diff options
Diffstat (limited to 'pkg/tcpip')
112 files changed, 11680 insertions, 1416 deletions
diff --git a/pkg/tcpip/BUILD b/pkg/tcpip/BUILD index 3c2b2b5ea..65d4d0cd8 100644 --- a/pkg/tcpip/BUILD +++ b/pkg/tcpip/BUILD @@ -6,6 +6,8 @@ package(licenses = ["notice"]) 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/prependable.go b/pkg/tcpip/buffer/prependable.go index 4287464f3..ba21f4eca 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 { @@ -72,3 +77,9 @@ func (p *Prependable) Prepend(size int) []byte { p.usedIdx -= size return p.View()[:size:size] } + +// DeepCopy copies p and the bytes backing it. +func (p Prependable) DeepCopy() Prependable { + p.buf = append(View(nil), p.buf...) + return p +} diff --git a/pkg/tcpip/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 02137e1c9..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) } diff --git a/pkg/tcpip/hash/jenkins/BUILD b/pkg/tcpip/hash/jenkins/BUILD index 0c5c20cea..e648efa71 100644 --- a/pkg/tcpip/hash/jenkins/BUILD +++ b/pkg/tcpip/hash/jenkins/BUILD @@ -7,9 +7,7 @@ go_library( name = "jenkins", srcs = ["jenkins.go"], importpath = "gvisor.dev/gvisor/pkg/tcpip/hash/jenkins", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], ) go_test( diff --git a/pkg/tcpip/header/BUILD b/pkg/tcpip/header/BUILD index 07d09abed..f1d837196 100644 --- a/pkg/tcpip/header/BUILD +++ b/pkg/tcpip/header/BUILD @@ -19,6 +19,7 @@ go_library( "ndp_neighbor_advert.go", "ndp_neighbor_solicit.go", "ndp_options.go", + "ndp_router_advert.go", "tcp.go", "udp.go", ], @@ -36,16 +37,29 @@ go_test( name = "header_x_test", size = "small", srcs = [ + "checksum_test.go", + "ipv6_test.go", "ipversion_test.go", "tcp_test.go", ], - deps = [":header"], + deps = [ + ":header", + "//pkg/tcpip", + "//pkg/tcpip/buffer", + "@com_github_google_go-cmp//cmp:go_default_library", + ], ) go_test( name = "header_test", size = "small", - srcs = ["ndp_test.go"], + srcs = [ + "eth_test.go", + "ndp_test.go", + ], embed = [":header"], - deps = ["//pkg/tcpip"], + deps = [ + "//pkg/tcpip", + "@com_github_google_go-cmp//cmp:go_default_library", + ], ) 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 e51c5098c..b4037b6c8 100644 --- a/pkg/tcpip/header/icmpv6.go +++ b/pkg/tcpip/header/icmpv6.go @@ -80,6 +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. @@ -125,7 +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) } @@ -190,7 +197,7 @@ 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/ipv6.go b/pkg/tcpip/header/ipv6.go index b125bbea5..fc671e439 100644 --- a/pkg/tcpip/header/ipv6.go +++ b/pkg/tcpip/header/ipv6.go @@ -90,9 +90,23 @@ const ( // 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" + + // IIDSize is the size of an interface identifier (IID), in bytes, as + // defined by RFC 4291 section 2.5.1. + IIDSize = 8 + + // IIDOffsetInIPv6Address is the offset, in bytes, from the start + // of an IPv6 address to the beginning of the interface identifier + // (IID) for auto-generated addresses. That is, all bytes before + // the IIDOffsetInIPv6Address-th byte are the prefix bytes, and all + // bytes including and after the IIDOffsetInIPv6Address-th byte are + // for the IID. + IIDOffsetInIPv6Address = 8 ) -// 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 { @@ -101,6 +115,15 @@ 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 { @@ -255,27 +278,43 @@ func SolicitedNodeAddr(addr tcpip.Address) tcpip.Address { return solicitedNodeMulticastPrefix + addr[len(addr)-3:] } +// EthernetAdddressToModifiedEUI64IntoBuf populates buf with a modified EUI-64 +// from a 48-bit Ethernet/MAC address, as per RFC 4291 section 2.5.1. +// +// buf MUST be at least 8 bytes. +func EthernetAdddressToModifiedEUI64IntoBuf(linkAddr tcpip.LinkAddress, buf []byte) { + buf[0] = linkAddr[0] ^ 2 + buf[1] = linkAddr[1] + buf[2] = linkAddr[2] + buf[3] = 0xFF + buf[4] = 0xFE + buf[5] = linkAddr[3] + buf[6] = linkAddr[4] + buf[7] = linkAddr[5] +} + +// EthernetAddressToModifiedEUI64 computes a modified EUI-64 from a 48-bit +// Ethernet/MAC address, as per RFC 4291 section 2.5.1. +func EthernetAddressToModifiedEUI64(linkAddr tcpip.LinkAddress) [IIDSize]byte { + var buf [IIDSize]byte + EthernetAdddressToModifiedEUI64IntoBuf(linkAddr, buf[:]) + return buf +} + // LinkLocalAddr computes the default IPv6 link-local address from a link-layer // (MAC) address. func LinkLocalAddr(linkAddr tcpip.LinkAddress) tcpip.Address { - // Convert a 48-bit MAC to an EUI-64 and then prepend the link-local - // header, FE80::. + // Convert a 48-bit MAC to a modified EUI-64 and then prepend the + // link-local header, FE80::. // // The conversion is very nearly: // aa:bb:cc:dd:ee:ff => FE80::Aabb:ccFF:FEdd:eeff // Note the capital A. The conversion aa->Aa involves a bit flip. - lladdrb := [16]byte{ - 0: 0xFE, - 1: 0x80, - 8: linkAddr[0] ^ 2, - 9: linkAddr[1], - 10: linkAddr[2], - 11: 0xFF, - 12: 0xFE, - 13: linkAddr[3], - 14: linkAddr[4], - 15: linkAddr[5], + lladdrb := [IPv6AddressSize]byte{ + 0: 0xFE, + 1: 0x80, } + EthernetAdddressToModifiedEUI64IntoBuf(linkAddr, lladdrb[IIDOffsetInIPv6Address:]) return tcpip.Address(lladdrb[:]) } diff --git a/pkg/tcpip/header/ipv6_test.go b/pkg/tcpip/header/ipv6_test.go new file mode 100644 index 000000000..42c5c6fc1 --- /dev/null +++ b/pkg/tcpip/header/ipv6_test.go @@ -0,0 +1,45 @@ +// 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_test + +import ( + "testing" + + "github.com/google/go-cmp/cmp" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/header" +) + +const linkAddr = tcpip.LinkAddress("\x02\x02\x03\x04\x05\x06") + +func TestEthernetAdddressToModifiedEUI64(t *testing.T) { + expectedIID := [header.IIDSize]byte{0, 2, 3, 255, 254, 4, 5, 6} + + if diff := cmp.Diff(expectedIID, header.EthernetAddressToModifiedEUI64(linkAddr)); diff != "" { + t.Errorf("EthernetAddressToModifiedEUI64(%s) mismatch (-want +got):\n%s", linkAddr, diff) + } + + var buf [header.IIDSize]byte + header.EthernetAdddressToModifiedEUI64IntoBuf(linkAddr, buf[:]) + if diff := cmp.Diff(expectedIID, buf); diff != "" { + t.Errorf("EthernetAddressToModifiedEUI64IntoBuf(%s, _) mismatch (-want +got):\n%s", linkAddr, diff) + } +} + +func TestLinkLocalAddr(t *testing.T) { + if got, want := header.LinkLocalAddr(linkAddr), tcpip.Address("\xfe\x80\x00\x00\x00\x00\x00\x00\x00\x02\x03\xff\xfe\x04\x05\x06"); got != want { + t.Errorf("got LinkLocalAddr(%s) = %s, want = %s", linkAddr, got, want) + } +} diff --git a/pkg/tcpip/header/ndp_neighbor_advert.go b/pkg/tcpip/header/ndp_neighbor_advert.go index 5c2b472c8..505c92668 100644 --- a/pkg/tcpip/header/ndp_neighbor_advert.go +++ b/pkg/tcpip/header/ndp_neighbor_advert.go @@ -18,6 +18,8 @@ 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 ( diff --git a/pkg/tcpip/header/ndp_neighbor_solicit.go b/pkg/tcpip/header/ndp_neighbor_solicit.go index 1dcb0fbc6..3a1b8e139 100644 --- a/pkg/tcpip/header/ndp_neighbor_solicit.go +++ b/pkg/tcpip/header/ndp_neighbor_solicit.go @@ -18,6 +18,8 @@ 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 ( diff --git a/pkg/tcpip/header/ndp_options.go b/pkg/tcpip/header/ndp_options.go index b28bde15b..06e0bace2 100644 --- a/pkg/tcpip/header/ndp_options.go +++ b/pkg/tcpip/header/ndp_options.go @@ -15,6 +15,11 @@ package header import ( + "encoding/binary" + "errors" + "math" + "time" + "gvisor.dev/gvisor/pkg/tcpip" ) @@ -27,15 +32,226 @@ const ( // 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 + + // NDPRecursiveDNSServerOptionType is the type of the Recursive DNS + // Server option, as per RFC 8106 section 5.1. + NDPRecursiveDNSServerOptionType = 25 + + // ndpRecursiveDNSServerLifetimeOffset is the start of the 4-byte + // Lifetime field within an NDPRecursiveDNSServer. + ndpRecursiveDNSServerLifetimeOffset = 2 + + // ndpRecursiveDNSServerAddressesOffset is the start of the addresses + // for IPv6 Recursive DNS Servers within an NDPRecursiveDNSServer. + ndpRecursiveDNSServerAddressesOffset = 6 + + // minNDPRecursiveDNSServerLength is the minimum NDP Recursive DNS + // Server option's length field value when it contains at least one + // IPv6 address. + minNDPRecursiveDNSServerLength = 3 + // 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 ) +var ( + // NDPInfiniteLifetime is a value that represents infinity for the + // 4-byte lifetime fields found in various NDP options. Its value is + // (2^32 - 1)s = 4294967295s. + // + // This is a variable instead of a constant so that tests can change + // this value to a smaller value. It should only be modified by tests. + NDPInfiniteLifetime = time.Second * math.MaxUint32 +) + +// 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") + ErrNDPInvalidLength = errors.New("NDP option's Length value is invalid as per relevant RFC") +) + +// 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 + + case NDPRecursiveDNSServerOptionType: + // RFC 8106 section 5.3.1 outlines that the RDNSS option + // must have a minimum length of 3 so it contains at + // least one IPv6 address. + if l < minNDPRecursiveDNSServerLength { + return nil, true, ErrNDPInvalidLength + } + + opt := NDPRecursiveDNSServer(body) + if len(opt.Addresses()) == 0 { + return nil, true, ErrNDPOptMalformedBody + } + + return opt, 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 @@ -75,15 +291,15 @@ func (b NDPOptions) Serialize(s NDPOptionsSerializer) int { return done } -// ndpOption is the set of functions to be implemented by all NDP option types. -type ndpOption interface { - // Type returns the type of this ndpOption. +// 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 this ndpOption, in bytes. + // Length returns the length of the body of the receiver, in bytes. Length() int - // serializeInto serializes this ndpOption into the provided byte + // serializeInto serializes the receiver into the provided byte // buffer. // // Note, the caller MUST provide a byte buffer with size of at least @@ -92,15 +308,15 @@ type ndpOption interface { // buffer is not of sufficient size. // // serializeInto will return the number of bytes that was used to - // serialize this ndpOption. Implementers must only use the number of - // bytes required to serialize this ndpOption. Callers MAY provide a + // 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 { +func paddedLength(o NDPOption) int { l := o.Length() if l == 0 { @@ -139,7 +355,7 @@ func paddedLength(o ndpOption) int { } // NDPOptionsSerializer is a serializer for NDP options. -type NDPOptionsSerializer []ndpOption +type NDPOptionsSerializer []NDPOption // Length returns the total number of bytes required to serialize. func (b NDPOptionsSerializer) Length() int { @@ -154,19 +370,220 @@ func (b NDPOptionsSerializer) Length() int { // 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. +// Type implements NDPOption.Type. func (o NDPTargetLinkLayerAddressOption) Type() uint8 { return NDPTargetLinkLayerAddressOptionType } -// Length implements ndpOption.Length. +// Length implements NDPOption.Length. func (o NDPTargetLinkLayerAddressOption) Length() int { return len(o) } -// serializeInto implements ndpOption.serializeInto. +// 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 +// NDPInfiniteLifetime. +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 +// NDPInfiniteLifetime. +// +// 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]) +} + +// Subnet returns the Prefix field and Prefix Length field represented in a +// tcpip.Subnet. +func (o NDPPrefixInformation) Subnet() tcpip.Subnet { + addrWithPrefix := tcpip.AddressWithPrefix{ + Address: o.Prefix(), + PrefixLen: int(o.PrefixLength()), + } + return addrWithPrefix.Subnet() +} + +// NDPRecursiveDNSServer is the NDP Recursive DNS Server option, as defined by +// RFC 8106 section 5.1. +// +// To make sure that the option meets its minimum length and does not end in the +// middle of a DNS server's IPv6 address, the length of a valid +// NDPRecursiveDNSServer must meet the following constraint: +// (Length - ndpRecursiveDNSServerAddressesOffset) % IPv6AddressSize == 0 +type NDPRecursiveDNSServer []byte + +// Type returns the type of an NDP Recursive DNS Server option. +// +// Type implements NDPOption.Type. +func (NDPRecursiveDNSServer) Type() uint8 { + return NDPRecursiveDNSServerOptionType +} + +// Length implements NDPOption.Length. +func (o NDPRecursiveDNSServer) Length() int { + return len(o) +} + +// serializeInto implements NDPOption.serializeInto. +func (o NDPRecursiveDNSServer) serializeInto(b []byte) int { + used := copy(b, o) + + // Zero out the reserved bytes that are before the Lifetime field. + for i := 0; i < ndpRecursiveDNSServerLifetimeOffset; i++ { + b[i] = 0 + } + + return used +} + +// Lifetime returns the length of time that the DNS server addresses +// in this option may be used for name resolution. +// +// Note, a value of 0 implies the addresses should no longer be used, +// and a value of infinity/forever is represented by NDPInfiniteLifetime. +// +// Lifetime may panic if o does not have enough bytes to hold the Lifetime +// field. +func (o NDPRecursiveDNSServer) Lifetime() time.Duration { + // The field is the time in seconds, as per RFC 8106 section 5.1. + return time.Second * time.Duration(binary.BigEndian.Uint32(o[ndpRecursiveDNSServerLifetimeOffset:])) +} + +// Addresses returns the recursive DNS server IPv6 addresses that may be +// used for name resolution. +// +// Note, some of the addresses returned MAY be link-local addresses. +// +// Addresses may panic if o does not hold valid IPv6 addresses. +func (o NDPRecursiveDNSServer) Addresses() []tcpip.Address { + l := len(o) + if l < ndpRecursiveDNSServerAddressesOffset { + return nil + } + + l -= ndpRecursiveDNSServerAddressesOffset + if l%IPv6AddressSize != 0 { + return nil + } + + buf := o[ndpRecursiveDNSServerAddressesOffset:] + var addrs []tcpip.Address + for len(buf) > 0 { + addr := tcpip.Address(buf[:IPv6AddressSize]) + if !IsV6UnicastAddress(addr) { + return nil + } + addrs = append(addrs, addr) + buf = buf[IPv6AddressSize:] + } + return addrs +} 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 index a431a6e61..2c439d70c 100644 --- a/pkg/tcpip/header/ndp_test.go +++ b/pkg/tcpip/header/ndp_test.go @@ -17,7 +17,9 @@ package header import ( "bytes" "testing" + "time" + "github.com/google/go-cmp/cmp" "gvisor.dev/gvisor/pkg/tcpip" ) @@ -35,18 +37,18 @@ func TestNDPNeighborSolicit(t *testing.T) { 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.Fatalf("got ns.TargetAddress = %s, want %s", 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.Fatalf("got ns.TargetAddress = %s, want %s", 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.Fatalf("got targetaddress buffer = %s, want %s", got, addr2) + t.Errorf("got targetaddress buffer = %s, want %s", got, addr2) } } @@ -64,59 +66,129 @@ func TestNDPNeighborAdvert(t *testing.T) { 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.Fatalf("got TargetAddress = %s, want %s", got, addr) + t.Errorf("got TargetAddress = %s, want %s", got, addr) } // Test getting the Router Flag. if got := na.RouterFlag(); !got { - t.Fatalf("got RouterFlag = false, want = true") + t.Errorf("got RouterFlag = false, want = true") } // Test getting the Solicited Flag. if got := na.SolicitedFlag(); got { - t.Fatalf("got SolicitedFlag = true, want = false") + t.Errorf("got SolicitedFlag = true, want = false") } // Test getting the Override Flag. if got := na.OverrideFlag(); !got { - t.Fatalf("got OverrideFlag = false, want = true") + 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.Fatalf("got TargetAddress = %s, want %s", 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.Fatalf("got targetaddress buffer = %s, want %s", 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.Fatalf("got RouterFlag = true, want = false") + t.Errorf("got RouterFlag = true, want = false") } // Test updating the Solicited Flag. na.SetSolicitedFlag(true) if got := na.SolicitedFlag(); !got { - t.Fatalf("got SolicitedFlag = false, want = true") + t.Errorf("got SolicitedFlag = false, want = true") } // Test updating the Override Flag. na.SetOverrideFlag(false) if got := na.OverrideFlag(); got { - t.Fatalf("got OverrideFlag = true, want = false") + t.Errorf("got OverrideFlag = true, want = false") } // Make sure flags got updated in the backing buffer. if got := b[ndpNAFlagsOffset]; got != 64 { - t.Fatalf("got flags byte = %d, want = 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) { @@ -159,6 +231,555 @@ func TestNDPTargetLinkLayerAddressOptionSerialize(t *testing.T) { 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) + } +} + +func TestNDPRecursiveDNSServerOptionSerialize(t *testing.T) { + b := []byte{ + 9, 8, + 1, 2, 4, 8, + 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, + } + 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} + expected := []byte{ + 25, 3, 0, 0, + 1, 2, 4, 8, + 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, + } + opts := NDPOptions(targetBuf) + serializer := NDPOptionsSerializer{ + NDPRecursiveDNSServer(b), + } + if got, want := opts.Serialize(serializer), len(expected); got != want { + t.Errorf("got Serialize = %d, want = %d", got, want) + } + if !bytes.Equal(targetBuf, expected) { + t.Fatalf("got targetBuf = %x, want = %x", targetBuf, expected) + } + + 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 != NDPRecursiveDNSServerOptionType { + t.Errorf("got Type = %d, want = %d", got, NDPRecursiveDNSServerOptionType) + } + + opt, ok := next.(NDPRecursiveDNSServer) + if !ok { + t.Fatalf("next (type = %T) cannot be casted to an NDPRecursiveDNSServer", next) + } + if got := opt.Type(); got != 25 { + t.Errorf("got Type = %d, want = 31", got) + } + if got := opt.Length(); got != 22 { + t.Errorf("got Length = %d, want = 22", got) + } + if got, want := opt.Lifetime(), 16909320*time.Second; got != want { + t.Errorf("got Lifetime = %s, want = %s", got, want) + } + want := []tcpip.Address{ + "\x00\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x0f", + } + if got := opt.Addresses(); !cmp.Equal(got, want) { + t.Errorf("got Addresses = %v, want = %v", 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) + } +} + +func TestNDPRecursiveDNSServerOption(t *testing.T) { + tests := []struct { + name string + buf []byte + lifetime time.Duration + addrs []tcpip.Address + }{ + { + "Valid1Addr", + []byte{ + 25, 3, 0, 0, + 0, 0, 0, 0, + 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, + }, + 0, + []tcpip.Address{ + "\x00\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x0f", + }, + }, + { + "Valid2Addr", + []byte{ + 25, 5, 0, 0, + 0, 0, 0, 0, + 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, + 17, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 16, + }, + 0, + []tcpip.Address{ + "\x00\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x0f", + "\x11\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x10", + }, + }, + { + "Valid3Addr", + []byte{ + 25, 7, 0, 0, + 0, 0, 0, 0, + 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, + 17, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 16, + 17, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 17, + }, + 0, + []tcpip.Address{ + "\x00\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x0f", + "\x11\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x10", + "\x11\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x11", + }, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + opts := NDPOptions(test.buf) + it, err := opts.Iter(true) + if err != nil { + t.Fatalf("got Iter = (_, %s), want = (_, nil)", err) + } + + // Iterator should get our 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 := next.Type(); got != NDPRecursiveDNSServerOptionType { + t.Fatalf("got Type %= %d, want = %d", got, NDPRecursiveDNSServerOptionType) + } + + opt, ok := next.(NDPRecursiveDNSServer) + if !ok { + t.Fatalf("next (type = %T) cannot be casted to an NDPRecursiveDNSServer", next) + } + if got := opt.Lifetime(); got != test.lifetime { + t.Errorf("got Lifetime = %d, want = %d", got, test.lifetime) + } + if got := opt.Addresses(); !cmp.Equal(got, test.addrs) { + t.Errorf("got Addresses = %v, want = %v", got, test.addrs) + } + + // 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, + }, + { + "InvalidRecursiveDNSServerCutsOffAddress", + []byte{ + 25, 4, 0, 0, + 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, + 0, 1, 2, 3, 4, 5, 6, 7, + }, + ErrNDPOptMalformedBody, + }, + { + "InvalidRecursiveDNSServerInvalidLengthField", + []byte{ + 25, 2, 0, 0, + 0, 0, 0, 0, + 0, 1, 2, 3, 4, 5, 6, 7, 8, + }, + ErrNDPInvalidLength, + }, + { + "RecursiveDNSServerTooSmall", + []byte{ + 25, 1, 0, 0, + 0, 0, 0, + }, + ErrNDPOptBufExhausted, + }, + { + "RecursiveDNSServerMulticast", + []byte{ + 25, 3, 0, 0, + 0, 0, 0, 0, + 255, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, + }, + ErrNDPOptMalformedBody, + }, + { + "RecursiveDNSServerUnspecified", + []byte{ + 25, 3, 0, 0, + 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + }, + ErrNDPOptMalformedBody, + }, + } + + 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/link/channel/BUILD b/pkg/tcpip/link/channel/BUILD index 97a794986..7dbc05754 100644 --- a/pkg/tcpip/link/channel/BUILD +++ b/pkg/tcpip/link/channel/BUILD @@ -6,7 +6,7 @@ go_library( name = "channel", srcs = ["channel.go"], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/channel", - visibility = ["//:sandbox"], + visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/link/channel/channel.go b/pkg/tcpip/link/channel/channel.go index 18adb2085..70188551f 100644 --- a/pkg/tcpip/link/channel/channel.go +++ b/pkg/tcpip/link/channel/channel.go @@ -25,10 +25,9 @@ import ( // PacketInfo holds all the information about an outbound packet. type PacketInfo struct { - Header buffer.View - Payload buffer.View - Proto tcpip.NetworkProtocolNumber - GSO *stack.GSO + Pkt tcpip.PacketBuffer + Proto tcpip.NetworkProtocolNumber + GSO *stack.GSO } // Endpoint is link layer endpoint that stores outbound packets in a channel @@ -65,14 +64,14 @@ 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 +95,7 @@ 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 } @@ -118,12 +117,55 @@ func (e *Endpoint) LinkAddress() tcpip.LinkAddress { } // WritePacket stores outbound packets into the channel. -func (e *Endpoint) WritePacket(_ *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (e *Endpoint) WritePacket(_ *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { p := PacketInfo{ - Header: hdr.View(), - Proto: protocol, - Payload: payload.ToView(), - GSO: gso, + Pkt: pkt, + Proto: protocol, + GSO: gso, + } + + select { + case e.C <- p: + default: + } + + return nil +} + +// WritePackets stores outbound packets into the channel. +func (e *Endpoint) WritePackets(_ *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + payloadView := pkts[0].Data.ToView() + n := 0 +packetLoop: + for _, pkt := range pkts { + off := pkt.DataOffset + size := pkt.DataSize + p := PacketInfo{ + Pkt: tcpip.PacketBuffer{ + Header: pkt.Header, + Data: buffer.NewViewFromBytes(payloadView[off : off+size]).ToVectorisedView(), + }, + Proto: protocol, + GSO: gso, + } + + select { + case e.C <- p: + n++ + default: + break packetLoop + } + } + + return n, nil +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *Endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { + p := PacketInfo{ + Pkt: tcpip.PacketBuffer{Data: vv}, + Proto: 0, + GSO: nil, } select { diff --git a/pkg/tcpip/link/fdbased/BUILD b/pkg/tcpip/link/fdbased/BUILD index 8fa9e3984..897c94821 100644 --- a/pkg/tcpip/link/fdbased/BUILD +++ b/pkg/tcpip/link/fdbased/BUILD @@ -14,9 +14,7 @@ go_library( "packet_dispatchers.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/fdbased", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go index f80ac3435..fa8a703d9 100644 --- a/pkg/tcpip/link/fdbased/endpoint.go +++ b/pkg/tcpip/link/fdbased/endpoint.go @@ -165,6 +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 @@ -242,7 +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 } } @@ -379,10 +386,11 @@ const ( // WritePacket writes outbound packets to the file descriptor. If it is not // currently writable, the packet is dropped. