -268,8 +269,8 @@ func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, i func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *header.TCPSynOptions) (*endpoint, *tcpip.Error) { // Create new endpoint. irs := s.sequenceNumber - cookie := l.createCookie(s.id, irs, encodeMSS(opts.MSS)) - ep, err := l.createConnectingEndpoint(s, cookie, irs, opts) + isn := generateSecureISN(s.id, l.stack.Seed()) + ep, err := l.createConnectingEndpoint(s, isn, irs, opts) if err != nil { return nil, err } @@ -288,9 +289,8 @@ func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *head // Perform the 3-way handshake. h := newHandshake(ep, seqnum.Size(ep.initialReceiveWindow())) - h.resetToSynRcvd(cookie, irs, opts) + h.resetToSynRcvd(isn, irs, opts) if err := h.execute(); err != nil { - ep.stack.Stats().TCP.FailedConnectionAttempts.Increment() ep.Close() if l.listenEP != nil { l.removePendingEndpoint(ep) @@ -298,7 +298,9 @@ func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *head return nil, err } ep.mu.Lock() + ep.stack.Stats().TCP.CurrentEstablished.Increment() ep.state = StateEstablished + ep.isConnectNotified = true ep.mu.Unlock() // Update the receive window scaling. We can't do it before the @@ -311,14 +313,14 @@ func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *head func (l *listenContext) addPendingEndpoint(n *endpoint) { l.pendingMu.Lock() - l.pendingEndpoints[n.id] = n + l.pendingEndpoints[n.ID] = n l.pending.Add(1) l.pendingMu.Unlock() } func (l *listenContext) removePendingEndpoint(n *endpoint) { l.pendingMu.Lock() - delete(l.pendingEndpoints, n.id) + delete(l.pendingEndpoints, n.ID) l.pending.Done() l.pendingMu.Unlock() } @@ -360,12 +362,19 @@ func (e *endpoint) handleSynSegment(ctx *listenContext, s *segment, opts *header defer decSynRcvdCount() defer e.decSynRcvdCount() defer s.decRef() + n, err := ctx.createEndpointAndPerformHandshake(s, opts) if err != nil { e.stack.Stats().TCP.FailedConnectionAttempts.Increment() + e.stats.FailedConnectionAttempts.Increment() return } ctx.removePendingEndpoint(n) + // Start the protocol goroutine. + wq := &waiter.Queue{} + n.startAcceptedLoop(wq) + e.stack.Stats().TCP.PassiveConnectionOpenings.Increment() + e.deliverAccepted(n) } @@ -399,6 +408,17 @@ func (e *endpoint) acceptQueueIsFull() bool { // handleListenSegment is called when a listening endpoint receives a segment // and needs to handle it. func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { + if s.flagsAreSet(header.TCPFlagSyn | header.TCPFlagAck) { + // RFC 793 section 3.4 page 35 (figure 12) outlines that a RST + // must be sent in response to a SYN-ACK while in the listen + // state to prevent completing a handshake from an old SYN. + e.sendTCP(&s.route, s.id, buffer.VectorisedView{}, e.ttl, e.sendTOS, header.TCPFlagRst, s.ackNumber, 0, 0, nil, nil) + return + } + + // TODO(b/143300739): Use the userMSS of the listening socket + // for accepted sockets. + switch s.flags { case header.TCPFlagSyn: opts := parseSynSegmentOptions(s) @@ -414,6 +434,7 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { } decSynRcvdCount() e.stack.Stats().TCP.ListenOverflowSynDrop.Increment() + e.stats.ReceiveErrors.ListenOverflowSynDrop.Increment() e.stack.Stats().DroppedPackets.Increment() return } else { @@ -421,6 +442,7 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { // is full then drop the syn. if e.acceptQueueIsFull() { e.stack.Stats().TCP.ListenOverflowSynDrop.Increment() + e.stats.ReceiveErrors.ListenOverflowSynDrop.Increment() e.stack.Stats().DroppedPackets.Increment() return } @@ -431,15 +453,14 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { // // Enable Timestamp option if the original syn did have // the timestamp option specified. - mss := mssForRoute(&s.route) synOpts := header.TCPSynOptions{ WS: -1, TS: opts.TS, TSVal: tcpTimeStamp(timeStampOffset()), TSEcr: opts.TSVal, - MSS: uint16(mss), + MSS: mssForRoute(&s.route), } - sendSynTCP(&s.route, s.id, header.TCPFlagSyn|header.TCPFlagAck, cookie, s.sequenceNumber+1, ctx.rcvWnd, synOpts) + e.sendSynTCP(&s.route, s.id, e.ttl, e.sendTOS, header.TCPFlagSyn|header.TCPFlagAck, cookie, s.sequenceNumber+1, ctx.rcvWnd, synOpts) e.stack.Stats().TCP.ListenOverflowSynCookieSent.Increment() } @@ -451,6 +472,7 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { // complete the connection at the time of retransmit if // the backlog has space. e.stack.Stats().TCP.ListenOverflowAckDrop.Increment() + e.stats.ReceiveErrors.ListenOverflowAckDrop.Increment() e.stack.Stats().DroppedPackets.Increment() return } @@ -505,6 +527,7 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { n, err := ctx.createConnectingEndpoint(s, s.ackNumber-1, s.sequenceNumber-1, rcvdSynOptions) if err != nil { e.stack.Stats().TCP.FailedConnectionAttempts.Increment() + e.stats.FailedConnectionAttempts.Increment() return } @@ -515,7 +538,9 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { n.tsOffset = 0 // Switch state to connected. + n.stack.Stats().TCP.CurrentEstablished.Increment() n.state = StateEstablished + n.isConnectNotified = true // Do the delivery in a separate goroutine so // that we don't block the listen loop in case @@ -526,6 +551,11 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { // number of goroutines as we do check before // entering here that there was at least some // space available in the backlog. + + // Start the protocol goroutine. + wq := &waiter.Queue{} + n.startAcceptedLoop(wq) + e.stack.Stats().TCP.PassiveConnectionOpenings.Increment() go e.deliverAccepted(n) } } @@ -536,7 +566,7 @@ func (e *endpoint) protocolListenLoop(rcvWnd seqnum.Size) *tcpip.Error { e.mu.Lock() v6only := e.v6only e.mu.Unlock() - ctx := newListenContext(e.stack, e, rcvWnd, v6only, e.netProto) + ctx := newListenContext(e.stack, e, rcvWnd, v6only, e.NetProto) defer func() { // Mark endpoint as closed. This will prevent goroutines running diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 21038a65a..a114c06c1 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -15,6 +15,7 @@ package tcp import ( + "encoding/binary" "sync" "time" @@ -22,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/hash/jenkins" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/seqnum" "gvisor.dev/gvisor/pkg/tcpip/stack" @@ -78,9 +80,6 @@ type handshake struct { // mss is the maximum segment size received from the peer. mss uint16 - // amss is the maximum segment size advertised by us to the peer. - amss uint16 - // sndWndScale is the send window scale, as defined in RFC 1323. A // negative value means no scaling is supported by the peer. sndWndScale int @@ -142,7 +141,32 @@ func (h *handshake) resetState() { h.flags = header.TCPFlagSyn h.ackNum = 0 h.mss = 0 - h.iss = seqnum.Value(uint32(b[0]) | uint32(b[1])<<8 | uint32(b[2])<<16 | uint32(b[3])<<24) + h.iss = generateSecureISN(h.ep.ID, h.ep.stack.Seed()) +} + +// generateSecureISN generates a secure Initial Sequence number based on the +// recommendation here https://tools.ietf.org/html/rfc6528#page-3. +func generateSecureISN(id stack.TransportEndpointID, seed uint32) seqnum.Value { + isnHasher := jenkins.Sum32(seed) + isnHasher.Write([]byte(id.LocalAddress)) + isnHasher.Write([]byte(id.RemoteAddress)) + portBuf := make([]byte, 2) + binary.LittleEndian.PutUint16(portBuf, id.LocalPort) + isnHasher.Write(portBuf) + binary.LittleEndian.PutUint16(portBuf, id.RemotePort) + isnHasher.Write(portBuf) + // The time period here is 64ns. This is similar to what linux uses + // generate a sequence number that overlaps less than one + // time per MSL (2 minutes). + // + // A 64ns clock ticks 10^9/64 = 15625000) times in a second. + // To wrap the whole 32 bit space would require + // 2^32/1562500 ~ 274 seconds. + // + // Which sort of guarantees that we won't reuse the ISN for a new + // connection for the same tuple for at least 274s. + isn := isnHasher.Sum32() + uint32(time.Now().UnixNano()>>6) + return seqnum.Value(isn) } // effectiveRcvWndScale returns the effective receive window scale to be used. @@ -238,6 +262,7 @@ func (h *handshake) synSentState(s *segment) *tcpip.Error { h.state = handshakeSynRcvd h.ep.mu.Lock() h.ep.state = StateSynRecv + ttl := h.ep.ttl h.ep.mu.Unlock() synOpts := header.TCPSynOptions{ WS: int(h.effectiveRcvWndScale()), @@ -251,8 +276,10 @@ func (h *handshake) synSentState(s *segment) *tcpip.Error { SACKPermitted: rcvSynOpts.SACKPermitted, MSS: h.ep.amss, } - sendSynTCP(&s.route, h.ep.id, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) - + if ttl == 0 { + ttl = s.route.DefaultTTL() + } + h.ep.sendSynTCP(&s.route, h.ep.ID, ttl, h.ep.sendTOS, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) return nil } @@ -296,7 +323,7 @@ func (h *handshake) synRcvdState(s *segment) *tcpip.Error { SACKPermitted: h.ep.sackPermitted, MSS: h.ep.amss, } - sendSynTCP(&s.route, h.ep.id, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) + h.ep.sendSynTCP(&s.route, h.ep.ID, h.ep.ttl, h.ep.sendTOS, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) return nil } @@ -383,6 +410,11 @@ func (h *handshake) resolveRoute() *tcpip.Error { switch index { case wakerForResolution: if _, err := h.ep.route.Resolve(resolutionWaker); err != tcpip.ErrWouldBlock { + if err == tcpip.ErrNoLinkAddress { + h.ep.stats.SendErrors.NoLinkAddr.Increment() + } else if err != nil { + h.ep.stats.SendErrors.NoRoute.Increment() + } // Either success (err == nil) or failure. return err } @@ -437,7 +469,7 @@ func (h *handshake) execute() *tcpip.Error { // Send the initial SYN segment and loop until the handshake is // completed. - h.ep.amss = mssForRoute(&h.ep.route) + h.ep.amss = calculateAdvertisedMSS(h.ep.userMSS, h.ep.route) synOpts := header.TCPSynOptions{ WS: h.rcvWndScale, @@ -460,7 +492,8 @@ func (h *handshake) execute() *tcpip.Error { synOpts.WS = -1 } } - sendSynTCP(&h.ep.route, h.ep.id, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) + h.ep.sendSynTCP(&h.ep.route, h.ep.ID, h.ep.ttl, h.ep.sendTOS, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) + for h.state != handshakeCompleted { switch index, _ := s.Fetch(true); index { case wakerForResend: @@ -469,7 +502,7 @@ func (h *handshake) execute() *tcpip.Error { return tcpip.ErrTimeout } rt.Reset(timeOut) - sendSynTCP(&h.ep.route, h.ep.id, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) + h.ep.sendSynTCP(&h.ep.route, h.ep.ID, h.ep.ttl, h.ep.sendTOS, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) case wakerForNotification: n := h.ep.fetchNotifications() @@ -579,24 +612,30 @@ func makeSynOptions(opts header.TCPSynOptions) []byte { return options[:offset] } -func sendSynTCP(r *stack.Route, id stack.TransportEndpointID, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts header.TCPSynOptions) *tcpip.Error { +func (e *endpoint) sendSynTCP(r *stack.Route, id stack.TransportEndpointID, ttl, tos uint8, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts header.TCPSynOptions) *tcpip.Error { options := makeSynOptions(opts) - err := sendTCP(r, id, buffer.VectorisedView{}, r.DefaultTTL(), flags, seq, ack, rcvWnd, options, nil) + // We ignore SYN send errors and let the callers re-attempt send. + if err := e.sendTCP(r, id, buffer.VectorisedView{}, ttl, tos, flags, seq, ack, rcvWnd, options, nil); err != nil { + e.stats.SendErrors.SynSendToNetworkFailed.Increment() + } putOptions(options) - return err + return nil } -// sendTCP sends a TCP segment with the provided options via the provided -// network endpoint and under the provided identity. -func sendTCP(r *stack.Route, id stack.TransportEndpointID, data buffer.VectorisedView, ttl uint8, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) *tcpip.Error { - optLen := len(opts) - // Allocate a buffer for the TCP header. - hdr := buffer.NewPrependable(header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen) - - if rcvWnd > 0xffff { - rcvWnd = 0xffff +func (e *endpoint) sendTCP(r *stack.Route, id stack.TransportEndpointID, data buffer.VectorisedView, ttl, tos uint8, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) *tcpip.Error { + if err := sendTCP(r, id, data, ttl, tos, flags, seq, ack, rcvWnd, opts, gso); err != nil { + e.stats.SendErrors.SegmentSendToNetworkFailed.Increment() + return err } + e.stats.SegmentsSent.Increment() + return nil +} +func buildTCPHdr(r *stack.Route, id stack.TransportEndpointID, d *stack.PacketDescriptor, data buffer.VectorisedView, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) { + optLen := len(opts) + hdr := &d.Hdr + packetSize := d.Size + off := d.Off // Initialize the header. tcp := header.TCP(hdr.Prepend(header.TCPMinimumSize + optLen)) tcp.Encode(&header.TCPFields{ @@ -610,7 +649,7 @@ func sendTCP(r *stack.Route, id stack.TransportEndpointID, data buffer.Vectorise }) copy(tcp[header.TCPMinimumSize:], opts) - length := uint16(hdr.UsedLength() + data.Size()) + length := uint16(hdr.UsedLength() + packetSize) xsum := r.PseudoHeaderChecksum(ProtocolNumber, length) // Only calculate the checksum if offloading isn't supported. if gso != nil && gso.NeedsCsum { @@ -620,16 +659,79 @@ func sendTCP(r *stack.Route, id stack.TransportEndpointID, data buffer.Vectorise // header and data and get the right sum of the TCP packet. tcp.SetChecksum(xsum) } else if r.Capabilities()&stack.CapabilityTXChecksumOffload == 0 { - xsum = header.ChecksumVV(data, xsum) + xsum = header.ChecksumVVWithOffset(data, xsum, off, packetSize) tcp.SetChecksum(^tcp.CalculateChecksum(xsum)) } +} + +func sendTCPBatch(r *stack.Route, id stack.TransportEndpointID, data buffer.VectorisedView, ttl, tos uint8, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) *tcpip.Error { + optLen := len(opts) + if rcvWnd > 0xffff { + rcvWnd = 0xffff + } + + mss := int(gso.MSS) + n := (data.Size() + mss - 1) / mss + + hdrs := stack.NewPacketDescriptors(n, header.TCPMinimumSize+int(r.MaxHeaderLength())+optLen) + + size := data.Size() + off := 0 + for i := 0; i < n; i++ { + packetSize := mss + if packetSize > size { + packetSize = size + } + size -= packetSize + hdrs[i].Off = off + hdrs[i].Size = packetSize + buildTCPHdr(r, id, &hdrs[i], data, flags, seq, ack, rcvWnd, opts, gso) + off += packetSize + seq = seq.Add(seqnum.Size(packetSize)) + } + if ttl == 0 { + ttl = r.DefaultTTL() + } + sent, err := r.WritePackets(gso, hdrs, data, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}) + if err != nil { + r.Stats().TCP.SegmentSendErrors.IncrementBy(uint64(n - sent)) + } + r.Stats().TCP.SegmentsSent.IncrementBy(uint64(sent)) + return err +} + +// sendTCP sends a TCP segment with the provided options via the provided +// network endpoint and under the provided identity. +func sendTCP(r *stack.Route, id stack.TransportEndpointID, data buffer.VectorisedView, ttl, tos uint8, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) *tcpip.Error { + optLen := len(opts) + if rcvWnd > 0xffff { + rcvWnd = 0xffff + } + + if r.Loop&stack.PacketLoop == 0 && gso != nil && gso.Type == stack.GSOSW && int(gso.MSS) < data.Size() { + return sendTCPBatch(r, id, data, ttl, tos, flags, seq, ack, rcvWnd, opts, gso) + } + + d := &stack.PacketDescriptor{ + Hdr: buffer.NewPrependable(header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen), + Off: 0, + Size: data.Size(), + } + buildTCPHdr(r, id, d, data, flags, seq, ack, rcvWnd, opts, gso) + + if ttl == 0 { + ttl = r.DefaultTTL() + } + if err := r.WritePacket(gso, d.Hdr, data, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}); err != nil { + r.Stats().TCP.SegmentSendErrors.Increment() + return err + } r.Stats().TCP.SegmentsSent.Increment() if (flags & header.TCPFlagRst) != 0 { r.Stats().TCP.ResetsSent.Increment() } - - return r.WritePacket(gso, hdr, data, ProtocolNumber, ttl) + return nil } // makeOptions makes an options slice. @@ -678,7 +780,7 @@ func (e *endpoint) sendRaw(data buffer.VectorisedView, flags byte, seq, ack seqn sackBlocks = e.sack.Blocks[:e.sack.NumBlocks] } options := e.makeOptions(sackBlocks) - err := sendTCP(&e.route, e.id, data, e.route.DefaultTTL(), flags, seq, ack, rcvWnd, options, e.gso) + err := e.sendTCP(&e.route, e.ID, data, e.ttl, e.sendTOS, flags, seq, ack, rcvWnd, options, e.gso) putOptions(options) return err } @@ -727,10 +829,26 @@ func (e *endpoint) handleClose() *tcpip.Error { func (e *endpoint) resetConnectionLocked(err *tcpip.Error) { // Only send a reset if the connection is being aborted for a reason // other than receiving a reset. + if e.state == StateEstablished || e.state == StateCloseWait { + e.stack.Stats().TCP.EstablishedResets.Increment() + e.stack.Stats().TCP.CurrentEstablished.Decrement() + } e.state = StateError - e.hardError = err + e.HardError = err if err != tcpip.ErrConnectionReset { - e.sendRaw(buffer.VectorisedView{}, header.TCPFlagAck|header.TCPFlagRst, e.snd.sndUna, e.rcv.rcvNxt, 0) + // The exact sequence number to be used for the RST is the same as the + // one used by Linux. We need to handle the case of window being shrunk + // which can cause sndNxt to be outside the acceptable window on the + // receiver. + // + // See: https://www.snellman.net/blog/archive/2016-02-01-tcp-rst/ for more + // information. + sndWndEnd := e.snd.sndUna.Add(e.snd.sndWnd) + resetSeqNum := sndWndEnd + if !sndWndEnd.LessThan(e.snd.sndNxt) || e.snd.sndNxt.Size(sndWndEnd) < (1<<e.snd.sndWndScale) { + resetSeqNum = e.snd.sndNxt + } + e.sendRaw(buffer.VectorisedView{}, header.TCPFlagAck|header.TCPFlagRst, resetSeqNum, e.rcv.rcvNxt, 0) } } @@ -744,6 +862,51 @@ func (e *endpoint) completeWorkerLocked() { } } +func (e *endpoint) handleReset(s *segment) (ok bool, err *tcpip.Error) { + if e.rcv.acceptable(s.sequenceNumber, 0) { + // RFC 793, page 37 states that "in all states + // except SYN-SENT, all reset (RST) segments are + // validated by checking their SEQ-fields." So + // we only process it if it's acceptable. + s.decRef() + e.mu.Lock() + switch e.state { + // In case of a RST in CLOSE-WAIT linux moves + // the socket to closed state with an error set + // to indicate EPIPE. + // + // Technically this seems to be at odds w/ RFC. + // As per https://tools.ietf.org/html/rfc793#section-2.7 + // page 69 the behavior for a segment arriving + // w/ RST bit set in CLOSE-WAIT is inlined below. + // + // ESTABLISHED + // FIN-WAIT-1 + // FIN-WAIT-2 + // CLOSE-WAIT + + // If the RST bit is set then, any outstanding RECEIVEs and + // SEND should receive "reset" responses. All segment queues + // should be flushed. Users should also receive an unsolicited + // general "connection reset" signal. Enter the CLOSED state, + // delete the TCB, and return. + case StateCloseWait: + e.state = StateClose + e.HardError = tcpip.ErrAborted + // We need to set this explicitly here because otherwise + // the port registrations will not be released till the + // endpoint is actively closed by the application. + e.workerCleanup = true + e.mu.Unlock() + return false, nil + default: + e.mu.Unlock() + return false, tcpip.ErrConnectionReset + } + } + return true, nil +} + // handleSegments pulls segments from the queue and processes them. It returns // no error if the protocol loop should continue, an error otherwise. func (e *endpoint) handleSegments() *tcpip.Error { @@ -761,14 +924,34 @@ func (e *endpoint) handleSegments() *tcpip.Error { } if s.flagIsSet(header.TCPFlagRst) { - if e.rcv.acceptable(s.sequenceNumber, 0) { - // RFC 793, page 37 states that "in all states - // except SYN-SENT, all reset (RST) segments are - // validated by checking their SEQ-fields." So - // we only process it if it's acceptable. - s.decRef() - return tcpip.ErrConnectionReset + if ok, err := e.handleReset(s); !ok { + return err } + } else if s.flagIsSet(header.TCPFlagSyn) { + // See: https://tools.ietf.org/html/rfc5961#section-4.1 + // 1) If the SYN bit is set, irrespective of the sequence number, TCP + // MUST send an ACK (also referred to as challenge ACK) to the remote + // peer: + // + // <SEQ=SND.NXT><ACK=RCV.NXT><CTL=ACK> + // + // After sending the acknowledgment, TCP MUST drop the unacceptable + // segment and stop processing further. + // + // By sending an ACK, the remote peer is challenged to confirm the loss + // of the previous connection and the request to start a new connection. + // A legitimate peer, after restart, would not have a TCB in the + // synchronized state. Thus, when the ACK arrives, the peer should send + // a RST segment back with the sequence number derived from the ACK + // field that caused the RST. + + // This RST will confirm that the remote peer has indeed closed the + // previous connection. Upon receipt of a valid RST, the local TCP + // endpoint MUST terminate its connection. The local TCP endpoint + // should then rely on SYN retransmission from the remote end to + // re-establish the connection. + + e.snd.sendAck() } else if s.flagIsSet(header.TCPFlagAck) { // Patch the window size in the segment according to the // send window scale. @@ -777,7 +960,15 @@ func (e *endpoint) handleSegments() *tcpip.Error { // RFC 793, page 41 states that "once in the ESTABLISHED // state all segments must carry current acknowledgment // information." - e.rcv.handleRcvdSegment(s) + drop, err := e.rcv.handleRcvdSegment(s) + if err != nil { + s.decRef() + return err + } + if drop { + s.decRef() + continue + } e.snd.handleRcvdSegment(s) } s.decRef() @@ -876,7 +1067,6 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { } e.mu.Unlock() - // When the protocol loop exits we should wake up our waiters. e.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut) } @@ -897,8 +1087,10 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { e.lastErrorMu.Unlock() e.mu.Lock() + e.stack.Stats().TCP.EstablishedResets.Increment() + e.stack.Stats().TCP.CurrentEstablished.Decrement() e.state = StateError - e.hardError = err + e.HardError = err // Lock released below. epilogue() @@ -920,6 +1112,10 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { // RTT itself. e.rcvAutoParams.prevCopied = initialRcvWnd e.rcvListMu.Unlock() + e.stack.Stats().TCP.CurrentEstablished.Increment() + e.mu.Lock() + e.state = StateEstablished + e.mu.Unlock() } e.keepalive.timer.init(&e.keepalive.waker) @@ -927,7 +1123,6 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { // Tell waiters that the endpoint is connected and writable. e.mu.Lock() - e.state = StateEstablished drained := e.drainDone != nil e.mu.Unlock() if drained { @@ -958,7 +1153,13 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { { w: &closeWaker, f: func() *tcpip.Error { - return tcpip.ErrConnectionAborted + // This means the socket is being closed due + // to the TCP_FIN_WAIT2 timeout was hit. Just + // mark the socket as closed. + e.mu.Lock() + e.state = StateClose + e.mu.Unlock() + return nil }, }, { @@ -1001,17 +1202,18 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { e.resetConnectionLocked(tcpip.ErrConnectionAborted) e.mu.Unlock() } + if n¬ifyClose != 0 && closeTimer == nil { - // Reset the connection 3 seconds after - // the endpoint has been closed. - // - // The timer could fire in background - // when the endpoint is drained. That's - // OK as the loop here will not honor - // the firing until the undrain arrives. - closeTimer = time.AfterFunc(3*time.Second, func() { - closeWaker.Assert() - }) + e.mu.Lock() + if e.state == 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() + }) + e.