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { if e.hdrSize > 0 { // Add ethernet header if needed. - eth := header.Ethernet(hdr.Prepend(header.EthernetMinimumSize)) + eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) + pkt.LinkHeader = buffer.View(eth) ethHdr := &header.EthernetFields{ DstAddr: r.RemoteLinkAddress, Type: protocol, @@ -397,17 +405,17 @@ 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 { - vnetHdr.hdrLen = uint16(hdr.UsedLength()) + vnetHdr.hdrLen = uint16(pkt.Header.UsedLength()) if gso.NeedsCsum { vnetHdr.flags = _VIRTIO_NET_HDR_F_NEEDS_CSUM vnetHdr.csumStart = header.EthernetMinimumSize + gso.L3HdrLen vnetHdr.csumOffset = gso.CsumOffset } - if gso.Type != stack.GSONone && uint16(payload.Size()) > gso.MSS { + if gso.Type != stack.GSONone && uint16(pkt.Data.Size()) > gso.MSS { switch gso.Type { case stack.GSOTCPv4: vnetHdr.gsoType = _VIRTIO_NET_HDR_GSO_TCPV4 @@ -420,18 +428,151 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen } } - return rawfile.NonBlockingWrite3(e.fds[0], vnetHdrBuf, hdr.View(), payload.ToView()) + return rawfile.NonBlockingWrite3(e.fds[0], vnetHdrBuf, pkt.Header.View(), pkt.Data.ToView()) + } + + if pkt.Data.Size() == 0 { + return rawfile.NonBlockingWrite(e.fds[0], pkt.Header.View()) } - if payload.Size() == 0 { - return rawfile.NonBlockingWrite(e.fds[0], hdr.View()) + return rawfile.NonBlockingWrite3(e.fds[0], pkt.Header.View(), pkt.Data.ToView(), nil) +} + +// WritePackets writes outbound packets to the file descriptor. If it is not +// currently writable, the packet is dropped. +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, 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++ } - return rawfile.NonBlockingWrite3(e.fds[0], hdr.View(), payload.ToView(), nil) + n := len(pkts) + + views := pkts[0].Data.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 pkts { + // TODO(b/134618279): Different packets may have different data + // in the future. We should handle this. + if !viewsEqual(pkts[i].Data.Views(), views) { + panic("All packets in pkts should have the same Data.") + } + + prevIovecIdx := iovecIdx + mmsgHdr := &mmsgHdrs[i] + mmsgHdr.Msg.Iov = &iovec[iovecIdx] + packetSize := pkts[i].DataSize + hdr := &pkts[i].Header + + off = pkts[i].DataOffset + 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 +} + +// viewsEqual tests whether v1 and v2 refer to the same backing bytes. +func viewsEqual(vs1, vs2 []buffer.View) bool { + return len(vs1) == len(vs2) && (len(vs1) == 0 || &vs1[0] == &vs2[0]) +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { + return rawfile.NonBlockingWrite(e.fds[0], vv.ToView()) } -// WriteRawPacket writes a raw packet directly to the file descriptor. -func (e *endpoint) WriteRawPacket(dest tcpip.Address, packet []byte) *tcpip.Error { +// InjectOutobund implements stack.InjectableEndpoint.InjectOutbound. +func (e *endpoint) InjectOutbound(dest tcpip.Address, packet []byte) *tcpip.Error { return rawfile.NonBlockingWrite(e.fds[0], packet) } @@ -468,9 +609,9 @@ 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..2066987eb 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) { @@ -168,7 +168,10 @@ func testWritePacket(t *testing.T, plen int, eth bool, gsoMaxSize uint32) { L3HdrLen: header.IPv4MaximumHeaderSize, } } - if err := c.ep.WritePacket(r, gso, hdr, payload.ToVectorisedView(), proto); err != nil { + if err := c.ep.WritePacket(r, gso, proto, tcpip.PacketBuffer{ + Header: hdr, + Data: payload.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed: %v", err) } @@ -258,7 +261,10 @@ func TestPreserveSrcAddress(t *testing.T) { // WritePacket panics given a prependable with anything less than // the minimum size of the ethernet header. hdr := buffer.NewPrependable(header.EthernetMinimumSize) - if err := c.ep.WritePacket(r, nil /* gso */, hdr, buffer.VectorisedView{}, proto); err != nil { + if err := c.ep.WritePacket(r, nil /* gso */, proto, tcpip.PacketBuffer{ + Header: hdr, + Data: buffer.VectorisedView{}, + }); err != nil { t.Fatalf("WritePacket failed: %v", err) } @@ -293,11 +299,12 @@ 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 +322,21 @@ 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..f35fcdff4 100644 --- a/pkg/tcpip/link/loopback/BUILD +++ b/pkg/tcpip/link/loopback/BUILD @@ -6,10 +6,11 @@ go_library( name = "loopback", srcs = ["loopback.go"], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/loopback", - visibility = ["//:sandbox"], + visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/header", "//pkg/tcpip/stack", ], ) diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go index b36629d2c..499cc608f 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) +func (e *endpoint) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { + views := make([]buffer.View, 1, 1+len(pkt.Data.Views())) + views[0] = pkt.Header.View() + views = append(views, pkt.Data.Views()...) - // Because we're immediately turning around and writing the packet back to the - // rx path, we intentionally don't preserve the remote and local link - // addresses from the stack.Route we're passed. - e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, vv) + // Because we're immediately turning around and writing the packet back + // to the rx path, we intentionally don't preserve the remote and local + // link addresses from the stack.Route we're passed. + e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, tcpip.PacketBuffer{ + Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.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, []tcpip.PacketBuffer, tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + panic("not implemented") +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { + // Reject the packet if it's shorter than an ethernet header. + if vv.Size() < header.EthernetMinimumSize { + return tcpip.ErrBadAddress + } + + // There should be an ethernet header at the beginning of vv. + linkHeader := header.Ethernet(vv.First()[:header.EthernetMinimumSize]) + vv.TrimFront(len(linkHeader)) + e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, linkHeader.Type(), tcpip.PacketBuffer{ + Data: vv, + LinkHeader: buffer.View(linkHeader), + }) + + return nil +} diff --git a/pkg/tcpip/link/muxed/BUILD b/pkg/tcpip/link/muxed/BUILD index 1bab380b0..1ac7948b6 100644 --- a/pkg/tcpip/link/muxed/BUILD +++ b/pkg/tcpip/link/muxed/BUILD @@ -7,9 +7,7 @@ go_library( name = "muxed", srcs = ["injectable.go"], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/muxed", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/link/muxed/injectable.go b/pkg/tcpip/link/muxed/injectable.go index 7c946101d..445b22c17 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, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + endpoint, ok := m.routes[r.RemoteAddress] + if !ok { + return 0, tcpip.ErrNoRoute + } + return endpoint.WritePackets(r, gso, pkts, protocol) } // WritePacket writes outbound packets to the appropriate LinkInjectableEndpoint // based on the RemoteAddress. HandleLocal only works if r.RemoteAddress has a // route registered in this endpoint. -func (m *InjectableEndpoint) WritePacket(r *stack.Route, _ *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (m *InjectableEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { if endpoint, ok := m.routes[r.RemoteAddress]; ok { - return endpoint.WritePacket(r, nil /* gso */, hdr, payload, protocol) + return endpoint.WritePacket(r, gso, protocol, pkt) } return tcpip.ErrNoRoute } -// WriteRawPacket writes outbound packets to the appropriate +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (m *InjectableEndpoint) WriteRawPacket(buffer.VectorisedView) *tcpip.Error { + // WriteRawPacket doesn't get a route or network address, so there's + // nowhere to write this. + return tcpip.ErrNoRoute +} + +// 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..63b249837 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) @@ -50,8 +50,10 @@ func TestInjectableEndpointDispatch(t *testing.T) { hdr.Prepend(1)[0] = 0xFA packetRoute := stack.Route{RemoteAddress: dstIP} - endpoint.WritePacket(&packetRoute, nil /* gso */, hdr, - buffer.NewViewFromBytes([]byte{0xFB}).ToVectorisedView(), ipv4.ProtocolNumber) + endpoint.WritePacket(&packetRoute, nil /* gso */, ipv4.ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buffer.NewViewFromBytes([]byte{0xFB}).ToVectorisedView(), + }) buf := make([]byte, 6500) bytesRead, err := sock.Read(buf) @@ -68,8 +70,10 @@ func TestInjectableEndpointDispatchHdrOnly(t *testing.T) { hdr := buffer.NewPrependable(1) hdr.Prepend(1)[0] = 0xFA packetRoute := stack.Route{RemoteAddress: dstIP} - endpoint.WritePacket(&packetRoute, nil /* gso */, hdr, - buffer.NewView(0).ToVectorisedView(), ipv4.ProtocolNumber) + endpoint.WritePacket(&packetRoute, nil /* gso */, ipv4.ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buffer.NewView(0).ToVectorisedView(), + }) buf := make([]byte, 6500) bytesRead, err := sock.Read(buf) if err != nil { diff --git a/pkg/tcpip/link/rawfile/BUILD b/pkg/tcpip/link/rawfile/BUILD index 2e8bc772a..d8211e93d 100644 --- a/pkg/tcpip/link/rawfile/BUILD +++ b/pkg/tcpip/link/rawfile/BUILD @@ -13,8 +13,9 @@ go_library( "rawfile_unsafe.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/rawfile", - visibility = [ - "//visibility:public", + visibility = ["//visibility:public"], + deps = [ + "//pkg/tcpip", + "@org_golang_x_sys//unix:go_default_library", ], - deps = ["//pkg/tcpip"], ) 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/BUILD b/pkg/tcpip/link/sharedmem/BUILD index 0a5ea3dc4..a4f9cdd69 100644 --- a/pkg/tcpip/link/sharedmem/BUILD +++ b/pkg/tcpip/link/sharedmem/BUILD @@ -12,9 +12,7 @@ go_library( "tx.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/sharedmem", - visibility = [ - "//:sandbox", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/log", "//pkg/tcpip", diff --git a/pkg/tcpip/link/sharedmem/pipe/BUILD b/pkg/tcpip/link/sharedmem/pipe/BUILD index 330ed5e94..6b5bc542c 100644 --- a/pkg/tcpip/link/sharedmem/pipe/BUILD +++ b/pkg/tcpip/link/sharedmem/pipe/BUILD @@ -12,7 +12,7 @@ go_library( "tx.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/sharedmem/pipe", - visibility = ["//:sandbox"], + visibility = ["//visibility:public"], ) go_test( diff --git a/pkg/tcpip/link/sharedmem/queue/BUILD b/pkg/tcpip/link/sharedmem/queue/BUILD index de1ce043d..8c9234d54 100644 --- a/pkg/tcpip/link/sharedmem/queue/BUILD +++ b/pkg/tcpip/link/sharedmem/queue/BUILD @@ -10,7 +10,7 @@ go_library( "tx.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/sharedmem/queue", - visibility = ["//:sandbox"], + visibility = ["//visibility:public"], deps = [ "//pkg/log", "//pkg/tcpip/link/sharedmem/pipe", diff --git a/pkg/tcpip/link/sharedmem/sharedmem.go b/pkg/tcpip/link/sharedmem/sharedmem.go index 9e71d4edf..080f9d667 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem.go +++ b/pkg/tcpip/link/sharedmem/sharedmem.go @@ -185,9 +185,10 @@ func (e *endpoint) LinkAddress() tcpip.LinkAddress { // WritePacket writes outbound packets to the file descriptor. If it is not // currently writable, the packet is dropped. -func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { // Add the ethernet header here. - eth := header.Ethernet(hdr.Prepend(header.EthernetMinimumSize)) + eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) + pkt.LinkHeader = buffer.View(eth) ethHdr := &header.EthernetFields{ DstAddr: r.RemoteLinkAddress, Type: protocol, @@ -199,10 +200,30 @@ func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, hdr buffer.Prependa } eth.Encode(ethHdr) - v := payload.ToView() + v := pkt.Data.ToView() // Transmit the packet. e.mu.Lock() - ok := e.tx.transmit(hdr.View(), v) + ok := e.tx.transmit(pkt.Header.View(), v) + e.mu.Unlock() + + if !ok { + return tcpip.ErrWouldBlock + } + + return nil +} + +// WritePackets implements stack.LinkEndpoint.WritePackets. +func (e *endpoint) WritePackets(r *stack.Route, _ *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + panic("not implemented") +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { + v := vv.ToView() + // Transmit the packet. + e.mu.Lock() + ok := e.tx.transmit(v, buffer.View{}) e.mu.Unlock() if !ok { @@ -253,8 +274,11 @@ 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..89603c48f 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() @@ -272,7 +273,10 @@ func TestSimpleSend(t *testing.T) { randomFill(buf) proto := tcpip.NetworkProtocolNumber(rand.Intn(0x10000)) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), proto); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, proto, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed: %v", err) } @@ -341,7 +345,9 @@ func TestPreserveSrcAddressInSend(t *testing.T) { hdr := buffer.NewPrependable(header.EthernetMinimumSize) proto := tcpip.NetworkProtocolNumber(rand.Intn(0x10000)) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buffer.VectorisedView{}, proto); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, proto, tcpip.PacketBuffer{ + Header: hdr, + }); err != nil { t.Fatalf("WritePacket failed: %v", err) } @@ -395,7 +401,10 @@ func TestFillTxQueue(t *testing.T) { for i := queuePipeSize / 40; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } @@ -410,7 +419,10 @@ func TestFillTxQueue(t *testing.T) { // Next attempt to write must fail. hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != want { + if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != want { t.Fatalf("WritePacket return unexpected result: got %v, want %v", err, want) } } @@ -435,7 +447,10 @@ func TestFillTxQueueAfterBadCompletion(t *testing.T) { // Send two packets so that the id slice has at least two slots. for i := 2; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } } @@ -455,7 +470,10 @@ func TestFillTxQueueAfterBadCompletion(t *testing.T) { ids := make(map[uint64]struct{}) for i := queuePipeSize / 40; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } @@ -470,7 +488,10 @@ func TestFillTxQueueAfterBadCompletion(t *testing.T) { // Next attempt to write must fail. hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != want { + if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != want { t.Fatalf("WritePacket return unexpected result: got %v, want %v", err, want) } } @@ -493,7 +514,10 @@ func TestFillTxMemory(t *testing.T) { ids := make(map[uint64]struct{}) for i := queueDataSize / bufferSize; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } @@ -509,7 +533,10 @@ func TestFillTxMemory(t *testing.T) { // Next attempt to write must fail. hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber) + err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }) if want := tcpip.ErrWouldBlock; err != want { t.Fatalf("WritePacket return unexpected result: got %v, want %v", err, want) } @@ -534,7 +561,10 @@ func TestFillTxMemoryWithMultiBuffer(t *testing.T) { // until there is only one buffer left. for i := queueDataSize/bufferSize - 1; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } @@ -547,7 +577,10 @@ func TestFillTxMemoryWithMultiBuffer(t *testing.T) { { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) uu := buffer.NewView(bufferSize).ToVectorisedView() - if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, hdr, uu, header.IPv4ProtocolNumber); err != want { + if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: uu, + }); err != want { t.Fatalf("WritePacket return unexpected result: got %v, want %v", err, want) } } @@ -555,7 +588,10 @@ func TestFillTxMemoryWithMultiBuffer(t *testing.T) { // Attempt to write the one-buffer packet again. It must succeed. { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } } diff --git a/pkg/tcpip/link/sniffer/BUILD b/pkg/tcpip/link/sniffer/BUILD index 1756114e6..d6ae0368a 100644 --- a/pkg/tcpip/link/sniffer/BUILD +++ b/pkg/tcpip/link/sniffer/BUILD @@ -9,9 +9,7 @@ go_library( "sniffer.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/sniffer", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/log", "//pkg/tcpip", diff --git a/pkg/tcpip/link/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go index e401dce44..3392b7edd 100644 --- a/pkg/tcpip/link/sniffer/sniffer.go +++ b/pkg/tcpip/link/sniffer/sniffer.go @@ -49,6 +49,13 @@ var LogPackets uint32 = 1 // LogPacketsToFile must be accessed atomically. var LogPacketsToFile uint32 = 1 +var transportProtocolMinSizes map[tcpip.TransportProtocolNumber]int = map[tcpip.TransportProtocolNumber]int{ + header.ICMPv4ProtocolNumber: header.IPv4MinimumSize, + header.ICMPv6ProtocolNumber: header.IPv6MinimumSize, + header.UDPProtocolNumber: header.UDPMinimumSize, + header.TCPProtocolNumber: header.TCPMinimumSize, +} + type endpoint struct { dispatcher stack.NetworkDispatcher lower stack.LinkEndpoint @@ -116,19 +123,19 @@ 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 +154,7 @@ 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 @@ -193,22 +200,19 @@ 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, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { if atomic.LoadUint32(&LogPackets) == 1 && e.file == nil { - logPacket("send", protocol, hdr.View(), gso) + logPacket("send", protocol, pkt.Header.View(), gso) } if e.file != nil && atomic.LoadUint32(&LogPacketsToFile) == 1 { - hdrBuf := hdr.View() - length := len(hdrBuf) + payload.Size() + hdrBuf := pkt.Header.View() + length := len(hdrBuf) + pkt.Data.Size() if length > int(e.maxPCAPLen) { length = int(e.maxPCAPLen) } buf := bytes.NewBuffer(make([]byte, 0, pcapPacketHeaderLen+length)) - if err := binary.Write(buf, binary.BigEndian, newPCAPPacketHeader(uint32(length), uint32(len(hdrBuf)+payload.Size()))); err != nil { + if err := binary.Write(buf, binary.BigEndian, newPCAPPacketHeader(uint32(length), uint32(len(hdrBuf)+pkt.Data.Size()))); err != nil { panic(err) } if len(hdrBuf) > length { @@ -218,26 +222,75 @@ 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(pkt.Data, 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, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { + e.dumpPacket(gso, protocol, pkt) + return e.lower.WritePacket(r, gso, protocol, pkt) +} + +// WritePackets implements the stack.LinkEndpoint interface. It is called by +// higher-level protocols to write packets; it just logs the packet and +// forwards the request to the lower endpoint. +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + view := pkts[0].Data.ToView() + for _, pkt := range pkts { + e.dumpPacket(gso, protocol, tcpip.PacketBuffer{ + Header: pkt.Header, + Data: view[pkt.DataOffset:][:pkt.DataSize].ToVectorisedView(), + }) + } + return e.lower.WritePackets(r, gso, pkts, protocol) +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { + if atomic.LoadUint32(&LogPackets) == 1 && e.file == nil { + logPacket("send", 0, buffer.View("[raw packet, no header available]"), nil /* gso */) + } + if e.file != nil && atomic.LoadUint32(&LogPacketsToFile) == 1 { + length := vv.Size() + if length > int(e.maxPCAPLen) { + length = int(e.maxPCAPLen) + } + + buf := bytes.NewBuffer(make([]byte, 0, pcapPacketHeaderLen+length)) + if err := binary.Write(buf, binary.BigEndian, newPCAPPacketHeader(uint32(length), uint32(vv.Size()))); err != nil { + panic(err) + } + logVectorisedView(vv, length, buf) if _, err := e.file.Write(buf.Bytes()); err != nil { panic(err) } } - return e.lower.WritePacket(r, gso, hdr, payload, protocol) + return e.lower.WriteRawPacket(vv) +} + +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. @@ -287,6 +340,13 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie return } + // We aren't guaranteed to have a transport header - it's possible for + // writes via raw endpoints to contain only network headers. + if minSize, ok := transportProtocolMinSizes[tcpip.TransportProtocolNumber(transProto)]; ok && len(b) < minSize { + log.Infof("%s %v -> %v transport protocol: %d, but no transport header found (possible raw packet)", prefix, src, dst, transProto) + return + } + // Figure out the transport layer info. transName := "unknown" srcPort := uint16(0) diff --git a/pkg/tcpip/link/tun/BUILD b/pkg/tcpip/link/tun/BUILD index 92dce8fac..a71a493fc 100644 --- a/pkg/tcpip/link/tun/BUILD +++ b/pkg/tcpip/link/tun/BUILD @@ -6,7 +6,5 @@ go_library( name = "tun", srcs = ["tun_unsafe.go"], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/tun", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], ) diff --git a/pkg/tcpip/link/waitable/BUILD b/pkg/tcpip/link/waitable/BUILD index 0746dc8ec..134837943 100644 --- a/pkg/tcpip/link/waitable/BUILD +++ b/pkg/tcpip/link/waitable/BUILD @@ -9,9 +9,7 @@ go_library( "waitable.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/waitable", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/gate", "//pkg/tcpip", diff --git a/pkg/tcpip/link/waitable/waitable.go b/pkg/tcpip/link/waitable/waitable.go index 5a1791cb5..a8de38979 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() } @@ -99,12 +99,36 @@ func (e *Endpoint) LinkAddress() tcpip.LinkAddress { // WritePacket implements stack.LinkEndpoint.WritePacket. It is called by // higher-level protocols to write packets. It only forwards packets to the // lower endpoint if Wait or WaitWrite haven't been called. -func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { if !e.writeGate.Enter() { return nil } - err := e.lower.WritePacket(r, gso, hdr, payload, protocol) + err := e.lower.WritePacket(r, gso, protocol, pkt) + e.writeGate.Leave() + return err +} + +// 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, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + if !e.writeGate.Enter() { + return len(pkts), nil + } + + n, err := e.lower.WritePackets(r, gso, pkts, protocol) + e.writeGate.Leave() + return n, err +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *Endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { + if !e.writeGate.Enter() { + return nil + } + + err := e.lower.WriteRawPacket(vv) e.writeGate.Leave() return err } diff --git a/pkg/tcpip/link/waitable/waitable_test.go b/pkg/tcpip/link/waitable/waitable_test.go index ae23c96b7..31b11a27a 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++ } @@ -65,7 +65,18 @@ func (e *countedEndpoint) LinkAddress() tcpip.LinkAddress { return e.linkAddr } -func (e *countedEndpoint) WritePacket(r *stack.Route, _ *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (e *countedEndpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { + e.writeCount++ + return nil +} + +// WritePackets implements stack.LinkEndpoint.WritePackets. +func (e *countedEndpoint) WritePackets(r *stack.Route, _ *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + e.writeCount += len(pkts) + return len(pkts), nil +} + +func (e *countedEndpoint) WriteRawPacket(buffer.VectorisedView) *tcpip.Error { e.writeCount++ return nil } @@ -78,21 +89,21 @@ func TestWaitWrite(t *testing.T) { wep := New(ep) // Write and check that it goes through. - wep.WritePacket(nil, nil /* gso */, buffer.Prependable{}, buffer.VectorisedView{}, 0) + wep.WritePacket(nil, nil /* gso */, 0, tcpip.PacketBuffer{}) if want := 1; ep.writeCount != want { t.Fatalf("Unexpected writeCount: got=%v, want=%v", ep.writeCount, want) } // Wait on dispatches, then try to write. It must go through. wep.WaitDispatch() - wep.WritePacket(nil, nil /* gso */, buffer.Prependable{}, buffer.VectorisedView{}, 0) + wep.WritePacket(nil, nil /* gso */, 0, tcpip.PacketBuffer{}) if want := 2; ep.writeCount != want { t.Fatalf("Unexpected writeCount: got=%v, want=%v", ep.writeCount, want) } // Wait on writes, then try to write. It must not go through. wep.WaitWrite() - wep.WritePacket(nil, nil /* gso */, buffer.Prependable{}, buffer.VectorisedView{}, 0) + wep.WritePacket(nil, nil /* gso */, 0, tcpip.PacketBuffer{}) if want := 2; ep.writeCount != want { t.Fatalf("Unexpected writeCount: got=%v, want=%v", ep.writeCount, want) } @@ -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/BUILD b/pkg/tcpip/network/arp/BUILD index df0d3a8c0..e7617229b 100644 --- a/pkg/tcpip/network/arp/BUILD +++ b/pkg/tcpip/network/arp/BUILD @@ -7,9 +7,7 @@ go_library( name = "arp", srcs = ["arp.go"], importpath = "gvisor.dev/gvisor/pkg/tcpip/network/arp", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index 6b1e854dc..da8482509 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, stack.NetworkHeaderParams, stack.PacketLooping) *tcpip.Error { +func (e *endpoint) WritePacket(*stack.Route, *stack.GSO, stack.NetworkHeaderParams, stack.PacketLooping, tcpip.PacketBuffer) *tcpip.Error { return tcpip.ErrNotSupported } -func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error { +// WritePackets implements stack.NetworkEndpoint.WritePackets. +func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, []tcpip.PacketBuffer, stack.NetworkHeaderParams, stack.PacketLooping) (int, *tcpip.Error) { + return 0, tcpip.ErrNotSupported +} + +func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, loop stack.PacketLooping, pkt tcpip.PacketBuffer) *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 @@ -97,23 +102,25 @@ 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) - pkt := header.ARP(hdr.Prepend(header.ARPSize)) - pkt.SetIPv4OverEthernet() - pkt.SetOp(header.ARPReply) - copy(pkt.HardwareAddressSender(), r.LocalLinkAddress[:]) - copy(pkt.ProtocolAddressSender(), h.ProtocolAddressTarget()) - copy(pkt.HardwareAddressTarget(), h.HardwareAddressSender()) - copy(pkt.ProtocolAddressTarget(), h.ProtocolAddressSender()) - e.linkEP.WritePacket(r, nil /* gso */, hdr, buffer.VectorisedView{}, ProtocolNumber) + packet := header.ARP(hdr.Prepend(header.ARPSize)) + packet.SetIPv4OverEthernet() + packet.SetOp(header.ARPReply) + copy(packet.HardwareAddressSender(), r.LocalLinkAddress[:]) + copy(packet.ProtocolAddressSender(), h.ProtocolAddressTarget()) + copy(packet.HardwareAddressTarget(), h.HardwareAddressSender()) + copy(packet.ProtocolAddressTarget(), h.ProtocolAddressSender()) + e.linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + }) fallthrough // also fill the cache from requests case header.ARPReply: addr := tcpip.Address(h.ProtocolAddressSender()) linkAddr := tcpip.LinkAddress(h.HardwareAddressSender()) - e.linkAddrCache.AddLinkAddress(e.nicid, addr, linkAddr) + e.linkAddrCache.AddLinkAddress(e.nicID, addr, linkAddr) } } @@ -130,12 +137,12 @@ 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 @@ -160,7 +167,9 @@ func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, linkEP stack. copy(h.ProtocolAddressSender(), localAddr) copy(h.ProtocolAddressTarget(), addr) - return linkEP.WritePacket(r, nil /* gso */, hdr, buffer.VectorisedView{}, ProtocolNumber) + return linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + }) } // ResolveStaticAddress implements stack.LinkAddressResolver. diff --git a/pkg/tcpip/network/arp/arp_test.go b/pkg/tcpip/network/arp/arp_test.go index 88b57ec03..8e6048a21 100644 --- a/pkg/tcpip/network/arp/arp_test.go +++ b/pkg/tcpip/network/arp/arp_test.go @@ -102,19 +102,21 @@ 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} { t.Run(strconv.Itoa(i), func(t *testing.T) { inject(address) - pkt := <-c.linkEP.C - if pkt.Proto != arp.ProtocolNumber { - t.Fatalf("expected ARP response, got network protocol number %d", pkt.Proto) + pi := <-c.linkEP.C + if pi.Proto != arp.ProtocolNumber { + t.Fatalf("expected ARP response, got network protocol number %d", pi.Proto) } - rep := header.ARP(pkt.Header) + rep := header.ARP(pi.Pkt.Header.View()) if !rep.IsValid() { - t.Fatalf("invalid ARP response len(pkt.Header)=%d", len(pkt.Header)) + t.Fatalf("invalid ARP response pi.Pkt.Header.UsedLength()=%d", pi.Pkt.Header.UsedLength()) } if got, want := tcpip.LinkAddress(rep.HardwareAddressSender()), stackLinkAddr; got != want { t.Errorf("got HardwareAddressSender = %s, want = %s", got, want) diff --git a/pkg/tcpip/network/fragmentation/BUILD b/pkg/tcpip/network/fragmentation/BUILD index 2cad0a0b6..acf1e022c 100644 --- a/pkg/tcpip/network/fragmentation/BUILD +++ b/pkg/tcpip/network/fragmentation/BUILD @@ -25,7 +25,7 @@ go_library( "reassembler_list.