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut) + } + e.mu.Unlock() } if n¬ifyKeepaliveChanged != 0 { @@ -1033,6 +1235,12 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { } } + if n¬ifyTickleWorker != 0 { + // Just a tickle notification. No need to do + // anything. + return nil + } + return nil }, }, @@ -1059,15 +1267,16 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { } e.rcvListMu.Unlock() - e.mu.RLock() + e.mu.Lock() if e.workerCleanup { e.notifyProtocolGoroutine(notifyClose) } - e.mu.RUnlock() // Main loop. Handle segments until both send and receive ends of the // connection have completed. - for !e.rcv.closed || !e.snd.closed || e.snd.sndUna != e.snd.sndNxtList { + + for e.state != StateTimeWait && e.state != StateClose && e.state != StateError { + e.mu.Unlock() e.workMu.Unlock() v, _ := s.Fetch(true) e.workMu.Lock() @@ -1083,15 +1292,156 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { return nil } + e.mu.Lock() + } + + state := e.state + e.mu.Unlock() + var reuseTW func() + if state == StateTimeWait { + // Disable close timer as we now entering real TIME_WAIT. + if closeTimer != nil { + closeTimer.Stop() + } + // Mark the current sleeper done so as to free all associated + // wakers. + s.Done() + // Wake up any waiters before we enter TIME_WAIT. + e.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut) + reuseTW = e.doTimeWait() } // Mark endpoint as closed. e.mu.Lock() if e.state != StateError { + e.stack.Stats().TCP.EstablishedResets.Increment() + e.stack.Stats().TCP.CurrentEstablished.Decrement() e.state = StateClose } + // Lock released below. epilogue() + // A new SYN was received during TIME_WAIT and we need to abort + // the timewait and redirect the segment to the listener queue + if reuseTW != nil { + reuseTW() + } + return nil } + +// handleTimeWaitSegments processes segments received during TIME_WAIT +// state. +func (e *endpoint) handleTimeWaitSegments() (extendTimeWait bool, reuseTW func()) { + checkRequeue := true + for i := 0; i < maxSegmentsPerWake; i++ { + s := e.segmentQueue.dequeue() + if s == nil { + checkRequeue = false + break + } + extTW, newSyn := e.rcv.handleTimeWaitSegment(s) + if newSyn { + info := e.EndpointInfo.TransportEndpointInfo + newID := info.ID + newID.RemoteAddress = "" + newID.RemotePort = 0 + netProtos := []tcpip.NetworkProtocolNumber{info.NetProto} + // If the local address is an IPv4 address then also + // look for IPv6 dual stack endpoints that might be + // listening on the local address. + if newID.LocalAddress.To4() != "" { + netProtos = []tcpip.NetworkProtocolNumber{header.IPv4ProtocolNumber, header.IPv6ProtocolNumber} + } + for _, netProto := range netProtos { + if listenEP := e.stack.FindTransportEndpoint(netProto, info.TransProto, newID, &s.route); listenEP != nil { + tcpEP := listenEP.(*endpoint) + if EndpointState(tcpEP.State()) == StateListen { + reuseTW = func() { + tcpEP.enqueueSegment(s) + } + // We explicitly do not decRef + // the segment as it's still + // valid and being reflected to + // a listening endpoint. + return false, reuseTW + } + } + } + } + if extTW { + extendTimeWait = true + } + s.decRef() + } + if checkRequeue && !e.segmentQueue.empty() { + e.newSegmentWaker.Assert() + } + return extendTimeWait, nil +} + +// doTimeWait is responsible for handling the TCP behaviour once a socket +// enters the TIME_WAIT state. Optionally it can return a closure that +// should be executed after releasing the endpoint registrations. This is +// done in cases where a new SYN is received during TIME_WAIT that carries +// a sequence number larger than one see on the connection. +func (e *endpoint) doTimeWait() (twReuse func()) { + // Trigger a 2 * MSL time wait state. During this period + // we will drop all incoming segments. + // NOTE: On Linux this is not configurable and is fixed at 60 seconds. + timeWaitDuration := DefaultTCPTimeWaitTimeout + + // Get the stack wide configuration. + var tcpTW tcpip.TCPTimeWaitTimeoutOption + if err := e.stack.TransportProtocolOption(ProtocolNumber, &tcpTW); err == nil { + timeWaitDuration = time.Duration(tcpTW) + } + + const newSegment = 1 + const notification = 2 + const timeWaitDone = 3 + + s := sleep.Sleeper{} + s.AddWaker(&e.newSegmentWaker, newSegment) + s.AddWaker(&e.notificationWaker, notification) + + var timeWaitWaker sleep.Waker + s.AddWaker(&timeWaitWaker, timeWaitDone) + timeWaitTimer := time.AfterFunc(timeWaitDuration, timeWaitWaker.Assert) + defer timeWaitTimer.Stop() + + for { + e.workMu.Unlock() + v, _ := s.Fetch(true) + e.workMu.Lock() + switch v { + case newSegment: + extendTimeWait, reuseTW := e.handleTimeWaitSegments() + if reuseTW != nil { + return reuseTW + } + if extendTimeWait { + timeWaitTimer.Reset(timeWaitDuration) + } + case notification: + n := e.fetchNotifications() + if n¬ifyClose != 0 { + return nil + } + if n¬ifyDrain != 0 { + for !e.segmentQueue.empty() { + // Ignore extending TIME_WAIT during a + // save. For sockets in TIME_WAIT we just + // terminate the TIME_WAIT early. + e.handleTimeWaitSegments() + } + close(e.drainDone) + <-e.undrain + return nil + } + case timeWaitDone: + return nil + } + } +} diff --git a/pkg/tcpip/transport/tcp/dual_stack_test.go b/pkg/tcpip/transport/tcp/dual_stack_test.go index c54610a87..dfaa4a559 100644 --- a/pkg/tcpip/transport/tcp/dual_stack_test.go +++ b/pkg/tcpip/transport/tcp/dual_stack_test.go @@ -42,7 +42,7 @@ func TestV4MappedConnectOnV6Only(t *testing.T) { } } -func testV4Connect(t *testing.T, c *context.Context) { +func testV4Connect(t *testing.T, c *context.Context, checkers ...checker.NetworkChecker) { // Start connection attempt. we, ch := waiter.NewChannelEntry(nil) c.WQ.EventRegister(&we, waiter.EventOut) @@ -55,12 +55,11 @@ func testV4Connect(t *testing.T, c *context.Context) { // Receive SYN packet. b := c.GetPacket() - checker.IPv4(t, b, - checker.TCP( - checker.DstPort(context.TestPort), - checker.TCPFlags(header.TCPFlagSyn), - ), - ) + synCheckers := append(checkers, checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagSyn), + )) + checker.IPv4(t, b, synCheckers...) tcp := header.TCP(header.IPv4(b).Payload()) c.IRS = seqnum.Value(tcp.SequenceNumber()) @@ -76,14 +75,13 @@ func testV4Connect(t *testing.T, c *context.Context) { }) // Receive ACK packet. - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(iss)+1), - ), - ) + ackCheckers := append(checkers, checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagAck), + checker.SeqNum(uint32(c.IRS)+1), + checker.AckNum(uint32(iss)+1), + )) + checker.IPv4(t, c.GetPacket(), ackCheckers...) // Wait for connection to be established. select { @@ -152,7 +150,7 @@ func TestV4ConnectWhenBoundToV4Mapped(t *testing.T) { testV4Connect(t, c) } -func testV6Connect(t *testing.T, c *context.Context) { +func testV6Connect(t *testing.T, c *context.Context, checkers ...checker.NetworkChecker) { // Start connection attempt to IPv6 address. we, ch := waiter.NewChannelEntry(nil) c.WQ.EventRegister(&we, waiter.EventOut) @@ -165,12 +163,11 @@ func testV6Connect(t *testing.T, c *context.Context) { // Receive SYN packet. b := c.GetV6Packet() - checker.IPv6(t, b, - checker.TCP( - checker.DstPort(context.TestPort), - checker.TCPFlags(header.TCPFlagSyn), - ), - ) + synCheckers := append(checkers, checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagSyn), + )) + checker.IPv6(t, b, synCheckers...) tcp := header.TCP(header.IPv6(b).Payload()) c.IRS = seqnum.Value(tcp.SequenceNumber()) @@ -186,14 +183,13 @@ func testV6Connect(t *testing.T, c *context.Context) { }) // Receive ACK packet. - checker.IPv6(t, c.GetV6Packet(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(iss)+1), - ), - ) + ackCheckers := append(checkers, checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagAck), + checker.SeqNum(uint32(c.IRS)+1), + checker.AckNum(uint32(iss)+1), + )) + checker.IPv6(t, c.GetV6Packet(), ackCheckers...) // Wait for connection to be established. select { diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 35b489c68..04c92c04c 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -15,6 +15,7 @@ package tcp import ( + "encoding/binary" "fmt" "math" "strings" @@ -26,6 +27,7 @@ import ( "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/hash/jenkins" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/iptables" "gvisor.dev/gvisor/pkg/tcpip/seqnum" @@ -119,6 +121,11 @@ const ( notifyReset notifyKeepaliveChanged notifyMSSChanged + // notifyTickleWorker is used to tickle the protocol main loop during a + // restore after we update the endpoint state to the correct one. This + // ensures the loop terminates if the final state of the endpoint is + // say TIME_WAIT. + notifyTickleWorker ) // SACKInfo holds TCP SACK related information for a given endpoint. @@ -170,6 +177,101 @@ type rcvBufAutoTuneParams struct { disabled bool } +// ReceiveErrors collect segment receive errors within transport layer. +type ReceiveErrors struct { + tcpip.ReceiveErrors + + // SegmentQueueDropped is the number of segments dropped due to + // a full segment queue. + SegmentQueueDropped tcpip.StatCounter + + // ChecksumErrors is the number of segments dropped due to bad checksums. + ChecksumErrors tcpip.StatCounter + + // ListenOverflowSynDrop is the number of times the listen queue overflowed + // and a SYN was dropped. + ListenOverflowSynDrop tcpip.StatCounter + + // ListenOverflowAckDrop is the number of times the final ACK + // in the handshake was dropped due to overflow. + ListenOverflowAckDrop tcpip.StatCounter + + // ZeroRcvWindowState is the number of times we advertised + // a zero receive window when rcvList is full. + ZeroRcvWindowState tcpip.StatCounter +} + +// SendErrors collect segment send errors within the transport layer. +type SendErrors struct { + tcpip.SendErrors + + // SegmentSendToNetworkFailed is the number of TCP segments failed to be sent + // to the network endpoint. + SegmentSendToNetworkFailed tcpip.StatCounter + + // SynSendToNetworkFailed is the number of TCP SYNs failed to be sent + // to the network endpoint. + SynSendToNetworkFailed tcpip.StatCounter + + // Retransmits is the number of TCP segments retransmitted. + Retransmits tcpip.StatCounter + + // FastRetransmit is the number of segments retransmitted in fast + // recovery. + FastRetransmit tcpip.StatCounter + + // Timeouts is the number of times the RTO expired. + Timeouts tcpip.StatCounter +} + +// Stats holds statistics about the endpoint. +type Stats struct { + // SegmentsReceived is the number of TCP segments received that + // the transport layer successfully parsed. + SegmentsReceived tcpip.StatCounter + + // SegmentsSent is the number of TCP segments sent. + SegmentsSent tcpip.StatCounter + + // FailedConnectionAttempts is the number of times we saw Connect and + // Accept errors. + FailedConnectionAttempts tcpip.StatCounter + + // ReceiveErrors collects segment receive errors within the + // transport layer. + ReceiveErrors ReceiveErrors + + // ReadErrors collects segment read errors from an endpoint read call. + ReadErrors tcpip.ReadErrors + + // SendErrors collects segment send errors within the transport layer. + SendErrors SendErrors + + // WriteErrors collects segment write errors from an endpoint write call. + WriteErrors tcpip.WriteErrors +} + +// IsEndpointStats is an empty method to implement the tcpip.EndpointStats +// marker interface. +func (*Stats) IsEndpointStats() {} + +// EndpointInfo holds useful information about a transport endpoint which +// can be queried by monitoring tools. +// +// +stateify savable +type EndpointInfo struct { + stack.TransportEndpointInfo + + // HardError is meaningful only when state is stateError. It stores the + // error to be returned when read/write syscalls are called and the + // endpoint is in this state. HardError is protected by endpoint mu. + HardError *tcpip.Error `state:".(string)"` +} + +// IsEndpointInfo is an empty method to implement the tcpip.EndpointInfo +// marker interface. +func (*EndpointInfo) IsEndpointInfo() {} + // endpoint represents a TCP endpoint. This struct serves as the interface // between users of the endpoint and the protocol implementation; it is legal to // have concurrent goroutines make calls into the endpoint, they are properly @@ -178,6 +280,8 @@ type rcvBufAutoTuneParams struct { // // +stateify savable type endpoint struct { + EndpointInfo + // workMu is used to arbitrate which goroutine may perform protocol // work. Only the main protocol goroutine is expected to call Lock() on // it, but other goroutines (e.g., send) may call TryLock() to eagerly @@ -186,9 +290,9 @@ type endpoint struct { // The following fields are initialized at creation time and do not // change throughout the lifetime of the endpoint. - stack *stack.Stack `state:"manual"` - netProto tcpip.NetworkProtocolNumber + stack *stack.Stack `state:"manual"` waiterQueue *waiter.Queue `state:"wait"` + uniqueID uint64 // lastError represents the last error that the endpoint reported; // access to it is protected by the following mutex. @@ -218,14 +322,19 @@ type endpoint struct { // The following fields are protected by the mutex. mu sync.RWMutex `state:"nosave"` - id stack.TransportEndpointID state EndpointState `state:".(EndpointState)"` + // origEndpointState is only used during a restore phase to save the + // endpoint state at restore time as the socket is moved to it's correct + // state. + origEndpointState EndpointState `state:"nosave"` + isPortReserved bool `state:"manual"` isRegistered bool boundNICID tcpip.NICID `state:"manual"` route stack.Route `state:"manual"` + ttl uint8 v6only bool isConnectNotified bool // TCP should never broadcast but Linux nevertheless supports enabling/ @@ -240,11 +349,6 @@ type endpoint struct { // address). effectiveNetProtos []tcpip.NetworkProtocolNumber `state:"manual"` - // hardError is meaningful only when state is stateError, it stores the - // error to be returned when read/write syscalls are called and the - // endpoint is in this state. hardError is protected by mu. - hardError *tcpip.Error `state:".(string)"` - // workerRunning specifies if a worker goroutine is running. workerRunning bool @@ -280,6 +384,9 @@ type endpoint struct { // reusePort is set to true if SO_REUSEPORT is enabled. reusePort bool + // bindToDevice is set to the NIC on which to bind or disabled if 0. + bindToDevice tcpip.NICID + // delay enables Nagle's algorithm. // // delay is a boolean (0 is false) and must be accessed atomically. @@ -315,7 +422,7 @@ type endpoint struct { // userMSS if non-zero is the MSS value explicitly set by the user // for this endpoint using the TCP_MAXSEG setsockopt. - userMSS int + userMSS uint16 // The following fields are used to manage the send buffer. When // segments are ready to be sent, they are added to sndQueue and the @@ -393,13 +500,49 @@ type endpoint struct { probe stack.TCPProbeFunc `state:"nosave"` // The following are only used to assist the restore run to re-connect. - bindAddress tcpip.Address connectingAddress tcpip.Address // amss is the advertised MSS to the peer by this endpoint. amss uint16 + // sendTOS represents IPv4 TOS or IPv6 TrafficClass, + // applied while sending packets. Defaults to 0 as on Linux. + sendTOS uint8 + gso *stack.GSO + + // TODO(b/142022063): Add ability to save and restore per endpoint stats. + stats Stats `state:"nosave"` + + // tcpLingerTimeout is the maximum amount of a time a socket + // a socket stays in TIME_WAIT state before being marked + // closed. + tcpLingerTimeout time.Duration + + // closed indicates that the user has called closed on the + // endpoint and at this point the endpoint is only around + // to complete the TCP shutdown. + closed bool +} + +// UniqueID implements stack.TransportEndpoint.UniqueID. +func (e *endpoint) UniqueID() uint64 { + return e.uniqueID +} + +// calculateAdvertisedMSS calculates the MSS to advertise. +// +// If userMSS is non-zero and is not greater than the maximum possible MSS for +// r, it will be used; otherwise, the maximum possible MSS will be used. +func calculateAdvertisedMSS(userMSS uint16, r stack.Route) uint16 { + // The maximum possible MSS is dependent on the route. + maxMSS := mssForRoute(&r) + + if userMSS != 0 && userMSS < maxMSS { + return userMSS + } + + return maxMSS } // StopWork halts packet processing. Only to be used in tests. @@ -427,10 +570,15 @@ type keepalive struct { waker sleep.Waker `state:"nosave"` } -func newEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) *endpoint { +func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) *endpoint { e := &endpoint{ - stack: stack, - netProto: netProto, + stack: s, + EndpointInfo: EndpointInfo{ + TransportEndpointInfo: stack.TransportEndpointInfo{ + NetProto: netProto, + TransProto: header.TCPProtocolNumber, + }, + }, waiterQueue: waiterQueue, state: StateInitial, rcvBufSize: DefaultReceiveBufferSize, @@ -443,29 +591,40 @@ func newEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waite interval: 75 * time.Second, count: 9, }, + uniqueID: s.UniqueID(), } var ss SendBufferSizeOption - if err := stack.TransportProtocolOption(ProtocolNumber, &ss); err == nil { + if err := s.TransportProtocolOption(ProtocolNumber, &ss); err == nil { e.sndBufSize = ss.Default } var rs ReceiveBufferSizeOption - if err := stack.TransportProtocolOption(ProtocolNumber, &rs); err == nil { + if err := s.TransportProtocolOption(ProtocolNumber, &rs); err == nil { e.rcvBufSize = rs.Default } var cs tcpip.CongestionControlOption - if err := stack.TransportProtocolOption(ProtocolNumber, &cs); err == nil { + if err := s.TransportProtocolOption(ProtocolNumber, &cs); err == nil { e.cc = cs } var mrb tcpip.ModerateReceiveBufferOption - if err := stack.TransportProtocolOption(ProtocolNumber, &mrb); err == nil { + if err := s.TransportProtocolOption(ProtocolNumber, &mrb); err == nil { e.rcvAutoParams.disabled = !bool(mrb) } - if p := stack.GetTCPProbe(); p != nil { + var de DelayEnabled + if err := s.TransportProtocolOption(ProtocolNumber, &de); err == nil && de { + e.SetSockOptInt(tcpip.DelayOption, 1) + } + + var tcpLT tcpip.TCPLingerTimeoutOption + if err := s.TransportProtocolOption(ProtocolNumber, &tcpLT); err == nil { + e.tcpLingerTimeout = time.Duration(tcpLT) + } + + if p := s.GetTCPProbe(); p != nil { e.probe = p } @@ -552,6 +711,13 @@ func (e *endpoint) notifyProtocolGoroutine(n uint32) { // with it. It must be called only once and with no other concurrent calls to // the endpoint. func (e *endpoint) Close() { + e.mu.Lock() + closed := e.closed + e.mu.Unlock() + if closed { + return + } + // Issue a shutdown so that the peer knows we won't send any more data // if we're connected, or stop accepting if we're listening. e.Shutdown(tcpip.ShutdownWrite | tcpip.ShutdownRead) @@ -564,14 +730,16 @@ func (e *endpoint) Close() { // in Listen() when trying to register. if e.state == StateListen && e.isPortReserved { if e.isRegistered { - e.stack.UnregisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e) + e.stack.StartTransportEndpointCleanup(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.bindToDevice) e.isRegistered = false } - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.bindToDevice) e.isPortReserved = false } + // Mark endpoint as closed. + e.closed = true // Either perform the local cleanup or kick the worker to make sure it // knows it needs to cleanup. tcpip.AddDanglingEndpoint(e) @@ -597,9 +765,7 @@ func (e *endpoint) closePendingAcceptableConnectionsLocked() { go func() { defer close(done) for n := range e.acceptedChan { - n.mu.Lock() - n.resetConnectionLocked(tcpip.ErrConnectionAborted) - n.mu.Unlock() + n.notifyProtocolGoroutine(notifyReset) n.Close() } }() @@ -625,16 +791,17 @@ func (e *endpoint) cleanupLocked() { e.workerCleanup = false if e.isRegistered { - e.stack.UnregisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e) + e.stack.StartTransportEndpointCleanup(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.bindToDevice) e.isRegistered = false } if e.isPortReserved { - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.bindToDevice) e.isPortReserved = false } e.route.Release() + e.stack.CompleteTransportEndpointCleanup(e) tcpip.DeleteDanglingEndpoint(e) } @@ -645,7 +812,9 @@ func (e *endpoint) initialReceiveWindow() int { if rcvWnd > math.MaxUint16 { rcvWnd = math.MaxUint16 } - routeWnd := InitialCwnd * int(mssForRoute(&e.route)) * 2 + + // Use the user supplied MSS, if available. + routeWnd := InitialCwnd * int(calculateAdvertisedMSS(e.userMSS, e.route)) * 2 if rcvWnd > routeWnd { rcvWnd = routeWnd } @@ -731,11 +900,12 @@ func (e *endpoint) Read(*tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, bufUsed := e.rcvBufUsed if s := e.state; !s.connected() && s != StateClose && bufUsed == 0 { e.rcvListMu.Unlock() - he := e.hardError + he := e.HardError e.mu.RUnlock() if s == StateError { return buffer.View{}, tcpip.ControlMessages{}, he } + e.stats.ReadErrors.InvalidEndpointState.Increment() return buffer.View{}, tcpip.ControlMessages{}, tcpip.ErrInvalidEndpointState } @@ -744,6 +914,9 @@ func (e *endpoint) Read(*tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, e.mu.RUnlock() + if err == tcpip.ErrClosedForReceive { + e.stats.ReadErrors.ReadClosed.Increment() + } return v, tcpip.ControlMessages{}, err } @@ -787,7 +960,7 @@ func (e *endpoint) isEndpointWritableLocked() (int, *tcpip.Error) { if !e.state.connected() { switch e.state { case StateError: - return 0, e.hardError + return 0, e.HardError default: return 0, tcpip.ErrClosedForSend } @@ -818,6 +991,7 @@ func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c if err != nil { e.sndBufMu.Unlock() e.mu.RUnlock() + e.stats.WriteErrors.WriteClosed.Increment() return 0, nil, err } @@ -852,6 +1026,7 @@ func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c if err != nil { e.sndBufMu.Unlock() e.mu.RUnlock() + e.stats.WriteErrors.WriteClosed.Increment() return 0, nil, err } @@ -863,7 +1038,7 @@ func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c } // Add data to the send queue. - s := newSegmentFromView(&e.route, e.id, v) + s := newSegmentFromView(&e.route, e.ID, v) e.sndBufUsed += len(v) e.sndBufInQueue += seqnum.Size(len(v)) e.sndQueue.PushBack(s) @@ -895,8 +1070,9 @@ func (e *endpoint) Peek(vec [][]byte) (int64, tcpip.ControlMessages, *tcpip.Erro // but has some pending unread data. if s := e.state; !s.connected() && s != StateClose { if s == StateError { - return 0, tcpip.ControlMessages{}, e.hardError + return 0, tcpip.ControlMessages{}, e.HardError } + e.stats.ReadErrors.InvalidEndpointState.Increment() return 0, tcpip.ControlMessages{}, tcpip.ErrInvalidEndpointState } @@ -905,6 +1081,7 @@ func (e *endpoint) Peek(vec [][]byte) (int64, tcpip.ControlMessages, *tcpip.Erro if e.rcvBufUsed == 0 { if e.rcvClosed || !e.state.connected() { + e.stats.ReadErrors.ReadClosed.Increment() return 0, tcpip.ControlMessages{}, tcpip.ErrClosedForReceive } return 0, tcpip.ControlMessages{}, tcpip.ErrWouldBlock @@ -1018,14 +1195,6 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error { e.sndBufMu.Unlock() return nil - default: - return nil - } -} - -// SetSockOpt sets a socket option. -func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { - switch v := opt.