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/network/fragmentation", - visibility = ["//:sandbox"], + visibility = ["//visibility:public"], deps = [ "//pkg/log", "//pkg/tcpip", @@ -44,11 +44,3 @@ go_test( embed = [":fragmentation"], deps = ["//pkg/tcpip/buffer"], ) - -filegroup( - name = "autogen", - srcs = [ - "reassembler_list.go", - ], - visibility = ["//:sandbox"], -) diff --git a/pkg/tcpip/network/ip_test.go b/pkg/tcpip/network/ip_test.go index f644a8b08..4144a7837 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) } @@ -150,27 +150,36 @@ func (*testObject) Wait() {} // WritePacket is called by network endpoints after producing a packet and // writing it to the link endpoint. This is used by the test object to verify // that the produced packet is as expected. -func (t *testObject) WritePacket(_ *stack.Route, _ *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (t *testObject) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { var prot tcpip.TransportProtocolNumber var srcAddr tcpip.Address var dstAddr tcpip.Address if t.v4 { - h := header.IPv4(hdr.View()) + h := header.IPv4(pkt.Header.View()) prot = tcpip.TransportProtocolNumber(h.Protocol()) srcAddr = h.SourceAddress() dstAddr = h.DestinationAddress() } else { - h := header.IPv6(hdr.View()) + h := header.IPv6(pkt.Header.View()) prot = tcpip.TransportProtocolNumber(h.NextHeader()) srcAddr = h.SourceAddress() dstAddr = h.DestinationAddress() } - t.checkValues(prot, payload, srcAddr, dstAddr) + t.checkValues(prot, pkt.Data, srcAddr, dstAddr) return nil } +// WritePackets implements stack.LinkEndpoint.WritePackets. +func (t *testObject) WritePackets(_ *stack.Route, _ *stack.GSO, pkt []tcpip.PacketBuffer, 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 +239,10 @@ func TestIPv4Send(t *testing.T) { if err != nil { t.Fatalf("could not find route: %v", err) } - if err := ep.WritePacket(&r, nil /* gso */, hdr, payload.ToVectorisedView(), stack.NetworkHeaderParams{Protocol: 123, TTL: 123, TOS: stack.DefaultTOS}, stack.PacketOut); err != nil { + if err := ep.WritePacket(&r, nil /* gso */, stack.NetworkHeaderParams{Protocol: 123, TTL: 123, TOS: stack.DefaultTOS}, stack.PacketOut, tcpip.PacketBuffer{ + Header: hdr, + Data: payload.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed: %v", err) } } @@ -270,7 +282,9 @@ 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 +372,9 @@ 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 +437,17 @@ 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 +480,10 @@ func TestIPv6Send(t *testing.T) { if err != nil { t.Fatalf("could not find route: %v", err) } - if err := ep.WritePacket(&r, nil /* gso */, hdr, payload.ToVectorisedView(), stack.NetworkHeaderParams{Protocol: 123, TTL: 123, TOS: stack.DefaultTOS}, stack.PacketOut); err != nil { + if err := ep.WritePacket(&r, nil /* gso */, stack.NetworkHeaderParams{Protocol: 123, TTL: 123, TOS: stack.DefaultTOS}, stack.PacketOut, tcpip.PacketBuffer{ + Header: hdr, + Data: payload.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed: %v", err) } } @@ -500,7 +523,9 @@ 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 +535,7 @@ 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 +587,7 @@ 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 +634,12 @@ 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/BUILD b/pkg/tcpip/network/ipv4/BUILD index 58e537aad..aeddfcdd4 100644 --- a/pkg/tcpip/network/ipv4/BUILD +++ b/pkg/tcpip/network/ipv4/BUILD @@ -10,9 +10,7 @@ go_library( "ipv4.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/network/ipv4", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go index 50b363dc4..32bf39e43 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 +99,11 @@ 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, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}); err != nil { + if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + Data: vv, + TransportHeader: buffer.View(pkt), + }); err != nil { sent.Dropped.Increment() return } @@ -104,19 +112,19 @@ 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 5cd895ff0..e645cf62c 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -47,7 +47,7 @@ const ( ) type endpoint struct { - nicid tcpip.NICID + nicID tcpip.NICID id stack.NetworkEndpointID prefixLen int linkEP stack.LinkEndpoint @@ -57,9 +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, @@ -89,7 +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. @@ -117,13 +117,14 @@ func (e *endpoint) GSOMaxSize() uint32 { } // writePacketFragments calls e.linkEP.WritePacket with each packet fragment to -// write. It assumes that the IP header is entirely in hdr but does not assume -// that only the IP header is in hdr. It assumes that the input packet's stated -// length matches the length of the hdr+payload. mtu includes the IP header and -// options. This does not support the DontFragment IP flag. -func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, mtu int) *tcpip.Error { +// write. It assumes that the IP header is entirely in pkt.Header but does not +// assume that only the IP header is in pkt.Header. It assumes that the input +// packet's stated length matches the length of the header+payload. mtu +// includes the IP header and options. This does not support the DontFragment +// IP flag. +func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, mtu int, pkt tcpip.PacketBuffer) *tcpip.Error { // This packet is too big, it needs to be fragmented. - ip := header.IPv4(hdr.View()) + ip := header.IPv4(pkt.Header.View()) flags := ip.Flags() // Update mtu to take into account the header, which will exist in all @@ -137,71 +138,85 @@ func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, hdr buff outerMTU := innerMTU + int(ip.HeaderLength()) offset := ip.FragmentOffset() - originalAvailableLength := hdr.AvailableLength() + originalAvailableLength := pkt.Header.AvailableLength() for i := 0; i < n; i++ { // Where possible, the first fragment that is sent has the same - // hdr.UsedLength() as the input packet. The link-layer endpoint may depends - // on this for looking at, eg, L4 headers. + // pkt.Header.UsedLength() as the input packet. The link-layer + // endpoint may depend on this for looking at, eg, L4 headers. h := ip if i > 0 { - hdr = buffer.NewPrependable(int(ip.HeaderLength()) + originalAvailableLength) - h = header.IPv4(hdr.Prepend(int(ip.HeaderLength()))) + pkt.Header = buffer.NewPrependable(int(ip.HeaderLength()) + originalAvailableLength) + h = header.IPv4(pkt.Header.Prepend(int(ip.HeaderLength()))) copy(h, ip[:ip.HeaderLength()]) } if i != n-1 { h.SetTotalLength(uint16(outerMTU)) h.SetFlagsFragmentOffset(flags|header.IPv4FlagMoreFragments, offset) } else { - h.SetTotalLength(uint16(h.HeaderLength()) + uint16(payload.Size())) + h.SetTotalLength(uint16(h.HeaderLength()) + uint16(pkt.Data.Size())) h.SetFlagsFragmentOffset(flags, offset) } h.SetChecksum(0) h.SetChecksum(^h.CalculateChecksum()) offset += uint16(innerMTU) if i > 0 { - newPayload := payload.Clone([]buffer.View{}) + newPayload := pkt.Data.Clone(nil) newPayload.CapLength(innerMTU) - if err := e.linkEP.WritePacket(r, gso, hdr, newPayload, ProtocolNumber); err != nil { + if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, tcpip.PacketBuffer{ + Header: pkt.Header, + Data: newPayload, + NetworkHeader: buffer.View(h), + }); err != nil { return err } r.Stats().IP.PacketsSent.Increment() - payload.TrimFront(newPayload.Size()) + pkt.Data.TrimFront(newPayload.Size()) continue } - // Special handling for the first fragment because it comes from the hdr. - if outerMTU >= hdr.UsedLength() { - // This fragment can fit all of hdr and possibly some of payload, too. - newPayload := payload.Clone([]buffer.View{}) - newPayloadLength := outerMTU - hdr.UsedLength() + // Special handling for the first fragment because it comes + // from the header. + if outerMTU >= pkt.Header.UsedLength() { + // This fragment can fit all of pkt.Header and possibly + // some of pkt.Data, too. + newPayload := pkt.Data.Clone(nil) + newPayloadLength := outerMTU - pkt.Header.UsedLength() newPayload.CapLength(newPayloadLength) - if err := e.linkEP.WritePacket(r, gso, hdr, newPayload, ProtocolNumber); err != nil { + if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, tcpip.PacketBuffer{ + Header: pkt.Header, + Data: newPayload, + NetworkHeader: buffer.View(h), + }); err != nil { return err } r.Stats().IP.PacketsSent.Increment() - payload.TrimFront(newPayloadLength) + pkt.Data.TrimFront(newPayloadLength) } else { - // The fragment is too small to fit all of hdr. - startOfHdr := hdr - startOfHdr.TrimBack(hdr.UsedLength() - outerMTU) + // The fragment is too small to fit all of pkt.Header. + startOfHdr := pkt.Header + startOfHdr.TrimBack(pkt.Header.UsedLength() - outerMTU) emptyVV := buffer.NewVectorisedView(0, []buffer.View{}) - if err := e.linkEP.WritePacket(r, gso, startOfHdr, emptyVV, ProtocolNumber); err != nil { + if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, tcpip.PacketBuffer{ + Header: startOfHdr, + Data: emptyVV, + NetworkHeader: buffer.View(h), + }); err != nil { return err } r.Stats().IP.PacketsSent.Increment() - // Add the unused bytes of hdr into the payload that remains to be sent. - restOfHdr := hdr.View()[outerMTU:] + // Add the unused bytes of pkt.Header into the pkt.Data + // that remains to be sent. + restOfHdr := pkt.Header.View()[outerMTU:] tmp := buffer.NewVectorisedView(len(restOfHdr), []buffer.View{buffer.NewViewFromBytes(restOfHdr)}) - tmp.Append(payload) - payload = tmp + tmp.Append(pkt.Data) + pkt.Data = tmp } } return nil } -// 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 { +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 @@ -219,41 +234,71 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen 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, params stack.NetworkHeaderParams, loop stack.PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error { + ip := e.addIPHeader(r, &pkt.Header, pkt.Data.Size(), params) + pkt.NetworkHeader = buffer.View(ip) if loop&stack.PacketLoop != 0 { - views := make([]buffer.View, 1, 1+len(payload.Views())) - views[0] = hdr.View() - views = append(views, payload.Views()...) - vv := buffer.NewVectorisedView(len(views[0])+payload.Size(), views) + // The inbound path expects the network header to still be in + // the PacketBuffer's Data field. + views := make([]buffer.View, 1, 1+len(pkt.Data.Views())) + views[0] = pkt.Header.View() + views = append(views, pkt.Data.Views()...) loopedR := r.MakeLoopedRoute() - e.HandlePacket(&loopedR, vv) + + e.HandlePacket(&loopedR, tcpip.PacketBuffer{ + Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), + }) + loopedR.Release() } if loop&stack.PacketOut == 0 { return nil } - if hdr.UsedLength()+payload.Size() > int(e.linkEP.MTU()) && (gso == nil || gso.Type == stack.GSONone) { - return e.writePacketFragments(r, gso, hdr, payload, int(e.linkEP.MTU())) + if pkt.Header.UsedLength()+pkt.Data.Size() > int(e.linkEP.MTU()) && (gso == nil || gso.Type == stack.GSONone) { + return e.writePacketFragments(r, gso, int(e.linkEP.MTU()), pkt) } - if err := e.linkEP.WritePacket(r, gso, hdr, payload, ProtocolNumber); err != nil { + if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, pkt); err != nil { return err } r.Stats().IP.PacketsSent.Increment() return nil } +// WritePackets implements stack.NetworkEndpoint.WritePackets. +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) { + if loop&stack.PacketLoop != 0 { + panic("multiple packets in local loop") + } + if loop&stack.PacketOut == 0 { + return len(pkts), nil + } + + for i := range pkts { + ip := e.addIPHeader(r, &pkts[i].Header, pkts[i].DataSize, params) + pkts[i].NetworkHeader = buffer.View(ip) + } + n, err := e.linkEP.WritePackets(r, gso, pkts, ProtocolNumber) + r.Stats().IP.PacketsSent.IncrementBy(uint64(n)) + return n, err +} + // WriteHeaderIncludedPacket writes a packet already containing a network // header through the given route. -func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error { +func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, loop stack.PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error { // The packet already has an IP header, but there are a few required // checks. - ip := header.IPv4(payload.First()) - if !ip.IsValid(payload.Size()) { + ip := header.IPv4(pkt.Data.First()) + if !ip.IsValid(pkt.Data.Size()) { return tcpip.ErrInvalidOptionValue } // Always set the total length. - ip.SetTotalLength(uint16(payload.Size())) + ip.SetTotalLength(uint16(pkt.Data.Size())) // Set the source address when zero. if ip.SourceAddress() == tcpip.Address(([]byte{0, 0, 0, 0})) { @@ -267,7 +312,7 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.Vect // Set the packet ID when zero. if ip.ID() == 0 { id := uint32(0) - if payload.Size() > header.IPv4MaximumHeaderSize+8 { + if pkt.Data.Size() > header.IPv4MaximumHeaderSize+8 { // Packets of 68 bytes or less are required by RFC 791 to not be // fragmented, so we only assign ids to larger packets. id = atomic.AddUint32(&e.protocol.ids[hashRoute(r, 0 /* protocol */, e.protocol.hashIV)%buckets], 1) @@ -280,35 +325,39 @@ 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, pkt.Clone()) } if loop&stack.PacketOut == 0 { return nil } - hdr := buffer.NewPrependableFromView(payload.ToView()) r.Stats().IP.PacketsSent.Increment() - return e.linkEP.WritePacket(r, nil /* gso */, hdr, buffer.VectorisedView{}, ProtocolNumber) + + ip = ip[:ip.HeaderLength()] + pkt.Header = buffer.NewPrependableFromView(buffer.View(ip)) + pkt.Data.TrimFront(int(ip.HeaderLength())) + return e.linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, pkt) } // HandlePacket is called by the link layer when new ipv4 packets arrive for // 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 vv.Size() == 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() @@ -316,10 +365,10 @@ func (e *endpoint) HandlePacket(r *stack.Route, vv buffer.VectorisedView) { 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 vv.size() causes a wrap - // around resulting in last being less than the offset. + // 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() @@ -327,7 +376,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, vv buffer.VectorisedView) { } var ready bool var err error - vv, ready, err = e.fragmentation.Process(hash.IPv4FragmentHash(h), h.FragmentOffset(), last, more, vv) + 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() @@ -340,11 +389,11 @@ 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. diff --git a/pkg/tcpip/network/ipv4/ipv4_test.go b/pkg/tcpip/network/ipv4/ipv4_test.go index 85ab0e3bc..e900f1b45 100644 --- a/pkg/tcpip/network/ipv4/ipv4_test.go +++ b/pkg/tcpip/network/ipv4/ipv4_test.go @@ -47,10 +47,6 @@ func TestExcludeBroadcast(t *testing.T) { t.Fatalf("CreateNIC 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, @@ -117,12 +113,12 @@ func makeHdrAndPayload(hdrLength int, extraLength int, viewSizes []int) (buffer. // comparePayloads compared the contents of all the packets against the contents // of the source packet. -func compareFragments(t *testing.T, packets []packetInfo, sourcePacketInfo packetInfo, mtu uint32) { +func compareFragments(t *testing.T, packets []tcpip.PacketBuffer, sourcePacketInfo tcpip.PacketBuffer, mtu uint32) { t.Helper() // Make a complete array of the sourcePacketInfo packet. source := header.IPv4(packets[0].Header.View()[:header.IPv4MinimumSize]) source = append(source, sourcePacketInfo.Header.View()...) - source = append(source, sourcePacketInfo.Payload.ToView()...) + source = append(source, sourcePacketInfo.Data.ToView()...) // Make a copy of the IP header, which will be modified in some fields to make // an expected header. @@ -136,7 +132,7 @@ func compareFragments(t *testing.T, packets []packetInfo, sourcePacketInfo packe for i, packet := range packets { // Confirm that the packet is valid. allBytes := packet.Header.View().ToVectorisedView() - allBytes.Append(packet.Payload) + allBytes.Append(packet.Data) ip := header.IPv4(allBytes.ToView()) if !ip.IsValid(len(ip)) { t.Errorf("IP packet is invalid:\n%s", hex.Dump(ip)) @@ -177,7 +173,7 @@ func compareFragments(t *testing.T, packets []packetInfo, sourcePacketInfo packe type errorChannel struct { *channel.Endpoint - Ch chan packetInfo + Ch chan tcpip.PacketBuffer packetCollectorErrors []*tcpip.Error } @@ -187,17 +183,11 @@ type errorChannel struct { func newErrorChannel(size int, mtu uint32, linkAddr tcpip.LinkAddress, packetCollectorErrors []*tcpip.Error) *errorChannel { return &errorChannel{ Endpoint: channel.New(size, mtu, linkAddr), - Ch: make(chan packetInfo, size), + Ch: make(chan tcpip.PacketBuffer, size), packetCollectorErrors: packetCollectorErrors, } } -// packetInfo holds all the information about an outbound packet. -type packetInfo struct { - Header buffer.Prependable - Payload buffer.VectorisedView -} - // Drain removes all outbound packets from the channel and counts them. func (e *errorChannel) Drain() int { c := 0 @@ -212,14 +202,9 @@ func (e *errorChannel) Drain() int { } // WritePacket stores outbound packets into the channel. -func (e *errorChannel) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { - p := packetInfo{ - Header: hdr, - Payload: payload, - } - +func (e *errorChannel) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { select { - case e.Ch <- p: + case e.Ch <- pkt: default: } @@ -296,18 +281,21 @@ func TestFragmentation(t *testing.T) { for _, ft := range fragTests { t.Run(ft.description, func(t *testing.T) { hdr, payload := makeHdrAndPayload(ft.hdrLength, ft.extraLength, ft.payloadViewsSizes) - source := packetInfo{ + source := tcpip.PacketBuffer{ Header: hdr, // Save the source payload because WritePacket will modify it. - Payload: payload.Clone([]buffer.View{}), + Data: payload.Clone(nil), } c := buildContext(t, nil, ft.mtu) - err := c.Route.WritePacket(ft.gso, hdr, payload, stack.NetworkHeaderParams{Protocol: tcp.ProtocolNumber, TTL: 42, TOS: stack.DefaultTOS}) + err := c.Route.WritePacket(ft.gso, stack.NetworkHeaderParams{Protocol: tcp.ProtocolNumber, TTL: 42, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + Data: payload, + }) if err != nil { t.Errorf("err got %v, want %v", err, nil) } - var results []packetInfo + var results []tcpip.PacketBuffer L: for { select { @@ -349,7 +337,10 @@ func TestFragmentationErrors(t *testing.T) { t.Run(ft.description, func(t *testing.T) { hdr, payload := makeHdrAndPayload(ft.hdrLength, header.IPv4MinimumSize, ft.payloadViewsSizes) c := buildContext(t, ft.packetCollectorErrors, ft.mtu) - err := c.Route.WritePacket(&stack.GSO{}, hdr, payload, stack.NetworkHeaderParams{Protocol: tcp.ProtocolNumber, TTL: 42, TOS: stack.DefaultTOS}) + err := c.Route.WritePacket(&stack.GSO{}, stack.NetworkHeaderParams{Protocol: tcp.ProtocolNumber, TTL: 42, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + Data: payload, + }) for i := 0; i < len(ft.packetCollectorErrors)-1; i++ { if got, want := ft.packetCollectorErrors[i], (*tcpip.Error)(nil); got != want { t.Errorf("ft.packetCollectorErrors[%d] got %v, want %v", i, got, want) @@ -455,7 +446,7 @@ func TestInvalidFragments(t *testing.T) { for _, tc := range testCases { t.Run(tc.name, func(t *testing.T) { - const nicid tcpip.NICID = 42 + const nicID tcpip.NICID = 42 s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocol{ ipv4.NewProtocol(), @@ -465,10 +456,12 @@ func TestInvalidFragments(t *testing.T) { 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)) + s.CreateNIC(nicID, sniffer.New(ep)) for _, pkt := range tc.packets { - ep.InjectLinkAddr(header.IPv4ProtocolNumber, remoteLinkAddr, buffer.NewVectorisedView(len(pkt), []buffer.View{pkt})) + 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 { diff --git a/pkg/tcpip/network/ipv6/BUILD b/pkg/tcpip/network/ipv6/BUILD index f06622a8b..e4e273460 100644 --- a/pkg/tcpip/network/ipv6/BUILD +++ b/pkg/tcpip/network/ipv6/BUILD @@ -10,9 +10,7 @@ go_library( "ipv6.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/network/ipv6", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index 7b638e9d0..1c3410618 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. @@ -61,19 +52,19 @@ 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 @@ -81,16 +72,34 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, vv buffer.V 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 iph.HopLimit() != ndpHopLimit { + if iph.HopLimit() != header.NDPHopLimit { + received.Invalid.Increment() + return + } + + if h.Code() != 0 { received.Invalid.Increment() return } @@ -104,9 +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() @@ -114,10 +123,10 @@ 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: @@ -171,7 +180,7 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, vv buffer.V // 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 { + 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 } @@ -180,9 +189,9 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, vv buffer.V header.NDPTargetLinkLayerAddressOption(r.LocalLinkAddress[:]), } hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv6NeighborAdvertMinimumSize + int(optsSerializer.Length())) - pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborAdvertSize)) - pkt.SetType(header.ICMPv6NeighborAdvert) - na := header.NDPNeighborAdvert(pkt.NDPPayload()) + 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) @@ -200,7 +209,7 @@ 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{})) // TODO(tamird/ghanan): there exists an explicit NDP option that is // used to update the neighbor table with link addresses for a @@ -209,7 +218,7 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, vv buffer.V // // Furthermore, the entirety of NDP handling here seems to be // contradicted by RFC 4861. - e.linkAddrCache.AddLinkAddress(e.nicid, r.RemoteAddress, r.RemoteLinkAddress) + e.linkAddrCache.AddLinkAddress(e.nicID, r.RemoteAddress, r.RemoteLinkAddress) // RFC 4861 Neighbor Discovery for IP version 6 (IPv6) // @@ -217,7 +226,9 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, vv buffer.V // // 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: ndpHopLimit, TOS: stack.DefaultTOS}); err != nil { + if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + }); err != nil { sent.Dropped.Increment() return } @@ -255,19 +266,19 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, vv buffer.V return } - // At this point we know that the targetAddress is not tentaive + // 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/140896005): Handle the scenario described above - // (inform the netstack integration that a duplicate address was - // was detected) + // 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) + 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: @@ -276,13 +287,16 @@ 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, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}); 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 */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + Data: pkt.Data, + }); err != nil { sent.Dropped.Increment() return } @@ -294,7 +308,7 @@ 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() @@ -306,8 +320,51 @@ 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() @@ -359,13 +416,15 @@ func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, linkEP stack. ip.Encode(&header.IPv6Fields{ PayloadLength: length, NextHeader: uint8(header.ICMPv6ProtocolNumber), - HopLimit: ndpHopLimit, + HopLimit: header.NDPHopLimit, SrcAddr: r.LocalAddress, DstAddr: r.RemoteAddress, }) // TODO(stijlist): count this in ICMP stats. - return linkEP.WritePacket(r, nil /* gso */, hdr, buffer.VectorisedView{}, ProtocolNumber) + return linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + }) } // ResolveStaticAddress implements stack.LinkAddressResolver. diff --git a/pkg/tcpip/network/ipv6/icmp_test.go b/pkg/tcpip/network/ipv6/icmp_test.go index dd3c4d7c4..335f634d5 100644 --- a/pkg/tcpip/network/ipv6/icmp_test.go +++ b/pkg/tcpip/network/ipv6/icmp_test.go @@ -30,7 +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") ) @@ -55,7 +55,7 @@ func (*stubLinkEndpoint) LinkAddress() tcpip.LinkAddress { return "" } -func (*stubLinkEndpoint) WritePacket(*stack.Route, *stack.GSO, buffer.Prependable, buffer.VectorisedView, tcpip.NetworkProtocolNumber) *tcpip.Error { +func (*stubLinkEndpoint) WritePacket(*stack.Route, *stack.GSO, tcpip.NetworkProtocolNumber, tcpip.PacketBuffer) *tcpip.Error { return nil } @@ -65,7 +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 { @@ -131,7 +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}, @@ -143,11 +143,13 @@ func TestICMPCounts(t *testing.T) { ip.Encode(&header.IPv6Fields{ PayloadLength: uint16(payloadLength), NextHeader: uint8(header.ICMPv6ProtocolNumber), - HopLimit: ndpHopLimit, + 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 { @@ -274,20 +276,22 @@ type routeArgs struct { func routeICMPv6Packet(t *testing.T, args routeArgs, fn func(*testing.T, header.ICMPv6)) { t.Helper() - pkt := <-args.src.C + pi := <-args.src.C { - views := []buffer.View{pkt.Header, pkt.Payload} - size := len(pkt.Header) + len(pkt.Payload) + views := []buffer.View{pi.Pkt.Header.View(), pi.Pkt.Data.ToView()} + size := pi.Pkt.Header.UsedLength() + pi.Pkt.Data.Size() vv := buffer.NewVectorisedView(size, views) - args.dst.InjectLinkAddr(pkt.Proto, args.dst.LinkAddress(), vv) + args.dst.InjectLinkAddr(pi.Proto, args.dst.LinkAddress(), tcpip.PacketBuffer{ + Data: vv, + }) } - if pkt.Proto != ProtocolNumber { - t.Errorf("unexpected protocol number %d", pkt.Proto) + if pi.Proto != ProtocolNumber { + t.Errorf("unexpected protocol number %d", pi.Proto) return } - ipv6 := header.IPv6(pkt.Header) + ipv6 := header.IPv6(pi.Pkt.Header.View()) transProto := tcpip.TransportProtocolNumber(ipv6.NextHeader()) if transProto != header.ICMPv6ProtocolNumber { t.Errorf("unexpected transport protocol number %d", transProto) @@ -359,3 +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 cd1e34085..e13f1fabf 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -43,7 +43,7 @@ const ( ) type endpoint struct { - nicid tcpip.NICID + nicID tcpip.NICID id stack.NetworkEndpointID prefixLen int linkEP stack.LinkEndpoint @@ -65,7 +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. @@ -97,9 +97,8 @@ 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, params stack.NetworkHeaderParams, 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, @@ -109,14 +108,26 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen 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, params stack.NetworkHeaderParams, loop stack.PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error { + ip := e.addIPHeader(r, &pkt.Header, pkt.Data.Size(), params) + pkt.NetworkHeader = buffer.View(ip) if loop&stack.PacketLoop != 0 { - views := make([]buffer.View, 1, 1+len(payload.Views())) - views[0] = hdr.View() - views = append(views, payload.Views()...) - vv := buffer.NewVectorisedView(len(views[0])+payload.Size(), views) + // The inbound path expects the network header to still be in + // the PacketBuffer's Data field. + views := make([]buffer.View, 1, 1+len(pkt.Data.Views())) + views[0] = pkt.Header.View() + views = append(views, pkt.Data.Views()...) loopedR := r.MakeLoopedRoute() - e.HandlePacket(&loopedR, vv) + + e.HandlePacket(&loopedR, tcpip.PacketBuffer{ + Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), + }) + loopedR.Release() } if loop&stack.PacketOut == 0 { @@ -124,36 +135,58 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen } r.Stats().IP.PacketsSent.Increment() - return e.linkEP.WritePacket(r, gso, hdr, payload, ProtocolNumber) + return e.linkEP.WritePacket(r, gso, ProtocolNumber, pkt) +} + +// WritePackets implements stack.LinkEndpoint.WritePackets. +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) { + if loop&stack.PacketLoop != 0 { + panic("not implemented") + } + if loop&stack.PacketOut == 0 { + return len(pkts), nil + } + + for i := range pkts { + hdr := &pkts[i].Header + size := pkts[i].DataSize + ip := e.addIPHeader(r, hdr, size, params) + pkts[i].NetworkHeader = buffer.View(ip) + } + + n, err := e.linkEP.WritePackets(r, gso, pkts, ProtocolNumber) + r.Stats().IP.PacketsSent.IncrementBy(uint64(n)) + return n, err } // WriteHeaderIncludedPacker implements stack.NetworkEndpoint. It is not yet // supported by IPv6. -func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error { - // TODO(b/119580726): Support IPv6 header-included packets. +func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, loop stack.PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error { + // TODO(b/146666412): Support IPv6 header-included packets. return tcpip.ErrNotSupported } // 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. @@ -188,9 +221,9 @@ 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, diff --git a/pkg/tcpip/network/ipv6/ipv6_test.go b/pkg/tcpip/network/ipv6/ipv6_test.go index deaa9b7f3..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 @@ -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..ab24372e7 --- /dev/null +++ b/pkg/tcpip/packet_buffer.go @@ -0,0 +1,67 @@ +// 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 + + // DataOffset is used for GSO output. It is the offset into the Data + // field where the payload of this packet starts. + DataOffset int + + // DataSize is used for GSO output. It is the size of this packet's + // payload. + DataSize int + + // Header holds the headers of outbound packets. As a packet is passed + // down the stack, each layer adds to Header. + Header buffer.Prependable + + // These fields are used by both inbound and outbound packets. They + // typically overlap with the Data and Header fields. + // + // The bytes backing these views are immutable. Each field may be nil + // if either it has not been set yet or no such header exists (e.g. + // packets sent via loopback may not have a link header). + // + // These fields may be Views into other slices (either Data or Header). + // SR dosen't support this, so deep copies are necessary in some cases. + LinkHeader buffer.View + NetworkHeader buffer.View + TransportHeader buffer.View +} + +// Clone makes a copy of pk. It clones the Data field, which creates a new +// VectorisedView but does not deep copy the underlying bytes. +// +// Clone also does not deep copy any of its other fields. +func (pk PacketBuffer) Clone() PacketBuffer { + pk.Data = pk.Data.Clone(nil) + return pk +} diff --git a/pkg/tcpip/packet_buffer_state.go b/pkg/tcpip/packet_buffer_state.go new file mode 100644 index 000000000..ad3cc24fa --- /dev/null +++ b/pkg/tcpip/packet_buffer_state.go @@ -0,0 +1,27 @@ +// 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.Header = pk.Header.DeepCopy() + pk.LinkHeader = append(buffer.View(nil), pk.LinkHeader...) + pk.NetworkHeader = append(buffer.View(nil), pk.NetworkHeader...) + pk.TransportHeader = append(buffer.View(nil), pk.TransportHeader...) +} diff --git a/pkg/tcpip/ports/BUILD b/pkg/tcpip/ports/BUILD index 11efb4e44..e156b01f6 100644 --- a/pkg/tcpip/ports/BUILD +++ b/pkg/tcpip/ports/BUILD @@ -1,5 +1,5 @@ -load("//tools/go_stateify:defs.bzl", "go_library") load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -7,7 +7,7 @@ go_library( name = "ports", srcs = ["ports.go"], importpath = "gvisor.dev/gvisor/pkg/tcpip/ports", - visibility = ["//:sandbox"], + visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", ], diff --git a/pkg/tcpip/ports/ports.go b/pkg/tcpip/ports/ports.go index 30cea8996..6c5e19e8f 100644 --- a/pkg/tcpip/ports/ports.go +++ b/pkg/tcpip/ports/ports.go @@ -41,6 +41,30 @@ type portDescriptor struct { port uint16 } +// Flags represents the type of port reservation. +// +// +stateify savable +type Flags struct { + // MostRecent represents UDP SO_REUSEADDR. + MostRecent bool + + // LoadBalanced indicates SO_REUSEPORT. + // + // LoadBalanced takes precidence over MostRecent. + LoadBalanced bool +} + +func (f Flags) bits() reuseFlag { + var rf reuseFlag + if f.MostRecent { + rf |= mostRecentFlag + } + if f.LoadBalanced { + rf |= loadBalancedFlag + } + return rf +} + // PortManager manages allocating, reserving and releasing ports. type PortManager struct { mu sync.RWMutex @@ -54,9 +78,59 @@ type PortManager struct { hint uint32 } +type reuseFlag int + +const ( + mostRecentFlag reuseFlag = 1 << iota + loadBalancedFlag + nextFlag + + flagMask = nextFlag - 1 +) + type portNode struct { - reuse bool - refs int + // refs stores the count for each possible flag combination. + refs [nextFlag]int +} + +func (p portNode) totalRefs() int { + var total int + for _, r := range p.refs { + total += r + } + return total +} + +// flagRefs returns the number of references with all specified flags. +func (p portNode) flagRefs(flags reuseFlag) int { + var total int + for i, r := range p.refs { + if reuseFlag(i)&flags == flags { + total += r + } + } + return total +} + +// allRefsHave returns if all references have all specified flags. +func (p portNode) allRefsHave(flags reuseFlag) bool { + for i, r := range p.refs { + if reuseFlag(i)&flags == flags && r > 0 { + return false + } + } + return true +} + +// intersectionRefs returns the set of flags shared by all references. +func (p portNode) intersectionRefs() reuseFlag { + intersection := flagMask + for i, r := range p.refs { + if r > 0 { + intersection &= reuseFlag(i) + } + } + return intersection } // deviceNode is never empty. When it has no elements, it is removed from the @@ -66,30 +140,44 @@ type deviceNode map[tcpip.NICID]portNode // 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 either of the port reuse flags is enabled on any of the nodes, all nodes +// sharing a port must share at least one reuse flag. This matches Linux's +// behavior. +func (d deviceNode) isAvailable(flags Flags, bindToDevice tcpip.NICID) bool { + flagBits := flags.bits() if bindToDevice == 0 { // Trying to binding all devices. - if !reuse { + if flagBits == 0 { // Can't bind because the (addr,port) is already bound. return false } + intersection := flagMask for _, p := range d { - if !p.reuse { - // Can't bind because the (addr,port) was previously bound without reuse. + i := p.intersectionRefs() + intersection &= i + if intersection&flagBits == 0 { + // Can't bind because the (addr,port) was + // previously bound without reuse. return false } } return true } + intersection := flagMask + if p, ok := d[0]; ok { - if !reuse || !p.reuse { + intersection = p.intersectionRefs() + if intersection&flagBits == 0 { return false } } if p, ok := d[bindToDevice]; ok { - if !reuse || !p.reuse { + i := p.intersectionRefs() + intersection &= i + if intersection&flagBits == 0 { return false } } @@ -103,12 +191,12 @@ type bindAddresses map[tcpip.Address]deviceNode // isAvailable checks whether an IP address is available to bind to. If the // address is the "any" address, check all other addresses. Otherwise, just // check against the "any" address and the provided address. -func (b bindAddresses) isAvailable(addr tcpip.Address, reuse bool, bindToDevice tcpip.NICID) bool { +func (b bindAddresses) isAvailable(addr tcpip.Address, flags Flags, 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) { + if !d.isAvailable(flags, bindToDevice) { return false } } @@ -117,14 +205,14 @@ func (b bindAddresses) isAvailable(addr tcpip.Address, reuse bool, bindToDevice // Check that there is no conflict with the "any" address. if d, ok := b[anyIPAddress]; ok { - if !d.isAvailable(reuse, bindToDevice) { + if !d.isAvailable(flags, bindToDevice) { return false } } // Check that this is no conflict with the provided address. if d, ok := b[addr]; ok { - if !d.isAvailable(reuse, bindToDevice) { + if !d.isAvailable(flags, bindToDevice) { return false } } @@ -190,17 +278,17 @@ func (s *PortManager) pickEphemeralPort(offset, count uint32, testPort func(p ui } // IsPortAvailable tests if the given port is available on all given protocols. -func (s *PortManager) IsPortAvailable(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool, bindToDevice tcpip.NICID) bool { +func (s *PortManager) IsPortAvailable(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, flags Flags, bindToDevice tcpip.NICID) bool { s.mu.Lock() defer s.mu.Unlock() - return s.isPortAvailableLocked(networks, transport, addr, port, reuse, bindToDevice) + return s.isPortAvailableLocked(networks, transport, addr, port, flags, bindToDevice) } -func (s *PortManager) isPortAvailableLocked(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool, bindToDevice tcpip.NICID) bool { +func (s *PortManager) isPortAvailableLocked(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, flags Flags, 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, bindToDevice) { + if !addrs.isAvailable(addr, flags, bindToDevice) { return false } } @@ -212,14 +300,14 @@ 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, bindToDevice tcpip.NICID) (reservedPort uint16, err *tcpip.Error) { +func (s *PortManager) ReservePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, flags Flags, bindToDevice tcpip.NICID) (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, bindToDevice) { + if !s.reserveSpecificPort(networks, transport, addr, port, flags, bindToDevice) { return 0, tcpip.ErrPortInUse } return port, nil @@ -227,15 +315,16 @@ func (s *PortManager) ReservePort(networks []tcpip.NetworkProtocolNumber, transp // A port wasn't specified, so try to find one. return s.PickEphemeralPort(func(p uint16) (bool, *tcpip.Error) { - return s.reserveSpecificPort(networks, transport, addr, p, reuse, bindToDevice), nil + return s.reserveSpecificPort(networks, transport, addr, p, flags, 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, bindToDevice tcpip.NICID) bool { - if !s.isPortAvailableLocked(networks, transport, addr, port, reuse, bindToDevice) { +func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, flags Flags, bindToDevice tcpip.NICID) bool { + if !s.isPortAvailableLocked(networks, transport, addr, port, flags, bindToDevice) { return false } + flagBits := flags.bits() // Reserve port on all network protocols. for _, network := range networks { @@ -250,12 +339,9 @@ func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber d = make(deviceNode) m[addr] = d } - if n, ok := d[bindToDevice]; ok { - n.refs++ - d[bindToDevice] = n - } else { - d[bindToDevice] = portNode{reuse: reuse, refs: 1} - } + n := d[bindToDevice] + n.refs[flagBits]++ + d[bindToDevice] = n } return true @@ -263,10 +349,12 @@ 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, bindToDevice tcpip.NICID) { +func (s *PortManager) ReleasePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, flags Flags, bindToDevice tcpip.NICID) { s.mu.Lock() defer s.mu.Unlock() + flagBits := flags.bits() + for _, network := range networks { desc := portDescriptor{network, transport, port} if m, ok := s.allocatedPorts[desc]; ok { @@ -278,9 +366,9 @@ func (s *PortManager) ReleasePort(networks []tcpip.NetworkProtocolNumber, transp if !ok { continue } - n.refs-- + n.refs[flagBits]-- d[bindToDevice] = n - if n.refs == 0 { + if n.refs == [nextFlag]int{} { delete(d, bindToDevice) } if len(d) == 0 { diff --git a/pkg/tcpip/ports/ports_test.go b/pkg/tcpip/ports/ports_test.go index 19f4833fc..d6969d050 100644 --- a/pkg/tcpip/ports/ports_test.go +++ b/pkg/tcpip/ports/ports_test.go @@ -33,7 +33,7 @@ type portReserveTestAction struct { port uint16 ip tcpip.Address want *tcpip.Error - reuse bool + flags Flags release bool device tcpip.NICID } @@ -50,7 +50,7 @@ func TestPortReservation(t *testing.T) { {port: 80, ip: fakeIPAddress1, want: nil}, /* N.B. Order of tests matters! */ {port: 80, ip: anyIPAddress, want: tcpip.ErrPortInUse}, - {port: 80, ip: fakeIPAddress, want: tcpip.ErrPortInUse, reuse: true}, + {port: 80, ip: fakeIPAddress, want: tcpip.ErrPortInUse, flags: Flags{LoadBalanced: true}}, }, }, { @@ -61,7 +61,7 @@ func TestPortReservation(t *testing.T) { /* release fakeIPAddress, but anyIPAddress is still inuse */ {port: 22, ip: fakeIPAddress, release: true}, {port: 22, ip: fakeIPAddress, want: tcpip.ErrPortInUse}, - {port: 22, ip: fakeIPAddress, want: tcpip.ErrPortInUse, reuse: true}, + {port: 22, ip: fakeIPAddress, want: tcpip.ErrPortInUse, flags: Flags{LoadBalanced: true}}, /* Release port 22 from any IP address, then try to reserve fake IP address on 22 */ {port: 22, ip: anyIPAddress, want: nil, release: true}, {port: 22, ip: fakeIPAddress, want: nil}, @@ -71,36 +71,36 @@ func TestPortReservation(t *testing.T) { actions: []portReserveTestAction{ {port: 00, ip: fakeIPAddress, want: nil}, {port: 00, ip: fakeIPAddress, want: nil}, - {port: 00, ip: fakeIPAddress, reuse: true, want: nil}, + {port: 00, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, want: nil}, }, }, { tname: "bind to ip with reuseport", actions: []portReserveTestAction{ - {port: 25, ip: fakeIPAddress, reuse: true, want: nil}, - {port: 25, ip: fakeIPAddress, reuse: true, want: nil}, + {port: 25, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, want: nil}, + {port: 25, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, want: nil}, - {port: 25, ip: fakeIPAddress, reuse: false, want: tcpip.ErrPortInUse}, - {port: 25, ip: anyIPAddress, reuse: false, want: tcpip.ErrPortInUse}, + {port: 25, ip: fakeIPAddress, flags: Flags{}, want: tcpip.ErrPortInUse}, + {port: 25, ip: anyIPAddress, flags: Flags{}, want: tcpip.ErrPortInUse}, - {port: 25, ip: anyIPAddress, reuse: true, want: nil}, + {port: 25, ip: anyIPAddress, flags: Flags{LoadBalanced: true}, want: nil}, }, }, { tname: "bind to inaddr any with reuseport", actions: []portReserveTestAction{ - {port: 24, ip: anyIPAddress, reuse: true, want: nil}, - {port: 24, ip: anyIPAddress, reuse: true, want: nil}, + {port: 24, ip: anyIPAddress, flags: Flags{LoadBalanced: true}, want: nil}, + {port: 24, ip: anyIPAddress, flags: Flags{LoadBalanced: true}, want: nil}, - {port: 24, ip: anyIPAddress, reuse: false, want: tcpip.ErrPortInUse}, - {port: 24, ip: fakeIPAddress, reuse: false, want: tcpip.ErrPortInUse}, + {port: 24, ip: anyIPAddress, flags: Flags{}, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, flags: Flags{}, want: tcpip.ErrPortInUse}, - {port: 24, ip: fakeIPAddress, reuse: true, want: nil}, - {port: 24, ip: fakeIPAddress, release: true, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, release: true, want: nil}, - {port: 24, ip: anyIPAddress, release: true}, - {port: 24, ip: anyIPAddress, reuse: false, want: tcpip.ErrPortInUse}, + {port: 24, ip: anyIPAddress, flags: Flags{LoadBalanced: true}, release: true}, + {port: 24, ip: anyIPAddress, flags: Flags{}, want: tcpip.ErrPortInUse}, - {port: 24, ip: anyIPAddress, release: true}, - {port: 24, ip: anyIPAddress, reuse: false, want: nil}, + {port: 24, ip: anyIPAddress, flags: Flags{LoadBalanced: true}, release: true}, + {port: 24, ip: anyIPAddress, flags: Flags{}, want: nil}, }, }, { tname: "bind twice with device fails", @@ -125,88 +125,152 @@ func TestPortReservation(t *testing.T) { 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, device: 123, flags: Flags{LoadBalanced: true}, want: tcpip.ErrPortInUse}, {port: 24, ip: fakeIPAddress, want: tcpip.ErrPortInUse}, - {port: 24, ip: fakeIPAddress, reuse: true, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, flags: Flags{LoadBalanced: 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: 123, flags: Flags{LoadBalanced: 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: 0, flags: Flags{LoadBalanced: true}, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 456, flags: Flags{LoadBalanced: 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}, + {port: 24, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, want: tcpip.ErrPortInUse}, }, }, { - tname: "bind with reuse", + tname: "bind with reuseport", actions: []portReserveTestAction{ - {port: 24, ip: fakeIPAddress, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{LoadBalanced: 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: 123, flags: Flags{LoadBalanced: true}, want: nil}, {port: 24, ip: fakeIPAddress, device: 0, want: tcpip.ErrPortInUse}, - {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, flags: Flags{LoadBalanced: true}, want: nil}, }, }, { - tname: "binding with reuse and device", + tname: "binding with reuseport and device", actions: []portReserveTestAction{ - {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, flags: Flags{LoadBalanced: 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: 123, flags: Flags{LoadBalanced: 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: 456, flags: Flags{LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, flags: Flags{LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, device: 789, flags: Flags{LoadBalanced: true}, want: nil}, {port: 24, ip: fakeIPAddress, device: 999, want: tcpip.ErrPortInUse}, }, }, { - tname: "mixing reuse and not reuse by binding to device", + tname: "mixing reuseport and not reuseport by binding to device", actions: []portReserveTestAction{ - {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, flags: Flags{LoadBalanced: 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: 789, flags: Flags{LoadBalanced: true}, want: nil}, {port: 24, ip: fakeIPAddress, device: 999, want: nil}, }, }, { - tname: "can't bind to 0 after mixing reuse and not reuse", + tname: "can't bind to 0 after mixing reuseport and not reuseport", actions: []portReserveTestAction{ - {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, flags: Flags{LoadBalanced: true}, want: nil}, {port: 24, ip: fakeIPAddress, device: 456, want: nil}, - {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 0, flags: Flags{LoadBalanced: 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}, + {port: 24, ip: fakeIPAddress, device: 123, flags: Flags{LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, flags: Flags{LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, device: 345, flags: Flags{}, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 789, flags: Flags{LoadBalanced: 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}, + {port: 24, ip: fakeIPAddress, device: 0, flags: Flags{LoadBalanced: true}, want: nil, release: true}, + {port: 24, ip: fakeIPAddress, device: 345, flags: Flags{}, want: nil}, }, }, { - tname: "bind twice with reuse once", + tname: "bind twice with reuseport once", actions: []portReserveTestAction{ - {port: 24, ip: fakeIPAddress, device: 123, reuse: false, want: nil}, - {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 123, flags: Flags{}, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, flags: Flags{LoadBalanced: 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}, + {port: 24, ip: fakeIPAddress, device: 123, flags: Flags{}, want: nil}, + {port: 24, ip: fakeIPAddress, device: 456, flags: Flags{}, 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}, + {port: 24, ip: fakeIPAddress, device: 345, flags: Flags{}, want: nil, release: true}, + {port: 9999, ip: fakeIPAddress, device: 123, flags: Flags{}, 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}, + {port: 24, ip: fakeIPAddress, device: 123, flags: Flags{}, want: nil, release: true}, + {port: 24, ip: fakeIPAddress, device: 456, flags: Flags{}, want: nil, release: true}, + }, + }, { + tname: "bind with reuseaddr", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true}, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 123, flags: Flags{MostRecent: true}, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 0, flags: Flags{MostRecent: true}, want: nil}, + }, + }, { + tname: "bind twice with reuseaddr once", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, flags: Flags{}, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, flags: Flags{MostRecent: true}, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind with reuseaddr and reuseport", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true, LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true, LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true, LoadBalanced: true}, want: nil}, + }, + }, { + tname: "bind with reuseaddr and reuseport, and then reuseaddr", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true, LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind with reuseaddr and reuseport, and then reuseport", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true, LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true}, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind with reuseaddr and reuseport twice, and then reuseaddr", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true, LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true, LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true}, want: nil}, + }, + }, { + tname: "bind with reuseaddr and reuseport twice, and then reuseport", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true, LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true, LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, want: nil}, + }, + }, { + tname: "bind with reuseaddr, and then reuseaddr and reuseport", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true, LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind with reuseport, and then reuseaddr and reuseport", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true, LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true}, want: tcpip.ErrPortInUse}, }, }, } { @@ -216,12 +280,12 @@ func TestPortReservation(t *testing.T) { for _, test := range test.actions { if test.release { - pm.ReleasePort(net, fakeTransNumber, test.ip, test.port, test.device) + pm.ReleasePort(net, fakeTransNumber, test.ip, test.port, test.flags, test.device) continue } - gotPort, err := pm.ReservePort(net, fakeTransNumber, test.ip, test.port, test.reuse, test.device) + gotPort, err := pm.ReservePort(net, fakeTransNumber, test.ip, test.port, test.flags, test.device) if err != test.want { - t.Fatalf("ReservePort(.., .., %s, %d, %t, %d) = %v, want %v", test.ip, test.port, test.reuse, test.device, err, test.want) + t.Fatalf("ReservePort(.., .., %s, %d, %+v, %d) = %v, want %v", test.ip, test.port, test.flags, 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) diff --git a/pkg/tcpip/sample/tun_tcp_connect/BUILD b/pkg/tcpip/sample/tun_tcp_connect/BUILD index a57752a7c..d7496fde6 100644 --- a/pkg/tcpip/sample/tun_tcp_connect/BUILD +++ b/pkg/tcpip/sample/tun_tcp_connect/BUILD @@ -5,6 +5,7 @@ package(licenses = ["notice"]) go_binary( name = "tun_tcp_connect", srcs = ["main.go"], + visibility = ["//:sandbox"], deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/sample/tun_tcp_echo/BUILD b/pkg/tcpip/sample/tun_tcp_echo/BUILD index dad8ef399..875561566 100644 --- a/pkg/tcpip/sample/tun_tcp_echo/BUILD +++ b/pkg/tcpip/sample/tun_tcp_echo/BUILD @@ -5,6 +5,7 @@ package(licenses = ["notice"]) go_binary( name = "tun_tcp_echo", srcs = ["main.go"], + visibility = ["//:sandbox"], deps = [ "//pkg/tcpip", "//pkg/tcpip/link/fdbased", diff --git a/pkg/tcpip/seqnum/BUILD b/pkg/tcpip/seqnum/BUILD index 29b7d761c..b31ddba2f 100644 --- a/pkg/tcpip/seqnum/BUILD +++ b/pkg/tcpip/seqnum/BUILD @@ -6,7 +6,5 @@ go_library( name = "seqnum", srcs = ["seqnum.go"], importpath = "gvisor.dev/gvisor/pkg/tcpip/seqnum", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], ) diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD index bfc03e90b..69077669a 100644 --- a/pkg/tcpip/stack/BUILD +++ b/pkg/tcpip/stack/BUILD @@ -31,9 +31,7 @@ go_library( "transport_demuxer.