(type) { case tcpip.DelayOption: if v == 0 { atomic.StoreUint32(&e.delay, 0) @@ -1037,6 +1206,16 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { } return nil + default: + return nil + } +} + +// SetSockOpt sets a socket option. +func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { + // Lower 2 bits represents ECN bits. RFC 3168, section 23.1 + const inetECNMask = 3 + switch v := opt.(type) { case tcpip.CorkOption: if v == 0 { atomic.StoreUint32(&e.cork, 0) @@ -1060,6 +1239,21 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.mu.Unlock() return nil + case tcpip.BindToDeviceOption: + e.mu.Lock() + defer e.mu.Unlock() + if v == "" { + e.bindToDevice = 0 + return nil + } + for nicID, nic := range e.stack.NICInfo() { + if nic.Name == string(v) { + e.bindToDevice = nicID + return nil + } + } + return tcpip.ErrUnknownDevice + case tcpip.QuickAckOption: if v == 0 { atomic.StoreUint32(&e.slowAck, 1) @@ -1074,14 +1268,14 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { return tcpip.ErrInvalidOptionValue } e.mu.Lock() - e.userMSS = int(userMSS) + e.userMSS = uint16(userMSS) e.mu.Unlock() e.notifyProtocolGoroutine(notifyMSSChanged) return nil case tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. - if e.netProto != header.IPv6ProtocolNumber { + if e.NetProto != header.IPv6ProtocolNumber { return tcpip.ErrInvalidEndpointState } @@ -1096,6 +1290,12 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.v6only = v != 0 return nil + case tcpip.TTLOption: + e.mu.Lock() + e.ttl = uint8(v) + e.mu.Unlock() + return nil + case tcpip.KeepaliveEnabledOption: e.keepalive.Lock() e.keepalive.enabled = v != 0 @@ -1164,6 +1364,45 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { // Linux returns ENOENT when an invalid congestion // control algorithm is specified. return tcpip.ErrNoSuchFile + + case tcpip.IPv4TOSOption: + e.mu.Lock() + // TODO(gvisor.dev/issue/995): ECN is not currently supported, + // ignore the bits for now. + e.sendTOS = uint8(v) & ^uint8(inetECNMask) + e.mu.Unlock() + return nil + + case tcpip.IPv6TrafficClassOption: + e.mu.Lock() + // TODO(gvisor.dev/issue/995): ECN is not currently supported, + // ignore the bits for now. + e.sendTOS = uint8(v) & ^uint8(inetECNMask) + e.mu.Unlock() + return nil + + case tcpip.TCPLingerTimeoutOption: + e.mu.Lock() + if v < 0 { + // Same as effectively disabling TCPLinger timeout. + v = 0 + } + var stkTCPLingerTimeout tcpip.TCPLingerTimeoutOption + if err := e.stack.TransportProtocolOption(header.TCPProtocolNumber, &stkTCPLingerTimeout); err != nil { + // We were unable to retrieve a stack config, just use + // the DefaultTCPLingerTimeout. + if v > tcpip.TCPLingerTimeoutOption(DefaultTCPLingerTimeout) { + stkTCPLingerTimeout = tcpip.TCPLingerTimeoutOption(DefaultTCPLingerTimeout) + } + } + // Cap it to the stack wide TCPLinger timeout. + if v > stkTCPLingerTimeout { + v = stkTCPLingerTimeout + } + e.tcpLingerTimeout = time.Duration(v) + e.mu.Unlock() + return nil + default: return nil } @@ -1190,6 +1429,7 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) { switch opt { case tcpip.ReceiveQueueSizeOption: return e.readyReceiveSize() + case tcpip.SendBufferSizeOption: e.sndBufMu.Lock() v := e.sndBufSize @@ -1202,8 +1442,16 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) { e.rcvListMu.Unlock() return v, nil + case tcpip.DelayOption: + var o int + if v := atomic.LoadUint32(&e.delay); v != 0 { + o = 1 + } + return o, nil + + default: + return -1, tcpip.ErrUnknownProtocolOption } - return -1, tcpip.ErrUnknownProtocolOption } // GetSockOpt implements tcpip.Endpoint.GetSockOpt. @@ -1224,13 +1472,6 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { *o = header.TCPDefaultMSS return nil - case *tcpip.DelayOption: - *o = 0 - if v := atomic.LoadUint32(&e.delay); v != 0 { - *o = 1 - } - return nil - case *tcpip.CorkOption: *o = 0 if v := atomic.LoadUint32(&e.cork); v != 0 { @@ -1260,6 +1501,16 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { } return nil + case *tcpip.BindToDeviceOption: + e.mu.RLock() + defer e.mu.RUnlock() + if nic, ok := e.stack.NICInfo()[e.bindToDevice]; ok { + *o = tcpip.BindToDeviceOption(nic.Name) + return nil + } + *o = "" + return nil + case *tcpip.QuickAckOption: *o = 1 if v := atomic.LoadUint32(&e.slowAck); v != 0 { @@ -1269,7 +1520,7 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { case *tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. - if e.netProto != header.IPv6ProtocolNumber { + if e.NetProto != header.IPv6ProtocolNumber { return tcpip.ErrUnknownProtocolOption } @@ -1283,6 +1534,12 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { } return nil + case *tcpip.TTLOption: + e.mu.Lock() + *o = tcpip.TTLOption(e.ttl) + e.mu.Unlock() + return nil + case *tcpip.TCPInfoOption: *o = tcpip.TCPInfoOption{} e.mu.RLock() @@ -1347,13 +1604,31 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { e.mu.Unlock() return nil + case *tcpip.IPv4TOSOption: + e.mu.RLock() + *o = tcpip.IPv4TOSOption(e.sendTOS) + e.mu.RUnlock() + return nil + + case *tcpip.IPv6TrafficClassOption: + e.mu.RLock() + *o = tcpip.IPv6TrafficClassOption(e.sendTOS) + e.mu.RUnlock() + return nil + + case *tcpip.TCPLingerTimeoutOption: + e.mu.Lock() + *o = tcpip.TCPLingerTimeoutOption(e.tcpLingerTimeout) + e.mu.Unlock() + return nil + default: return tcpip.ErrUnknownProtocolOption } } func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress) (tcpip.NetworkProtocolNumber, *tcpip.Error) { - netProto := e.netProto + netProto := e.NetProto if header.IsV4MappedAddress(addr.Addr) { // Fail if using a v4 mapped address on a v6only endpoint. if e.v6only { @@ -1369,7 +1644,7 @@ func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress) (tcpip.NetworkProtocol // Fail if we're bound to an address length different from the one we're // checking. - if l := len(e.id.LocalAddress); l != 0 && len(addr.Addr) != 0 && l != len(addr.Addr) { + if l := len(e.ID.LocalAddress); l != 0 && len(addr.Addr) != 0 && l != len(addr.Addr) { return 0, tcpip.ErrInvalidEndpointState } @@ -1383,7 +1658,12 @@ func (*endpoint) Disconnect() *tcpip.Error { // Connect connects the endpoint to its peer. func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { - return e.connect(addr, true, true) + err := e.connect(addr, true, true) + if err != nil && !err.IgnoreStats() { + e.stack.Stats().TCP.FailedConnectionAttempts.Increment() + e.stats.FailedConnectionAttempts.Increment() + } + return err } // connect connects the endpoint to its peer. In the normal non-S/R case, the @@ -1392,14 +1672,9 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { // created (so no new handshaking is done); for stack-accepted connections not // yet accepted by the app, they are restored without running the main goroutine // here. -func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (err *tcpip.Error) { +func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tcpip.Error { e.mu.Lock() defer e.mu.Unlock() - defer func() { - if err != nil && !err.IgnoreStats() { - e.stack.Stats().TCP.FailedConnectionAttempts.Increment() - } - }() connectingAddr := addr.Addr @@ -1419,7 +1694,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (er return tcpip.ErrAlreadyConnected } - nicid := addr.NIC + nicID := addr.NIC switch e.state { case StateBound: // If we're already bound to a NIC but the caller is requesting @@ -1428,11 +1703,11 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (er break } - if nicid != 0 && nicid != e.boundNICID { + if nicID != 0 && nicID != e.boundNICID { return tcpip.ErrNoRoute } - nicid = e.boundNICID + nicID = e.boundNICID case StateInitial: // Nothing to do. We'll eventually fill-in the gaps in the ID (if any) @@ -1444,29 +1719,29 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (er return tcpip.ErrAlreadyConnecting case StateError: - return e.hardError + return e.HardError default: return tcpip.ErrInvalidEndpointState } // Find a route to the desired destination. - r, err := e.stack.FindRoute(nicid, e.id.LocalAddress, addr.Addr, netProto, false /* multicastLoop */) + r, err := e.stack.FindRoute(nicID, e.ID.LocalAddress, addr.Addr, netProto, false /* multicastLoop */) if err != nil { return err } defer r.Release() - origID := e.id + origID := e.ID netProtos := []tcpip.NetworkProtocolNumber{netProto} - e.id.LocalAddress = r.LocalAddress - e.id.RemoteAddress = r.RemoteAddress - e.id.RemotePort = addr.Port + e.ID.LocalAddress = r.LocalAddress + e.ID.RemoteAddress = r.RemoteAddress + e.ID.RemotePort = addr.Port - if e.id.LocalPort != 0 { + if e.ID.LocalPort != 0 { // The endpoint is bound to a port, attempt to register it. - err := e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, e.id, e, e.reusePort) + err := e.stack.RegisterTransportEndpoint(nicID, netProtos, ProtocolNumber, e.ID, e, e.reusePort, e.bindToDevice) if err != nil { return err } @@ -1475,20 +1750,35 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (er // one. Make sure that it isn't one that will result in the same // address/port for both local and remote (otherwise this // endpoint would be trying to connect to itself). - sameAddr := e.id.LocalAddress == e.id.RemoteAddress - if _, err := e.stack.PickEphemeralPort(func(p uint16) (bool, *tcpip.Error) { - if sameAddr && p == e.id.RemotePort { + sameAddr := e.ID.LocalAddress == e.ID.RemoteAddress + + // Calculate a port offset based on the destination IP/port and + // src IP to ensure that for a given tuple (srcIP, destIP, + // destPort) the offset used as a starting point is the same to + // ensure that we can cycle through the port space effectively. + h := jenkins.Sum32(e.stack.Seed()) + h.Write([]byte(e.ID.LocalAddress)) + h.Write([]byte(e.ID.RemoteAddress)) + portBuf := make([]byte, 2) + binary.LittleEndian.PutUint16(portBuf, e.ID.RemotePort) + h.Write(portBuf) + portOffset := h.Sum32() + + if _, err := e.stack.PickEphemeralPortStable(portOffset, func(p uint16) (bool, *tcpip.Error) { + if sameAddr && p == e.ID.RemotePort { return false, nil } - if !e.stack.IsPortAvailable(netProtos, ProtocolNumber, e.id.LocalAddress, p, false) { + // 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, false /* reusePort */, e.bindToDevice) { return false, nil } - id := e.id + id := e.ID id.LocalPort = p - switch e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, id, e, e.reusePort) { + switch e.stack.RegisterTransportEndpoint(nicID, netProtos, ProtocolNumber, id, e, e.reusePort, e.bindToDevice) { case nil: - e.id = id + e.ID = id return true, nil case tcpip.ErrPortInUse: return false, nil @@ -1504,14 +1794,14 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (er // 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.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, origID.LocalAddress, origID.LocalPort, e.bindToDevice) e.isPortReserved = false } e.isRegistered = true e.state = StateConnecting e.route = r.Clone() - e.boundNICID = nicid + e.boundNICID = nicID e.effectiveNetProtos = netProtos e.connectingAddress = connectingAddr @@ -1523,7 +1813,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (er e.segmentQueue.mu.Lock() for _, l := range []segmentList{e.segmentQueue.list, e.sndQueue, e.snd.writeList} { for s := l.Front(); s != nil; s = s.Next() { - s.id = e.id + s.id = e.ID s.route = r.Clone() e.sndWaker.Assert() } @@ -1531,6 +1821,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (er e.segmentQueue.mu.Unlock() e.snd.updateMaxPayloadSize(int(e.route.MTU()), 0) e.state = StateEstablished + e.stack.Stats().TCP.CurrentEstablished.Increment() } if run { @@ -1551,9 +1842,8 @@ func (*endpoint) ConnectEndpoint(tcpip.Endpoint) *tcpip.Error { // peer. func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { e.mu.Lock() - defer e.mu.Unlock() e.shutdownFlags |= flags - + finQueued := false switch { case e.state.connected(): // Close for read. @@ -1568,6 +1858,7 @@ func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { // the connection with a RST. if (e.shutdownFlags&tcpip.ShutdownWrite) != 0 && rcvBufUsed > 0 { e.notifyProtocolGoroutine(notifyReset) + e.mu.Unlock() return nil } } @@ -1583,17 +1874,14 @@ func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { } // Queue fin segment. - s := newSegmentFromView(&e.route, e.id, nil) + s := newSegmentFromView(&e.route, e.ID, nil) e.sndQueue.PushBack(s) e.sndBufInQueue++ - + finQueued = true // Mark endpoint as closed. e.sndClosed = true e.sndBufMu.Unlock() - - // Tell protocol goroutine to close. - e.sndCloseWaker.Assert() } case e.state == StateListen: @@ -1601,24 +1889,37 @@ func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { if flags&tcpip.ShutdownRead != 0 { e.notifyProtocolGoroutine(notifyClose) } - default: + e.mu.Unlock() return tcpip.ErrNotConnected } - + e.mu.Unlock() + if finQueued { + if e.workMu.TryLock() { + e.handleClose() + e.workMu.Unlock() + } else { + // Tell protocol goroutine to close. + e.sndCloseWaker.Assert() + } + } return nil } // Listen puts the endpoint in "listen" mode, which allows it to accept // new connections. -func (e *endpoint) Listen(backlog int) (err *tcpip.Error) { +func (e *endpoint) Listen(backlog int) *tcpip.Error { + err := e.listen(backlog) + if err != nil && !err.IgnoreStats() { + e.stack.Stats().TCP.FailedConnectionAttempts.Increment() + e.stats.FailedConnectionAttempts.Increment() + } + return err +} + +func (e *endpoint) listen(backlog int) *tcpip.Error { e.mu.Lock() defer e.mu.Unlock() - defer func() { - if err != nil && !err.IgnoreStats() { - e.stack.Stats().TCP.FailedConnectionAttempts.Increment() - } - }() // Allow the backlog to be adjusted if the endpoint is not shutting down. // When the endpoint shuts down, it sets workerCleanup to true, and from @@ -1644,11 +1945,12 @@ func (e *endpoint) Listen(backlog int) (err *tcpip.Error) { // Endpoint must be bound before it can transition to listen mode. if e.state != StateBound { + e.stats.ReadErrors.InvalidEndpointState.Increment() return tcpip.ErrInvalidEndpointState } // Register the endpoint. - if err := e.stack.RegisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.reusePort); err != nil { + if err := e.stack.RegisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.reusePort, e.bindToDevice); err != nil { return err } @@ -1692,12 +1994,7 @@ func (e *endpoint) Accept() (tcpip.Endpoint, *waiter.Queue, *tcpip.Error) { return nil, nil, tcpip.ErrWouldBlock } - // Start the protocol goroutine. - wq := &waiter.Queue{} - n.startAcceptedLoop(wq) - e.stack.Stats().TCP.PassiveConnectionOpenings.Increment() - - return n, wq, nil + return n, n.waiterQueue, nil } // Bind binds the endpoint to a specific local port and optionally address. @@ -1712,7 +2009,7 @@ func (e *endpoint) Bind(addr tcpip.FullAddress) (err *tcpip.Error) { return tcpip.ErrAlreadyBound } - e.bindAddress = addr.Addr + e.BindAddr = addr.Addr netProto, err := e.checkV4Mapped(&addr) if err != nil { return err @@ -1729,26 +2026,26 @@ func (e *endpoint) Bind(addr tcpip.FullAddress) (err *tcpip.Error) { } } - port, err := e.stack.ReservePort(netProtos, ProtocolNumber, addr.Addr, addr.Port, e.reusePort) + port, err := e.stack.ReservePort(netProtos, ProtocolNumber, addr.Addr, addr.Port, e.reusePort, e.bindToDevice) if err != nil { return err } e.isPortReserved = true e.effectiveNetProtos = netProtos - e.id.LocalPort = port + e.ID.LocalPort = port // Any failures beyond this point must remove the port registration. - defer func() { + defer func(bindToDevice tcpip.NICID) { if err != nil { - e.stack.ReleasePort(netProtos, ProtocolNumber, addr.Addr, port) + e.stack.ReleasePort(netProtos, ProtocolNumber, addr.Addr, port, bindToDevice) e.isPortReserved = false e.effectiveNetProtos = nil - e.id.LocalPort = 0 - e.id.LocalAddress = "" + e.ID.LocalPort = 0 + e.ID.LocalAddress = "" e.boundNICID = 0 } - }() + }(e.bindToDevice) // If an address is specified, we must ensure that it's one of our // local addresses. @@ -1759,7 +2056,7 @@ func (e *endpoint) Bind(addr tcpip.FullAddress) (err *tcpip.Error) { } e.boundNICID = nic - e.id.LocalAddress = addr.Addr + e.ID.LocalAddress = addr.Addr } // Mark endpoint as bound. @@ -1774,8 +2071,8 @@ func (e *endpoint) GetLocalAddress() (tcpip.FullAddress, *tcpip.Error) { defer e.mu.RUnlock() return tcpip.FullAddress{ - Addr: e.id.LocalAddress, - Port: e.id.LocalPort, + Addr: e.ID.LocalAddress, + Port: e.ID.LocalPort, NIC: e.boundNICID, }, nil } @@ -1790,19 +2087,20 @@ func (e *endpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Error) { } return tcpip.FullAddress{ - Addr: e.id.RemoteAddress, - Port: e.id.RemotePort, + Addr: e.ID.RemoteAddress, + Port: e.ID.RemotePort, NIC: e.boundNICID, }, nil } // HandlePacket is called by the stack when new packets arrive to this transport // endpoint. -func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, vv buffer.VectorisedView) { - s := newSegment(r, id, vv) +func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt tcpip.PacketBuffer) { + s := newSegment(r, id, pkt) if !s.parse() { e.stack.Stats().MalformedRcvdPackets.Increment() e.stack.Stats().TCP.InvalidSegmentsReceived.Increment() + e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() s.decRef() return } @@ -1810,27 +2108,34 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, vv if !s.csumValid { e.stack.Stats().MalformedRcvdPackets.Increment() e.stack.Stats().TCP.ChecksumErrors.Increment() + e.stats.ReceiveErrors.ChecksumErrors.Increment() s.decRef() return } e.stack.Stats().TCP.ValidSegmentsReceived.Increment() + e.stats.SegmentsReceived.Increment() if (s.flags & header.TCPFlagRst) != 0 { e.stack.Stats().TCP.ResetsReceived.Increment() } + e.enqueueSegment(s) +} + +func (e *endpoint) enqueueSegment(s *segment) { // Send packet to worker goroutine. if e.segmentQueue.enqueue(s) { e.newSegmentWaker.Assert() } else { // The queue is full, so we drop the segment. e.stack.Stats().DroppedPackets.Increment() + e.stats.ReceiveErrors.SegmentQueueDropped.Increment() s.decRef() } } // HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket. -func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, vv buffer.VectorisedView) { +func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, pkt tcpip.PacketBuffer) { switch typ { case stack.ControlPacketTooBig: e.sndBufMu.Lock() @@ -1874,6 +2179,7 @@ func (e *endpoint) readyToRead(s *segment) { // that a subsequent read of the segment will correctly trigger // a non-zero notification. if avail := e.receiveBufferAvailableLocked(); avail>>e.rcv.rcvWndScale == 0 { + e.stats.ReceiveErrors.ZeroRcvWindowState.Increment() e.zeroWindow = true } e.rcvList.PushBack(s) @@ -2026,7 +2332,7 @@ func (e *endpoint) completeState() stack.TCPEndpointState { // Copy EndpointID. e.mu.Lock() - s.ID = stack.TCPEndpointID(e.id) + s.ID = stack.TCPEndpointID(e.ID) e.mu.Unlock() // Copy endpoint rcv state. @@ -2119,11 +2425,7 @@ func (e *endpoint) completeState() stack.TCPEndpointState { return s } -func (e *endpoint) initGSO() { - if e.route.Capabilities()&stack.CapabilityGSO == 0 { - return - } - +func (e *endpoint) initHardwareGSO() { gso := &stack.GSO{} switch e.route.NetProto { case header.IPv4ProtocolNumber: @@ -2133,7 +2435,7 @@ func (e *endpoint) initGSO() { gso.Type = stack.GSOTCPv6 gso.L3HdrLen = header.IPv6MinimumSize default: - panic(fmt.Sprintf("Unknown netProto: %v", e.netProto)) + panic(fmt.Sprintf("Unknown netProto: %v", e.NetProto)) } gso.NeedsCsum = true gso.CsumOffset = header.TCPChecksumOffset @@ -2141,6 +2443,18 @@ func (e *endpoint) initGSO() { e.gso = gso } +func (e *endpoint) initGSO() { + if e.route.Capabilities()&stack.CapabilityHardwareGSO != 0 { + e.initHardwareGSO() + } else if e.route.Capabilities()&stack.CapabilitySoftwareGSO != 0 { + e.gso = &stack.GSO{ + MaxSize: e.route.GSOMaxSize(), + Type: stack.GSOSW, + NeedsCsum: false, + } + } +} + // State implements tcpip.Endpoint.State. It exports the endpoint's protocol // state for diagnostics. func (e *endpoint) State() uint32 { @@ -2149,6 +2463,37 @@ func (e *endpoint) State() uint32 { return uint32(e.state) } +// Info returns a copy of the endpoint info. +func (e *endpoint) Info() tcpip.EndpointInfo { + e.mu.RLock() + // Make a copy of the endpoint info. + ret := e.EndpointInfo + e.mu.RUnlock() + return &ret +} + +// Stats returns a pointer to the endpoint stats. +func (e *endpoint) Stats() tcpip.EndpointStats { + return &e.stats +} + +// Wait implements stack.TransportEndpoint.Wait. +func (e *endpoint) Wait() { + waitEntry, notifyCh := waiter.NewChannelEntry(nil) + e.waiterQueue.EventRegister(&waitEntry, waiter.EventHUp) + defer e.waiterQueue.EventUnregister(&waitEntry) + for { + e.mu.Lock() + running := e.workerRunning + e.mu.Unlock() + if !running { + break + } + <-notifyCh + } +} + func mssForRoute(r *stack.Route) uint16 { + // TODO(b/143359391): Respect TCP Min and Max size. return uint16(r.MTU() - header.TCPMinimumSize) } diff --git a/pkg/tcpip/transport/tcp/endpoint_state.go b/pkg/tcpip/transport/tcp/endpoint_state.go index 831389ec7..7aa4c3f0e 100644 --- a/pkg/tcpip/transport/tcp/endpoint_state.go +++ b/pkg/tcpip/transport/tcp/endpoint_state.go @@ -55,7 +55,7 @@ func (e *endpoint) beforeSave() { case StateEstablished, StateSynSent, StateSynRecv, StateFinWait1, StateFinWait2, StateTimeWait, StateCloseWait, StateLastAck, StateClosing: if e.route.Capabilities()&stack.CapabilitySaveRestore == 0 { if e.route.Capabilities()&stack.CapabilityDisconnectOk == 0 { - panic(tcpip.ErrSaveRejection{fmt.Errorf("endpoint cannot be saved in connected state: local %v:%d, remote %v:%d", e.id.LocalAddress, e.id.LocalPort, e.id.RemoteAddress, e.id.RemotePort)}) + panic(tcpip.ErrSaveRejection{fmt.Errorf("endpoint cannot be saved in connected state: local %v:%d, remote %v:%d", e.ID.LocalAddress, e.ID.LocalPort, e.ID.RemoteAddress, e.ID.RemotePort)}) } e.resetConnectionLocked(tcpip.ErrConnectionAborted) e.mu.Unlock() @@ -78,7 +78,7 @@ func (e *endpoint) beforeSave() { } fallthrough case StateError, StateClose: - for e.state == StateError && e.workerRunning { + for (e.state == StateError || e.state == StateClose) && e.workerRunning { e.mu.Unlock() time.Sleep(100 * time.Millisecond) e.mu.Lock() @@ -165,6 +165,12 @@ func (e *endpoint) loadState(state EndpointState) { // afterLoad is invoked by stateify. func (e *endpoint) afterLoad() { + // Freeze segment queue before registering to prevent any segments + // from being delivered while it is being restored. + e.origEndpointState = e.state + // Restore the endpoint to InitialState as it will be moved to + // its origEndpointState during Resume. + e.state = StateInitial stack.StackFromEnv.RegisterRestoredEndpoint(e) } @@ -173,8 +179,8 @@ func (e *endpoint) Resume(s *stack.Stack) { e.stack = s e.segmentQueue.setLimit(MaxUnprocessedSegments) e.workMu.Init() + state := e.origEndpointState - state := e.state switch state { case StateInitial, StateBound, StateListen, StateConnecting, StateEstablished: var ss SendBufferSizeOption @@ -189,12 +195,13 @@ func (e *endpoint) Resume(s *stack.Stack) { } bind := func() { - e.state = StateInitial - if len(e.bindAddress) == 0 { - e.bindAddress = e.id.LocalAddress + if len(e.BindAddr) == 0 { + e.BindAddr = e.ID.LocalAddress } - if err := e.Bind(tcpip.FullAddress{Addr: e.bindAddress, Port: e.id.LocalPort}); err != nil { - panic("endpoint binding failed: " + 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)) } } @@ -202,21 +209,31 @@ func (e *endpoint) Resume(s *stack.Stack) { case StateEstablished, StateFinWait1, StateFinWait2, StateTimeWait, StateCloseWait, StateLastAck, StateClosing: bind() if len(e.connectingAddress) == 0 { - e.connectingAddress = e.id.RemoteAddress + e.connectingAddress = e.ID.RemoteAddress // This endpoint is accepted by netstack but not yet by // the app. If the endpoint is IPv6 but the remote // address is IPv4, we need to connect as IPv6 so that // dual-stack mode can be properly activated. - if e.netProto == header.IPv6ProtocolNumber && len(e.id.RemoteAddress) != header.IPv6AddressSize { - e.connectingAddress = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xff\xff" + e.id.RemoteAddress + if e.NetProto == header.IPv6ProtocolNumber && len(e.ID.RemoteAddress) != header.IPv6AddressSize { + e.connectingAddress = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xff\xff" + e.ID.RemoteAddress } } // Reset the scoreboard to reinitialize the sack information as // we do not restore SACK information. e.scoreboard.Reset() - if err := e.connect(tcpip.FullAddress{NIC: e.boundNICID, Addr: e.connectingAddress, Port: e.id.RemotePort}, false, e.workerRunning); err != tcpip.ErrConnectStarted { + if err := e.connect(tcpip.FullAddress{NIC: e.boundNICID, Addr: e.connectingAddress, Port: e.ID.RemotePort}, false, e.workerRunning); err != tcpip.ErrConnectStarted { panic("endpoint connecting failed: " + err.String()) } + e.mu.Lock() + e.state = e.origEndpointState + closed := e.closed + e.mu.Unlock() + e.notifyProtocolGoroutine(notifyTickleWorker) + if state == StateFinWait2 && closed { + // If the endpoint has been closed then make sure we notify so + // that the FIN_WAIT2 timer is started after a restore. + e.notifyProtocolGoroutine(notifyClose) + } connectedLoading.Done() case StateListen: tcpip.AsyncLoading.Add(1) @@ -236,7 +253,7 @@ func (e *endpoint) Resume(s *stack.Stack) { connectedLoading.Wait() listenLoading.Wait() bind() - if err := e.Connect(tcpip.FullAddress{NIC: e.boundNICID, Addr: e.connectingAddress, Port: e.id.RemotePort}); err != tcpip.ErrConnectStarted { + if err := e.Connect(tcpip.FullAddress{NIC: e.boundNICID, Addr: e.connectingAddress, Port: e.ID.RemotePort}); err != tcpip.ErrConnectStarted { panic("endpoint connecting failed: " + err.String()) } connectingLoading.Done() @@ -263,8 +280,12 @@ func (e *endpoint) Resume(s *stack.Stack) { tcpip.AsyncLoading.Done() }() } - fallthrough + e.state = StateClose + e.stack.CompleteTransportEndpointCleanup(e) + tcpip.DeleteDanglingEndpoint(e) case StateError: + e.state = StateError + e.stack.CompleteTransportEndpointCleanup(e) tcpip.DeleteDanglingEndpoint(e) } } @@ -288,21 +309,21 @@ func (e *endpoint) loadLastError(s string) { } // saveHardError is invoked by stateify. -func (e *endpoint) saveHardError() string { - if e.hardError == nil { +func (e *EndpointInfo) saveHardError() string { + if e.HardError == nil { return "" } - return e.hardError.String() + return e.HardError.String() } // loadHardError is invoked by stateify. -func (e *endpoint) loadHardError(s string) { +func (e *EndpointInfo) loadHardError(s string) { if s == "" { return } - e.hardError = loadError(s) + e.HardError = loadError(s) } var messageToError map[string]*tcpip.Error diff --git a/pkg/tcpip/transport/tcp/forwarder.go b/pkg/tcpip/transport/tcp/forwarder.go index 63666f0b3..4983bca81 100644 --- a/pkg/tcpip/transport/tcp/forwarder.go +++ b/pkg/tcpip/transport/tcp/forwarder.go @@ -18,7 +18,6 @@ import ( "sync" "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/seqnum" "gvisor.dev/gvisor/pkg/tcpip/stack" @@ -63,8 +62,8 @@ func NewForwarder(s *stack.Stack, rcvWnd, maxInFlight int, handler func(*Forward // // This function is expected to be passed as an argument to the // stack.SetTransportProtocolHandler function. -func (f *Forwarder) HandlePacket(r *stack.Route, id stack.TransportEndpointID, netHeader buffer.View, vv buffer.VectorisedView) bool { - s := newSegment(r, id, vv) +func (f *Forwarder) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt tcpip.PacketBuffer) bool { + s := newSegment(r, id, pkt) defer s.decRef() // We only care about well-formed SYN packets. diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index d5d8ab96a..89b965c23 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -23,6 +23,7 @@ package tcp import ( "strings" "sync" + "time" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" @@ -54,12 +55,23 @@ const ( // MaxUnprocessedSegments is the maximum number of unprocessed segments // that can be queued for a given endpoint. MaxUnprocessedSegments = 300 + + // DefaultTCPLingerTimeout is the amount of time that sockets linger in + // FIN_WAIT_2 state before being marked closed. + DefaultTCPLingerTimeout = 60 * time.Second + + // DefaultTCPTimeWaitTimeout is the amount of time that sockets linger + // in TIME_WAIT state before being marked closed. + DefaultTCPTimeWaitTimeout = 60 * time.Second ) // SACKEnabled option can be used to enable SACK support in the TCP // protocol. See: https://tools.ietf.org/html/rfc2018. type SACKEnabled bool +// DelayEnabled option can be used to enable Nagle's algorithm in the TCP protocol. +type DelayEnabled bool + // SendBufferSizeOption allows the default, min and max send buffer sizes for // TCP endpoints to be queried or configured. type SendBufferSizeOption struct { @@ -84,11 +96,14 @@ const ( type protocol struct { mu sync.Mutex sackEnabled bool + delayEnabled bool sendBufferSize SendBufferSizeOption recvBufferSize ReceiveBufferSizeOption congestionControl string availableCongestionControl []string moderateReceiveBuffer bool + tcpLingerTimeout time.Duration + tcpTimeWaitTimeout time.Duration } // Number returns the tcp protocol number. @@ -97,7 +112,7 @@ func (*protocol) Number() tcpip.TransportProtocolNumber { } // NewEndpoint creates a new tcp endpoint. -func (*protocol) NewEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { +func (p *protocol) NewEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { return newEndpoint(stack, netProto, waiterQueue), nil } @@ -126,8 +141,8 @@ func (*protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) { // a reset is sent in response to any incoming segment except another reset. In // particular, SYNs addressed to a non-existent connection are rejected by this // means." -func (*protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, netHeader buffer.View, vv buffer.VectorisedView) bool { - s := newSegment(r, id, vv) +func (*protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt tcpip.PacketBuffer) bool { + s := newSegment(r, id, pkt) defer s.decRef() if !s.parse() || !s.csumValid { @@ -153,7 +168,7 @@ func replyWithReset(s *segment) { ack := s.sequenceNumber.Add(s.logicalLen()) - sendTCP(&s.route, s.id, buffer.VectorisedView{}, s.route.DefaultTTL(), header.TCPFlagRst|header.TCPFlagAck, seq, ack, 0, nil /* options */, nil /* gso */) + sendTCP(&s.route, s.id, buffer.VectorisedView{}, s.route.DefaultTTL(), stack.DefaultTOS, header.TCPFlagRst|header.TCPFlagAck, seq, ack, 0 /* rcvWnd */, nil /* options */, nil /* gso */) } // SetOption implements TransportProtocol.SetOption. @@ -165,6 +180,12 @@ func (p *protocol) SetOption(option interface{}) *tcpip.Error { p.mu.Unlock() return nil + case DelayEnabled: + p.mu.Lock() + p.delayEnabled = bool(v) + p.mu.Unlock() + return nil + case SendBufferSizeOption: if v.Min <= 0 || v.Default < v.Min || v.Default > v.Max { return tcpip.ErrInvalidOptionValue @@ -202,6 +223,24 @@ func (p *protocol) SetOption(option interface{}) *tcpip.Error { p.mu.Unlock() return nil + case tcpip.TCPLingerTimeoutOption: + if v < 0 { + v = 0 + } + p.mu.Lock() + p.tcpLingerTimeout = time.Duration(v) + p.mu.Unlock() + return nil + + case tcpip.TCPTimeWaitTimeoutOption: + if v < 0 { + v = 0 + } + p.mu.Lock() + p.tcpTimeWaitTimeout = time.Duration(v) + p.mu.Unlock() + return nil + default: return tcpip.ErrUnknownProtocolOption } @@ -216,6 +255,12 @@ func (p *protocol) Option(option interface{}) *tcpip.Error { p.mu.Unlock() return nil + case *DelayEnabled: + p.mu.Lock() + *v = DelayEnabled(p.delayEnabled) + p.mu.Unlock() + return nil + case *SendBufferSizeOption: p.mu.Lock() *v = p.sendBufferSize @@ -246,6 +291,18 @@ func (p *protocol) Option(option interface{}) *tcpip.Error { p.mu.Unlock() return nil + case *tcpip.TCPLingerTimeoutOption: + p.mu.Lock() + *v = tcpip.TCPLingerTimeoutOption(p.tcpLingerTimeout) + p.mu.Unlock() + return nil + + case *tcpip.TCPTimeWaitTimeoutOption: + p.mu.Lock() + *v = tcpip.TCPTimeWaitTimeoutOption(p.tcpTimeWaitTimeout) + p.mu.Unlock() + return nil + default: return tcpip.ErrUnknownProtocolOption } @@ -258,5 +315,7 @@ func NewProtocol() stack.TransportProtocol { recvBufferSize: ReceiveBufferSizeOption{MinBufferSize, DefaultReceiveBufferSize, MaxBufferSize}, congestionControl: ccReno, availableCongestionControl: []string{ccReno, ccCubic}, + tcpLingerTimeout: DefaultTCPLingerTimeout, + tcpTimeWaitTimeout: DefaultTCPTimeWaitTimeout, } } diff --git a/pkg/tcpip/transport/tcp/rcv.go b/pkg/tcpip/transport/tcp/rcv.go index e90f9a7d9..068b90fb6 100644 --- a/pkg/tcpip/transport/tcp/rcv.go +++ b/pkg/tcpip/transport/tcp/rcv.go @@ -18,6 +18,7 @@ import ( "container/heap" "time" + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/seqnum" ) @@ -209,6 +210,11 @@ func (r *receiver) consumeSegment(s *segment, segSeq seqnum.Value, segLen seqnum switch r.ep.state { case StateFinWait1: r.ep.state = StateFinWait2 + // Notify protocol goroutine that we have received an + // ACK to our FIN so that it can start the FIN_WAIT2 + // timer to abort connection if the other side does + // not close within 2MSL. + r.ep.notifyProtocolGoroutine(notifyClose) case StateClosing: r.ep.state = StateTimeWait case StateLastAck: @@ -253,23 +259,105 @@ func (r *receiver) updateRTT() { r.ep.rcvListMu.Unlock() } -// handleRcvdSegment handles TCP segments directed at the connection managed by -// r as they arrive. It is called by the protocol main loop. -func (r *receiver) handleRcvdSegment(s *segment) { +func (r *receiver) handleRcvdSegmentClosing(s *segment, state EndpointState, closed bool) (drop bool, err *tcpip.Error) { + r.ep.rcvListMu.Lock() + rcvClosed := r.ep.rcvClosed || r.closed + r.ep.rcvListMu.Unlock() + + // If we are in one of the shutdown states then we need to do + // additional checks before we try and process the segment. + switch state { + case StateCloseWait, StateClosing, StateLastAck: + if !s.sequenceNumber.LessThanEq(r.rcvNxt) { + s.decRef() + // Just drop the segment as we have + // already received a FIN and this + // segment is after the sequence number + // for the FIN. + return true, nil + } + fallthrough + case StateFinWait1: + fallthrough + case StateFinWait2: + // If we are closed for reads (either due to an + // incoming FIN or the user calling shutdown(.., + // SHUT_RD) then any data past the rcvNxt should + // trigger a RST. + endDataSeq := s.sequenceNumber.Add(seqnum.Size(s.data.Size())) + if rcvClosed && r.rcvNxt.LessThan(endDataSeq) { + s.decRef() + return true, tcpip.ErrConnectionAborted + } + if state == StateFinWait1 { + break + } + + // If it's a retransmission of an old data segment + // or a pure ACK then allow it. + if s.sequenceNumber.Add(s.logicalLen()).LessThanEq(r.rcvNxt) || + s.logicalLen() == 0 { + break + } + + // In FIN-WAIT2 if the socket is fully + // closed(not owned by application on our end + // then the only acceptable segment is a + // FIN. Since FIN can technically also carry + // data we verify that the segment carrying a + // FIN ends at exactly e.rcvNxt+1. + // + // From RFC793 page 25. + // + // For sequence number purposes, the SYN is + // considered to occur before the first actual + // data octet of the segment in which it occurs, + // while the FIN is considered to occur after + // the last actual data octet in a segment in + // which it occurs. + if closed && (!s.flagIsSet(header.TCPFlagFin) || s.sequenceNumber.Add(s.logicalLen()) != r.rcvNxt+1) { + s.decRef() + return true, tcpip.ErrConnectionAborted + } + } + // We don't care about receive processing anymore if the receive side // is closed. - if r.closed { - return + // + // NOTE: We still want to permit a FIN as it's possible only our + // end has closed and the peer is yet to send a FIN. Hence we + // compare only the payload. + segEnd := s.sequenceNumber.Add(seqnum.Size(s.data.Size())) + if rcvClosed && !segEnd.LessThanEq(r.rcvNxt) { + return true, nil + } + return false, nil +} + +// handleRcvdSegment handles TCP segments directed at the connection managed by +// r as they arrive. It is called by the protocol main loop. +func (r *receiver) handleRcvdSegment(s *segment) (drop bool, err *tcpip.Error) { + r.ep.mu.RLock() + state := r.ep.state + closed := r.ep.closed + r.ep.mu.RUnlock() + + if state != StateEstablished { + drop, err := r.handleRcvdSegmentClosing(s, state, closed) + if drop || err != nil { + return drop, err + } } segLen := seqnum.Size(s.data.Size()) segSeq := s.sequenceNumber // If the sequence number range is outside the acceptable range, just - // send an ACK. This is according to RFC 793, page 37. + // send an ACK and stop further processing of the segment. + // This is according to RFC 793, page 68. if !r.acceptable(segSeq, segLen) { r.ep.snd.sendAck() - return + return true, nil } // Defer segment processing if it can't be consumed now. @@ -288,7 +376,7 @@ func (r *receiver) handleRcvdSegment(s *segment) { // have to retransmit. r.ep.snd.sendAck() } - return + return false, nil } // Since we consumed a segment update the receiver's RTT estimate @@ -315,4 +403,67 @@ func (r *receiver) handleRcvdSegment(s *segment) { r.pendingBufUsed -= s.logicalLen() s.decRef() } + return false, nil +} + +// handleTimeWaitSegment handles inbound segments received when the endpoint +// has entered the TIME_WAIT state. +func (r *receiver) handleTimeWaitSegment(s *segment) (resetTimeWait bool, newSyn bool) { + segSeq := s.sequenceNumber + segLen := seqnum.Size(s.data.Size()) + + // Just silently drop any RST packets in TIME_WAIT. We do not support + // TIME_WAIT assasination as a result we confirm w/ fix 1 as described + // in https://tools.ietf.org/html/rfc1337#section-3. + if s.flagIsSet(header.TCPFlagRst) { + return false, false + } + + // If it's a SYN and the sequence number is higher than any seen before + // for this connection then try and redirect it to a listening endpoint + // if available. + // + // RFC 1122: + // "When a connection is [...] on TIME-WAIT state [...] + // [a TCP] MAY accept a new SYN from the remote TCP to + // reopen the connection directly, if it: + + // (1) assigns its initial sequence number for the new + // connection to be larger than the largest sequence + // number it used on the previous connection incarnation, + // and + + // (2) returns to TIME-WAIT state if the SYN turns out + // to be an old duplicate". + if s.flagIsSet(header.TCPFlagSyn) && r.rcvNxt.LessThan(segSeq) { + + return false, true + } + + // Drop the segment if it does not contain an ACK. + if !s.flagIsSet(header.TCPFlagAck) { + return false, false + } + + // Update Timestamp if required. See RFC7323, section-4.3. + if r.ep.sendTSOk && s.parsedOptions.TS { + r.ep.updateRecentTimestamp(s.parsedOptions.TSVal, r.ep.snd.maxSentAck, segSeq) + } + + if segSeq.Add(1) == r.rcvNxt && s.flagIsSet(header.TCPFlagFin) { + // If it's a FIN-ACK then resetTimeWait and send an ACK, as it + // indicates our final ACK could have been lost. + r.ep.snd.sendAck() + return true, false + } + + // If the sequence number range is outside the acceptable range or + // carries data then just send an ACK. This is according to RFC 793, + // page 37. + // + // NOTE: In TIME_WAIT the only acceptable sequence number is rcvNxt. + if segSeq != r.rcvNxt || segLen != 0 { + r.ep.snd.sendAck() + } + return false, false } diff --git a/pkg/tcpip/transport/tcp/segment.go b/pkg/tcpip/transport/tcp/segment.go index ea725d513..1c10da5ca 100644 --- a/pkg/tcpip/transport/tcp/segment.go +++ b/pkg/tcpip/transport/tcp/segment.go @@ -18,6 +18,7 @@ import ( "sync/atomic" "time" + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/seqnum" @@ -60,13 +61,13 @@ type segment struct { xmitTime time.Time `state:".(unixTime)"` } -func newSegment(r *stack.Route, id stack.TransportEndpointID, vv buffer.VectorisedView) *segment { +func newSegment(r *stack.Route, id stack.TransportEndpointID, pkt tcpip.PacketBuffer) *segment { s := &segment{ refCnt: 1, id: id, route: r.Clone(), } - s.data = vv.Clone(s.views[:]) + s.data = pkt.Data.Clone(s.views[:]) s.rcvdTime = time.Now() return s } @@ -99,8 +100,14 @@ func (s *segment) clone() *segment { return t } -func (s *segment) flagIsSet(flag uint8) bool { - return (s.flags & flag) != 0 +// flagIsSet checks if at least one flag in flags is set in s.flags. +func (s *segment) flagIsSet(flags uint8) bool { + return s.flags&flags != 0 +} + +// flagsAreSet checks if all flags in flags are set in s.flags. +func (s *segment) flagsAreSet(flags uint8) bool { + return s.flags&flags == flags } func (s *segment) decRef() { diff --git a/pkg/tcpip/transport/tcp/snd.go b/pkg/tcpip/transport/tcp/snd.go index 735edfe55..d3f7c9125 100644 --- a/pkg/tcpip/transport/tcp/snd.go +++ b/pkg/tcpip/transport/tcp/snd.go @@ -417,6 +417,7 @@ func (s *sender) resendSegment() { s.fr.rescueRxt = seg.sequenceNumber.Add(seqnum.Size(seg.data.Size())) - 1 s.sendSegment(seg) s.ep.stack.Stats().TCP.FastRetransmit.Increment() + s.ep.stats.SendErrors.FastRetransmit.Increment() // Run SetPipe() as per RFC 6675 section 5 Step 4.4 s.SetPipe() @@ -435,6 +436,7 @@ func (s *sender) retransmitTimerExpired() bool { } s.ep.stack.Stats().TCP.Timeouts.Increment() + s.ep.stats.SendErrors.Timeouts.Increment() // Give up if we've waited more than a minute since the last resend. if s.rto >= 60*time.Second { @@ -672,6 +674,7 @@ func (s *sender) maybeSendSegment(seg *segment, limit int, end seqnum.Value) (se default: s.ep.state = StateFinWait1 } + s.ep.stack.Stats().TCP.CurrentEstablished.Decrement() s.ep.mu.Unlock() } else { // We're sending a non-FIN segment. @@ -1188,6 +1191,7 @@ func (s *sender) handleRcvdSegment(seg *segment) { func (s *sender) sendSegment(seg *segment) *tcpip.Error { if !seg.xmitTime.IsZero() { s.ep.stack.Stats().TCP.Retransmits.Increment() + s.ep.stats.SendErrors.Retransmits.Increment() if s.sndCwnd < s.sndSsthresh { s.ep.stack.Stats().TCP.SlowStartRetransmits.Increment() } diff --git a/pkg/tcpip/transport/tcp/tcp_noracedetector_test.go b/pkg/tcpip/transport/tcp/tcp_noracedetector_test.go index 9fa97528b..782d7b42c 100644 --- a/pkg/tcpip/transport/tcp/tcp_noracedetector_test.go +++ b/pkg/tcpip/transport/tcp/tcp_noracedetector_test.go @@ -500,6 +500,14 @@ func TestRetransmit(t *testing.T) { t.Errorf("got stats.TCP.Retransmits.Value = %v, want = %v", got, want) } + if got, want := c.EP.Stats().(*tcp.Stats).SendErrors.Timeouts.Value(), uint64(1); got != want { + t.Errorf("got EP SendErrors.Timeouts.Value = %v, want = %v", got, want) + } + + if got, want := c.EP.Stats().(*tcp.Stats).SendErrors.Retransmits.Value(), uint64(1); got != want { + t.Errorf("got EP stats SendErrors.Retransmits.Value = %v, want = %v", got, want) + } + if got, want := c.Stack().Stats().TCP.SlowStartRetransmits.Value(), uint64(1); got != want { t.Errorf("got stats.TCP.SlowStartRetransmits.Value = %v, want = %v", got, want) } diff --git a/pkg/tcpip/transport/tcp/tcp_sack_test.go b/pkg/tcpip/transport/tcp/tcp_sack_test.go index 4e7f1a740..afea124ec 100644 --- a/pkg/tcpip/transport/tcp/tcp_sack_test.go +++ b/pkg/tcpip/transport/tcp/tcp_sack_test.go @@ -520,10 +520,18 @@ func TestSACKRecovery(t *testing.T) { t.Errorf("got stats.TCP.FastRetransmit.Value = %v, want = %v", got, want) } + if got, want := c.EP.Stats().(*tcp.Stats).SendErrors.FastRetransmit.Value(), uint64(1); got != want { + t.Errorf("got EP stats SendErrors.FastRetransmit = %v, want = %v", got, want) + } + if got, want := c.Stack().Stats().TCP.Retransmits.Value(), uint64(4); got != want { t.Errorf("got stats.TCP.Retransmits.Value = %v, want = %v", got, want) } + if got, want := c.EP.Stats().(*tcp.Stats).SendErrors.Retransmits.Value(), uint64(4); got != want { + t.Errorf("got EP stats Stats.SendErrors.Retransmits = %v, want = %v", got, want) + } + c.CheckNoPacketTimeout("More packets received than expected during recovery after partial ack for this cwnd.", 50*time.Millisecond) // Acknowledge all pending data to recover point. diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index 2be094876..84579ce52 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -99,7 +99,10 @@ func TestConnectDoesNotIncrementFailedConnectionAttempts(t *testing.T) { c.CreateConnected(789, 30000, -1 /* epRcvBuf */) if got := stats.TCP.FailedConnectionAttempts.Value(); got != want { - t.Errorf("got stats.TCP.FailedConnectionOpenings.Value() = %v, want = %v", got, want) + t.Errorf("got stats.TCP.FailedConnectionAttempts.Value() = %v, want = %v", got, want) + } + if got := c.EP.Stats().(*tcp.Stats).FailedConnectionAttempts.Value(); got != want { + t.Errorf("got EP stats.FailedConnectionAttempts = %v, want = %v", got, want) } } @@ -122,6 +125,9 @@ func TestActiveFailedConnectionAttemptIncrement(t *testing.T) { if got := stats.TCP.FailedConnectionAttempts.Value(); got != want { t.Errorf("got stats.TCP.FailedConnectionAttempts.Value() = %v, want = %v", got, want) } + if got := c.EP.Stats().(*tcp.Stats).FailedConnectionAttempts.Value(); got != want { + t.Errorf("got EP stats FailedConnectionAttempts = %v, want = %v", got, want) + } } func TestTCPSegmentsSentIncrement(t *testing.T) { @@ -136,6 +142,9 @@ func TestTCPSegmentsSentIncrement(t *testing.T) { if got := stats.TCP.SegmentsSent.Value(); got != want { t.Errorf("got stats.TCP.SegmentsSent.Value() = %v, want = %v", got, want) } + if got := c.EP.Stats().(*tcp.Stats).SegmentsSent.Value(); got != want { + t.Errorf("got EP stats SegmentsSent.Value() = %v, want = %v", got, want) + } } func TestTCPResetsSentIncrement(t *testing.T) { @@ -197,17 +206,18 @@ func TestTCPResetSentForACKWhenNotUsingSynCookies(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() + // Set TCPLingerTimeout to 5 seconds so that sockets are marked closed wq := &waiter.Queue{} ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, wq) if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) + t.Fatalf("NewEndpoint failed: %s", err) } if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) + t.Fatalf("Bind failed: %s", err) } if err := ep.Listen(10); err != nil { - t.Fatalf("Listen failed: %v", err) + t.Fatalf("Listen failed: %s", err) } // Send a SYN request. @@ -247,7 +257,7 @@ func TestTCPResetSentForACKWhenNotUsingSynCookies(t *testing.T) { case <-ch: c.EP, _, err = ep.Accept() if err != nil { - t.Fatalf("Accept failed: %v", err) + t.Fatalf("Accept failed: %s", err) } case <-time.After(1 * time.Second): @@ -255,6 +265,13 @@ func TestTCPResetSentForACKWhenNotUsingSynCookies(t *testing.T) { } } + // Lower stackwide TIME_WAIT timeout so that the reservations + // are released instantly on Close. + tcpTW := tcpip.TCPTimeWaitTimeoutOption(1 * time.Millisecond) + if err := c.Stack().SetTransportProtocolOption(tcp.ProtocolNumber, tcpTW); err != nil { + t.Fatalf("e.stack.SetTransportProtocolOption(%d, %s) = %s", tcp.ProtocolNumber, tcpTW, err) + } + c.EP.Close() checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), @@ -276,6 +293,11 @@ func TestTCPResetSentForACKWhenNotUsingSynCookies(t *testing.T) { // Get the ACK to the FIN we just sent. c.GetPacket() + // Since an active close was done we need to wait for a little more than + // tcpLingerTimeout for the port reservations to be released and the + // socket to move to a CLOSED state. + time.Sleep(20 * time.Millisecond) + // Now resend the same ACK, this ACK should generate a RST as there // should be no endpoint in SYN-RCVD state and we are not using // syn-cookies yet. The reason we send the same ACK is we need a valid @@ -367,6 +389,13 @@ func TestConnectResetAfterClose(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() + // Set TCPLinger to 3 seconds so that sockets are marked closed + // after 3 second in FIN_WAIT2 state. + tcpLingerTimeout := 3 * time.Second + if err := c.Stack().SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.TCPLingerTimeoutOption(tcpLingerTimeout)); err != nil { + t.Fatalf("c.stack.SetTransportProtocolOption(tcp, tcpip.TCPLingerTimeoutOption(%d) failed: %s", tcpLingerTimeout, err) + } + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) ep := c.EP c.EP = nil @@ -387,12 +416,24 @@ func TestConnectResetAfterClose(t *testing.T) { DstPort: c.Port, Flags: header.TCPFlagAck, SeqNum: 790, - AckNum: c.IRS.Add(1), + AckNum: c.IRS.Add(2), + RcvWnd: 30000, + }) + + // Wait for the ep to give up waiting for a FIN. + time.Sleep(tcpLingerTimeout + 1*time.Second) + + // Now send an ACK and it should trigger a RST as the endpoint should + // not exist anymore. + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + Flags: header.TCPFlagAck, + SeqNum: 790, + AckNum: c.IRS.Add(2), RcvWnd: 30000, }) - // Wait for the ep to give up waiting for a FIN, and send a RST. - time.Sleep(3 * time.Second) for { b := c.GetPacket() tcpHdr := header.TCP(header.IPv4(b).Payload()) @@ -404,7 +445,7 @@ func TestConnectResetAfterClose(t *testing.T) { checker.IPv4(t, b, checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), + checker.SeqNum(uint32(c.IRS)+2), checker.AckNum(790), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagRst), ), @@ -465,6 +506,347 @@ func TestSimpleReceive(t *testing.T) { ) } +// TestUserSuppliedMSSOnConnectV4 tests that the user supplied MSS is used when +// creating a new active IPv4 TCP socket. It should be present in the sent TCP +// SYN segment. +func TestUserSuppliedMSSOnConnectV4(t *testing.T) { + const mtu = 5000 + const maxMSS = mtu - header.IPv4MinimumSize - header.TCPMinimumSize + tests := []struct { + name string + setMSS uint16 + expMSS uint16 + }{ + { + "EqualToMaxMSS", + maxMSS, + maxMSS, + }, + { + "LessThanMTU", + maxMSS - 1, + maxMSS - 1, + }, + { + "GreaterThanMTU", + maxMSS + 1, + maxMSS, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + c := context.New(t, mtu) + defer c.Cleanup() + + c.Create(-1) + + // Set the MSS socket option. + opt := tcpip.MaxSegOption(test.setMSS) + if err := c.EP.SetSockOpt(opt); err != nil { + t.Fatalf("SetSockOpt(%#v) failed: %s", opt, err) + } + + // Get expected window size. + rcvBufSize, err := c.EP.GetSockOptInt(tcpip.ReceiveBufferSizeOption) + if err != nil { + t.Fatalf("GetSockOpt(%v) failed: %s", tcpip.ReceiveBufferSizeOption, err) + } + ws := tcp.FindWndScale(seqnum.Size(rcvBufSize)) + + // Start connection attempt to IPv4 address. + if err := c.EP.Connect(tcpip.FullAddress{Addr: context.TestAddr, Port: context.TestPort}); err != tcpip.ErrConnectStarted { + t.Fatalf("Unexpected return value from Connect: %v", err) + } + + // Receive SYN packet with our user supplied MSS. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagSyn), + checker.TCPSynOptions(header.TCPSynOptions{MSS: test.expMSS, WS: ws}))) + }) + } +} + +// TestUserSuppliedMSSOnConnectV6 tests that the user supplied MSS is used when +// creating a new active IPv6 TCP socket. It should be present in the sent TCP +// SYN segment. +func TestUserSuppliedMSSOnConnectV6(t *testing.T) { + const mtu = 5000 + const maxMSS = mtu - header.IPv6MinimumSize - header.TCPMinimumSize + tests := []struct { + name string + setMSS uint16 + expMSS uint16 + }{ + { + "EqualToMaxMSS", + maxMSS, + maxMSS, + }, + { + "LessThanMTU", + maxMSS - 1, + maxMSS - 1, + }, + { + "GreaterThanMTU", + maxMSS + 1, + maxMSS, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + c := context.New(t, mtu) + defer c.Cleanup() + + c.CreateV6Endpoint(true) + + // Set the MSS socket option. + opt := tcpip.MaxSegOption(test.setMSS) + if err := c.EP.SetSockOpt(opt); err != nil { + t.Fatalf("SetSockOpt(%#v) failed: %s", opt, err) + } + + // Get expected window size. + rcvBufSize, err := c.EP.GetSockOptInt(tcpip.ReceiveBufferSizeOption) + if err != nil { + t.Fatalf("GetSockOpt(%v) failed: %s", tcpip.ReceiveBufferSizeOption, err) + } + ws := tcp.FindWndScale(seqnum.Size(rcvBufSize)) + + // Start connection attempt to IPv6 address. + if err := c.EP.Connect(tcpip.FullAddress{Addr: context.TestV6Addr, Port: context.TestPort}); err != tcpip.ErrConnectStarted { + t.Fatalf("Unexpected return value from Connect: %v", err) + } + + // Receive SYN packet with our user supplied MSS. + checker.IPv6(t, c.GetV6Packet(), checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagSyn), + checker.TCPSynOptions(header.TCPSynOptions{MSS: test.expMSS, WS: ws}))) + }) + } +} + +func TestSendRstOnListenerRxSynAckV4(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.Create(-1) + + if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatal("Bind failed:", err) + } + + if err := c.EP.Listen(10); err != nil { + t.Fatal("Listen failed:", err) + } + + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn | header.TCPFlagAck, + SeqNum: 100, + AckNum: 200, + }) + + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagRst), + checker.SeqNum(200))) +} + +func TestSendRstOnListenerRxSynAckV6(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.CreateV6Endpoint(true) + + if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatal("Bind failed:", err) + } + + if err := c.EP.Listen(10); err != nil { + t.Fatal("Listen failed:", err) + } + + c.SendV6Packet(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn | header.TCPFlagAck, + SeqNum: 100, + AckNum: 200, + }) + + checker.IPv6(t, c.GetV6Packet(), checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagRst), + checker.SeqNum(200))) +} + +func TestTOSV4(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &c.WQ) + if err != nil { + t.Fatalf("NewEndpoint failed: %s", err) + } + c.EP = ep + + const tos = 0xC0 + if err := c.EP.SetSockOpt(tcpip.IPv4TOSOption(tos)); err != nil { + t.Errorf("SetSockOpt(%#v) failed: %s", tcpip.IPv4TOSOption(tos), err) + } + + var v tcpip.IPv4TOSOption + if err := c.EP.GetSockOpt(&v); err != nil { + t.Errorf("GetSockopt failed: %s", err) + } + + if want := tcpip.IPv4TOSOption(tos); v != want { + t.Errorf("got GetSockOpt(...) = %#v, want = %#v", v, want) + } + + testV4Connect(t, c, checker.TOS(tos, 0)) + + data := []byte{1, 2, 3} + view := buffer.NewView(len(data)) + copy(view, data) + + if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { + t.Fatalf("Write failed: %s", err) + } + + // Check that data is received. + b := c.GetPacket() + checker.IPv4(t, b, + checker.PayloadLen(len(data)+header.TCPMinimumSize), + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS)+1), + checker.AckNum(790), // Acknum is initial sequence number + 1 + checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), + ), + checker.TOS(tos, 0), + ) + + if p := b[header.IPv4MinimumSize+header.TCPMinimumSize:]; !bytes.Equal(data, p) { + t.Errorf("got data = %x, want = %x", p, data) + } +} + +func TestTrafficClassV6(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.CreateV6Endpoint(false) + + const tos = 0xC0 + if err := c.EP.SetSockOpt(tcpip.IPv6TrafficClassOption(tos)); err != nil { + t.Errorf("SetSockOpt(%#v) failed: %s", tcpip.IPv6TrafficClassOption(tos), err) + } + + var v tcpip.IPv6TrafficClassOption + if err := c.EP.GetSockOpt(&v); err != nil { + t.Fatalf("GetSockopt failed: %s", err) + } + + if want := tcpip.IPv6TrafficClassOption(tos); v != want { + t.Errorf("got GetSockOpt(...) = %#v, want = %#v", v, want) + } + + // Test the connection request. + testV6Connect(t, c, checker.TOS(tos, 0)) + + data := []byte{1, 2, 3} + view := buffer.NewView(len(data)) + copy(view, data) + + if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { + t.Fatalf("Write failed: %s", err) + } + + // Check that data is received. + b := c.GetV6Packet() + checker.IPv6(t, b, + checker.PayloadLen(len(data)+header.TCPMinimumSize), + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS)+1), + checker.AckNum(790), + checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), + ), + checker.TOS(tos, 0), + ) + + if p := b[header.IPv6MinimumSize+header.TCPMinimumSize:]; !bytes.Equal(data, p) { + t.Errorf("got data = %x, want = %x", p, data) + } +} + +func TestConnectBindToDevice(t *testing.T) { + for _, test := range []struct { + name string + device string + want tcp.EndpointState + }{ + {"RightDevice", "nic1", tcp.StateEstablished}, + {"WrongDevice", "nic2", tcp.StateSynSent}, + {"AnyDevice", "", tcp.StateEstablished}, + } { + t.Run(test.name, func(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.Create(-1) + bindToDevice := tcpip.BindToDeviceOption(test.device) + c.EP.SetSockOpt(bindToDevice) + // Start connection attempt. + waitEntry, _ := waiter.NewChannelEntry(nil) + c.WQ.EventRegister(&waitEntry, waiter.EventOut) + defer c.WQ.EventUnregister(&waitEntry) + + if err := c.EP.Connect(tcpip.FullAddress{Addr: context.TestAddr, Port: context.TestPort}); err != tcpip.ErrConnectStarted { + t.Fatalf("Unexpected return value from Connect: %v", err) + } + + // Receive SYN packet. + b := c.GetPacket() + checker.IPv4(t, b, + checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagSyn), + ), + ) + if got, want := tcp.EndpointState(c.EP.State()), tcp.StateSynSent; got != want { + t.Fatalf("Unexpected endpoint state: want %v, got %v", want, got) + } + tcpHdr := header.TCP(header.IPv4(b).Payload()) + c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) + + iss := seqnum.Value(789) + rcvWnd := seqnum.Size(30000) + c.SendPacket(nil, &context.Headers{ + SrcPort: tcpHdr.DestinationPort(), + DstPort: tcpHdr.SourcePort(), + Flags: header.TCPFlagSyn | header.TCPFlagAck, + SeqNum: iss, + AckNum: c.IRS.Add(1), + RcvWnd: rcvWnd, + TCPOpts: nil, + }) + + c.GetPacket() + if got, want := tcp.EndpointState(c.EP.State()), test.want; got != want { + t.Fatalf("Unexpected endpoint state: want %v, got %v", want, got) + } + }) + } +} + func TestOutOfOrderReceive(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() @@ -760,8 +1142,7 @@ func TestRstOnCloseWithUnreadDataFinConvertRst(t *testing.T) { checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagRst), - // We shouldn't consume a sequence number on RST. - checker.SeqNum(uint32(c.IRS)+1), + checker.SeqNum(uint32(c.IRS)+2), )) // The RST puts the endpoint into an error state. if got, want := tcp.EndpointState(c.EP.State()), tcp.StateError; got != want { @@ -797,6 +1178,10 @@ func TestShutdownRead(t *testing.T) { if _, _, err := c.EP.Read(nil); err != tcpip.ErrClosedForReceive { t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrClosedForReceive) } + var want uint64 = 1 + if got := c.EP.Stats().(*tcp.Stats).ReadErrors.ReadClosed.Value(); got != want { + t.Fatalf("got EP stats Stats.ReadErrors.ReadClosed got %v want %v", got, want) + } } func TestFullWindowReceive(t *testing.T) { @@ -853,6 +1238,11 @@ func TestFullWindowReceive(t *testing.T) { t.Fatalf("got data = %v, want = %v", v, data) } + var want uint64 = 1 + if got := c.EP.Stats().(*tcp.Stats).ReceiveErrors.ZeroRcvWindowState.Value(); got != want { + t.Fatalf("got EP stats ReceiveErrors.ZeroRcvWindowState got %v want %v", got, want) + } + // Check that we get an ACK for the newly non-zero window. checker.IPv4(t, c.GetPacket(), checker.TCP( @@ -1444,7 +1834,7 @@ func TestDelay(t *testing.T) { c.CreateConnected(789, 30000, -1 /* epRcvBuf */) - c.EP.SetSockOpt(tcpip.DelayOption(1)) + c.EP.SetSockOptInt(tcpip.DelayOption, 1) var allData []byte for i, data := range [][]byte{{0}, {1, 2, 3, 4}, {5, 6, 7}, {8, 9}, {10}, {11}} { @@ -1492,7 +1882,7 @@ func TestUndelay(t *testing.T) { c.CreateConnected(789, 30000, -1 /* epRcvBuf */) - c.EP.SetSockOpt(tcpip.DelayOption(1)) + c.EP.SetSockOptInt(tcpip.DelayOption, 1) allData := [][]byte{{0}, {1, 2, 3}} for i, data := range allData { @@ -1525,7 +1915,7 @@ func TestUndelay(t *testing.T) { // Check that we don't get the second packet yet. c.CheckNoPacketTimeout("delayed second packet transmitted", 100*time.Millisecond) - c.EP.SetSockOpt(tcpip.DelayOption(0)) + c.EP.SetSockOptInt(tcpip.DelayOption, 0) // Check that data is received. second := c.GetPacket() @@ -1562,7 +1952,7 @@ func TestMSSNotDelayed(t *testing.T) { fn func(tcpip.Endpoint) }{ {"no-op", func(tcpip.Endpoint) {}}, - {"delay", func(ep tcpip.Endpoint) { ep.SetSockOpt(tcpip.DelayOption(1)) }}, + {"delay", func(ep tcpip.Endpoint) { ep.SetSockOptInt(tcpip.DelayOption, 1) }}, {"cork", func(ep tcpip.Endpoint) { ep.SetSockOpt(tcpip.CorkOption(1)) }}, } @@ -1692,6 +2082,34 @@ func TestSendGreaterThanMTU(t *testing.T) { testBrokenUpWrite(t, c, maxPayload) } +func TestSetTTL(t *testing.T) { + for _, wantTTL := range []uint8{1, 2, 50, 64, 128, 254, 255} { + t.Run(fmt.Sprintf("TTL:%d", wantTTL), func(t *testing.T) { + c := context.New(t, 65535) + defer c.Cleanup() + + var err *tcpip.Error + c.EP, err = c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) + if err != nil { + t.Fatalf("NewEndpoint failed: %v", err) + } + + if err := c.EP.SetSockOpt(tcpip.TTLOption(wantTTL)); err != nil { + t.Fatalf("SetSockOpt failed: %v", err) + } + + if err := c.EP.Connect(tcpip.FullAddress{Addr: context.TestAddr, Port: context.TestPort}); err != tcpip.ErrConnectStarted { + t.Fatalf("Unexpected return value from Connect: %v", err) + } + + // Receive SYN packet. + b := c.GetPacket() + + checker.IPv4(t, b, checker.TTL(wantTTL)) + }) + } +} + func TestActiveSendMSSLessThanMTU(t *testing.T) { const maxPayload = 100 c := context.New(t, 65535) @@ -1997,6 +2415,13 @@ loop: t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrConnectionReset) } } + // Expect the state to be StateError and subsequent Reads to fail with HardError. + if _, _, err := c.EP.Read(nil); err != tcpip.ErrConnectionReset { + t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrConnectionReset) + } + if tcp.EndpointState(c.EP.State()) != tcp.StateError { + t.Fatalf("got EP state is not StateError") + } } func TestSendOnResetConnection(t *testing.T) { @@ -2568,6 +2993,17 @@ func TestReceivedValidSegmentCountIncrement(t *testing.T) { if got := stats.TCP.ValidSegmentsReceived.Value(); got != want { t.Errorf("got stats.TCP.ValidSegmentsReceived.Value() = %v, want = %v", got, want) } + if got := c.EP.Stats().(*tcp.Stats).SegmentsReceived.Value(); got != want { + t.Errorf("got EP stats Stats.SegmentsReceived = %v, want = %v", got, want) + } + // Ensure there were no errors during handshake. If these stats have + // incremented, then the connection should not have been established. + if got := c.EP.Stats().(*tcp.Stats).SendErrors.NoRoute.Value(); got != 0 { + t.Errorf("got EP stats Stats.SendErrors.NoRoute = %v, want = %v", got, 0) + } + if got := c.EP.Stats().(*tcp.Stats).SendErrors.NoLinkAddr.Value(); got != 0 { + t.Errorf("got EP stats Stats.SendErrors.NoLinkAddr = %v, want = %v", got, 0) + } } func TestReceivedInvalidSegmentCountIncrement(t *testing.T) { @@ -2592,6 +3028,9 @@ func TestReceivedInvalidSegmentCountIncrement(t *testing.T) { if got := stats.TCP.InvalidSegmentsReceived.Value(); got != want { t.Errorf("got stats.TCP.InvalidSegmentsReceived.Value() = %v, want = %v", got, want) } + if got := c.EP.Stats().(*tcp.Stats).ReceiveErrors.MalformedPacketsReceived.Value(); got != want { + t.Errorf("got EP Stats.ReceiveErrors.MalformedPacketsReceived stats = %v, want = %v", got, want) + } } func TestReceivedIncorrectChecksumIncrement(t *testing.T) { @@ -2618,6 +3057,9 @@ func TestReceivedIncorrectChecksumIncrement(t *testing.T) { if got := stats.TCP.ChecksumErrors.Value(); got != want { t.Errorf("got stats.TCP.ChecksumErrors.Value() = %d, want = %d", got, want) } + if got := c.EP.Stats().(*tcp.Stats).ReceiveErrors.ChecksumErrors.Value(); got != want { + t.Errorf("got EP stats Stats.ReceiveErrors.ChecksumErrors = %d, want = %d", got, want) + } } func TestReceivedSegmentQueuing(t *testing.T) { @@ -2674,6 +3116,13 @@ func TestReadAfterClosedState(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() + // Set TCPTimeWaitTimeout to 1 seconds so that sockets are marked closed + // after 1 second in TIME_WAIT state. + tcpTimeWaitTimeout := 1 * time.Second + if err := c.Stack().SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.TCPTimeWaitTimeoutOption(tcpTimeWaitTimeout)); err != nil { + t.Fatalf("c.stack.SetTransportProtocolOption(tcp, tcpip.TCPTimeWaitTimeout(%d) failed: %s", tcpTimeWaitTimeout, err) + } + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) we, ch := waiter.NewChannelEntry(nil) @@ -2681,12 +3130,12 @@ func TestReadAfterClosedState(t *testing.T) { defer c.WQ.EventUnregister(&we) if _, _, err := c.EP.Read(nil); err != tcpip.ErrWouldBlock { - t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrWouldBlock) + t.Fatalf("got c.EP.Read(nil) = %v, want = %s", err, tcpip.ErrWouldBlock) } // Shutdown immediately for write, check that we get a FIN. if err := c.EP.Shutdown(tcpip.ShutdownWrite); err != nil { - t.Fatalf("Shutdown failed: %v", err) + t.Fatalf("Shutdown failed: %s", err) } checker.IPv4(t, c.GetPacket(), @@ -2724,10 +3173,9 @@ func TestReadAfterClosedState(t *testing.T) { ), ) - // Give the stack the chance to transition to closed state. Note that since - // both the sender and receiver are now closed, we effectively skip the - // TIME-WAIT state. - time.Sleep(1 * time.Second) + // Give the stack the chance to transition to closed state from + // TIME_WAIT. + time.Sleep(tcpTimeWaitTimeout * 2) if got, want := tcp.EndpointState(c.EP.State()), tcp.StateClose; got != want { t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) @@ -2744,7 +3192,7 @@ func TestReadAfterClosedState(t *testing.T) { peekBuf := make([]byte, 10) n, _, err := c.EP.Peek([][]byte{peekBuf}) if err != nil { - t.Fatalf("Peek failed: %v", err) + t.Fatalf("Peek failed: %s", err) } peekBuf = peekBuf[:n] @@ -2755,7 +3203,7 @@ func TestReadAfterClosedState(t *testing.T) { // Receive data. v, _, err := c.EP.Read(nil) if err != nil { - t.Fatalf("Read failed: %v", err) + t.Fatalf("Read failed: %s", err) } if !bytes.Equal(data, v) { @@ -2765,11 +3213,11 @@ func TestReadAfterClosedState(t *testing.T) { // Now that we drained the queue, check that functions fail with the // right error code. if _, _, err := c.EP.Read(nil); err != tcpip.ErrClosedForReceive { - t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrClosedForReceive) + t.Fatalf("got c.EP.Read(nil) = %v, want = %s", err, tcpip.ErrClosedForReceive) } if _, _, err := c.EP.Peek([][]byte{peekBuf}); err != tcpip.ErrClosedForReceive { - t.Fatalf("got c.EP.Peek(...) = %v, want = %v", err, tcpip.ErrClosedForReceive) + t.Fatalf("got c.EP.Peek(...) = %v, want = %s", err, tcpip.ErrClosedForReceive) } } @@ -2970,6 +3418,62 @@ func TestMinMaxBufferSizes(t *testing.T) { checkSendBufferSize(t, ep, tcp.DefaultSendBufferSize*30) } +func TestBindToDeviceOption(t *testing.T) { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}}) + + ep, err := s.NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) + if err != nil { + t.Fatalf("NewEndpoint failed; %v", err) + } + defer ep.Close() + + if err := s.CreateNamedNIC(321, "my_device", loopback.New()); err != nil { + t.Errorf("CreateNamedNIC failed: %v", err) + } + + // Make an nameless NIC. + if err := s.CreateNIC(54321, loopback.New()); err != nil { + t.Errorf("CreateNIC failed: %v", err) + } + + // strPtr is used instead of taking the address of string literals, which is + // a compiler error. + strPtr := func(s string) *string { + return &s + } + + testActions := []struct { + name string + setBindToDevice *string + setBindToDeviceError *tcpip.Error + getBindToDevice tcpip.BindToDeviceOption + }{ + {"GetDefaultValue", nil, nil, ""}, + {"BindToNonExistent", strPtr("non_existent_device"), tcpip.ErrUnknownDevice, ""}, + {"BindToExistent", strPtr("my_device"), nil, "my_device"}, + {"UnbindToDevice", strPtr(""), nil, ""}, + } + for _, testAction := range testActions { + t.Run(testAction.name, func(t *testing.T) { + if testAction.setBindToDevice != nil { + bindToDevice := tcpip.BindToDeviceOption(*testAction.setBindToDevice) + if got, want := ep.SetSockOpt(bindToDevice), testAction.setBindToDeviceError; got != want { + t.Errorf("SetSockOpt(%v) got %v, want %v", bindToDevice, got, want) + } + } + bindToDevice := tcpip.BindToDeviceOption("to be modified by GetSockOpt") + if ep.GetSockOpt(&bindToDevice) != nil { + t.Errorf("GetSockOpt got %v, want %v", ep.GetSockOpt(&bindToDevice), nil) + } + if got, want := bindToDevice, testAction.getBindToDevice; got != want { + t.Errorf("bindToDevice got %q, want %q", got, want) + } + }) + } +} + func makeStack() (*stack.Stack, *tcpip.Error) { s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocol{ @@ -3880,7 +4384,8 @@ func TestListenBacklogFullSynCookieInUse(t *testing.T) { // Send a SYN request. irs := seqnum.Value(789) c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, + // pick a different src port for new SYN. + SrcPort: context.TestPort + 1, DstPort: context.StackPort, Flags: header.TCPFlagSyn, SeqNum: irs, @@ -4014,6 +4519,9 @@ func TestPassiveFailedConnectionAttemptIncrement(t *testing.T) { if got := stats.TCP.ListenOverflowSynDrop.Value(); got != want { t.Errorf("got stats.TCP.ListenOverflowSynDrop.Value() = %v, want = %v", got, want) } + if got := c.EP.Stats().