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/stack", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/ilist", "//pkg/rand", @@ -73,6 +71,7 @@ go_test( "//pkg/tcpip/transport/icmp", "//pkg/tcpip/transport/udp", "//pkg/waiter", + "@com_github_google_go-cmp//cmp:go_default_library", ], ) @@ -86,11 +85,3 @@ go_test( "//pkg/tcpip", ], ) - -filegroup( - name = "autogen", - srcs = [ - "linkaddrentry_list.go", - ], - visibility = ["//:sandbox"], -) diff --git a/pkg/tcpip/stack/ndp.go b/pkg/tcpip/stack/ndp.go index bed60d7b1..90664ba8a 100644 --- a/pkg/tcpip/stack/ndp.go +++ b/pkg/tcpip/stack/ndp.go @@ -38,6 +38,34 @@ const ( // 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 + + // defaultDiscoverOnLinkPrefixes is the default configuration for + // whether or not to discover on-link prefixes from incoming Router + // Advertisements' Prefix Information option, as a host. + // + // Default = true. + defaultDiscoverOnLinkPrefixes = true + + // defaultAutoGenGlobalAddresses is the default configuration for + // whether or not to generate global IPv6 addresses in response to + // receiving a new Prefix Information option with its Autonomous + // Address AutoConfiguration flag set, as a host. + // + // Default = true. + defaultAutoGenGlobalAddresses = true + // minimumRetransmitTimer is the minimum amount of time to wait between // sending NDP Neighbor solicitation messages. Note, RFC 4861 does // not impose a minimum Retransmit Timer, but we do here to make sure @@ -49,8 +77,117 @@ const ( // // 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 + + // MaxDiscoveredOnLinkPrefixes is the maximum number of discovered + // on-link prefixes. The stack should stop discovering new on-link + // prefixes after discovering MaxDiscoveredOnLinkPrefixes on-link + // prefixes. + // + // Max = 10. + MaxDiscoveredOnLinkPrefixes = 10 + + // validPrefixLenForAutoGen is the expected prefix length that an + // address can be generated for. Must be 64 bits as the interface + // identifier (IID) is 64 bits and an IPv6 address is 128 bits, so + // 128 - 64 = 64. + validPrefixLenForAutoGen = 64 ) +var ( + // MinPrefixInformationValidLifetimeForUpdate is the minimum Valid + // Lifetime to update the valid lifetime of a generated address by + // SLAAC. + // + // This is exported as a variable (instead of a constant) so tests + // can update it to a smaller value. + // + // Min = 2hrs. + MinPrefixInformationValidLifetimeForUpdate = 2 * time.Hour +) + +// 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 if the newly discovered + // router should be remembered. + // + // 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 + + // OnDefaultRouterInvalidated will be called when a discovered default + // router that was remembered is invalidated. + // + // 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) + + // OnOnLinkPrefixDiscovered will be called when a new on-link prefix is + // discovered. Implementations must return true if the newly discovered + // on-link prefix should be remembered. + // + // This function is not permitted to block indefinitely. This function + // is also not permitted to call into the stack. + OnOnLinkPrefixDiscovered(nicID tcpip.NICID, prefix tcpip.Subnet) bool + + // OnOnLinkPrefixInvalidated will be called when a discovered on-link + // prefix that was remembered is invalidated. + // + // This function is not permitted to block indefinitely. This function + // is also not permitted to call into the stack. + OnOnLinkPrefixInvalidated(nicID tcpip.NICID, prefix tcpip.Subnet) + + // OnAutoGenAddress will be called when a new prefix with its + // autonomous address-configuration flag set has been received and SLAAC + // has been performed. Implementations may prevent the stack from + // assigning the address to the NIC by returning false. + // + // This function is not permitted to block indefinitely. It must not + // call functions on the stack itself. + OnAutoGenAddress(tcpip.NICID, tcpip.AddressWithPrefix) bool + + // OnAutoGenAddressInvalidated will be called when an auto-generated + // address (as part of SLAAC) has been invalidated. + // + // This function is not permitted to block indefinitely. It must not + // call functions on the stack itself. + OnAutoGenAddressInvalidated(tcpip.NICID, tcpip.AddressWithPrefix) + + // OnRecursiveDNSServerOption will be called when an NDP option with + // recursive DNS servers has been received. Note, addrs may contain + // link-local addresses. + // + // It is up to the caller to use the DNS Servers only for their valid + // lifetime. OnRecursiveDNSServerOption may be called for new or + // already known DNS servers. If called with known DNS servers, their + // valid lifetimes must be refreshed to lifetime (it may be increased, + // decreased, or completely invalidated when lifetime = 0). + OnRecursiveDNSServerOption(nicID tcpip.NICID, addrs []tcpip.Address, lifetime time.Duration) +} + // NDPConfigurations is the NDP configurations for the netstack. type NDPConfigurations struct { // The number of Neighbor Solicitation messages to send when doing @@ -64,6 +201,31 @@ type NDPConfigurations struct { // // 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 + + // DiscoverOnLinkPrefixes determines whether or not on-link prefixes + // will be discovered from Router Advertisements' Prefix Information + // option. This configuration is ignored if HandleRAs is false. + DiscoverOnLinkPrefixes bool + + // AutoGenGlobalAddresses determines whether or not global IPv6 + // addresses will be generated for a NIC in response to receiving a new + // Prefix Information option with its Autonomous Address + // AutoConfiguration flag set, as a host, as per RFC 4862 (SLAAC). + // + // Note, if an address was already generated for some unique prefix, as + // part of SLAAC, this option does not affect whether or not the + // lifetime(s) of the generated address changes; this option only + // affects the generation of new addresses as part of SLAAC. + AutoGenGlobalAddresses bool } // DefaultNDPConfigurations returns an NDPConfigurations populated with @@ -72,6 +234,10 @@ func DefaultNDPConfigurations() NDPConfigurations { return NDPConfigurations{ DupAddrDetectTransmits: defaultDupAddrDetectTransmits, RetransmitTimer: defaultRetransmitTimer, + HandleRAs: defaultHandleRAs, + DiscoverDefaultRouters: defaultDiscoverDefaultRouters, + DiscoverOnLinkPrefixes: defaultDiscoverOnLinkPrefixes, + AutoGenGlobalAddresses: defaultAutoGenGlobalAddresses, } } @@ -88,8 +254,24 @@ func (c *NDPConfigurations) validate() { // 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 + + // The on-link prefixes discovered through Router Advertisements' Prefix + // Information option. + onLinkPrefixes map[tcpip.Subnet]onLinkPrefixState + + // The addresses generated by SLAAC. + autoGenAddresses map[tcpip.Address]autoGenAddressState } // dadState holds the Duplicate Address Detection timer and channel to signal @@ -105,13 +287,84 @@ type dadState struct { done *bool } +// defaultRouterState holds data associated with a default router discovered by +// a Router Advertisement (RA). +type defaultRouterState struct { + invalidationTimer *time.Timer + + // Used to inform 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 + // inform the timer using doNotInvalidate to not invalidate R, so that + // once T2 obtains the lock, it will see that it is set to true and do + // nothing further. + doNotInvalidate *bool +} + +// onLinkPrefixState holds data associated with an on-link prefix discovered by +// a Router Advertisement's Prefix Information option (PI) when the NDP +// configurations was configured to do so. +type onLinkPrefixState struct { + invalidationTimer *time.Timer + + // Used to signal the timer not to invalidate the on-link prefix (P) in + // a race condition (T1 is a goroutine that handles a PI for P and T2 + // is the goroutine that handles P's invalidation timer firing): + // T1: Receive a new PI for P + // T1: Obtain the NIC's lock before processing the PI + // T2: P's invalidation timer fires, and gets blocked on obtaining the + // NIC's lock + // T1: Refreshes/extends P's lifetime & releases NIC's lock + // T2: Obtains NIC's lock & invalidates P immediately + // + // To resolve this, T1 will check to see if the timer already fired, and + // inform the timer using doNotInvalidate to not invalidate P, so that + // once T2 obtains the lock, it will see that it is set to true and do + // nothing further. + doNotInvalidate *bool +} + +// autoGenAddressState holds data associated with an address generated via +// SLAAC. +type autoGenAddressState struct { + invalidationTimer *time.Timer + + // Used to signal the timer not to invalidate the SLAAC address (A) in + // a race condition (T1 is a goroutine that handles a PI for A and T2 + // is the goroutine that handles A's invalidation timer firing): + // T1: Receive a new PI for A + // T1: Obtain the NIC's lock before processing the PI + // T2: A's invalidation timer fires, and gets blocked on obtaining the + // NIC's lock + // T1: Refreshes/extends A's lifetime & releases NIC's lock + // T2: Obtains NIC's lock & invalidates A immediately + // + // To resolve this, T1 will check to see if the timer already fired, and + // inform the timer using doNotInvalidate to not invalidate A, so that + // once T2 obtains the lock, it will see that it is set to true and do + // nothing further. + doNotInvalidate *bool + + // Nonzero only when the address is not valid forever (invalidationTimer + // is not nil). + validUntil time.Time +} + // startDuplicateAddressDetection performs Duplicate Address Detection. // // This function must only be called by IPv6 addresses that are currently // tentative. // -// The NIC that ndp belongs to (n) MUST be locked. -func (ndp *ndpState) startDuplicateAddressDetection(n *NIC, addr tcpip.Address, ref *referencedNetworkEndpoint) *tcpip.Error { +// 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 @@ -127,13 +380,13 @@ func (ndp *ndpState) startDuplicateAddressDetection(n *NIC, addr tcpip.Address, // 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, n.ID())) + panic(fmt.Sprintf("ndpdad: already performing DAD for addr %s on NIC(%d)", addr, ndp.nic.ID())) } - remaining := n.stack.ndpConfigs.DupAddrDetectTransmits + remaining := ndp.configs.DupAddrDetectTransmits { - done, err := ndp.doDuplicateAddressDetection(n, addr, remaining, ref) + done, err := ndp.doDuplicateAddressDetection(addr, remaining, ref) if err != nil { return err } @@ -146,42 +399,59 @@ func (ndp *ndpState) startDuplicateAddressDetection(n *NIC, addr tcpip.Address, var done bool var timer *time.Timer - timer = time.AfterFunc(n.stack.ndpConfigs.RetransmitTimer, func() { - n.mu.Lock() - defer n.mu.Unlock() + timer = time.AfterFunc(ndp.configs.RetransmitTimer, func() { + var d bool + var err *tcpip.Error - 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 it this function - // obtained the NIC lock. Simply return here and do - // nothing further. - return - } + // 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() - ref, ok := n.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, n.ID())) - } + 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 + } - if done, err := ndp.doDuplicateAddressDetection(n, addr, remaining, ref); err != nil || done { - if err != nil { - log.Printf("ndpdad: Error occured during DAD iteration for addr (%s) on NIC(%d); err = %s", addr, n.ID(), err) + 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())) } - ndp.stopDuplicateAddressDetection(addr) - return - } + d, err = ndp.doDuplicateAddressDetection(addr, remaining, ref) + if err != nil || d { + delete(ndp.dad, addr) - timer.Reset(n.stack.ndpConfigs.RetransmitTimer) - remaining-- + 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{ @@ -204,11 +474,11 @@ func (ndp *ndpState) startDuplicateAddressDetection(n *NIC, addr tcpip.Address, // 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(n *NIC, addr tcpip.Address, remaining uint8, ref *referencedNetworkEndpoint) (bool, *tcpip.Error) { +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, n.ID())) + panic(fmt.Sprintf("ndpdad: addr %s is not tentative on NIC(%d)", addr, ndp.nic.ID())) } if remaining == 0 { @@ -219,17 +489,17 @@ func (ndp *ndpState) doDuplicateAddressDetection(n *NIC, addr tcpip.Address, rem // Send a new NS. snmc := header.SolicitedNodeAddr(addr) - snmcRef, ok := n.endpoints[NetworkEndpointID{snmc}] + 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", n.ID(), snmc, addr)) + 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, n.linkEP.LinkAddress(), snmcRef, false, false) + 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)) @@ -239,7 +509,9 @@ func (ndp *ndpState) doDuplicateAddressDetection(n *NIC, addr tcpip.Address, rem 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: r.DefaultTTL(), TOS: DefaultTOS}); err != nil { + if err := r.WritePacket(nil, NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + }); err != nil { sent.Dropped.Increment() return false, err } @@ -275,5 +547,611 @@ func (ndp *ndpState) stopDuplicateAddressDetection(addr tcpip.Address) { delete(ndp.dad, addr) - return + // 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. Inform the timer not to invalidate the + // router when it obtains the lock as we just + // got a new RA that refreshes its lifetime to a + // non-zero value. See + // defaultRouterState.doNotInvalidate for more + // details. + *rtr.doNotInvalidate = true + } + + 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/141556115): Do (RetransTimer, ReachableTime)) Parameter + // Discovery. + + // We know the options is valid as far as wire format is concerned since + // we got the Router Advertisement, as documented by this fn. Given this + // we do not check the iterator for errors on calls to Next. + it, _ := ra.Options().Iter(false) + for opt, done, _ := it.Next(); !done; opt, done, _ = it.Next() { + switch opt := opt.(type) { + case header.NDPRecursiveDNSServer: + if ndp.nic.stack.ndpDisp == nil { + continue + } + + ndp.nic.stack.ndpDisp.OnRecursiveDNSServerOption(ndp.nic.ID(), opt.Addresses(), opt.Lifetime()) + + case header.NDPPrefixInformation: + prefix := opt.Subnet() + + // Is the prefix a link-local? + if header.IsV6LinkLocalAddress(prefix.ID()) { + // ...Yes, skip as per RFC 4861 section 6.3.4, + // and RFC 4862 section 5.5.3.b (for SLAAC). + continue + } + + // Is the Prefix Length 0? + if prefix.Prefix() == 0 { + // ...Yes, skip as this is an invalid prefix + // as all IPv6 addresses cannot be on-link. + continue + } + + if opt.OnLinkFlag() { + ndp.handleOnLinkPrefixInformation(opt) + } + + if opt.AutonomousAddressConfigurationFlag() { + ndp.handleAutonomousPrefixInformation(opt) + } + } + + // TODO(b/141556115): Do (MTU) Parameter Discovery. + } +} + +// invalidateDefaultRouter invalidates a discovered default router. +// +// The NIC that ndp belongs to 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 + *rtr.doNotInvalidate = true + rtr.doNotInvalidate = nil + + delete(ndp.defaultRouters, ip) + + // Let the integrator know a discovered default router is invalidated. + if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil { + 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 MUST be locked. +func (ndp *ndpState) rememberDefaultRouter(ip tcpip.Address, rl time.Duration) { + ndpDisp := ndp.nic.stack.ndpDisp + if ndpDisp == nil { + return + } + + // Inform the integrator when we discovered a default router. + if !ndpDisp.OnDefaultRouterDiscovered(ndp.nic.ID(), ip) { + // 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.doNotInvalidate for more + // details. + var doNotInvalidate bool + + 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() + + if doNotInvalidate { + doNotInvalidate = false + return + } + + ndp.invalidateDefaultRouter(ip) + }), + doNotInvalidate: &doNotInvalidate, + } +} + +// rememberOnLinkPrefix remembers a newly discovered on-link prefix with IPv6 +// address with prefix prefix with lifetime l. +// +// The prefix identified by prefix MUST NOT already be known. +// +// The NIC that ndp belongs to MUST be locked. +func (ndp *ndpState) rememberOnLinkPrefix(prefix tcpip.Subnet, l time.Duration) { + ndpDisp := ndp.nic.stack.ndpDisp + if ndpDisp == nil { + return + } + + // Inform the integrator when we discovered an on-link prefix. + if !ndpDisp.OnOnLinkPrefixDiscovered(ndp.nic.ID(), prefix) { + // Informed by the integrator to not remember the prefix, do + // nothing further. + return + } + + // Used to signal the timer not to invalidate the on-link prefix (P) in + // a race condition. See onLinkPrefixState.doNotInvalidate for more + // details. + var doNotInvalidate bool + var timer *time.Timer + + // Only create a timer if the lifetime is not infinite. + if l < header.NDPInfiniteLifetime { + timer = ndp.prefixInvalidationCallback(prefix, l, &doNotInvalidate) + } + + ndp.onLinkPrefixes[prefix] = onLinkPrefixState{ + invalidationTimer: timer, + doNotInvalidate: &doNotInvalidate, + } +} + +// invalidateOnLinkPrefix invalidates a discovered on-link prefix. +// +// The NIC that ndp belongs to MUST be locked. +func (ndp *ndpState) invalidateOnLinkPrefix(prefix tcpip.Subnet) { + s, ok := ndp.onLinkPrefixes[prefix] + + // Is the on-link prefix still discovered? + if !ok { + // ...Nope, do nothing further. + return + } + + if s.invalidationTimer != nil { + s.invalidationTimer.Stop() + s.invalidationTimer = nil + *s.doNotInvalidate = true + } + + s.doNotInvalidate = nil + + delete(ndp.onLinkPrefixes, prefix) + + // Let the integrator know a discovered on-link prefix is invalidated. + if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil { + ndpDisp.OnOnLinkPrefixInvalidated(ndp.nic.ID(), prefix) + } +} + +// prefixInvalidationCallback returns a new on-link prefix invalidation timer +// for prefix that fires after vl. +// +// doNotInvalidate is used to signal the timer when it fires at the same time +// that a prefix's valid lifetime gets refreshed. See +// onLinkPrefixState.doNotInvalidate for more details. +func (ndp *ndpState) prefixInvalidationCallback(prefix tcpip.Subnet, vl time.Duration, doNotInvalidate *bool) *time.Timer { + return time.AfterFunc(vl, func() { + ndp.nic.stack.mu.Lock() + defer ndp.nic.stack.mu.Unlock() + ndp.nic.mu.Lock() + defer ndp.nic.mu.Unlock() + + if *doNotInvalidate { + *doNotInvalidate = false + return + } + + ndp.invalidateOnLinkPrefix(prefix) + }) +} + +// handleOnLinkPrefixInformation handles a Prefix Information option with +// its on-link flag set, as per RFC 4861 section 6.3.4. +// +// handleOnLinkPrefixInformation assumes that the prefix this pi is for is +// not the link-local prefix and the on-link flag is set. +// +// The NIC that ndp belongs to MUST be locked. +func (ndp *ndpState) handleOnLinkPrefixInformation(pi header.NDPPrefixInformation) { + prefix := pi.Subnet() + prefixState, ok := ndp.onLinkPrefixes[prefix] + vl := pi.ValidLifetime() + + if !ok && vl == 0 { + // Don't know about this prefix but it has a zero valid + // lifetime, so just ignore. + return + } + + if !ok && vl != 0 { + // This is a new on-link prefix we are discovering + // + // Only remember it if we currently know about less than + // MaxDiscoveredOnLinkPrefixes on-link prefixes. + if ndp.configs.DiscoverOnLinkPrefixes && len(ndp.onLinkPrefixes) < MaxDiscoveredOnLinkPrefixes { + ndp.rememberOnLinkPrefix(prefix, vl) + } + return + } + + if ok && vl == 0 { + // We know about the on-link prefix, but it is + // no longer to be considered on-link, so + // invalidate it. + ndp.invalidateOnLinkPrefix(prefix) + return + } + + // This is an already discovered on-link prefix with a + // new non-zero valid lifetime. + // Update the invalidation timer. + timer := prefixState.invalidationTimer + + if timer == nil && vl >= header.NDPInfiniteLifetime { + // Had infinite valid lifetime before and + // continues to have an invalid lifetime. Do + // nothing further. + return + } + + if timer != nil && !timer.Stop() { + // If we reach this point, then we know the timer alread fired + // after we took the NIC lock. Inform the timer to not + // invalidate the prefix once it obtains the lock as we just + // got a new PI that refreshes its lifetime to a non-zero value. + // See onLinkPrefixState.doNotInvalidate for more details. + *prefixState.doNotInvalidate = true + } + + if vl >= header.NDPInfiniteLifetime { + // Prefix is now valid forever so we don't need + // an invalidation timer. + prefixState.invalidationTimer = nil + ndp.onLinkPrefixes[prefix] = prefixState + return + } + + if timer != nil { + // We already have a timer so just reset it to + // expire after the new valid lifetime. + timer.Reset(vl) + return + } + + // We do not have a timer so just create a new one. + prefixState.invalidationTimer = ndp.prefixInvalidationCallback(prefix, vl, prefixState.doNotInvalidate) + ndp.onLinkPrefixes[prefix] = prefixState +} + +// handleAutonomousPrefixInformation handles a Prefix Information option with +// its autonomous flag set, as per RFC 4862 section 5.5.3. +// +// handleAutonomousPrefixInformation assumes that the prefix this pi is for is +// not the link-local prefix and the autonomous flag is set. +// +// The NIC