(*tcp.Stats).ReceiveErrors.ListenOverflowSynDrop.Value(); got != want { + t.Errorf("got EP stats Stats.ReceiveErrors.ListenOverflowSynDrop = %v, want = %v", got, want) + } we, ch := waiter.NewChannelEntry(nil) c.WQ.EventRegister(&we, waiter.EventIn) @@ -4052,6 +4560,14 @@ func TestEndpointBindListenAcceptState(t *testing.T) { t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) } + // Expect InvalidEndpointState errors on a read at this point. + if _, _, err := ep.Read(nil); err != tcpip.ErrInvalidEndpointState { + t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrInvalidEndpointState) + } + if got := ep.Stats().(*tcp.Stats).ReadErrors.InvalidEndpointState.Value(); got != 1 { + t.Fatalf("got EP stats Stats.ReadErrors.InvalidEndpointState got %v want %v", got, 1) + } + if err := ep.Listen(10); err != nil { t.Fatalf("Listen failed: %v", err) } @@ -4083,6 +4599,9 @@ func TestEndpointBindListenAcceptState(t *testing.T) { if got, want := tcp.EndpointState(aep.State()), tcp.StateEstablished; got != want { t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) } + if err := aep.Connect(tcpip.FullAddress{Addr: context.TestAddr, Port: context.TestPort}); err != tcpip.ErrAlreadyConnected { + t.Errorf("Unexpected error attempting to call connect on an established endpoint, got: %v, want: %v", err, tcpip.ErrAlreadyConnected) + } // Listening endpoint remains in listen state. if got, want := tcp.EndpointState(ep.State()), tcp.StateListen; got != want { t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) @@ -4370,3 +4889,590 @@ func TestReceiveBufferAutoTuning(t *testing.T) { payloadSize *= 2 } } + +func TestDelayEnabled(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + checkDelayOption(t, c, false, 0) // Delay is disabled by default. + + for _, v := range []struct { + delayEnabled tcp.DelayEnabled + wantDelayOption int + }{ + {delayEnabled: false, wantDelayOption: 0}, + {delayEnabled: true, wantDelayOption: 1}, + } { + c := context.New(t, defaultMTU) + defer c.Cleanup() + if err := c.Stack().SetTransportProtocolOption(tcp.ProtocolNumber, v.delayEnabled); err != nil { + t.Fatalf("SetTransportProtocolOption(tcp, %t) failed: %v", v.delayEnabled, err) + } + checkDelayOption(t, c, v.delayEnabled, v.wantDelayOption) + } +} + +func checkDelayOption(t *testing.T, c *context.Context, wantDelayEnabled tcp.DelayEnabled, wantDelayOption int) { + t.Helper() + + var gotDelayEnabled tcp.DelayEnabled + if err := c.Stack().TransportProtocolOption(tcp.ProtocolNumber, &gotDelayEnabled); err != nil { + t.Fatalf("TransportProtocolOption(tcp, &gotDelayEnabled) failed: %v", err) + } + if gotDelayEnabled != wantDelayEnabled { + t.Errorf("TransportProtocolOption(tcp, &gotDelayEnabled) got %t, want %t", gotDelayEnabled, wantDelayEnabled) + } + + ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, new(waiter.Queue)) + if err != nil { + t.Fatalf("NewEndPoint(tcp, ipv4, new(waiter.Queue)) failed: %v", err) + } + gotDelayOption, err := ep.GetSockOptInt(tcpip.DelayOption) + if err != nil { + t.Fatalf("ep.GetSockOptInt(tcpip.DelayOption) failed: %v", err) + } + if gotDelayOption != wantDelayOption { + t.Errorf("ep.GetSockOptInt(tcpip.DelayOption) got: %d, want: %d", gotDelayOption, wantDelayOption) + } +} + +func TestTCPLingerTimeout(t *testing.T) { + c := context.New(t, 1500 /* mtu */) + defer c.Cleanup() + + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) + + testCases := []struct { + name string + tcpLingerTimeout time.Duration + want time.Duration + }{ + {"NegativeLingerTimeout", -123123, 0}, + {"ZeroLingerTimeout", 0, 0}, + {"InRangeLingerTimeout", 10 * time.Second, 10 * time.Second}, + // Values > stack's TCPLingerTimeout are capped to the stack's + // value. Defaults to tcp.DefaultTCPLingerTimeout(60 seconds) + {"AboveMaxLingerTimeout", 65 * time.Second, 60 * time.Second}, + } + for _, tc := range testCases { + t.Run(tc.name, func(t *testing.T) { + if err := c.EP.SetSockOpt(tcpip.TCPLingerTimeoutOption(tc.tcpLingerTimeout)); err != nil { + t.Fatalf("SetSockOpt(%s) = %s", tc.tcpLingerTimeout, err) + } + var v tcpip.TCPLingerTimeoutOption + if err := c.EP.GetSockOpt(&v); err != nil { + t.Fatalf("GetSockOpt(tcpip.TCPLingerTimeoutOption) = %s", err) + } + if got, want := time.Duration(v), tc.want; got != want { + t.Fatalf("unexpected linger timeout got: %s, want: %s", got, want) + } + }) + } +} + +func TestTCPTimeWaitRSTIgnored(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + wq := &waiter.Queue{} + ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, wq) + if err != nil { + t.Fatalf("NewEndpoint failed: %s", err) + } + if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatalf("Bind failed: %s", err) + } + + if err := ep.Listen(10); err != nil { + t.Fatalf("Listen failed: %s", err) + } + + // Send a SYN request. + iss := seqnum.Value(789) + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: iss, + }) + + // Receive the SYN-ACK reply. + b := c.GetPacket() + tcpHdr := header.TCP(header.IPv4(b).Payload()) + c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) + + ackHeaders := &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: iss + 1, + AckNum: c.IRS + 1, + } + + // Send ACK. + c.SendPacket(nil, ackHeaders) + + // Try to accept the connection. + we, ch := waiter.NewChannelEntry(nil) + wq.EventRegister(&we, waiter.EventIn) + defer wq.EventUnregister(&we) + + c.EP, _, err = ep.Accept() + if err == tcpip.ErrWouldBlock { + // Wait for connection to be established. + select { + case <-ch: + c.EP, _, err = ep.Accept() + if err != nil { + t.Fatalf("Accept failed: %s", err) + } + + case <-time.After(1 * time.Second): + t.Fatalf("Timed out waiting for accept") + } + } + + c.EP.Close() + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+1)), + checker.AckNum(uint32(iss)+1), + checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) + + finHeaders := &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck | header.TCPFlagFin, + SeqNum: iss + 1, + AckNum: c.IRS + 2, + } + + c.SendPacket(nil, finHeaders) + + // Get the ACK to the FIN we just sent. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+2)), + checker.AckNum(uint32(iss)+2), + checker.TCPFlags(header.TCPFlagAck))) + + // Now send a RST and this should be ignored and not + // generate an ACK. + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagRst, + SeqNum: iss + 1, + AckNum: c.IRS + 2, + }) + + c.CheckNoPacketTimeout("unexpected packet received in TIME_WAIT state", 1*time.Second) + + // Out of order ACK should generate an immediate ACK in + // TIME_WAIT. + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: iss + 1, + AckNum: c.IRS + 3, + }) + + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+2)), + checker.AckNum(uint32(iss)+2), + checker.TCPFlags(header.TCPFlagAck))) +} + +func TestTCPTimeWaitOutOfOrder(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + wq := &waiter.Queue{} + ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, wq) + if err != nil { + t.Fatalf("NewEndpoint failed: %s", err) + } + if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatalf("Bind failed: %s", err) + } + + if err := ep.Listen(10); err != nil { + t.Fatalf("Listen failed: %s", err) + } + + // Send a SYN request. + iss := seqnum.Value(789) + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: iss, + }) + + // Receive the SYN-ACK reply. + b := c.GetPacket() + tcpHdr := header.TCP(header.IPv4(b).Payload()) + c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) + + ackHeaders := &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: iss + 1, + AckNum: c.IRS + 1, + } + + // Send ACK. + c.SendPacket(nil, ackHeaders) + + // Try to accept the connection. + we, ch := waiter.NewChannelEntry(nil) + wq.EventRegister(&we, waiter.EventIn) + defer wq.EventUnregister(&we) + + c.EP, _, err = ep.Accept() + if err == tcpip.ErrWouldBlock { + // Wait for connection to be established. + select { + case <-ch: + c.EP, _, err = ep.Accept() + if err != nil { + t.Fatalf("Accept failed: %s", err) + } + + case <-time.After(1 * time.Second): + t.Fatalf("Timed out waiting for accept") + } + } + + c.EP.Close() + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+1)), + checker.AckNum(uint32(iss)+1), + checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) + + finHeaders := &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck | header.TCPFlagFin, + SeqNum: iss + 1, + AckNum: c.IRS + 2, + } + + c.SendPacket(nil, finHeaders) + + // Get the ACK to the FIN we just sent. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+2)), + checker.AckNum(uint32(iss)+2), + checker.TCPFlags(header.TCPFlagAck))) + + // Out of order ACK should generate an immediate ACK in + // TIME_WAIT. + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: iss + 1, + AckNum: c.IRS + 3, + }) + + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+2)), + checker.AckNum(uint32(iss)+2), + checker.TCPFlags(header.TCPFlagAck))) +} + +func TestTCPTimeWaitNewSyn(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + wq := &waiter.Queue{} + ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, wq) + if err != nil { + t.Fatalf("NewEndpoint failed: %s", err) + } + if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatalf("Bind failed: %s", err) + } + + if err := ep.Listen(10); err != nil { + t.Fatalf("Listen failed: %s", err) + } + + // Send a SYN request. + iss := seqnum.Value(789) + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: iss, + }) + + // Receive the SYN-ACK reply. + b := c.GetPacket() + tcpHdr := header.TCP(header.IPv4(b).Payload()) + c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) + + ackHeaders := &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: iss + 1, + AckNum: c.IRS + 1, + } + + // Send ACK. + c.SendPacket(nil, ackHeaders) + + // Try to accept the connection. + we, ch := waiter.NewChannelEntry(nil) + wq.EventRegister(&we, waiter.EventIn) + defer wq.EventUnregister(&we) + + c.EP, _, err = ep.Accept() + if err == tcpip.ErrWouldBlock { + // Wait for connection to be established. + select { + case <-ch: + c.EP, _, err = ep.Accept() + if err != nil { + t.Fatalf("Accept failed: %s", err) + } + + case <-time.After(1 * time.Second): + t.Fatalf("Timed out waiting for accept") + } + } + + c.EP.Close() + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+1)), + checker.AckNum(uint32(iss)+1), + checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) + + finHeaders := &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck | header.TCPFlagFin, + SeqNum: iss + 1, + AckNum: c.IRS + 2, + } + + c.SendPacket(nil, finHeaders) + + // Get the ACK to the FIN we just sent. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+2)), + checker.AckNum(uint32(iss)+2), + checker.TCPFlags(header.TCPFlagAck))) + + // Send a SYN request w/ sequence number lower than + // the highest sequence number sent. We just reuse + // the same number. + iss = seqnum.Value(789) + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: iss, + }) + + c.CheckNoPacketTimeout("unexpected packet received in response to SYN", 1*time.Second) + + // Send a SYN request w/ sequence number higher than + // the highest sequence number sent. + iss = seqnum.Value(792) + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: iss, + }) + + // Receive the SYN-ACK reply. + b = c.GetPacket() + tcpHdr = header.TCP(header.IPv4(b).Payload()) + c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) + + ackHeaders = &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: iss + 1, + AckNum: c.IRS + 1, + } + + // Send ACK. + c.SendPacket(nil, ackHeaders) + + // Try to accept the connection. + c.EP, _, err = ep.Accept() + if err == tcpip.ErrWouldBlock { + // Wait for connection to be established. + select { + case <-ch: + c.EP, _, err = ep.Accept() + if err != nil { + t.Fatalf("Accept failed: %s", err) + } + + case <-time.After(1 * time.Second): + t.Fatalf("Timed out waiting for accept") + } + } +} + +func TestTCPTimeWaitDuplicateFINExtendsTimeWait(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + // Set TCPTimeWaitTimeout to 5 seconds so that sockets are marked closed + // after 5 seconds in TIME_WAIT state. + tcpTimeWaitTimeout := 5 * time.Second + if err := c.Stack().SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.TCPTimeWaitTimeoutOption(tcpTimeWaitTimeout)); err != nil { + t.Fatalf("c.stack.SetTransportProtocolOption(tcp, tcpip.TCPLingerTimeoutOption(%d) failed: %s", tcpTimeWaitTimeout, err) + } + + wq := &waiter.Queue{} + ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, wq) + if err != nil { + t.Fatalf("NewEndpoint failed: %s", err) + } + if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatalf("Bind failed: %s", err) + } + + if err := ep.Listen(10); err != nil { + t.Fatalf("Listen failed: %s", err) + } + + // Send a SYN request. + iss := seqnum.Value(789) + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: iss, + }) + + // Receive the SYN-ACK reply. + b := c.GetPacket() + tcpHdr := header.TCP(header.IPv4(b).Payload()) + c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) + + ackHeaders := &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: iss + 1, + AckNum: c.IRS + 1, + } + + // Send ACK. + c.SendPacket(nil, ackHeaders) + + // Try to accept the connection. + we, ch := waiter.NewChannelEntry(nil) + wq.EventRegister(&we, waiter.EventIn) + defer wq.EventUnregister(&we) + + c.EP, _, err = ep.Accept() + if err == tcpip.ErrWouldBlock { + // Wait for connection to be established. + select { + case <-ch: + c.EP, _, err = ep.Accept() + if err != nil { + t.Fatalf("Accept failed: %s", err) + } + + case <-time.After(1 * time.Second): + t.Fatalf("Timed out waiting for accept") + } + } + + c.EP.Close() + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+1)), + checker.AckNum(uint32(iss)+1), + checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) + + finHeaders := &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck | header.TCPFlagFin, + SeqNum: iss + 1, + AckNum: c.IRS + 2, + } + + c.SendPacket(nil, finHeaders) + + // Get the ACK to the FIN we just sent. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+2)), + checker.AckNum(uint32(iss)+2), + checker.TCPFlags(header.TCPFlagAck))) + + time.Sleep(2 * time.Second) + + // Now send a duplicate FIN. This should cause the TIME_WAIT to extend + // by another 5 seconds and also send us a duplicate ACK as it should + // indicate that the final ACK was potentially lost. + c.SendPacket(nil, finHeaders) + + // Get the ACK to the FIN we just sent. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+2)), + checker.AckNum(uint32(iss)+2), + checker.TCPFlags(header.TCPFlagAck))) + + // Sleep for 4 seconds so at this point we are 1 second past the + // original tcpLingerTimeout of 5 seconds. + time.Sleep(4 * time.Second) + + // Send an ACK and it should not generate any packet as the socket + // should still be in TIME_WAIT for another another 5 seconds due + // to the duplicate FIN we sent earlier. + *ackHeaders = *finHeaders + ackHeaders.SeqNum = ackHeaders.SeqNum + 1 + ackHeaders.Flags = header.TCPFlagAck + c.SendPacket(nil, ackHeaders) + + c.CheckNoPacketTimeout("unexpected packet received from endpoint in TIME_WAIT", 1*time.Second) + // Now sleep for another 2 seconds so that we are past the + // extended TIME_WAIT of 7 seconds (2 + 5). + time.Sleep(2 * time.Second) + + // Resend the same ACK. + c.SendPacket(nil, ackHeaders) + + // Receive the RST that should be generated as there is no valid + // endpoint. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(ackHeaders.AckNum)), + checker.AckNum(uint32(ackHeaders.SeqNum)), + checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck))) +} diff --git a/pkg/tcpip/transport/tcp/testing/context/context.go b/pkg/tcpip/transport/tcp/testing/context/context.go index d3f1d2cdf..4854e719d 100644 --- a/pkg/tcpip/transport/tcp/testing/context/context.go +++ b/pkg/tcpip/transport/tcp/testing/context/context.go @@ -158,7 +158,14 @@ func New(t *testing.T, mtu uint32) *Context { if testing.Verbose() { wep = sniffer.New(ep) } - if err := s.CreateNIC(1, wep); err != nil { + if err := s.CreateNamedNIC(1, "nic1", wep); err != nil { + t.Fatalf("CreateNIC failed: %v", err) + } + wep2 := stack.LinkEndpoint(channel.New(1000, mtu, "")) + if testing.Verbose() { + wep2 = sniffer.New(channel.New(1000, mtu, "")) + } + if err := s.CreateNamedNIC(2, "nic2", wep2); err != nil { t.Fatalf("CreateNIC failed: %v", err) } @@ -295,7 +302,9 @@ func (c *Context) SendICMPPacket(typ header.ICMPv4Type, code uint8, p1, p2 []byt copy(icmp[header.ICMPv4PayloadOffset:], p2) // Inject packet. - c.linkEP.Inject(ipv4.ProtocolNumber, buf.ToVectorisedView()) + c.linkEP.InjectInbound(ipv4.ProtocolNumber, tcpip.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) } // BuildSegment builds a TCP segment based on the given Headers and payload. @@ -343,13 +352,17 @@ func (c *Context) BuildSegment(payload []byte, h *Headers) buffer.VectorisedView // SendSegment sends a TCP segment that has already been built and written to a // buffer.VectorisedView. func (c *Context) SendSegment(s buffer.VectorisedView) { - c.linkEP.Inject(ipv4.ProtocolNumber, s) + c.linkEP.InjectInbound(ipv4.ProtocolNumber, tcpip.PacketBuffer{ + Data: s, + }) } // SendPacket builds and sends a TCP segment(with the provided payload & TCP // headers) in an IPv4 packet via the link layer endpoint. func (c *Context) SendPacket(payload []byte, h *Headers) { - c.linkEP.Inject(ipv4.ProtocolNumber, c.BuildSegment(payload, h)) + c.linkEP.InjectInbound(ipv4.ProtocolNumber, tcpip.PacketBuffer{ + Data: c.BuildSegment(payload, h), + }) } // SendAck sends an ACK packet. @@ -511,7 +524,9 @@ func (c *Context) SendV6Packet(payload []byte, h *Headers) { t.SetChecksum(^t.CalculateChecksum(xsum)) // Inject packet. - c.linkEP.Inject(ipv6.ProtocolNumber, buf.ToVectorisedView()) + c.linkEP.InjectInbound(ipv6.ProtocolNumber, tcpip.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) } // CreateConnected creates a connected TCP endpoint. @@ -588,12 +603,8 @@ func (c *Context) Connect(iss seqnum.Value, rcvWnd seqnum.Size, options []byte) c.Port = tcpHdr.SourcePort() } -// CreateConnectedWithRawOptions creates a connected TCP endpoint and sends -// the specified option bytes as the Option field in the initial SYN packet. -// -// It also sets the receive buffer for the endpoint to the specified -// value in epRcvBuf. -func (c *Context) CreateConnectedWithRawOptions(iss seqnum.Value, rcvWnd seqnum.Size, epRcvBuf int, options []byte) { +// Create creates a TCP endpoint. +func (c *Context) Create(epRcvBuf int) { // Create TCP endpoint. var err *tcpip.Error c.EP, err = c.s.NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &c.WQ) @@ -606,6 +617,15 @@ func (c *Context) CreateConnectedWithRawOptions(iss seqnum.Value, rcvWnd seqnum. c.t.Fatalf("SetSockOpt failed failed: %v", err) } } +} + +// CreateConnectedWithRawOptions creates a connected TCP endpoint and sends +// the specified option bytes as the Option field in the initial SYN packet. +// +// It also sets the receive buffer for the endpoint to the specified +// value in epRcvBuf. +func (c *Context) CreateConnectedWithRawOptions(iss seqnum.Value, rcvWnd seqnum.Size, epRcvBuf int, options []byte) { + c.Create(epRcvBuf) c.Connect(iss, rcvWnd, options) } diff --git a/pkg/tcpip/transport/udp/BUILD b/pkg/tcpip/transport/udp/BUILD index c1ca22b35..c9460aa0d 100644 --- a/pkg/tcpip/transport/udp/BUILD +++ b/pkg/tcpip/transport/udp/BUILD @@ -1,10 +1,9 @@ load("@io_bazel_rules_go//go:def.bzl", "go_test") - -package(licenses = ["notice"]) - load("//tools/go_generics:defs.bzl", "go_template_instance") load("//tools/go_stateify:defs.bzl", "go_library") +package(licenses = ["notice"]) + go_template_instance( name = "udp_packet_list", out = "udp_packet_list.go", @@ -52,6 +51,7 @@ go_test( "//pkg/tcpip/checker", "//pkg/tcpip/header", "//pkg/tcpip/link/channel", + "//pkg/tcpip/link/loopback", "//pkg/tcpip/link/sniffer", "//pkg/tcpip/network/ipv4", "//pkg/tcpip/network/ipv6", diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 0bec7e62d..5270f24df 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -31,9 +31,6 @@ type udpPacket struct { senderAddress tcpip.FullAddress data buffer.VectorisedView `state:".(buffer.VectorisedView)"` timestamp int64 - // views is used as buffer for data when its length is large - // enough to store a VectorisedView. - views [8]buffer.View `state:"nosave"` } // EndpointState represents the state of a UDP endpoint. @@ -49,6 +46,22 @@ const ( StateClosed ) +// String implements fmt.Stringer.String. +func (s EndpointState) String() string { + switch s { + case StateInitial: + return "INITIAL" + case StateBound: + return "BOUND" + case StateConnected: + return "CONNECTING" + case StateClosed: + return "CLOSED" + default: + return "UNKNOWN" + } +} + // endpoint represents a UDP endpoint. This struct serves as the interface // between users of the endpoint and the protocol implementation; it is legal to // have concurrent goroutines make calls into the endpoint, they are properly @@ -58,11 +71,13 @@ const ( // // +stateify savable type endpoint struct { + stack.TransportEndpointInfo + // The following fields are initialized at creation time and do not // change throughout the lifetime of the endpoint. stack *stack.Stack `state:"manual"` - netProto tcpip.NetworkProtocolNumber waiterQueue *waiter.Queue + uniqueID uint64 // The following fields are used to manage the receive queue, and are // protected by rcvMu. @@ -76,20 +91,23 @@ type endpoint struct { // The following fields are protected by the mu mutex. mu sync.RWMutex `state:"nosave"` sndBufSize int - id stack.TransportEndpointID state EndpointState - bindNICID tcpip.NICID - regNICID tcpip.NICID route stack.Route `state:"manual"` dstPort uint16 v6only bool + ttl uint8 multicastTTL uint8 multicastAddr tcpip.Address multicastNICID tcpip.NICID multicastLoop bool reusePort bool + bindToDevice tcpip.NICID broadcast bool + // sendTOS represents IPv4 TOS or IPv6 TrafficClass, + // applied while sending packets. Defaults to 0 as on Linux. + sendTOS uint8 + // shutdownFlags represent the current shutdown state of the endpoint. shutdownFlags tcpip.ShutdownFlags @@ -104,6 +122,9 @@ type endpoint struct { // IPv4 when IPv6 endpoint is bound or connected to an IPv4 mapped // address). effectiveNetProtos []tcpip.NetworkProtocolNumber + + // TODO(b/142022063): Add ability to save and restore per endpoint stats. + stats tcpip.TransportEndpointStats `state:"nosave"` } // +stateify savable @@ -112,10 +133,13 @@ type multicastMembership struct { multicastAddr tcpip.Address } -func newEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) *endpoint { +func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) *endpoint { return &endpoint{ - stack: stack, - netProto: netProto, + stack: s, + TransportEndpointInfo: stack.TransportEndpointInfo{ + NetProto: netProto, + TransProto: header.UDPProtocolNumber, + }, waiterQueue: waiterQueue, // RFC 1075 section 5.4 recommends a TTL of 1 for membership // requests. @@ -133,9 +157,16 @@ func newEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waite multicastLoop: true, rcvBufSizeMax: 32 * 1024, sndBufSize: 32 * 1024, + state: StateInitial, + uniqueID: s.UniqueID(), } } +// UniqueID implements stack.TransportEndpoint.UniqueID. +func (e *endpoint) UniqueID() uint64 { + return e.uniqueID +} + // Close puts the endpoint in a closed state and frees all resources // associated with it. func (e *endpoint) Close() { @@ -144,12 +175,12 @@ func (e *endpoint) Close() { switch e.state { case StateBound, StateConnected: - e.stack.UnregisterTransportEndpoint(e.regNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e) - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort) + e.stack.UnregisterTransportEndpoint(e.RegisterNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.bindToDevice) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.bindToDevice) } for _, mem := range e.multicastMemberships { - e.stack.LeaveGroup(e.netProto, mem.nicID, mem.multicastAddr) + e.stack.LeaveGroup(e.NetProto, mem.nicID, mem.multicastAddr) } e.multicastMemberships = nil @@ -189,6 +220,7 @@ func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMess if e.rcvList.Empty() { err := tcpip.ErrWouldBlock if e.rcvClosed { + e.stats.ReadErrors.ReadClosed.Increment() err = tcpip.ErrClosedForReceive } e.rcvMu.Unlock() @@ -250,33 +282,56 @@ func (e *endpoint) prepareForWrite(to *tcpip.FullAddress) (retry bool, err *tcpi // connectRoute establishes a route to the specified interface or the // configured multicast interface if no interface is specified and the // specified address is a multicast address. -func (e *endpoint) connectRoute(nicid tcpip.NICID, addr tcpip.FullAddress, netProto tcpip.NetworkProtocolNumber) (stack.Route, tcpip.NICID, *tcpip.Error) { - localAddr := e.id.LocalAddress +func (e *endpoint) connectRoute(nicID tcpip.NICID, addr tcpip.FullAddress, netProto tcpip.NetworkProtocolNumber) (stack.Route, tcpip.NICID, *tcpip.Error) { + localAddr := e.ID.LocalAddress if isBroadcastOrMulticast(localAddr) { // A packet can only originate from a unicast address (i.e., an interface). localAddr = "" } if header.IsV4MulticastAddress(addr.Addr) || header.IsV6MulticastAddress(addr.Addr) { - if nicid == 0 { - nicid = e.multicastNICID + if nicID == 0 { + nicID = e.multicastNICID } - if localAddr == "" && nicid == 0 { + if localAddr == "" && nicID == 0 { localAddr = e.multicastAddr } } // Find a route to the desired destination. - r, err := e.stack.FindRoute(nicid, localAddr, addr.Addr, netProto, e.multicastLoop) + r, err := e.stack.FindRoute(nicID, localAddr, addr.Addr, netProto, e.multicastLoop) if err != nil { return stack.Route{}, 0, err } - return r, nicid, nil + return r, nicID, nil } // Write writes data to the endpoint's peer. This method does not block // if the data cannot be written. func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { + n, ch, err := e.write(p, opts) + switch err { + case nil: + e.stats.PacketsSent.Increment() + case tcpip.ErrMessageTooLong, tcpip.ErrInvalidOptionValue: + e.stats.WriteErrors.InvalidArgs.Increment() + case tcpip.ErrClosedForSend: + e.stats.WriteErrors.WriteClosed.Increment() + case tcpip.ErrInvalidEndpointState: + e.stats.WriteErrors.InvalidEndpointState.Increment() + case tcpip.ErrNoLinkAddress: + e.stats.SendErrors.NoLinkAddr.Increment() + case tcpip.ErrNoRoute, tcpip.ErrBroadcastDisabled, tcpip.ErrNetworkUnreachable: + // Errors indicating any problem with IP routing of the packet. + e.stats.SendErrors.NoRoute.Increment() + default: + // For all other errors when writing to the network layer. + e.stats.SendErrors.SendToNetworkFailed.Increment() + } + return n, ch, err +} + +func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { // MSG_MORE is unimplemented. (This also means that MSG_EOR is a no-op.) if opts.More { return 0, nil, tcpip.ErrInvalidOptionValue @@ -327,13 +382,13 @@ func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c } else { // Reject destination address if it goes through a different // NIC than the endpoint was bound to. - nicid := to.NIC - if e.bindNICID != 0 { - if nicid != 0 && nicid != e.bindNICID { + nicID := to.NIC + if e.BindNICID != 0 { + if nicID != 0 && nicID != e.BindNICID { return 0, nil, tcpip.ErrNoRoute } - nicid = e.bindNICID + nicID = e.BindNICID } if to.Addr == header.IPv4Broadcast && !e.broadcast { @@ -345,7 +400,7 @@ func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c return 0, nil, err } - r, _, err := e.connectRoute(nicid, *to, netProto) + r, _, err := e.connectRoute(nicID, *to, netProto) if err != nil { return 0, nil, err } @@ -373,12 +428,16 @@ func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c return 0, nil, tcpip.ErrMessageTooLong } - ttl := route.DefaultTTL() + ttl := e.ttl + useDefaultTTL := ttl == 0 + if header.IsV4MulticastAddress(route.RemoteAddress) || header.IsV6MulticastAddress(route.RemoteAddress) { ttl = e.multicastTTL + // Multicast allows a 0 TTL. + useDefaultTTL = false } - if err := sendUDP(route, buffer.View(v).ToVectorisedView(), e.id.LocalPort, dstPort, ttl); err != nil { + if err := sendUDP(route, buffer.View(v).ToVectorisedView(), e.ID.LocalPort, dstPort, ttl, useDefaultTTL, e.sendTOS); err != nil { return 0, nil, err } return int64(len(v)), nil, nil @@ -399,7 +458,7 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { switch v := opt.(type) { case tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. - if e.netProto != header.IPv6ProtocolNumber { + if e.NetProto != header.IPv6ProtocolNumber { return tcpip.ErrInvalidEndpointState } @@ -413,6 +472,11 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.v6only = v != 0 + case tcpip.TTLOption: + e.mu.Lock() + e.ttl = uint8(v) + e.mu.Unlock() + case tcpip.MulticastTTLOption: e.mu.Lock() e.multicastTTL = uint8(v) @@ -447,7 +511,7 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { } } - if e.bindNICID != 0 && e.bindNICID != nic { + if e.BindNICID != 0 && e.BindNICID != nic { return tcpip.ErrInvalidEndpointState } @@ -474,7 +538,7 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { } } } else { - nicID = e.stack.CheckLocalAddress(nicID, e.netProto, v.InterfaceAddr) + nicID = e.stack.CheckLocalAddress(nicID, e.NetProto, v.InterfaceAddr) } if nicID == 0 { return tcpip.ErrUnknownDevice @@ -491,7 +555,7 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { } } - if err := e.stack.JoinGroup(e.netProto, nicID, v.MulticastAddr); err != nil { + if err := e.stack.JoinGroup(e.NetProto, nicID, v.MulticastAddr); err != nil { return err } @@ -512,7 +576,7 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { } } } else { - nicID = e.stack.CheckLocalAddress(nicID, e.netProto, v.InterfaceAddr) + nicID = e.stack.CheckLocalAddress(nicID, e.NetProto, v.InterfaceAddr) } if nicID == 0 { return tcpip.ErrUnknownDevice @@ -534,7 +598,7 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { return tcpip.ErrBadLocalAddress } - if err := e.stack.LeaveGroup(e.netProto, nicID, v.MulticastAddr); err != nil { + if err := e.stack.LeaveGroup(e.NetProto, nicID, v.MulticastAddr); err != nil { return err } @@ -551,12 +615,39 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.reusePort = v != 0 e.mu.Unlock() + case tcpip.BindToDeviceOption: + e.mu.Lock() + defer e.mu.Unlock() + if v == "" { + e.bindToDevice = 0 + return nil + } + for nicID, nic := range e.stack.NICInfo() { + if nic.Name == string(v) { + e.bindToDevice = nicID + return nil + } + } + return tcpip.ErrUnknownDevice + case tcpip.BroadcastOption: e.mu.Lock() e.broadcast = v != 0 e.mu.Unlock() return nil + + case tcpip.IPv4TOSOption: + e.mu.Lock() + e.sendTOS = uint8(v) + e.mu.Unlock() + return nil + + case tcpip.IPv6TrafficClassOption: + e.mu.Lock() + e.sendTOS = uint8(v) + e.mu.Unlock() + return nil } return nil } @@ -598,7 +689,7 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { case *tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. - if e.netProto != header.IPv6ProtocolNumber { + if e.NetProto != header.IPv6ProtocolNumber { return tcpip.ErrUnknownProtocolOption } @@ -612,6 +703,12 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { } return nil + case *tcpip.TTLOption: + e.mu.Lock() + *o = tcpip.TTLOption(e.ttl) + e.mu.Unlock() + return nil + case *tcpip.MulticastTTLOption: e.mu.Lock() *o = tcpip.MulticastTTLOption(e.multicastTTL) @@ -646,6 +743,16 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { } return nil + case *tcpip.BindToDeviceOption: + e.mu.RLock() + defer e.mu.RUnlock() + if nic, ok := e.stack.NICInfo()[e.bindToDevice]; ok { + *o = tcpip.BindToDeviceOption(nic.Name) + return nil + } + *o = tcpip.BindToDeviceOption("") + return nil + case *tcpip.KeepaliveEnabledOption: *o = 0 return nil @@ -661,6 +768,18 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { } return nil + case *tcpip.IPv4TOSOption: + e.mu.RLock() + *o = tcpip.IPv4TOSOption(e.sendTOS) + e.mu.RUnlock() + return nil + + case *tcpip.IPv6TrafficClassOption: + e.mu.RLock() + *o = tcpip.IPv6TrafficClassOption(e.sendTOS) + e.mu.RUnlock() + return nil + default: return tcpip.ErrUnknownProtocolOption } @@ -668,7 +787,7 @@ 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) *tcpip.Error { +func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort uint16, ttl uint8, useDefaultTTL bool, tos uint8) *tcpip.Error { // Allocate a buffer for the UDP header. hdr := buffer.NewPrependable(header.UDPMinimumSize + int(r.MaxHeaderLength())) @@ -691,14 +810,21 @@ func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort u udp.SetChecksum(^udp.CalculateChecksum(xsum)) } + if useDefaultTTL { + ttl = r.DefaultTTL() + } + if err := r.WritePacket(nil /* gso */, hdr, data, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}); err != nil { + r.Stats().UDP.PacketSendErrors.Increment() + return err + } + // Track count of packets sent. r.Stats().UDP.PacketsSent.Increment() - - return r.WritePacket(nil /* gso */, hdr, data, ProtocolNumber, ttl) + return nil } func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress, allowMismatch bool) (tcpip.NetworkProtocolNumber, *tcpip.Error) { - netProto := e.netProto + netProto := e.NetProto if len(addr.Addr) == 0 { return netProto, nil } @@ -715,14 +841,14 @@ func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress, allowMismatch bool) (t } // Fail if we are bound to an IPv6 address. - if !allowMismatch && len(e.id.LocalAddress) == 16 { + if !allowMismatch && len(e.ID.LocalAddress) == 16 { return 0, tcpip.ErrNetworkUnreachable } } // Fail if we're bound to an address length different from the one we're // checking. - if l := len(e.id.LocalAddress); l != 0 && l != len(addr.Addr) { + if l := len(e.ID.LocalAddress); l != 0 && l != len(addr.Addr) { return 0, tcpip.ErrInvalidEndpointState } @@ -739,27 +865,27 @@ func (e *endpoint) Disconnect() *tcpip.Error { } id := stack.TransportEndpointID{} // Exclude ephemerally bound endpoints. - if e.bindNICID != 0 || e.id.LocalAddress == "" { + if e.BindNICID != 0 || e.ID.LocalAddress == "" { var err *tcpip.Error id = stack.TransportEndpointID{ - LocalPort: e.id.LocalPort, - LocalAddress: e.id.LocalAddress, + LocalPort: e.ID.LocalPort, + LocalAddress: e.ID.LocalAddress, } - id, err = e.registerWithStack(e.regNICID, e.effectiveNetProtos, id) + id, err = e.registerWithStack(e.RegisterNICID, e.effectiveNetProtos, id) if err != nil { return err } e.state = StateBound } else { - if e.id.LocalPort != 0 { + if e.ID.LocalPort != 0 { // Release the ephemeral port. - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.bindToDevice) } e.state = StateInitial } - e.stack.UnregisterTransportEndpoint(e.regNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e) - e.id = id + e.stack.UnregisterTransportEndpoint(e.RegisterNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.bindToDevice) + e.ID = id e.route.Release() e.route = stack.Route{} e.dstPort = 0 @@ -781,33 +907,33 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { e.mu.Lock() defer e.mu.Unlock() - nicid := addr.NIC + nicID := addr.NIC var localPort uint16 switch e.state { case StateInitial: case StateBound, StateConnected: - localPort = e.id.LocalPort - if e.bindNICID == 0 { + localPort = e.ID.LocalPort + if e.BindNICID == 0 { break } - if nicid != 0 && nicid != e.bindNICID { + if nicID != 0 && nicID != e.BindNICID { return tcpip.ErrInvalidEndpointState } - nicid = e.bindNICID + nicID = e.BindNICID default: return tcpip.ErrInvalidEndpointState } - r, nicid, err := e.connectRoute(nicid, addr, netProto) + r, nicID, err := e.connectRoute(nicID, addr, netProto) if err != nil { return err } defer r.Release() id := stack.TransportEndpointID{ - LocalAddress: e.id.LocalAddress, + LocalAddress: e.ID.LocalAddress, LocalPort: localPort, RemotePort: addr.Port, RemoteAddress: r.RemoteAddress, @@ -828,20 +954,20 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { } } - id, err = e.registerWithStack(nicid, netProtos, id) + id, err = e.registerWithStack(nicID, netProtos, id) if err != nil { return err } // Remove the old registration. - if e.id.LocalPort != 0 { - e.stack.UnregisterTransportEndpoint(e.regNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e) + if e.ID.LocalPort != 0 { + e.stack.UnregisterTransportEndpoint(e.RegisterNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.bindToDevice) } - e.id = id + e.ID = id e.route = r.Clone() e.dstPort = addr.Port - e.regNICID = nicid + e.RegisterNICID = nicID e.effectiveNetProtos = netProtos e.state = StateConnected @@ -896,18 +1022,18 @@ func (*endpoint) Accept() (tcpip.Endpoint, *waiter.Queue, *tcpip.Error) { return nil, nil, tcpip.ErrNotSupported } -func (e *endpoint) registerWithStack(nicid tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, id stack.TransportEndpointID) (stack.TransportEndpointID, *tcpip.Error) { - if e.id.LocalPort == 0 { - port, err := e.stack.ReservePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, e.reusePort) +func (e *endpoint) registerWithStack(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, id stack.TransportEndpointID) (stack.TransportEndpointID, *tcpip.Error) { + if e.ID.LocalPort == 0 { + port, err := e.stack.ReservePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, e.reusePort, e.bindToDevice) if err != nil { return id, err } id.LocalPort = port } - err := e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, id, e, e.reusePort) + err := e.stack.RegisterTransportEndpoint(nicID, netProtos, ProtocolNumber, id, e, e.reusePort, e.bindToDevice) if err != nil { - e.stack.ReleasePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort) + e.stack.ReleasePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, e.bindToDevice) } return id, err } @@ -935,11 +1061,11 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) *tcpip.Error { } } - nicid := addr.NIC + nicID := addr.NIC if len(addr.Addr) != 0 && !isBroadcastOrMulticast(addr.Addr) { // A local unicast address was specified, verify that it's valid. - nicid = e.stack.CheckLocalAddress(addr.NIC, netProto, addr.Addr) - if nicid == 0 { + nicID = e.stack.CheckLocalAddress(addr.NIC, netProto, addr.Addr) + if nicID == 0 { return tcpip.ErrBadLocalAddress } } @@ -948,13 +1074,13 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) *tcpip.Error { LocalPort: addr.Port, LocalAddress: addr.Addr, } - id, err = e.registerWithStack(nicid, netProtos, id) + id, err = e.registerWithStack(nicID, netProtos, id) if err != nil { return err } - e.id = id - e.regNICID = nicid + e.ID = id + e.RegisterNICID = nicID e.effectiveNetProtos = netProtos // Mark endpoint as bound. @@ -979,7 +1105,7 @@ func (e *endpoint) Bind(addr tcpip.FullAddress) *tcpip.Error { } // Save the effective NICID generated by bindLocked. - e.bindNICID = e.regNICID + e.BindNICID = e.RegisterNICID return nil } @@ -990,9 +1116,9 @@ func (e *endpoint) GetLocalAddress() (tcpip.FullAddress, *tcpip.Error) { defer e.mu.RUnlock() return tcpip.FullAddress{ - NIC: e.regNICID, - Addr: e.id.LocalAddress, - Port: e.id.LocalPort, + NIC: e.RegisterNICID, + Addr: e.ID.LocalAddress, + Port: e.ID.LocalPort, }, nil } @@ -1006,9 +1132,9 @@ func (e *endpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Error) { } return tcpip.FullAddress{ - NIC: e.regNICID, - Addr: e.id.RemoteAddress, - Port: e.id.RemotePort, + NIC: e.RegisterNICID, + Addr: e.ID.RemoteAddress, + Port: e.ID.RemotePort, }, nil } @@ -1032,42 +1158,52 @@ func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { // HandlePacket is called by the stack when new packets arrive to this transport // endpoint. -func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, vv buffer.VectorisedView) { +func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt tcpip.PacketBuffer) { // Get the header then trim it from the view. - hdr := header.UDP(vv.First()) - if int(hdr.Length()) > vv.Size() { + hdr := header.UDP(pkt.Data.First()) + if int(hdr.Length()) > pkt.Data.Size() { // Malformed packet. e.stack.Stats().UDP.MalformedPacketsReceived.Increment() + e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() return } - vv.TrimFront(header.UDPMinimumSize) + pkt.Data.TrimFront(header.UDPMinimumSize) e.rcvMu.Lock() e.stack.Stats().UDP.PacketsReceived.Increment() + e.stats.PacketsReceived.Increment() // Drop the packet if our buffer is currently full. - if !e.rcvReady || e.rcvClosed || e.rcvBufSize >= e.rcvBufSizeMax { + if !e.rcvReady || e.rcvClosed { + e.rcvMu.Unlock() e.stack.Stats().UDP.ReceiveBufferErrors.Increment() + e.stats.ReceiveErrors.ClosedReceiver.Increment() + return + } + + if e.rcvBufSize >= e.rcvBufSizeMax { e.rcvMu.Unlock() + e.stack.Stats().UDP.ReceiveBufferErrors.Increment() + e.stats.ReceiveErrors.ReceiveBufferOverflow.Increment() return } wasEmpty := e.rcvBufSize == 0 // Push new packet into receive list and increment the buffer size. - pkt := &udpPacket{ + packet := &udpPacket{ senderAddress: tcpip.FullAddress{ NIC: r.NICID(), Addr: id.RemoteAddress, Port: hdr.SourcePort(), }, } - pkt.data = vv.Clone(pkt.views[:]) - e.rcvList.PushBack(pkt) - e.rcvBufSize += vv.Size() + packet.data = pkt.Data + e.rcvList.PushBack(packet) + e.rcvBufSize += pkt.Data.Size() - pkt.timestamp = e.stack.NowNanoseconds() + packet.timestamp = e.stack.NowNanoseconds() e.rcvMu.Unlock() @@ -1078,7 +1214,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, vv } // HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket. -func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, vv buffer.VectorisedView) { +func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, pkt tcpip.PacketBuffer) { } // State implements tcpip.Endpoint.State. @@ -1088,6 +1224,23 @@ func (e *endpoint) State() uint32 { return uint32(e.state) } +// Info returns a copy of the endpoint info. +func (e *endpoint) Info() tcpip.EndpointInfo { + e.mu.RLock() + // Make a copy of the endpoint info. + ret := e.TransportEndpointInfo + e.mu.RUnlock() + return &ret +} + +// Stats returns a pointer to the endpoint stats. +func (e *endpoint) Stats() tcpip.EndpointStats { + return &e.stats +} + +// Wait implements tcpip.Endpoint.Wait. +func (*endpoint) Wait() {} + func isBroadcastOrMulticast(a tcpip.Address) bool { return a == header.IPv4Broadcast || header.IsV4MulticastAddress(a) || header.IsV6MulticastAddress(a) } diff --git a/pkg/tcpip/transport/udp/endpoint_state.go b/pkg/tcpip/transport/udp/endpoint_state.go index be46e6d4e..b227e353b 100644 --- a/pkg/tcpip/transport/udp/endpoint_state.go +++ b/pkg/tcpip/transport/udp/endpoint_state.go @@ -72,7 +72,7 @@ func (e *endpoint) Resume(s *stack.Stack) { e.stack = s for _, m := range e.multicastMemberships { - if err := e.stack.JoinGroup(e.netProto, m.nicID, m.multicastAddr); err != nil { + if err := e.stack.JoinGroup(e.NetProto, m.nicID, m.multicastAddr); err != nil { panic(err) } } @@ -93,13 +93,13 @@ func (e *endpoint) Resume(s *stack.Stack) { var err *tcpip.Error if e.state == StateConnected { - e.route, err = e.stack.FindRoute(e.regNICID, e.id.LocalAddress, e.id.RemoteAddress, netProto, e.multicastLoop) + e.route, err = e.stack.FindRoute(e.RegisterNICID, e.ID.LocalAddress, e.ID.RemoteAddress, netProto, e.multicastLoop) if err != nil { panic(err) } - } else if len(e.id.LocalAddress) != 0 && !isBroadcastOrMulticast(e.id.LocalAddress) { // stateBound + } else if len(e.ID.LocalAddress) != 0 && !isBroadcastOrMulticast(e.ID.LocalAddress) { // stateBound // A local unicast address is specified, verify that it's valid. - if e.stack.CheckLocalAddress(e.regNICID, netProto, e.id.LocalAddress) == 0 { + if e.stack.CheckLocalAddress(e.RegisterNICID, netProto, e.ID.LocalAddress) == 0 { panic(tcpip.ErrBadLocalAddress) } } @@ -107,9 +107,9 @@ func (e *endpoint) Resume(s *stack.Stack) { // Our saved state had a port, but we don't actually have a // reservation. We need to remove the port from our state, but still // pass it to the reservation machinery. - id := e.id - e.id.LocalPort = 0 - e.id, err = e.registerWithStack(e.regNICID, e.effectiveNetProtos, id) + id := e.ID + e.ID.LocalPort = 0 + e.ID, err = e.registerWithStack(e.RegisterNICID, e.effectiveNetProtos, id) if err != nil { panic(err) } diff --git a/pkg/tcpip/transport/udp/forwarder.go b/pkg/tcpip/transport/udp/forwarder.go index a9edc2c8d..fc706ede2 100644 --- a/pkg/tcpip/transport/udp/forwarder.go +++ b/pkg/tcpip/transport/udp/forwarder.go @@ -16,7 +16,6 @@ package udp import ( "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/waiter" ) @@ -44,12 +43,12 @@ func NewForwarder(s *stack.Stack, handler func(*ForwarderRequest)) *Forwarder { // // This function is expected to be passed as an argument to the // stack.SetTransportProtocolHandler function. -func (f *Forwarder) HandlePacket(r *stack.Route, id stack.TransportEndpointID, netHeader buffer.View, vv buffer.VectorisedView) bool { +func (f *Forwarder) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt tcpip.PacketBuffer) bool { f.handler(&ForwarderRequest{ stack: f.stack, route: r, id: id, - vv: vv, + pkt: pkt, }) return true @@ -62,7 +61,7 @@ type ForwarderRequest struct { stack *stack.Stack route *stack.Route id stack.TransportEndpointID - vv buffer.VectorisedView + pkt tcpip.PacketBuffer } // ID returns the 4-tuple (src address, src port, dst address, dst port) that @@ -74,15 +73,15 @@ func (r *ForwarderRequest) ID() stack.TransportEndpointID { // CreateEndpoint creates a connected UDP endpoint for the session request. func (r *ForwarderRequest) CreateEndpoint(queue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { ep := newEndpoint(r.stack, r.route.NetProto, queue) - if err := r.stack.RegisterTransportEndpoint(r.route.NICID(), []tcpip.NetworkProtocolNumber{r.route.NetProto}, ProtocolNumber, r.id, ep, ep.reusePort); err != nil { + if err := r.stack.RegisterTransportEndpoint(r.route.NICID(), []tcpip.NetworkProtocolNumber{r.route.NetProto}, ProtocolNumber, r.id, ep, ep.reusePort, ep.bindToDevice); err != nil { ep.Close() return nil, err } - ep.id = r.id + ep.ID = r.id ep.route = r.route.Clone() ep.dstPort = r.id.RemotePort - ep.regNICID = r.route.NICID() + ep.RegisterNICID = r.route.NICID() ep.state = StateConnected @@ -90,7 +89,7 @@ func (r *ForwarderRequest) CreateEndpoint(queue *waiter.Queue) (tcpip.Endpoint, ep.rcvReady = true ep.rcvMu.Unlock() - ep.HandlePacket(r.route, r.id, r.vv) + ep.HandlePacket(r.route, r.id, r.pkt) return ep, nil } diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go index f5cc932dd..43f11b700 100644 --- a/pkg/tcpip/transport/udp/protocol.go +++ b/pkg/tcpip/transport/udp/protocol.go @@ -66,10 +66,10 @@ func (*protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) { // HandleUnknownDestinationPacket handles packets targeted at this protocol but // that don't match any existing endpoint. -func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, netHeader buffer.View, vv buffer.VectorisedView) bool { +func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt tcpip.PacketBuffer) bool { // Get the header then trim it from the view. - hdr := header.UDP(vv.First()) - if int(hdr.Length()) > vv.Size() { + hdr := header.UDP(pkt.Data.First()) + if int(hdr.Length()) > pkt.Data.Size() { // Malformed packet. r.Stack().Stats().UDP.MalformedPacketsReceived.Increment() return true @@ -116,13 +116,18 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans } headerLen := int(r.MaxHeaderLength()) + header.ICMPv4MinimumSize available := int(mtu) - headerLen - payloadLen := len(netHeader) + vv.Size() + payloadLen := len(pkt.NetworkHeader) + pkt.Data.Size() if payloadLen > available { payloadLen = available } - payload := buffer.NewVectorisedView(len(netHeader), []buffer.View{netHeader}) - payload.Append(vv) + // The buffers used by pkt may be used elsewhere in the system. + // For example, a raw or packet socket may use what UDP + // considers an unreachable destination. Thus we deep copy pkt + // to prevent multiple ownership and SR errors. + newNetHeader := append(buffer.View(nil), pkt.NetworkHeader...) + payload := newNetHeader.ToVectorisedView() + payload.Append(pkt.Data.ToView().ToVectorisedView()) payload.CapLength(payloadLen) hdr := buffer.NewPrependable(headerLen) @@ -130,7 +135,7 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans pkt.SetType(header.ICMPv4DstUnreachable) pkt.SetCode(header.ICMPv4PortUnreachable) pkt.SetChecksum(header.ICMPv4Checksum(pkt, payload)) - r.WritePacket(nil /* gso */, hdr, payload, header.ICMPv4ProtocolNumber, r.DefaultTTL()) + r.WritePacket(nil /* gso */, hdr, payload, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}) case header.IPv6AddressSize: if !r.Stack().AllowICMPMessage() { @@ -151,12 +156,12 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans } headerLen := int(r.MaxHeaderLength()) + header.ICMPv6DstUnreachableMinimumSize available := int(mtu) - headerLen - payloadLen := len(netHeader) + vv.Size() + payloadLen := len(pkt.NetworkHeader) + pkt.Data.Size() if payloadLen > available { payloadLen = available } - payload := buffer.NewVectorisedView(len(netHeader), []buffer.View{netHeader}) - payload.Append(vv) + payload := buffer.NewVectorisedView(len(pkt.NetworkHeader), []buffer.View{pkt.NetworkHeader}) + payload.Append(pkt.Data) payload.CapLength(payloadLen) hdr := buffer.NewPrependable(headerLen) @@ -164,7 +169,7 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans pkt.SetType(header.ICMPv6DstUnreachable) pkt.SetCode(header.ICMPv6PortUnreachable) pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, payload)) - r.WritePacket(nil /* gso */, hdr, payload, header.ICMPv6ProtocolNumber, r.DefaultTTL()) + r.WritePacket(nil /* gso */, hdr, payload, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}) } return true } diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index 2ec27be4d..30ee9801b 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -17,7 +17,6 @@ package udp_test import ( "bytes" "fmt" - "math" "math/rand" "testing" "time" @@ -27,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/checker" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/link/channel" + "gvisor.dev/gvisor/pkg/tcpip/link/loopback" "gvisor.dev/gvisor/pkg/tcpip/link/sniffer" "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" @@ -384,18 +384,21 @@ func (c *testContext) injectPacket(flow testFlow, payload []byte) { h := flow.header4Tuple(incoming) if flow.isV4() { - c.injectV4Packet(payload, &h) + c.injectV4Packet(payload, &h, true /* valid */) } else { - c.injectV6Packet(payload, &h) + c.injectV6Packet(payload, &h, true /* valid */) } } // injectV6Packet creates a V6 test packet with the given payload and header -// values, and injects it into the link endpoint. -func (c *testContext) injectV6Packet(payload []byte, h *header4Tuple) { +// 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) { // Allocate a buffer for data and headers. buf := buffer.NewView(header.UDPMinimumSize + header.IPv6MinimumSize + len(payload)) - copy(buf[len(buf)-len(payload):], payload) + payloadStart := len(buf) - len(payload) + copy(buf[payloadStart:], payload) // Initialize the IP header. ip := header.IPv6(buf) @@ -409,10 +412,16 @@ func (c *testContext) injectV6Packet(payload []byte, h *header4Tuple) { // 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: uint16(header.UDPMinimumSize + len(payload)), + Length: l, }) // Calculate the UDP pseudo-header checksum. @@ -423,15 +432,22 @@ func (c *testContext) injectV6Packet(payload []byte, h *header4Tuple) { u.SetChecksum(^u.CalculateChecksum(xsum)) // Inject packet. - c.linkEP.Inject(ipv6.ProtocolNumber, buf.ToVectorisedView()) + c.linkEP.InjectInbound(ipv6.ProtocolNumber, tcpip.PacketBuffer{ + Data: buf.ToVectorisedView(), + NetworkHeader: buffer.View(ip), + TransportHeader: buffer.View(u), + }) } -// injectV6Packet creates a V4 test packet with the given payload and header -// values, and injects it into the link endpoint. -func (c *testContext) injectV4Packet(payload []byte, h *header4Tuple) { +// 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) { // Allocate a buffer for data and headers. buf := buffer.NewView(header.UDPMinimumSize + header.IPv4MinimumSize + len(payload)) - copy(buf[len(buf)-len(payload):], payload) + payloadStart := len(buf) - len(payload) + copy(buf[payloadStart:], payload) // Initialize the IP header. ip := header.IPv4(buf) @@ -461,7 +477,12 @@ func (c *testContext) injectV4Packet(payload []byte, h *header4Tuple) { u.SetChecksum(^u.CalculateChecksum(xsum)) // Inject packet. - c.linkEP.Inject(ipv4.ProtocolNumber, buf.ToVectorisedView()) + + c.linkEP.InjectInbound(ipv4.ProtocolNumber, tcpip.PacketBuffer{ + Data: buf.ToVectorisedView(), + NetworkHeader: buffer.View(ip), + TransportHeader: buffer.View(u), + }) } func newPayload() []byte { @@ -476,94 +497,67 @@ func newMinPayload(minSize int) []byte { return b } -func TestBindPortReuse(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpoint(ipv6.ProtocolNumber) - - var eps [5]tcpip.Endpoint - reusePortOpt := tcpip.ReusePortOption(1) - - pollChannel := make(chan tcpip.Endpoint) - for i := 0; i < len(eps); i++ { - // Try to receive the data. - wq := waiter.Queue{} - we, ch := waiter.NewChannelEntry(nil) - wq.EventRegister(&we, waiter.EventIn) - defer wq.EventUnregister(&we) - defer close(ch) - - var err *tcpip.Error - eps[i], err = c.s.NewEndpoint(udp.ProtocolNumber, ipv6.ProtocolNumber, &wq) - if err != nil { - c.t.Fatalf("NewEndpoint failed: %v", err) - } - - go func(ep tcpip.Endpoint) { - for range ch { - pollChannel <- ep - } - }(eps[i]) +func TestBindToDeviceOption(t *testing.T) { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}}) - defer eps[i].Close() - if err := eps[i].SetSockOpt(reusePortOpt); err != nil { - c.t.Fatalf("SetSockOpt failed failed: %v", err) - } - if err := eps[i].Bind(tcpip.FullAddress{Addr: stackV6Addr, Port: stackPort}); err != nil { - t.Fatalf("ep.Bind(...) failed: %v", err) - } + ep, err := s.NewEndpoint(udp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) + if err != nil { + t.Fatalf("NewEndpoint failed; %v", err) } + defer ep.Close() - npackets := 100000 - nports := 10000 - ports := make(map[uint16]tcpip.Endpoint) - stats := make(map[tcpip.Endpoint]int) - for i := 0; i < npackets; i++ { - // Send a packet. - port := uint16(i % nports) - payload := newPayload() - c.injectV6Packet(payload, &header4Tuple{ - srcAddr: tcpip.FullAddress{Addr: testV6Addr, Port: testPort + port}, - dstAddr: tcpip.FullAddress{Addr: stackV6Addr, Port: stackPort}, - }) + if err := s.CreateNamedNIC(321, "my_device", loopback.New()); err != nil { + t.Errorf("CreateNamedNIC failed: %v", err) + } - var addr tcpip.FullAddress - ep := <-pollChannel - _, _, err := ep.Read(&addr) - if err != nil { - c.t.Fatalf("Read failed: %v", err) - } - stats[ep]++ - if i < nports { - ports[uint16(i)] = ep - } else { - // Check that all packets from one client are handled - // by the same socket. - if ports[port] != ep { - t.Fatalf("Port mismatch") - } - } + // Make an nameless NIC. + if err := s.CreateNIC(54321, loopback.New()); err != nil { + t.Errorf("CreateNIC failed: %v", err) } - if len(stats) != len(eps) { - t.Fatalf("Only %d(expected %d) sockets received packets", len(stats), len(eps)) + // strPtr is used instead of taking the address of string literals, which is + // a compiler error. + strPtr := func(s string) *string { + return &s } - // Check that a packet distribution is fair between sockets. - for _, c := range stats { - n := float64(npackets) / float64(len(eps)) - // The deviation is less than 10%. - if math.Abs(float64(c)-n) > n/10 { - t.Fatal(c, n) - } + testActions := []struct { + name string + setBindToDevice *string + setBindToDeviceError *tcpip.Error + getBindToDevice tcpip.BindToDeviceOption + }{ + {"GetDefaultValue", nil, nil, ""}, + {"BindToNonExistent", strPtr("non_existent_device"), tcpip.ErrUnknownDevice, ""}, + {"BindToExistent", strPtr("my_device"), nil, "my_device"}, + {"UnbindToDevice", strPtr(""), nil, ""}, + } + for _, testAction := range testActions { + t.Run(testAction.name, func(t *testing.T) { + if testAction.setBindToDevice != nil { + bindToDevice := tcpip.BindToDeviceOption(*testAction.setBindToDevice) + if got, want := ep.SetSockOpt(bindToDevice), testAction.setBindToDeviceError; got != want { + t.Errorf("SetSockOpt(%v) got %v, want %v", bindToDevice, got, want) + } + } + bindToDevice := tcpip.BindToDeviceOption("to be modified by GetSockOpt") + if ep.GetSockOpt(&bindToDevice) != nil { + t.Errorf("GetSockOpt got %v, want %v", ep.GetSockOpt(&bindToDevice), nil) + } + if got, want := bindToDevice, testAction.getBindToDevice; got != want { + t.Errorf("bindToDevice got %q, want %q", got, want) + } + }) } } -// testRead sends a packet of the given test flow into the stack by injecting it -// into the link endpoint. It then reads it from the UDP endpoint and verifies -// its correctness. -func testRead(c *testContext, flow testFlow) { +// testReadInternal sends a packet of the given test flow into the stack by +// injecting it into the link endpoint. It then attempts to read it from the +// UDP endpoint and depending on if this was expected to succeed verifies its +// correctness. +func testReadInternal(c *testContext, flow testFlow, packetShouldBeDropped, expectReadError bool) { c.t.Helper() payload := newPayload() @@ -574,6 +568,9 @@ func testRead(c *testContext, flow testFlow) { c.wq.EventRegister(&we, waiter.EventIn) defer c.wq.EventUnregister(&we) + // Take a snapshot of the stats to validate them at the end of the test. + epstats := c.ep.Stats().(*tcpip.TransportEndpointStats).Clone() + var addr tcpip.FullAddress v, _, err := c.ep.Read(&addr) if err == tcpip.ErrWouldBlock { @@ -581,25 +578,55 @@ func testRead(c *testContext, flow testFlow) { select { case <-ch: v, _, err = c.ep.Read(&addr) - if err != nil { - c.t.Fatalf("Read failed: %v", err) - } - case <-time.After(1 * time.Second): - c.t.Fatalf("Timed out waiting for data") + case <-time.After(300 * time.Millisecond): + if packetShouldBeDropped { + return // expected to time out + } + c.t.Fatal("timed out waiting for data") } } + if expectReadError && err != nil { + c.checkEndpointReadStats(1, epstats, err) + return + } + + if err != nil { + c.t.Fatal("Read failed:", err) + } + + if packetShouldBeDropped { + c.t.Fatalf("Read unexpectedly received data from %s", addr.Addr) + } + // Check the peer address. h := flow.header4Tuple(incoming) if addr.Addr != h.srcAddr.Addr { - c.t.Fatalf("Unexpected remote address: got %v, want %v", addr.Addr, h.srcAddr) + c.t.Fatalf("unexpected remote address: got %s, want %s", addr.Addr, h.srcAddr) } // Check the payload. if !bytes.Equal(payload, v) { - c.t.Fatalf("Bad payload: got %x, want %x", v, payload) + c.t.Fatalf("bad payload: got %x, want %x", v, payload) } + c.checkEndpointReadStats(1, epstats, err) +} + +// testRead sends a packet of the given test flow into the stack by injecting it +// into the link endpoint. It then reads it from the UDP endpoint and verifies +// its correctness. +func testRead(c *testContext, flow testFlow) { + c.t.Helper() + testReadInternal(c, flow, false /* packetShouldBeDropped */, false /* expectReadError */) +} + +// testFailingRead sends a packet of the given test flow into the stack by +// injecting it into the link endpoint. It then tries to read it from the UDP +// endpoint and expects this to fail. +func testFailingRead(c *testContext, flow testFlow, expectReadError bool) { + c.t.Helper() + testReadInternal(c, flow, true /* packetShouldBeDropped */, expectReadError) } func TestBindEphemeralPort(t *testing.T) { @@ -771,13 +798,17 @@ func TestReadOnBoundToMulticast(t *testing.T) { c.t.Fatal("SetSockOpt failed:", err) } + // Check that we receive multicast packets but not unicast or broadcast + // ones. testRead(c, flow) + testFailingRead(c, broadcast, false /* expectReadError */) + testFailingRead(c, unicastV4, false /* expectReadError */) }) } } // TestV4ReadOnBoundToBroadcast checks that an endpoint can bind to a broadcast -// address and receive broadcast data on it. +// address and can receive only broadcast data. func TestV4ReadOnBoundToBroadcast(t *testing.T) { for _, flow := range []testFlow{broadcast, broadcastIn6} { t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { @@ -792,8 +823,31 @@ func TestV4ReadOnBoundToBroadcast(t *testing.T) { c.t.Fatalf("Bind failed: %s", err) } - // Test acceptance. + // Check that we receive broadcast packets but not unicast ones. + testRead(c, flow) + testFailingRead(c, unicastV4, false /* expectReadError */) + }) + } +} + +// TestV4ReadBroadcastOnBoundToWildcard checks that an endpoint can bind to ANY +// and receive broadcast and unicast data. +func TestV4ReadBroadcastOnBoundToWildcard(t *testing.T) { + for _, flow := range []testFlow{broadcast, broadcastIn6} { + t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpointForFlow(flow) + + // Bind to wildcard. + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + c.t.Fatalf("Bind failed: %s (", err) + } + + // Check that we receive both broadcast and unicast packets. testRead(c, flow) + testRead(c, unicastV4) }) } } @@ -802,7 +856,8 @@ func TestV4ReadOnBoundToBroadcast(t *testing.T) { // and verifies it fails with the provided error code. func testFailingWrite(c *testContext, flow testFlow, wantErr *tcpip.Error) { c.t.Helper() - + // Take a snapshot of the stats to validate them at the end of the test. + epstats := c.ep.Stats().(*tcpip.TransportEndpointStats).Clone() h := flow.header4Tuple(outgoing) writeDstAddr := flow.mapAddrIfApplicable(h.dstAddr.Addr) @@ -810,6 +865,7 @@ func testFailingWrite(c *testContext, flow testFlow, wantErr *tcpip.Error) { _, _, gotErr := c.ep.Write(tcpip.SlicePayload(payload), tcpip.WriteOptions{ To: &tcpip.FullAddress{Addr: writeDstAddr, Port: h.dstAddr.Port}, }) + c.checkEndpointWriteStats(1, epstats, gotErr) if gotErr != wantErr { c.t.Fatalf("Write returned unexpected error: got %v, want %v", gotErr, wantErr) } @@ -835,6 +891,8 @@ func testWriteWithoutDestination(c *testContext, flow testFlow, checkers ...chec func testWriteInternal(c *testContext, flow testFlow, setDest bool, checkers ...checker.NetworkChecker) uint16 { c.t.Helper() + // Take a snapshot of the stats to validate them at the end of the test. + epstats := c.ep.Stats().(*tcpip.TransportEndpointStats).Clone() writeOpts := tcpip.WriteOptions{} if setDest { @@ -852,7 +910,7 @@ func testWriteInternal(c *testContext, flow testFlow, setDest bool, checkers ... if n != int64(len(payload)) { c.t.Fatalf("Bad number of bytes written: got %v, want %v", n, len(payload)) } - + c.checkEndpointWriteStats(1, epstats, err) // Received the packet and check the payload. b := c.getPacketAndVerify(flow, checkers...) var udp header.UDP @@ -921,6 +979,10 @@ func TestDualWriteConnectedToV6(t *testing.T) { // Write to V4 mapped address. testFailingWrite(c, unicastV4in6, tcpip.ErrNetworkUnreachable) + const want = 1 + if got := c.ep.Stats().(*tcpip.TransportEndpointStats).SendErrors.NoRoute.Value(); got != want { + c.t.Fatalf("Endpoint stat not updated. got %d want %d", got, want) + } } func TestDualWriteConnectedToV4Mapped(t *testing.T) { @@ -1183,6 +1245,109 @@ func TestTTL(t *testing.T) { } } +func TestSetTTL(t *testing.T) { + for _, flow := range []testFlow{unicastV4, unicastV4in6, unicastV6, unicastV6Only, broadcast, broadcastIn6} { + t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { + for _, wantTTL := range []uint8{1, 2, 50, 64, 128, 254, 255} { + t.Run(fmt.Sprintf("TTL:%d", wantTTL), func(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpointForFlow(flow) + + if err := c.ep.SetSockOpt(tcpip.TTLOption(wantTTL)); err != nil { + c.t.Fatalf("SetSockOpt failed: %v", err) + } + + var p stack.NetworkProtocol + if flow.isV4() { + p = ipv4.NewProtocol() + } else { + p = ipv6.NewProtocol() + } + ep, err := p.NewEndpoint(0, tcpip.AddressWithPrefix{}, nil, nil, nil) + if err != nil { + t.Fatal(err) + } + ep.Close() + + testWrite(c, flow, checker.TTL(wantTTL)) + }) + } + }) + } +} + +func TestTOSV4(t *testing.T) { + for _, flow := range []testFlow{unicastV4, multicastV4, broadcast} { + t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpointForFlow(flow) + + const tos = 0xC0 + var v tcpip.IPv4TOSOption + if err := c.ep.GetSockOpt(&v); err != nil { + c.t.Errorf("GetSockopt failed: %s", err) + } + // Test for expected default value. + if v != 0 { + c.t.Errorf("got GetSockOpt(...) = %#v, want = %#v", v, 0) + } + + if err := c.ep.SetSockOpt(tcpip.IPv4TOSOption(tos)); err != nil { + c.t.Errorf("SetSockOpt(%#v) failed: %s", tcpip.IPv4TOSOption(tos), err) + } + + if err := c.ep.GetSockOpt(&v); err != nil { + c.t.Errorf("GetSockopt failed: %s", err) + } + + if want := tcpip.IPv4TOSOption(tos); v != want { + c.t.Errorf("got GetSockOpt(...) = %#v, want = %#v", v, want) + } + + testWrite(c, flow, checker.TOS(tos, 0)) + }) + } +} + +func TestTOSV6(t *testing.T) { + for _, flow := range []testFlow{unicastV4in6, unicastV6, unicastV6Only, multicastV4in6, multicastV6, broadcastIn6} { + t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpointForFlow(flow) + + const tos = 0xC0 + var v tcpip.IPv6TrafficClassOption + if err := c.ep.GetSockOpt(&v); err != nil { + c.t.Errorf("GetSockopt failed: %s", err) + } + // Test for expected default value. + if v != 0 { + c.t.Errorf("got GetSockOpt(...) = %#v, want = %#v", v, 0) + } + + if err := c.ep.SetSockOpt(tcpip.IPv6TrafficClassOption(tos)); err != nil { + c.t.Errorf("SetSockOpt failed: %s", err) + } + + if err := c.ep.GetSockOpt(&v); err != nil { + c.t.Errorf("GetSockopt failed: %s", err) + } + + if want := tcpip.IPv6TrafficClassOption(tos); v != want { + c.t.Errorf("got GetSockOpt(...) = %#v, want = %#v", v, want) + } + + testWrite(c, flow, checker.TOS(tos, 0)) + }) + } +} + func TestMulticastInterfaceOption(t *testing.T) { for _, flow := range []testFlow{multicastV4, multicastV4in6, multicastV6, multicastV6Only} { t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { @@ -1396,3 +1561,117 @@ func TestV6UnknownDestination(t *testing.T) { }) } } + +// TestIncrementMalformedPacketsReceived verifies if the malformed received +// global and endpoint stats get incremented. +func TestIncrementMalformedPacketsReceived(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: %v", err) + } + + payload := newPayload() + c.t.Helper() + h := unicastV6.header4Tuple(incoming) + c.injectV6Packet(payload, &h, false /* !valid */) + + var want uint64 = 1 + if got := c.s.Stats().UDP.MalformedPacketsReceived.Value(); got != want { + t.Errorf("got stats.UDP.MalformedPacketsReceived.Value() = %v, want = %v", 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) + } +} + +// TestShutdownRead verifies endpoint read shutdown and error +// stats increment on packet receive. +func TestShutdownRead(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: %v", err) + } + + if err := c.ep.Connect(tcpip.FullAddress{Addr: testV6Addr, Port: testPort}); err != nil { + c.t.Fatalf("Connect failed: %v", err) + } + + if err := c.ep.Shutdown(tcpip.ShutdownRead); err != nil { + t.Fatalf("Shutdown failed: %v", err) + } + + testFailingRead(c, unicastV6, true /* expectReadError */) + + var want uint64 = 1 + if got := c.s.Stats().UDP.ReceiveBufferErrors.Value(); got != want { + t.Errorf("got stats.UDP.ReceiveBufferErrors.Value() = %v, want = %v", got, want) + } + if got := c.ep.Stats().(*tcpip.TransportEndpointStats).ReceiveErrors.ClosedReceiver.Value(); got != want { + t.Errorf("got EP Stats.ReceiveErrors.ClosedReceiver stats = %v, want = %v", got, want) + } +} + +// TestShutdownWrite verifies endpoint write shutdown and error +// stats increment on packet write. +func TestShutdownWrite(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpoint(ipv6.ProtocolNumber) + + if err := c.ep.Connect(tcpip.FullAddress{Addr: testV6Addr, Port: testPort}); err != nil { + c.t.Fatalf("Connect failed: %v", err) + } + + if err := c.ep.Shutdown(tcpip.ShutdownWrite); err != nil { + t.Fatalf("Shutdown failed: %v", err) + } + + testFailingWrite(c, unicastV6, tcpip.ErrClosedForSend) +} + +func (c *testContext) checkEndpointWriteStats(incr uint64, want tcpip.TransportEndpointStats, err *tcpip.Error) { + got := c.ep.Stats().(*tcpip.TransportEndpointStats).Clone() + switch err { + case nil: + want.PacketsSent.IncrementBy(incr) + case tcpip.ErrMessageTooLong, tcpip.ErrInvalidOptionValue: + want.WriteErrors.InvalidArgs.IncrementBy(incr) + case tcpip.ErrClosedForSend: + want.WriteErrors.WriteClosed.IncrementBy(incr) + case tcpip.ErrInvalidEndpointState: + want.WriteErrors.InvalidEndpointState.IncrementBy(incr) + case tcpip.ErrNoLinkAddress: + want.SendErrors.NoLinkAddr.IncrementBy(incr) + case tcpip.ErrNoRoute, tcpip.ErrBroadcastDisabled, tcpip.ErrNetworkUnreachable: + want.SendErrors.NoRoute.IncrementBy(incr) + default: + want.SendErrors.SendToNetworkFailed.IncrementBy(incr) + } + if got != want { + c.t.Errorf("Endpoint stats not matching for error %s got %+v want %+v", err, got, want) + } +} + +func (c *testContext) checkEndpointReadStats(incr uint64, want tcpip.TransportEndpointStats, err *tcpip.Error) { + got := c.ep.Stats().(*tcpip.TransportEndpointStats).Clone() + switch err { + case nil, tcpip.ErrWouldBlock: + case tcpip.ErrClosedForReceive: + want.ReadErrors.ReadClosed.IncrementBy(incr) + default: + c.t.Errorf("Endpoint error missing stats update err %v", err) + } + if got != want { + c.t.Errorf("Endpoint stats not matching for error %s got %+v want %+v", err, got, want) + } +} |