that ndp belongs to and its associated stack MUST be locked. +func (ndp *ndpState) handleAutonomousPrefixInformation(pi header.NDPPrefixInformation) { + vl := pi.ValidLifetime() + pl := pi.PreferredLifetime() + + // If the preferred lifetime is greater than the valid lifetime, + // silently ignore the Prefix Information option, as per RFC 4862 + // section 5.5.3.c. + if pl > vl { + return + } + + prefix := pi.Subnet() + + // Check if we already have an auto-generated address for prefix. + for _, ref := range ndp.nic.endpoints { + if ref.protocol != header.IPv6ProtocolNumber { + continue + } + + if ref.configType != slaac { + continue + } + + addr := ref.ep.ID().LocalAddress + refAddrWithPrefix := tcpip.AddressWithPrefix{Address: addr, PrefixLen: ref.ep.PrefixLen()} + if refAddrWithPrefix.Subnet() != prefix { + continue + } + + // + // At this point, we know we are refreshing a SLAAC generated + // IPv6 address with the prefix, prefix. Do the work as outlined + // by RFC 4862 section 5.5.3.e. + // + + addrState, ok := ndp.autoGenAddresses[addr] + if !ok { + panic(fmt.Sprintf("must have an autoGenAddressess entry for the SLAAC generated IPv6 address %s", addr)) + } + + // TODO(b/143713887): Handle deprecating auto-generated address + // after the preferred lifetime. + + // As per RFC 4862 section 5.5.3.e, the valid lifetime of the + // address generated by SLAAC is as follows: + // + // 1) If the received Valid Lifetime is greater than 2 hours or + // greater than RemainingLifetime, set the valid lifetime of + // the address to the advertised Valid Lifetime. + // + // 2) If RemainingLifetime is less than or equal to 2 hours, + // ignore the advertised Valid Lifetime. + // + // 3) Otherwise, reset the valid lifetime of the address to 2 + // hours. + + // Handle the infinite valid lifetime separately as we do not + // keep a timer in this case. + if vl >= header.NDPInfiniteLifetime { + if addrState.invalidationTimer != nil { + // Valid lifetime was finite before, but now it + // is valid forever. + if !addrState.invalidationTimer.Stop() { + *addrState.doNotInvalidate = true + } + addrState.invalidationTimer = nil + addrState.validUntil = time.Time{} + ndp.autoGenAddresses[addr] = addrState + } + + return + } + + var effectiveVl time.Duration + var rl time.Duration + + // If the address was originally set to be valid forever, + // assume the remaining time to be the maximum possible value. + if addrState.invalidationTimer == nil { + rl = header.NDPInfiniteLifetime + } else { + rl = time.Until(addrState.validUntil) + } + + if vl > MinPrefixInformationValidLifetimeForUpdate || vl > rl { + effectiveVl = vl + } else if rl <= MinPrefixInformationValidLifetimeForUpdate { + ndp.autoGenAddresses[addr] = addrState + return + } else { + effectiveVl = MinPrefixInformationValidLifetimeForUpdate + } + + if addrState.invalidationTimer == nil { + addrState.invalidationTimer = ndp.autoGenAddrInvalidationTimer(addr, effectiveVl, addrState.doNotInvalidate) + } else { + if !addrState.invalidationTimer.Stop() { + *addrState.doNotInvalidate = true + } + addrState.invalidationTimer.Reset(effectiveVl) + } + + addrState.validUntil = time.Now().Add(effectiveVl) + ndp.autoGenAddresses[addr] = addrState + return + } + + // We do not already have an address within the prefix, prefix. Do the + // work as outlined by RFC 4862 section 5.5.3.d if n is configured + // to auto-generated global addresses by SLAAC. + + // Are we configured to auto-generate new global addresses? + if !ndp.configs.AutoGenGlobalAddresses { + return + } + + // If we do not already have an address for this prefix and the valid + // lifetime is 0, no need to do anything further, as per RFC 4862 + // section 5.5.3.d. + if vl == 0 { + return + } + + // Make sure the prefix is valid (as far as its length is concerned) to + // generate a valid IPv6 address from an interface identifier (IID), as + // per RFC 4862 sectiion 5.5.3.d. + if prefix.Prefix() != validPrefixLenForAutoGen { + return + } + + // Only attempt to generate an interface-specific IID if we have a valid + // link address. + // + // TODO(b/141011931): Validate a LinkEndpoint's link address + // (provided by LinkEndpoint.LinkAddress) before reaching this + // point. + linkAddr := ndp.nic.linkEP.LinkAddress() + if !header.IsValidUnicastEthernetAddress(linkAddr) { + return + } + + // Generate an address within prefix from the modified EUI-64 of ndp's + // NIC's Ethernet MAC address. + addrBytes := make([]byte, header.IPv6AddressSize) + copy(addrBytes[:header.IIDOffsetInIPv6Address], prefix.ID()[:header.IIDOffsetInIPv6Address]) + header.EthernetAdddressToModifiedEUI64IntoBuf(linkAddr, addrBytes[header.IIDOffsetInIPv6Address:]) + addr := tcpip.Address(addrBytes) + addrWithPrefix := tcpip.AddressWithPrefix{ + Address: addr, + PrefixLen: validPrefixLenForAutoGen, + } + + // If the nic already has this address, do nothing further. + if ndp.nic.hasPermanentAddrLocked(addr) { + return + } + + // Inform the integrator that we have a new SLAAC address. + ndpDisp := ndp.nic.stack.ndpDisp + if ndpDisp == nil { + return + } + if !ndpDisp.OnAutoGenAddress(ndp.nic.ID(), addrWithPrefix) { + // Informed by the integrator not to add the address. + return + } + + if _, err := ndp.nic.addAddressLocked(tcpip.ProtocolAddress{ + Protocol: header.IPv6ProtocolNumber, + AddressWithPrefix: addrWithPrefix, + }, FirstPrimaryEndpoint, permanent, slaac); err != nil { + panic(err) + } + + // Setup the timers to deprecate and invalidate this newly generated + // address. + + // TODO(b/143713887): Handle deprecating auto-generated addresses + // after the preferred lifetime. + + var doNotInvalidate bool + var vTimer *time.Timer + if vl < header.NDPInfiniteLifetime { + vTimer = ndp.autoGenAddrInvalidationTimer(addr, vl, &doNotInvalidate) + } + + ndp.autoGenAddresses[addr] = autoGenAddressState{ + invalidationTimer: vTimer, + doNotInvalidate: &doNotInvalidate, + validUntil: time.Now().Add(vl), + } +} + +// invalidateAutoGenAddress invalidates an auto-generated address. +// +// The NIC that ndp belongs to MUST be locked. +func (ndp *ndpState) invalidateAutoGenAddress(addr tcpip.Address) { + if !ndp.cleanupAutoGenAddrResourcesAndNotify(addr) { + return + } + + ndp.nic.removePermanentAddressLocked(addr) +} + +// cleanupAutoGenAddrResourcesAndNotify cleans up an invalidated auto-generated +// address's resources from ndp. If the stack has an NDP dispatcher, it will +// be notified that addr has been invalidated. +// +// Returns true if ndp had resources for addr to cleanup. +// +// The NIC that ndp belongs to MUST be locked. +func (ndp *ndpState) cleanupAutoGenAddrResourcesAndNotify(addr tcpip.Address) bool { + state, ok := ndp.autoGenAddresses[addr] + + if !ok { + return false + } + + if state.invalidationTimer != nil { + state.invalidationTimer.Stop() + state.invalidationTimer = nil + *state.doNotInvalidate = true + } + + state.doNotInvalidate = nil + + delete(ndp.autoGenAddresses, addr) + + if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil { + ndpDisp.OnAutoGenAddressInvalidated(ndp.nic.ID(), tcpip.AddressWithPrefix{ + Address: addr, + PrefixLen: validPrefixLenForAutoGen, + }) + } + + return true +} + +// autoGenAddrInvalidationTimer returns a new invalidation timer for an +// auto-generated address that fires after vl. +// +// doNotInvalidate is used to inform the timer when it fires at the same time +// that an auto-generated address's valid lifetime gets refreshed. See +// autoGenAddrState.doNotInvalidate for more details. +func (ndp *ndpState) autoGenAddrInvalidationTimer(addr tcpip.Address, vl time.Duration, doNotInvalidate *bool) *time.Timer { + return time.AfterFunc(vl, func() { + ndp.nic.mu.Lock() + defer ndp.nic.mu.Unlock() + + if *doNotInvalidate { + *doNotInvalidate = false + return + } + + ndp.invalidateAutoGenAddress(addr) + }) } diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go index cc789d70f..666f86c33 100644 --- a/pkg/tcpip/stack/ndp_test.go +++ b/pkg/tcpip/stack/ndp_test.go @@ -15,9 +15,12 @@ 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" @@ -29,12 +32,46 @@ import ( ) 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" - linkAddr1 = "\x01\x02\x03\x04\x05\x06" - linkAddr2 = "\x01\x02\x03\x04\x05\x07" + 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 = 100 * time.Millisecond ) +var ( + llAddr1 = header.LinkLocalAddr(linkAddr1) + llAddr2 = header.LinkLocalAddr(linkAddr2) + llAddr3 = header.LinkLocalAddr(linkAddr3) +) + +// prefixSubnetAddr returns a prefix (Address + Length), the prefix's equivalent +// tcpip.Subnet, and an address where the lower half of the address is composed +// of the EUI-64 of linkAddr if it is a valid unicast ethernet address. +func prefixSubnetAddr(offset uint8, linkAddr tcpip.LinkAddress) (tcpip.AddressWithPrefix, tcpip.Subnet, tcpip.AddressWithPrefix) { + prefixBytes := []byte{1, 2, 3, 4, 5, 6, 7, 8 + offset, 0, 0, 0, 0, 0, 0, 0, 0} + prefix := tcpip.AddressWithPrefix{ + Address: tcpip.Address(prefixBytes), + PrefixLen: 64, + } + + subnet := prefix.Subnet() + + var addr tcpip.AddressWithPrefix + if header.IsValidUnicastEthernetAddress(linkAddr) { + addrBytes := []byte(subnet.ID()) + header.EthernetAdddressToModifiedEUI64IntoBuf(linkAddr, addrBytes[header.IIDOffsetInIPv6Address:]) + addr = tcpip.AddressWithPrefix{ + Address: tcpip.Address(addrBytes), + PrefixLen: 64, + } + } + + return prefix, subnet, addr +} + // 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) { @@ -42,7 +79,7 @@ func TestDADDisabled(t *testing.T) { NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, } - e := channel.New(10, 1280, linkAddr1) + e := channel.New(0, 1280, linkAddr1) s := stack.New(opts) if err := s.CreateNIC(1, e); err != nil { t.Fatalf("CreateNIC(_) = %s", err) @@ -68,6 +105,160 @@ func TestDADDisabled(t *testing.T) { } } +// 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 +} + +type ndpPrefixEvent struct { + nicID tcpip.NICID + prefix tcpip.Subnet + // true if prefix was discovered, false if invalidated. + discovered bool +} + +type ndpAutoGenAddrEventType int + +const ( + newAddr ndpAutoGenAddrEventType = iota + invalidatedAddr +) + +type ndpAutoGenAddrEvent struct { + nicID tcpip.NICID + addr tcpip.AddressWithPrefix + eventType ndpAutoGenAddrEventType +} + +type ndpRDNSS struct { + addrs []tcpip.Address + lifetime time.Duration +} + +type ndpRDNSSEvent struct { + nicID tcpip.NICID + rdnss ndpRDNSS +} + +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 + prefixC chan ndpPrefixEvent + rememberPrefix bool + autoGenAddrC chan ndpAutoGenAddrEvent + rdnssC chan ndpRDNSSEvent +} + +// 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 { + if c := n.routerC; c != nil { + c <- ndpRouterEvent{ + nicID, + addr, + true, + } + } + + return n.rememberRouter +} + +// Implements stack.NDPDispatcher.OnDefaultRouterInvalidated. +func (n *ndpDispatcher) OnDefaultRouterInvalidated(nicID tcpip.NICID, addr tcpip.Address) { + if c := n.routerC; c != nil { + c <- ndpRouterEvent{ + nicID, + addr, + false, + } + } +} + +// Implements stack.NDPDispatcher.OnOnLinkPrefixDiscovered. +func (n *ndpDispatcher) OnOnLinkPrefixDiscovered(nicID tcpip.NICID, prefix tcpip.Subnet) bool { + if c := n.prefixC; c != nil { + c <- ndpPrefixEvent{ + nicID, + prefix, + true, + } + } + + return n.rememberPrefix +} + +// Implements stack.NDPDispatcher.OnOnLinkPrefixInvalidated. +func (n *ndpDispatcher) OnOnLinkPrefixInvalidated(nicID tcpip.NICID, prefix tcpip.Subnet) { + if c := n.prefixC; c != nil { + c <- ndpPrefixEvent{ + nicID, + prefix, + false, + } + } +} + +func (n *ndpDispatcher) OnAutoGenAddress(nicID tcpip.NICID, addr tcpip.AddressWithPrefix) bool { + if c := n.autoGenAddrC; c != nil { + c <- ndpAutoGenAddrEvent{ + nicID, + addr, + newAddr, + } + } + return true +} + +func (n *ndpDispatcher) OnAutoGenAddressInvalidated(nicID tcpip.NICID, addr tcpip.AddressWithPrefix) { + if c := n.autoGenAddrC; c != nil { + c <- ndpAutoGenAddrEvent{ + nicID, + addr, + invalidatedAddr, + } + } +} + +// Implements stack.NDPDispatcher.OnRecursiveDNSServerOption. +func (n *ndpDispatcher) OnRecursiveDNSServerOption(nicID tcpip.NICID, addrs []tcpip.Address, lifetime time.Duration) { + if c := n.rdnssC; c != nil { + c <- ndpRDNSSEvent{ + nicID, + ndpRDNSS{ + addrs, + lifetime, + }, + } + } +} + // 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 @@ -88,9 +279,17 @@ func TestDADResolve(t *testing.T) { } for _, test := range tests { + test := test + t.Run(test.name, func(t *testing.T) { + t.Parallel() + + 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 @@ -107,8 +306,7 @@ func TestDADResolve(t *testing.T) { stat := s.Stats().ICMP.V6PacketsSent.NeighborSolicit - // Should have sent an NDP NS almost immediately. - time.Sleep(100 * time.Millisecond) + // Should have sent an NDP NS immediately. if got := stat.Value(); got != 1 { t.Fatalf("got NeighborSolicit = %d, want = 1", got) @@ -124,16 +322,10 @@ func TestDADResolve(t *testing.T) { t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want) } - // Wait for the remaining time - 500ms, to make sure - // the address is still not resolved. Note, we subtract - // 600ms because we already waited for 100ms earlier, - // so our remaining time is 100ms less than the expected - // time. - // (X - 100ms) - 500ms = X - 600ms - // - // TODO(b/140896005): Use events from the netstack to - // be signalled before DAD resolves. - time.Sleep(test.expectedRetransmitTimer*time.Duration(test.dupAddrDetectTransmits) - 600*time.Millisecond) + // 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) @@ -142,13 +334,30 @@ func TestDADResolve(t *testing.T) { t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want) } - // Wait for the remaining time + 250ms, at which point - // the address should be resolved. Note, the remaining - // time is 500ms. See above comments. - // - // TODO(b/140896005): Use events from the netstack to - // know immediately when DAD completes. - time.Sleep(750 * time.Millisecond) + // 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) @@ -172,7 +381,8 @@ func TestDADResolve(t *testing.T) { } // Check NDP packet. - checker.IPv6(t, p.Header.ToVectorisedView().First(), + checker.IPv6(t, p.Pkt.Header.View().ToVectorisedView().First(), + checker.TTL(header.NDPHopLimit), checker.NDPNS( checker.NDPNSTargetAddress(addr1))) } @@ -250,13 +460,18 @@ func TestDADFail(t *testing.T) { 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: stack.DefaultNDPConfigurations(), + NDPConfigs: ndpConfigs, + NDPDisp: &ndpDisp, } opts.NDPConfigs.RetransmitTimer = time.Second * 2 - e := channel.New(10, 1280, linkAddr1) + e := channel.New(0, 1280, linkAddr1) s := stack.New(opts) if err := s.CreateNIC(1, e); err != nil { t.Fatalf("CreateNIC(_) = %s", err) @@ -279,15 +494,37 @@ func TestDADFail(t *testing.T) { // Receive a packet to simulate multiple nodes owning or // attempting to own the same address. hdr := test.makeBuf(addr1) - e.Inject(header.IPv6ProtocolNumber, hdr.View().ToVectorisedView()) + 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 3 seconds to make sure that DAD did not resolve - time.Sleep(3 * time.Second) + // 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) @@ -302,13 +539,20 @@ func TestDADFail(t *testing.T) { // 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, } - opts.NDPConfigs.RetransmitTimer = time.Second - opts.NDPConfigs.DupAddrDetectTransmits = 2 - e := channel.New(10, 1280, linkAddr1) + e := channel.New(0, 1280, linkAddr1) s := stack.New(opts) if err := s.CreateNIC(1, e); err != nil { t.Fatalf("CreateNIC(_) = %s", err) @@ -332,11 +576,27 @@ func TestDADStop(t *testing.T) { t.Fatalf("RemoveAddress(_, %s) = %s", addr1, err) } - // Wait for the time to normally resolve - // DupAddrDetectTransmits(2) * RetransmitTimer(1s) = 2s. - // An extra 250ms is added to make sure that if DAD was still running - // it resolves and the check below fails. - time.Sleep(2*time.Second + 250*time.Millisecond) + // 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) @@ -350,3 +610,1448 @@ func TestDADStop(t *testing.T) { 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(0, 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) + } + }) + } +} + +// raBufWithOpts returns a valid NDP Router Advertisement with options. +// +// Note, raBufWithOpts does not populate any of the RA fields other than the +// Router Lifetime. +func raBufWithOpts(ip tcpip.Address, rl uint16, optSer header.NDPOptionsSerializer) tcpip.PacketBuffer { + icmpSize := header.ICMPv6HeaderSize + header.NDPRAMinimumSize + int(optSer.Length()) + hdr := buffer.NewPrependable(header.IPv6MinimumSize + icmpSize) + pkt := header.ICMPv6(hdr.Prepend(icmpSize)) + pkt.SetType(header.ICMPv6RouterAdvert) + pkt.SetCode(0) + ra := header.NDPRouterAdvert(pkt.NDPPayload()) + opts := ra.Options() + opts.Serialize(optSer) + // 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()} +} + +// 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 { + return raBufWithOpts(ip, rl, header.NDPOptionsSerializer{}) +} + +// raBufWithPI returns a valid NDP Router Advertisement with a single Prefix +// Information option. +// +// Note, raBufWithPI does not populate any of the RA fields other than the +// Router Lifetime. +func raBufWithPI(ip tcpip.Address, rl uint16, prefix tcpip.AddressWithPrefix, onLink, auto bool, vl, pl uint32) tcpip.PacketBuffer { + flags := uint8(0) + if onLink { + // The OnLink flag is the 7th bit in the flags byte. + flags |= 1 << 7 + } + if auto { + // The Address Auto-Configuration flag is the 6th bit in the + // flags byte. + flags |= 1 << 6 + } + + // A valid header.NDPPrefixInformation must be 30 bytes. + buf := [30]byte{} + // The first byte in a header.NDPPrefixInformation is the Prefix Length + // field. + buf[0] = uint8(prefix.PrefixLen) + // The 2nd byte within a header.NDPPrefixInformation is the Flags field. + buf[1] = flags + // The Valid Lifetime field starts after the 2nd byte within a + // header.NDPPrefixInformation. + binary.BigEndian.PutUint32(buf[2:], vl) + // The Preferred Lifetime field starts after the 6th byte within a + // header.NDPPrefixInformation. + binary.BigEndian.PutUint32(buf[6:], pl) + // The Prefix Address field starts after the 14th byte within a + // header.NDPPrefixInformation. + copy(buf[14:], prefix.Address) + return raBufWithOpts(ip, rl, header.NDPOptionsSerializer{ + header.NDPPrefixInformation(buf[:]), + }) +} + +// 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) { + t.Parallel() + + ndpDisp := ndpDispatcher{ + routerC: make(chan ndpRouterEvent, 1), + } + e := channel.New(0, 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") + default: + } + }) + } +} + +// Check e to make sure that the event is for addr on nic with ID 1, and the +// discovered flag set to discovered. +func checkRouterEvent(e ndpRouterEvent, addr tcpip.Address, discovered bool) string { + return cmp.Diff(ndpRouterEvent{nicID: 1, addr: addr, discovered: discovered}, e, cmp.AllowUnexported(e)) +} + +// TestRouterDiscoveryDispatcherNoRemember tests that the stack does not +// remember a discovered router when the dispatcher asks it not to. +func TestRouterDiscoveryDispatcherNoRemember(t *testing.T) { + t.Parallel() + + ndpDisp := ndpDispatcher{ + routerC: make(chan ndpRouterEvent, 1), + } + e := channel.New(0, 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) + } + + // Receive an RA for a router we should not remember. + const lifetimeSeconds = 1 + e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, lifetimeSeconds)) + select { + case e := <-ndpDisp.routerC: + if diff := checkRouterEvent(e, llAddr2, true); diff != "" { + t.Errorf("router event mismatch (-want +got):\n%s", diff) + } + default: + t.Fatal("expected router discovery event") + } + + // Wait for the invalidation time plus some buffer 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(lifetimeSeconds*time.Second + defaultTimeout): + } +} + +func TestRouterDiscovery(t *testing.T) { + t.Parallel() + + ndpDisp := ndpDispatcher{ + routerC: make(chan ndpRouterEvent, 1), + rememberRouter: true, + } + e := channel.New(0, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + HandleRAs: true, + DiscoverDefaultRouters: true, + }, + NDPDisp: &ndpDisp, + }) + + expectRouterEvent := func(addr tcpip.Address, discovered bool) { + t.Helper() + + select { + case e := <-ndpDisp.routerC: + if diff := checkRouterEvent(e, addr, discovered); diff != "" { + t.Errorf("router event mismatch (-want +got):\n%s", diff) + } + default: + t.Fatal("expected router discovery event") + } + } + + expectAsyncRouterInvalidationEvent := func(addr tcpip.Address, timeout time.Duration) { + t.Helper() + + select { + case e := <-ndpDisp.routerC: + if diff := checkRouterEvent(e, addr, false); diff != "" { + t.Errorf("router event mismatch (-want +got):\n%s", diff) + } + case <-time.After(timeout): + t.Fatal("timed out 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") + default: + } + + // Rx an RA from lladdr2 with a huge lifetime. + e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, 1000)) + expectRouterEvent(llAddr2, true) + + // Rx an RA from another router (lladdr3) with non-zero lifetime. + l3Lifetime := time.Duration(6) + e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr3, uint16(l3Lifetime))) + expectRouterEvent(llAddr3, true) + + // 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") + default: + } + + // 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. + expectAsyncRouterInvalidationEvent(llAddr2, l2Lifetime*time.Second+defaultTimeout) + + // Rx an RA from lladdr2 with huge lifetime. + e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, 1000)) + expectRouterEvent(llAddr2, true) + + // Rx an RA from lladdr2 with zero lifetime. It should be invalidated. + e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, 0)) + expectRouterEvent(llAddr2, false) + + // 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. + expectAsyncRouterInvalidationEvent(llAddr3, l3Lifetime*time.Second+defaultTimeout) +} + +// TestRouterDiscoveryMaxRouters tests that only +// stack.MaxDiscoveredDefaultRouters discovered routers are remembered. +func TestRouterDiscoveryMaxRouters(t *testing.T) { + t.Parallel() + + ndpDisp := ndpDispatcher{ + routerC: make(chan ndpRouterEvent, 1), + rememberRouter: true, + } + e := channel.New(0, 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) + } + + // 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 { + select { + case e := <-ndpDisp.routerC: + if diff := checkRouterEvent(e, llAddr, true); diff != "" { + t.Errorf("router event mismatch (-want +got):\n%s", diff) + } + default: + t.Fatal("expected 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") + default: + } + } + } +} + +// TestNoPrefixDiscovery tests that prefix discovery will not be performed if +// configured not to. +func TestNoPrefixDiscovery(t *testing.T) { + prefix := tcpip.AddressWithPrefix{ + Address: tcpip.Address("\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x00"), + PrefixLen: 64, + } + + // Being configured to discover prefixes 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 + // prefix 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), DiscoverOnLinkPrefixes(%t), Forwarding(%t)", handle, discover, forwarding), func(t *testing.T) { + t.Parallel() + + ndpDisp := ndpDispatcher{ + prefixC: make(chan ndpPrefixEvent, 1), + } + e := channel.New(0, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + HandleRAs: handle, + DiscoverOnLinkPrefixes: discover, + }, + NDPDisp: &ndpDisp, + }) + s.SetForwarding(forwarding) + + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + // Rx an RA with prefix with non-zero lifetime. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, 10, 0)) + + select { + case <-ndpDisp.prefixC: + t.Fatal("unexpectedly discovered a prefix when configured not to") + default: + } + }) + } +} + +// Check e to make sure that the event is for prefix on nic with ID 1, and the +// discovered flag set to discovered. +func checkPrefixEvent(e ndpPrefixEvent, prefix tcpip.Subnet, discovered bool) string { + return cmp.Diff(ndpPrefixEvent{nicID: 1, prefix: prefix, discovered: discovered}, e, cmp.AllowUnexported(e)) +} + +// TestPrefixDiscoveryDispatcherNoRemember tests that the stack does not +// remember a discovered on-link prefix when the dispatcher asks it not to. +func TestPrefixDiscoveryDispatcherNoRemember(t *testing.T) { + t.Parallel() + + prefix, subnet, _ := prefixSubnetAddr(0, "") + + ndpDisp := ndpDispatcher{ + prefixC: make(chan ndpPrefixEvent, 1), + } + e := channel.New(0, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + HandleRAs: true, + DiscoverDefaultRouters: false, + DiscoverOnLinkPrefixes: true, + }, + NDPDisp: &ndpDisp, + }) + + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + // Receive an RA with prefix that we should not remember. + const lifetimeSeconds = 1 + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, lifetimeSeconds, 0)) + select { + case e := <-ndpDisp.prefixC: + if diff := checkPrefixEvent(e, subnet, true); diff != "" { + t.Errorf("prefix event mismatch (-want +got):\n%s", diff) + } + default: + t.Fatal("expected prefix discovery event") + } + + // Wait for the invalidation time plus some buffer to make sure we do + // not actually receive any invalidation events as we should not have + // remembered the prefix in the first place. + select { + case <-ndpDisp.prefixC: + t.Fatal("should not have received any prefix events") + case <-time.After(lifetimeSeconds*time.Second + defaultTimeout): + } +} + +func TestPrefixDiscovery(t *testing.T) { + t.Parallel() + + prefix1, subnet1, _ := prefixSubnetAddr(0, "") + prefix2, subnet2, _ := prefixSubnetAddr(1, "") + prefix3, subnet3, _ := prefixSubnetAddr(2, "") + + ndpDisp := ndpDispatcher{ + prefixC: make(chan ndpPrefixEvent, 1), + rememberPrefix: true, + } + e := channel.New(0, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + HandleRAs: true, + DiscoverOnLinkPrefixes: true, + }, + NDPDisp: &ndpDisp, + }) + + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + expectPrefixEvent := func(prefix tcpip.Subnet, discovered bool) { + t.Helper() + + select { + case e := <-ndpDisp.prefixC: + if diff := checkPrefixEvent(e, prefix, discovered); diff != "" { + t.Errorf("prefix event mismatch (-want +got):\n%s", diff) + } + default: + t.Fatal("expected prefix discovery event") + } + } + + // Receive an RA with prefix1 in an NDP Prefix Information option (PI) + // with zero valid lifetime. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, false, 0, 0)) + select { + case <-ndpDisp.prefixC: + t.Fatal("unexpectedly discovered a prefix with 0 lifetime") + default: + } + + // Receive an RA with prefix1 in an NDP Prefix Information option (PI) + // with non-zero lifetime. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, false, 100, 0)) + expectPrefixEvent(subnet1, true) + + // Receive an RA with prefix2 in a PI. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix2, true, false, 100, 0)) + expectPrefixEvent(subnet2, true) + + // Receive an RA with prefix3 in a PI. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix3, true, false, 100, 0)) + expectPrefixEvent(subnet3, true) + + // Receive an RA with prefix1 in a PI with lifetime = 0. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, false, 0, 0)) + expectPrefixEvent(subnet1, false) + + // Receive an RA with prefix2 in a PI with lesser lifetime. + lifetime := uint32(2) + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix2, true, false, lifetime, 0)) + select { + case <-ndpDisp.prefixC: + t.Fatal("unexpectedly received prefix event when updating lifetime") + default: + } + + // Wait for prefix2's most recent invalidation timer plus some buffer to + // expire. + select { + case e := <-ndpDisp.prefixC: + if diff := checkPrefixEvent(e, subnet2, false); diff != "" { + t.Errorf("prefix event mismatch (-want +got):\n%s", diff) + } + case <-time.After(time.Duration(lifetime)*time.Second + defaultTimeout): + t.Fatal("timed out waiting for prefix discovery event") + } + + // Receive RA to invalidate prefix3. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix3, true, false, 0, 0)) + expectPrefixEvent(subnet3, false) +} + +func TestPrefixDiscoveryWithInfiniteLifetime(t *testing.T) { + // Update the infinite lifetime value to a smaller value so we can test + // that when we receive a PI with such a lifetime value, we do not + // invalidate the prefix. + const testInfiniteLifetimeSeconds = 2 + const testInfiniteLifetime = testInfiniteLifetimeSeconds * time.Second + saved := header.NDPInfiniteLifetime + header.NDPInfiniteLifetime = testInfiniteLifetime + defer func() { + header.NDPInfiniteLifetime = saved + }() + + prefix := tcpip.AddressWithPrefix{ + Address: tcpip.Address("\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x00"), + PrefixLen: 64, + } + subnet := prefix.Subnet() + + ndpDisp := ndpDispatcher{ + prefixC: make(chan ndpPrefixEvent, 1), + rememberPrefix: true, + } + e := channel.New(0, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + HandleRAs: true, + DiscoverOnLinkPrefixes: true, + }, + NDPDisp: &ndpDisp, + }) + + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + expectPrefixEvent := func(prefix tcpip.Subnet, discovered bool) { + t.Helper() + + select { + case e := <-ndpDisp.prefixC: + if diff := checkPrefixEvent(e, prefix, discovered); diff != "" { + t.Errorf("prefix event mismatch (-want +got):\n%s", diff) + } + default: + t.Fatal("expected prefix discovery event") + } + } + + // Receive an RA with prefix in an NDP Prefix Information option (PI) + // with infinite valid lifetime which should not get invalidated. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, testInfiniteLifetimeSeconds, 0)) + expectPrefixEvent(subnet, true) + select { + case <-ndpDisp.prefixC: + t.Fatal("unexpectedly invalidated a prefix with infinite lifetime") + case <-time.After(testInfiniteLifetime + defaultTimeout): + } + + // Receive an RA with finite lifetime. + // The prefix should get invalidated after 1s. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, testInfiniteLifetimeSeconds-1, 0)) + select { + case e := <-ndpDisp.prefixC: + if diff := checkPrefixEvent(e, subnet, false); diff != "" { + t.Errorf("prefix event mismatch (-want +got):\n%s", diff) + } + case <-time.After(testInfiniteLifetime): + t.Fatal("timed out waiting for prefix discovery event") + } + + // Receive an RA with finite lifetime. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, testInfiniteLifetimeSeconds-1, 0)) + expectPrefixEvent(subnet, true) + + // Receive an RA with prefix with an infinite lifetime. + // The prefix should not be invalidated. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, testInfiniteLifetimeSeconds, 0)) + select { + case <-ndpDisp.prefixC: + t.Fatal("unexpectedly invalidated a prefix with infinite lifetime") + case <-time.After(testInfiniteLifetime + defaultTimeout): + } + + // Receive an RA with a prefix with a lifetime value greater than the + // set infinite lifetime value. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, testInfiniteLifetimeSeconds+1, 0)) + select { + case <-ndpDisp.prefixC: + t.Fatal("unexpectedly invalidated a prefix with infinite lifetime") + case <-time.After((testInfiniteLifetimeSeconds+1)*time.Second + defaultTimeout): + } + + // Receive an RA with 0 lifetime. + // The prefix should get invalidated. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, 0, 0)) + expectPrefixEvent(subnet, false) +} + +// TestPrefixDiscoveryMaxRouters tests that only +// stack.MaxDiscoveredOnLinkPrefixes discovered on-link prefixes are remembered. +func TestPrefixDiscoveryMaxOnLinkPrefixes(t *testing.T) { + t.Parallel() + + ndpDisp := ndpDispatcher{ + prefixC: make(chan ndpPrefixEvent, stack.MaxDiscoveredOnLinkPrefixes+3), + rememberPrefix: true, + } + e := channel.New(0, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + HandleRAs: true, + DiscoverDefaultRouters: false, + DiscoverOnLinkPrefixes: true, + }, + NDPDisp: &ndpDisp, + }) + + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + optSer := make(header.NDPOptionsSerializer, stack.MaxDiscoveredOnLinkPrefixes+2) + prefixes := [stack.MaxDiscoveredOnLinkPrefixes + 2]tcpip.Subnet{} + + // Receive an RA with 2 more than the max number of discovered on-link + // prefixes. + for i := 0; i < stack.MaxDiscoveredOnLinkPrefixes+2; i++ { + prefixAddr := [16]byte{1, 2, 3, 4, 5, 6, 7, 8, 0, 0, 0, 0, 0, 0, 0, 0} + prefixAddr[7] = byte(i) + prefix := tcpip.AddressWithPrefix{ + Address: tcpip.Address(prefixAddr[:]), + PrefixLen: 64, + } + prefixes[i] = prefix.Subnet() + buf := [30]byte{} + buf[0] = uint8(prefix.PrefixLen) + buf[1] = 128 + binary.BigEndian.PutUint32(buf[2:], 10) + copy(buf[14:], prefix.Address) + + optSer[i] = header.NDPPrefixInformation(buf[:]) + } + + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithOpts(llAddr1, 0, optSer)) + for i := 0; i < stack.MaxDiscoveredOnLinkPrefixes+2; i++ { + if i < stack.MaxDiscoveredOnLinkPrefixes { + select { + case e := <-ndpDisp.prefixC: + if diff := checkPrefixEvent(e, prefixes[i], true); diff != "" { + t.Errorf("prefix event mismatch (-want +got):\n%s", diff) + } + default: + t.Fatal("expected prefix discovery event") + } + } else { + select { + case <-ndpDisp.prefixC: + t.Fatal("should not have discovered a new prefix after we already discovered the max number of prefixes") + default: + } + } + } +} + +// Checks to see if list contains an IPv6 address, item. +func contains(list []tcpip.ProtocolAddress, item tcpip.AddressWithPrefix) bool { + protocolAddress := tcpip.ProtocolAddress{ + Protocol: header.IPv6ProtocolNumber, + AddressWithPrefix: item, + } + + for _, i := range list { + if i == protocolAddress { + return true + } + } + + return false +} + +// TestNoAutoGenAddr tests that SLAAC is not performed when configured not to. +func TestNoAutoGenAddr(t *testing.T) { + prefix, _, _ := prefixSubnetAddr(0, "") + + // Being configured to auto-generate addresses means handle and + // autogen 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, autogen = + // true and forwarding = false (the required configuration to do + // SLAAC) - that will done in other tests. + for i := 0; i < 7; i++ { + handle := i&1 != 0 + autogen := i&2 != 0 + forwarding := i&4 == 0 + + t.Run(fmt.Sprintf("HandleRAs(%t), AutoGenAddr(%t), Forwarding(%t)", handle, autogen, forwarding), func(t *testing.T) { + t.Parallel() + + ndpDisp := ndpDispatcher{ + autoGenAddrC: make(chan ndpAutoGenAddrEvent, 1), + } + e := channel.New(0, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + HandleRAs: handle, + AutoGenGlobalAddresses: autogen, + }, + NDPDisp: &ndpDisp, + }) + s.SetForwarding(forwarding) + + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + // Rx an RA with prefix with non-zero lifetime. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, false, true, 10, 0)) + + select { + case <-ndpDisp.autoGenAddrC: + t.Fatal("unexpectedly auto-generated an address when configured not to") + default: + } + }) + } +} + +// Check e to make sure that the event is for addr on nic with ID 1, and the +// event type is set to eventType. +func checkAutoGenAddrEvent(e ndpAutoGenAddrEvent, addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType) string { + return cmp.Diff(ndpAutoGenAddrEvent{nicID: 1, addr: addr, eventType: eventType}, e, cmp.AllowUnexported(e)) +} + +// TestAutoGenAddr tests that an address is properly generated and invalidated +// when configured to do so. +func TestAutoGenAddr(t *testing.T) { + const newMinVL = 2 + newMinVLDuration := newMinVL * time.Second + saved := stack.MinPrefixInformationValidLifetimeForUpdate + defer func() { + stack.MinPrefixInformationValidLifetimeForUpdate = saved + }() + stack.MinPrefixInformationValidLifetimeForUpdate = newMinVLDuration + + prefix1, _, addr1 := prefixSubnetAddr(0, linkAddr1) + prefix2, _, addr2 := prefixSubnetAddr(1, linkAddr1) + + ndpDisp := ndpDispatcher{ + autoGenAddrC: make(chan ndpAutoGenAddrEvent, 1), + } + e := channel.New(0, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + }) + + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + expectAutoGenAddrEvent := func(addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType) { + t.Helper() + + select { + case e := <-ndpDisp.autoGenAddrC: + if diff := checkAutoGenAddrEvent(e, addr, eventType); diff != "" { + t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff) + } + default: + t.Fatal("expected addr auto gen event") + } + } + + // Receive an RA with prefix1 in an NDP Prefix Information option (PI) + // with zero valid lifetime. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, 0, 0)) + select { + case <-ndpDisp.autoGenAddrC: + t.Fatal("unexpectedly auto-generated an address with 0 lifetime") + default: + } + + // Receive an RA with prefix1 in an NDP Prefix Information option (PI) + // with non-zero lifetime. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, 100, 0)) + expectAutoGenAddrEvent(addr1, newAddr) + if !contains(s.NICInfo()[1].ProtocolAddresses, addr1) { + t.Fatalf("Should have %s in the list of addresses", addr1) + } + + // Receive an RA with prefix2 in an NDP Prefix Information option (PI) + // with preferred lifetime > valid lifetime + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, 5, 6)) + select { + case <-ndpDisp.autoGenAddrC: + t.Fatal("unexpectedly auto-generated an address with preferred lifetime > valid lifetime") + default: + } + + // Receive an RA with prefix2 in a PI. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix2, true, true, 100, 0)) + expectAutoGenAddrEvent(addr2, newAddr) + if !contains(s.NICInfo()[1].ProtocolAddresses, addr1) { + t.Fatalf("Should have %s in the list of addresses", addr1) + } + if !contains(s.NICInfo()[1].ProtocolAddresses, addr2) { + t.Fatalf("Should have %s in the list of addresses", addr2) + } + + // Refresh valid lifetime for addr of prefix1. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, newMinVL, 0)) + select { + case <-ndpDisp.autoGenAddrC: + t.Fatal("unexpectedly auto-generated an address when we already have an address for a prefix") + default: + } + + // Wait for addr of prefix1 to be invalidated. + select { + case e := <-ndpDisp.autoGenAddrC: + if diff := checkAutoGenAddrEvent(e, addr1, invalidatedAddr); diff != "" { + t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff) + } + case <-time.After(newMinVLDuration + defaultTimeout): + t.Fatal("timed out waiting for addr auto gen event") + } + if contains(s.NICInfo()[1].ProtocolAddresses, addr1) { + t.Fatalf("Should not have %s in the list of addresses", addr1) + } + if !contains(s.NICInfo()[1].ProtocolAddresses, addr2) { + t.Fatalf("Should have %s in the list of addresses", addr2) + } +} + +// TestAutoGenAddrValidLifetimeUpdates tests that the valid lifetime of an +// auto-generated address only gets updated when required to, as specified in +// RFC 4862 section 5.5.3.e. +func TestAutoGenAddrValidLifetimeUpdates(t *testing.T) { + const infiniteVL = 4294967295 + const newMinVL = 5 + saved := stack.MinPrefixInformationValidLifetimeForUpdate + defer func() { + stack.MinPrefixInformationValidLifetimeForUpdate = saved + }() + stack.MinPrefixInformationValidLifetimeForUpdate = newMinVL * time.Second + + prefix, _, addr := prefixSubnetAddr(0, linkAddr1) + + tests := []struct { + name string + ovl uint32 + nvl uint32 + evl uint32 + }{ + // Should update the VL to the minimum VL for updating if the + // new VL is less than newMinVL but was originally greater than + // it. + { + "LargeVLToVLLessThanMinVLForUpdate", + 9999, + 1, + newMinVL, + }, + { + "LargeVLTo0", + 9999, + 0, + newMinVL, + }, + { + "InfiniteVLToVLLessThanMinVLForUpdate", + infiniteVL, + 1, + newMinVL, + }, + { + "InfiniteVLTo0", + infiniteVL, + 0, + newMinVL, + }, + + // Should not update VL if original VL was less than newMinVL + // and the new VL is also less than newMinVL. + { + "ShouldNotUpdateWhenBothOldAndNewAreLessThanMinVLForUpdate", + newMinVL - 1, + newMinVL - 3, + newMinVL - 1, + }, + + // Should take the new VL if the new VL is greater than the + // remaining time or is greater than newMinVL. + { + "MorethanMinVLToLesserButStillMoreThanMinVLForUpdate", + newMinVL + 5, + newMinVL + 3, + newMinVL + 3, + }, + { + "SmallVLToGreaterVLButStillLessThanMinVLForUpdate", + newMinVL - 3, + newMinVL - 1, + newMinVL - 1, + }, + { + "SmallVLToGreaterVLThatIsMoreThaMinVLForUpdate", + newMinVL - 3, + newMinVL + 1, + newMinVL + 1, + }, + } + + const delta = 500 * time.Millisecond + + // This Run will not return until the parallel tests finish. + // + // We need this because we need to do some teardown work after the + // parallel tests complete. + // + // See https://godoc.org/testing#hdr-Subtests_and_Sub_benchmarks for + // more details. + t.Run("group", func(t *testing.T) { + for _, test := range tests { + test := test + + t.Run(test.name, func(t *testing.T) { + t.Parallel() + + ndpDisp := ndpDispatcher{ + autoGenAddrC: make(chan ndpAutoGenAddrEvent, 10), + } + e := channel.New(10, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + }) + + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + // Receive an RA with prefix with initial VL, + // test.ovl. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, test.ovl, 0)) + select { + case e := <-ndpDisp.autoGenAddrC: + if diff := checkAutoGenAddrEvent(e, addr, newAddr); diff != "" { + t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff) + } + default: + t.Fatal("expected addr auto gen event") + } + + // Receive an new RA with prefix with new VL, + // test.nvl. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, test.nvl, 0)) + + // + // Validate that the VL for the address got set + // to test.evl. + // + + // Make sure we do not get any invalidation + // events until atleast 500ms (delta) before + // test.evl. + select { + case <-ndpDisp.autoGenAddrC: + t.Fatalf("unexpectedly received an auto gen addr event") + case <-time.After(time.Duration(test.evl)*time.Second - delta): + } + + // Wait for another second (2x delta), but now + // we expect the invalidation event. + select { + case e := <-ndpDisp.autoGenAddrC: + if diff := checkAutoGenAddrEvent(e, addr, invalidatedAddr); diff != "" { + t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff) + } + + case <-time.After(2 * delta): + t.Fatal("timeout waiting for addr auto gen event") + } + }) + } + }) +} + +// TestAutoGenAddrRemoval tests that when auto-generated addresses are removed +// by the user, its resources will be cleaned up and an invalidation event will +// be sent to the integrator. +func TestAutoGenAddrRemoval(t *testing.T) { + t.Parallel() + + prefix, _, addr := prefixSubnetAddr(0, linkAddr1) + + ndpDisp := ndpDispatcher{ + autoGenAddrC: make(chan ndpAutoGenAddrEvent, 1), + } + e := channel.New(0, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + }) + + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + expectAutoGenAddrEvent := func(addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType) { + t.Helper() + + select { + case e := <-ndpDisp.autoGenAddrC: + if diff := checkAutoGenAddrEvent(e, addr, eventType); diff != "" { + t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff) + } + default: + t.Fatal("expected addr auto gen event") + } + } + + // Receive a PI to auto-generate an address. + const lifetimeSeconds = 1 + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, lifetimeSeconds, 0)) + expectAutoGenAddrEvent(addr, newAddr) + + // Removing the address should result in an invalidation event + // immediately. + if err := s.RemoveAddress(1, addr.Address); err != nil { + t.Fatalf("RemoveAddress(_, %s) = %s", addr.Address, err) + } + expectAutoGenAddrEvent(addr, invalidatedAddr) + + // Wait for the original valid lifetime to make sure the original timer + // got stopped/cleaned up. + select { + case <-ndpDisp.autoGenAddrC: + t.Fatalf("unexpectedly received an auto gen addr event") + case <-time.After(lifetimeSeconds*time.Second + defaultTimeout): + } +} + +// TestAutoGenAddrStaticConflict tests that if SLAAC generates an address that +// is already assigned to the NIC, the static address remains. +func TestAutoGenAddrStaticConflict(t *testing.T) { + t.Parallel() + + prefix, _, addr := prefixSubnetAddr(0, linkAddr1) + + ndpDisp := ndpDispatcher{ + autoGenAddrC: make(chan ndpAutoGenAddrEvent, 1), + } + e := channel.New(0, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + }) + + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + // Add the address as a static address before SLAAC tries to add it. + if err := s.AddProtocolAddress(1, tcpip.ProtocolAddress{Protocol: header.IPv6ProtocolNumber, AddressWithPrefix: addr}); err != nil { + t.Fatalf("AddAddress(_, %d, %s) = %s", header.IPv6ProtocolNumber, addr.Address, err) + } + if !contains(s.NICInfo()[1].ProtocolAddresses, addr) { + t.Fatalf("Should have %s in the list of addresses", addr1) + } + + // Receive a PI where the generated address will be the same as the one + // that we already have assigned statically. + const lifetimeSeconds = 1 + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, lifetimeSeconds, 0)) + select { + case <-ndpDisp.autoGenAddrC: + t.Fatal("unexpectedly received an auto gen addr event for an address we already have statically") + default: + } + if !contains(s.NICInfo()[1].ProtocolAddresses, addr) { + t.Fatalf("Should have %s in the list of addresses", addr1) + } + + // Should not get an invalidation event after the PI's invalidation + // time. + select { + case <-ndpDisp.autoGenAddrC: + t.Fatal("unexpectedly received an auto gen addr event") + case <-time.After(lifetimeSeconds*time.Second + defaultTimeout): + } + if !contains(s.NICInfo()[1].ProtocolAddresses, addr) { + t.Fatalf("Should have %s in the list of addresses", addr1) + } +} + +// TestNDPRecursiveDNSServerDispatch tests that we properly dispatch an event +// to the integrator when an RA is received with the NDP Recursive DNS Server +// option with at least one valid address. +func TestNDPRecursiveDNSServerDispatch(t *testing.T) { + t.Parallel() + + tests := []struct { + name string + opt header.NDPRecursiveDNSServer + expected *ndpRDNSS + }{ + { + "Unspecified", + header.NDPRecursiveDNSServer([]byte{ + 0, 0, + 0, 0, 0, 2, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + }), + nil, + }, + { + "Multicast", + header.NDPRecursiveDNSServer([]byte{ + 0, 0, + 0, 0, 0, 2, + 255, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, + }), + nil, + }, + { + "OptionTooSmall", + header.NDPRecursiveDNSServer([]byte{ + 0, 0, + 0, 0, 0, 2, + 1, 2, 3, 4, 5, 6, 7, 8, + }), + nil, + }, + { + "0Addresses", + header.NDPRecursiveDNSServer([]byte{ + 0, 0, + 0, 0, 0, 2, + }), + nil, + }, + { + "Valid1Address", + header.NDPRecursiveDNSServer([]byte{ + 0, 0, + 0, 0, 0, 2, + 1, 2, 3, 4, 5, 6, 7, 8, 0, 0, 0, 0, 0, 0, 0, 1, + }), + &ndpRDNSS{ + []tcpip.Address{ + "\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x01", + }, + 2 * time.Second, + }, + }, + { + "Valid2Addresses", + header.NDPRecursiveDNSServer([]byte{ + 0, 0, + 0, 0, 0, 1, + 1, 2, 3, 4, 5, 6, 7, 8, 0, 0, 0, 0, 0, 0, 0, 1, + 1, 2, 3, 4, 5, 6, 7, 8, 0, 0, 0, 0, 0, 0, 0, 2, + }), + &ndpRDNSS{ + []tcpip.Address{ + "\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x01", + "\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x02", + }, + time.Second, + }, + }, + { + "Valid3Addresses", + header.NDPRecursiveDNSServer([]byte{ + 0, 0, + 0, 0, 0, 0, + 1, 2, 3, 4, 5, 6, 7, 8, 0, 0, 0, 0, 0, 0, 0, 1, + 1, 2, 3, 4, 5, 6, 7, 8, 0, 0, 0, 0, 0, 0, 0, 2, + 1, 2, 3, 4, 5, 6, 7, 8, 0, 0, 0, 0, 0, 0, 0, 3, + }), + &ndpRDNSS{ + []tcpip.Address{ + "\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x01", + "\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x02", + "\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x03", + }, + 0, + }, + }, + } + + for _, test := range tests { + test := test + + t.Run(test.name, func(t *testing.T) { + t.Parallel() + + ndpDisp := ndpDispatcher{ + // We do not expect more than a single RDNSS + // event at any time for this test. + rdnssC: make(chan ndpRDNSSEvent, 1), + } + e := channel.New(0, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + HandleRAs: true, + }, + NDPDisp: &ndpDisp, + }) + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithOpts(llAddr1, 0, header.NDPOptionsSerializer{test.opt})) + + if test.expected != nil { + select { + case e := <-ndpDisp.rdnssC: + if e.nicID != 1 { + t.Errorf("got rdnss nicID = %d, want = 1", e.nicID) + } + if diff := cmp.Diff(e.rdnss.addrs, test.expected.addrs); diff != "" { + t.Errorf("rdnss addrs mismatch (-want +got):\n%s", diff) + } + if e.rdnss.lifetime != test.expected.lifetime { + t.Errorf("got rdnss lifetime = %s, want = %s", e.rdnss.lifetime, test.expected.lifetime) + } + default: + t.Fatal("expected an RDNSS option event") + } + } + + // Should have no more RDNSS options. + select { + case e := <-ndpDisp.rdnssC: + t.Fatalf("unexpectedly got a new RDNSS option event: %+v", e) + default: + } + }) + } +} diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 43f4ad91e..e8401c673 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" @@ -37,13 +36,20 @@ type NIC struct { 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 } @@ -78,16 +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, - 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{}, @@ -99,9 +115,24 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, loopback }, }, ndp: ndpState{ - dad: make(map[tcpip.Address]dadState), + configs: stack.ndpConfigs, + dad: make(map[tcpip.Address]dadState), + defaultRouters: make(map[tcpip.Address]defaultRouterState), + onLinkPrefixes: make(map[tcpip.Subnet]onLinkPrefixState), + autoGenAddresses: make(map[tcpip.Address]autoGenAddressState), }, } + 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 @@ -126,11 +157,50 @@ 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 @@ -166,18 +236,7 @@ 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 } @@ -186,6 +245,20 @@ func (n *NIC) primaryEndpoint(protocol tcpip.NetworkProtocolNumber) *referencedN return nil } +// hasPermanentAddrLocked returns true if n has a permanent (including currently +// tentative) address, addr. +func (n *NIC) hasPermanentAddrLocked(addr tcpip.Address) bool { + ref, ok := n.endpoints[NetworkEndpointID{addr}] + + if !ok { + return false + } + + kind := ref.getKind() + + return kind == permanent || kind == permanentTentative +} + func (n *NIC) getRef(protocol tcpip.NetworkProtocolNumber, dst tcpip.Address) *referencedNetworkEndpoint { return n.getRefOrCreateTemp(protocol, dst, CanBePrimaryEndpoint, n.promiscuous) } @@ -277,7 +350,7 @@ func (n *NIC) getRefOrCreateTemp(protocol tcpip.NetworkProtocolNumber, address t Address: address, PrefixLen: netProto.DefaultPrefixLen(), }, - }, peb, temporary) + }, peb, temporary, static) n.mu.Unlock() return ref @@ -291,9 +364,31 @@ func (n *NIC) addPermanentAddressLocked(protocolAddress tcpip.ProtocolAddress, p // 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 @@ -304,10 +399,10 @@ func (n *NIC) addPermanentAddressLocked(protocolAddress tcpip.ProtocolAddress, p } } - return n.addAddressLocked(protocolAddress, peb, permanent) + return n.addAddressLocked(protocolAddress, peb, permanent, static) } -func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior, kind networkEndpointKind) (*referencedNetworkEndpoint, *tcpip.Error) { +func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior, kind networkEndpointKind, configType networkEndpointConfigType) (*referencedNetworkEndpoint, *tcpip.Error) { // TODO(b/141022673): Validate IP address before adding them. // Sanity check. @@ -337,11 +432,12 @@ func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb Primar } ref := &referencedNetworkEndpoint{ - refs: 1, - ep: ep, - nic: n, - protocol: protocolAddress.Protocol, - kind: kind, + refs: 1, + ep: ep, + nic: n, + protocol: protocolAddress.Protocol, + kind: kind, + configType: configType, } // Set up cache if link address resolution exists for this protocol. @@ -362,22 +458,11 @@ 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 - } - - switch peb { - case CanBePrimaryEndpoint: - l.PushBack(ref) - case FirstPrimaryEndpoint: - l.PushFront(ref) - } + n.insertPrimaryEndpointLocked(ref, peb) // If we are adding a tentative IPv6 address, start DAD. if isIPv6Unicast && kind == permanentTentative { - if err := n.ndp.startDuplicateAddressDetection(n, protocolAddress.AddressWithPrefix.Address, ref); err != nil { + if err := n.ndp.startDuplicateAddressDetection(protocolAddress.AddressWithPrefix.Address, ref); err != nil { return nil, err } } @@ -430,8 +515,7 @@ 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) + for _, ref := range list { // Don't include tentative, expired or tempory endpoints // to avoid confusion and prevent the caller from using // those. @@ -496,6 +580,19 @@ 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() @@ -513,9 +610,12 @@ 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() @@ -540,9 +640,18 @@ func (n *NIC) removePermanentAddressLocked(addr tcpip.Address) *tcpip.Error { 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) + if isIPv6Unicast { + // If we are removing a tentative IPv6 unicast address, stop + // DAD. + if kind == permanentTentative { + n.ndp.stopDuplicateAddressDetection(addr) + } + + // If we are removing an address generated via SLAAC, cleanup + // its SLAAC resources and notify the integrator. + if r.configType == slaac { + n.ndp.cleanupAutoGenAddrResourcesAndNotify(addr) + } } r.setKind(permanentExpired) @@ -585,6 +694,11 @@ 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 { @@ -635,10 +749,10 @@ 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() } @@ -648,9 +762,9 @@ 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 { @@ -658,19 +772,39 @@ 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()) + 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 } @@ -698,31 +832,34 @@ 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() + pkt.Header = buffer.NewPrependableFromView(pkt.Data.First()) + pkt.Data.RemoveFirst() // TODO(b/128629022): use route.WritePacket. - if err := n.linkEP.WritePacket(&r, nil /* gso */, hdr, vv, protocol); err != nil { + if err := n.linkEP.WritePacket(&r, nil /* gso */, protocol, pkt); err != nil { r.Stats().IP.OutgoingPacketErrors.Increment() } else { n.stats.Tx.Packets.Increment() - n.stats.Tx.Bytes.IncrementBy(uint64(hdr.UsedLength() + vv.Size())) + n.stats.Tx.Bytes.IncrementBy(uint64(pkt.Header.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() @@ -734,41 +871,41 @@ 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. - 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.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 @@ -779,17 +916,17 @@ 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.stack.demux.deliverControlPacket(n, net, trans, typ, extra, vv, id) { + if n.stack.demux.deliverControlPacket(n, net, trans, typ, extra, pkt, id) { return } } @@ -838,6 +975,26 @@ func (n *NIC) dupTentativeAddrDetected(addr tcpip.Address) *tcpip.Error { 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 ( @@ -857,7 +1014,7 @@ const ( // removing the permanent address from the NIC. permanent - // An expired permanent endoint is a permanent endoint that had its address + // An expired permanent endpoint is a permanent endpoint that had its address // removed from the NIC, and it is waiting to be removed once no more routes // hold a reference to it. This is achieved by decreasing its reference count // by 1. If its address is re-added before the endpoint is removed, its type @@ -873,8 +1030,50 @@ 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 networkEndpointConfigType int32 + +const ( + // A statically configured endpoint is an address that was added by + // some user-specified action (adding an explicit address, joining a + // multicast group). + static networkEndpointConfigType = iota + + // A slaac configured endpoint is an IPv6 endpoint that was + // added by SLAAC as per RFC 4862 section 5.5.3. + slaac +) + type referencedNetworkEndpoint struct { - ilist.Entry ep NetworkEndpoint nic *NIC protocol tcpip.NetworkProtocolNumber @@ -889,6 +1088,10 @@ type referencedNetworkEndpoint struct { // networkEndpointKind must only be accessed using {get,set}Kind(). kind networkEndpointKind + + // configType is the method that was used to configure this endpoint. + // This must never change after the endpoint is added to a NIC. + configType networkEndpointConfigType } func (r *referencedNetworkEndpoint) getKind() networkEndpointKind { diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go index 9d6157f22..61fd46d66 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. @@ -182,12 +232,17 @@ type NetworkEndpoint interface { MaxHeaderLength() uint16 // WritePacket writes a packet to the given destination address and - // protocol. - WritePacket(r *Route, gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params NetworkHeaderParams, loop PacketLooping) *tcpip.Error + // protocol. It sets pkt.NetworkHeader. pkt.TransportHeader must have + // already been set. + WritePacket(r *Route, gso *GSO, params NetworkHeaderParams, loop PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error + + // WritePackets writes packets to the given destination address and + // protocol. pkts must not be zero length. + WritePackets(r *Route, gso *GSO, pkts []tcpip.PacketBuffer, params NetworkHeaderParams, loop PacketLooping) (int, *tcpip.Error) // WriteHeaderIncludedPacket writes a packet that includes a network // header to the given destination address. - WriteHeaderIncludedPacket(r *Route, payload buffer.VectorisedView, loop PacketLooping) *tcpip.Error + WriteHeaderIncludedPacket(r *Route, loop PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error // ID returns the network protocol endpoint ID. ID() *NetworkEndpointID @@ -199,8 +254,10 @@ 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() @@ -225,7 +282,7 @@ 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 @@ -242,9 +299,15 @@ 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 @@ -266,12 +329,18 @@ 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 @@ -293,13 +362,27 @@ type LinkEndpoint interface { // link endpoint. LinkAddress() tcpip.LinkAddress - // WritePacket writes a packet with the given protocol through the given - // route. + // WritePacket writes a packet with the given protocol through the + // given route. It sets pkt.LinkHeader if a link layer header exists. + // pkt.NetworkHeader and pkt.TransportHeader must have already been + // set. // // To participate in transparent bridging, a LinkEndpoint implementation // should call eth.Encode with header.EthernetFields.SrcAddr set to // r.LocalLinkAddress if it is provided. - WritePacket(r *Route, gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error + WritePacket(r *Route, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error + + // WritePackets writes packets with the given protocol through the + // given route. pkts must not be zero length. + // + // Right now, WritePackets is used only when the software segmentation + // offload is enabled. If it will be used for something else, it may + // require to change syscall filters. + WritePackets(r *Route, gso *GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) + + // WriteRawPacket writes a packet directly to the link. The packet + // should already have an ethernet header. + WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error // Attach attaches the data link layer endpoint to the network-layer // dispatcher of the stack. @@ -324,13 +407,14 @@ 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 @@ -359,10 +443,10 @@ 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 @@ -373,17 +457,22 @@ 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. @@ -394,8 +483,14 @@ 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. @@ -423,3 +518,7 @@ 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 e72373964..34307ae07 100644 --- a/pkg/tcpip/stack/route.go +++ b/pkg/tcpip/stack/route.go @@ -17,7 +17,6 @@ package stack import ( "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" ) @@ -47,8 +46,8 @@ 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 @@ -69,7 +68,7 @@ func makeRoute(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr tcpip LocalLinkAddress: localLinkAddr, RemoteAddress: remoteAddr, ref: ref, - loop: loop, + Loop: loop, } } @@ -154,34 +153,54 @@ func (r *Route) IsResolutionRequired() bool { } // WritePacket writes the packet through the given route. -func (r *Route) WritePacket(gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params NetworkHeaderParams) *tcpip.Error { +func (r *Route) WritePacket(gso *GSO, params NetworkHeaderParams, pkt tcpip.PacketBuffer) *tcpip.Error { if !r.ref.isValidForOutgoing() { return tcpip.ErrInvalidEndpointState } - err := r.ref.ep.WritePacket(r, gso, hdr, payload, params, r.loop) + err := r.ref.ep.WritePacket(r, gso, params, r.Loop, pkt) if err != nil { r.Stats().IP.OutgoingPacketErrors.Increment() } else { r.ref.nic.stats.Tx.Packets.Increment() - r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(hdr.UsedLength() + payload.Size())) + r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(pkt.Header.UsedLength() + pkt.Data.Size())) } return err } +// WritePackets writes the set of packets through the given route. +func (r *Route) WritePackets(gso *GSO, pkts []tcpip.PacketBuffer, params NetworkHeaderParams) (int, *tcpip.Error) { + if !r.ref.isValidForOutgoing() { + return 0, tcpip.ErrInvalidEndpointState + } + + n, err := r.ref.ep.WritePackets(r, gso, pkts, params, r.Loop) + if err != nil { + r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(len(pkts) - n)) + } + r.ref.nic.stats.Tx.Packets.IncrementBy(uint64(n)) + payloadSize := 0 + for i := 0; i < n; i++ { + r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(pkts[i].Header.UsedLength())) + payloadSize += pkts[i].DataSize + } + 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 { +func (r *Route) WriteHeaderIncludedPacket(pkt tcpip.PacketBuffer) *tcpip.Error { if !r.ref.isValidForOutgoing() { return tcpip.ErrInvalidEndpointState } - if err := r.ref.ep.WriteHeaderIncludedPacket(r, payload, r.loop); err != nil { + if err := r.ref.ep.WriteHeaderIncludedPacket(r, r.Loop, pkt); err != nil { r.Stats().IP.OutgoingPacketErrors.Increment() return err } r.ref.nic.stats.Tx.Packets.Increment() - r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(payload.Size())) + r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(pkt.Data.Size())) return nil } diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index a199bc1cc..0e88643a4 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -22,6 +22,7 @@ package stack import ( "encoding/binary" "sync" + "sync/atomic" "time" "golang.org/x/time/rate" @@ -50,7 +51,7 @@ const ( 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 @@ -344,6 +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 { @@ -351,10 +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 @@ -362,9 +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 @@ -394,14 +402,31 @@ type Stack struct { // by the stack. icmpRateLimiter *ICMPRateLimiter - // portSeed is a one-time random value initialized at stack startup + // 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. - portSeed uint32 + seed uint32 - // ndpConfigs is the NDP configurations used by interfaces. + // 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. @@ -425,16 +450,35 @@ 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 NDP configurations used by interfaces. + // 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 @@ -481,22 +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(), - portSeed: generateRandUint32(), - ndpConfigs: opts.NDPConfigs, + 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. @@ -514,8 +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) @@ -523,6 +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 @@ -584,7 +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 @@ -650,12 +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] @@ -666,6 +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 { @@ -988,13 +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 } @@ -1053,35 +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) } } @@ -1100,6 +1167,31 @@ func (s *Stack) UnregisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip s.demux.unregisterEndpoint(netProtos, protocol, id, ep, bindToDevice) } +// 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() + + 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. @@ -1121,6 +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() { @@ -1135,6 +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) + vv := buffer.View(fakeHeader).ToVectorisedView() + vv.Append(payload) + + if err := nic.linkEP.WriteRawPacket(vv); 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. @@ -1286,12 +1544,47 @@ func (s *Stack) DupTentativeAddrDetected(id tcpip.NICID, addr tcpip.Address) *tc return nic.dupTentativeAddrDetected(addr) } -// PortSeed returns a 32 bit value that can be used as a seed value for port -// picking. +// 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) PortSeed() uint32 { - return s.portSeed +func (s *Stack) Seed() uint32 { + return s.seed } func generateRandUint32() uint32 { diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index 10fd1065f..8fc034ca1 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 { @@ -119,32 +122,38 @@ func (f *fakeNetworkEndpoint) Capabilities() stack.LinkEndpointCapabilities { return f.ep.Capabilities() } -func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) *tcpip.Error { +func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, loop stack.PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error { // Increment the sent packet count in the protocol descriptor. f.proto.sendPacketCount[int(r.RemoteAddress[0])%len(f.proto.sendPacketCount)]++ // Add the protocol's header to the packet and send it to the link // endpoint. - b := hdr.Prepend(fakeNetHeaderLen) + b := pkt.Header.Prepend(fakeNetHeaderLen) b[0] = r.RemoteAddress[0] b[1] = f.id.LocalAddress[0] b[2] = byte(params.Protocol) if loop&stack.PacketLoop != 0 { - views := make([]buffer.View, 1, 1+len(payload.Views())) - views[0] = hdr.View() - views = append(views, payload.Views()...) - vv := buffer.NewVectorisedView(len(views[0])+payload.Size(), views) - f.HandlePacket(r, vv) + views := make([]buffer.View, 1, 1+len(pkt.Data.Views())) + views[0] = pkt.Header.View() + views = append(views, pkt.Data.Views()...) + f.HandlePacket(r, tcpip.PacketBuffer{ + Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), + }) } if loop&stack.PacketOut == 0 { return nil } - return f.ep.WritePacket(r, gso, hdr, payload, fakeNetNumber) + return f.ep.WritePacket(r, gso, fakeNetNumber, pkt) +} + +// WritePackets implements stack.LinkEndpoint.WritePackets. +func (f *fakeNetworkEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) { + panic("not implemented") } -func (*fakeNetworkEndpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error { +func (*fakeNetworkEndpoint) WriteHeaderIncludedPacket(r *stack.Route, loop stack.PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error { return tcpip.ErrNotSupported } @@ -189,9 +198,9 @@ 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 +260,9 @@ 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 +272,9 @@ 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 +284,9 @@ 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 +295,9 @@ 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 +307,9 @@ 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 +329,10 @@ func sendTo(s *stack.Stack, addr tcpip.Address, payload buffer.View) *tcpip.Erro func send(r stack.Route, payload buffer.View) *tcpip.Error { hdr := buffer.NewPrependable(int(r.MaxHeaderLength())) - return r.WritePacket(nil /* gso */, hdr, payload.ToVectorisedView(), stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}) + return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + Data: payload.ToVectorisedView(), + }) } func testSendTo(t *testing.T, s *stack.Stack, addr tcpip.Address, ep *channel.Endpoint, payload buffer.View) { @@ -365,7 +387,9 @@ 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 +684,11 @@ 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 +721,7 @@ 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 +733,7 @@ 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 +750,13 @@ 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 +764,7 @@ 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 +779,20 @@ 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 +807,10 @@ 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 +828,10 @@ 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 +847,10 @@ 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 +858,17 @@ 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 { @@ -1637,12 +1661,12 @@ 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) } @@ -1650,7 +1674,7 @@ 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{ @@ -1659,17 +1683,17 @@ 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) } @@ -1686,24 +1710,24 @@ 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) } @@ -1714,7 +1738,7 @@ 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{ @@ -1724,17 +1748,17 @@ 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) } @@ -1753,7 +1777,7 @@ 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) @@ -1761,7 +1785,7 @@ func TestAddProtocolAddressWithOptions(t *testing.T) { } } - gotAddresses := s.AllAddresses()[nicid] + gotAddresses := s.AllAddresses()[nicID] verifyAddresses(t, expectedAddresses, gotAddresses) } @@ -1787,7 +1811,9 @@ 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) } @@ -1847,7 +1873,9 @@ 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: @@ -1864,3 +1892,297 @@ 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 92267ce4d..67c21be42 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" ) @@ -41,6 +41,31 @@ type transportEndpoints struct { rawEndpoints []RawTransportEndpoint } +// 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, bindToDevice tcpip.NICID) { + eps.mu.Lock() + defer eps.mu.Unlock() + epsByNic, ok := eps.endpoints[id] + if !ok { + return + } + 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 @@ -48,9 +73,19 @@ type endpointsByNic struct { 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, vv buffer.VectorisedView) { +func (epsByNic *endpointsByNic) handlePacket(r *Route, id TransportEndpointID, pkt tcpip.PacketBuffer) { epsByNic.mu.RLock() mpep, ok := epsByNic.endpoints[r.ref.nic.ID()] @@ -64,18 +99,17 @@ func (epsByNic *endpointsByNic) handlePacket(r *Route, id TransportEndpointID, v // 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, vv) + 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, vv) + 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, vv buffer.VectorisedView) { +func (epsByNic *endpointsByNic) handleControlPacket(n *NIC, id TransportEndpointID, typ ControlType, extra uint32, pkt tcpip.PacketBuffer) { epsByNic.mu.RLock() defer epsByNic.mu.RUnlock() @@ -91,7 +125,7 @@ func (epsByNic *endpointsByNic) handleControlPacket(n *NIC, id TransportEndpoint // 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, vv) + selectEndpoint(id, mpep, epsByNic.seed).HandleControlPacket(id, typ, extra, pkt) } // registerEndpoint returns true if it succeeds. It fails and returns @@ -127,21 +161,6 @@ func (epsByNic *endpointsByNic) unregisterEndpoint(bindToDevice tcpip.NICID, t T return len(epsByNic.endpoints) == 0 } -// 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, bindToDevice tcpip.NICID) { - eps.mu.Lock() - defer eps.mu.Unlock() - epsByNic, ok := eps.endpoints[id] - if !ok { - return - } - if !epsByNic.unregisterEndpoint(bindToDevice, ep) { - return - } - delete(eps.endpoints, id) -} - // transportDemuxer demultiplexes packets targeted at a transport endpoint // (i.e., after they've been parsed by the network layer). It does two levels // of demultiplexing: first based on the network and transport protocols, then @@ -183,14 +202,27 @@ func (d *transportDemuxer) registerEndpoint(netProtos []tcpip.NetworkProtocolNum // multiPortEndpoint is a container for TransportEndpoints which are bound to // the same pair of address and port. 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 // 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). // // This is similar to val % n, but faster. @@ -224,22 +256,40 @@ func selectEndpoint(id TransportEndpointID, mpep *multiPortEndpoint, seed uint32 return mpep.endpointsArr[idx] } -func (ep *multiPortEndpoint) handlePacketAll(r *Route, id TransportEndpointID, vv buffer.VectorisedView) { +func (ep *multiPortEndpoint) handlePacketAll(r *Route, id TransportEndpointID, pkt tcpip.PacketBuffer) { ep.mu.RLock() for i, endpoint := range ep.endpointsArr { - // HandlePacket modifies vv, so each endpoint needs its own copy except for - // the final one. + // 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, vv) + endpoint.HandlePacket(r, id, pkt) break } - vvCopy := buffer.NewView(vv.Size()) - copy(vvCopy, vv.ToView()) - endpoint.HandlePacket(r, id, vvCopy.ToVectorisedView()) + endpoint.HandlePacket(r, id, pkt.Clone()) } ep.mu.RUnlock() // Don't use defer for performance reasons. } +// 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() + } +} + +// 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 { @@ -257,6 +307,15 @@ func (ep *multiPortEndpoint) singleRegisterEndpoint(t TransportEndpoint, reusePo // 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 } @@ -330,9 +389,9 @@ 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 { +// then, if matches are found, delivers the packet to them. Returns true if +// the packet no longer needs to be handled. +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 @@ -341,13 +400,38 @@ func (d *transportDemuxer) deliverPacket(r *Route, protocol tcpip.TransportProto eps.mu.RLock() // 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. + // If the packet is a UDP broadcast or multicast, then find all matching + // transport endpoints. If the packet is a TCP packet with a non-unicast + // source or destination address, then do nothing further and instruct + // the caller to do the same. var destEps []*endpointsByNic - if protocol == header.UDPProtocolNumber && isMulticastOrBroadcast(id.LocalAddress) { - destEps = d.findAllEndpointsLocked(eps, vv, id) - } else if ep := d.findEndpointLocked(eps, vv, id); ep != nil { - destEps = append(destEps, ep) + switch protocol { + case header.UDPProtocolNumber: + if isMulticastOrBroadcast(id.LocalAddress) { + destEps = d.findAllEndpointsLocked(eps, id) + break + } + + if ep := d.findEndpointLocked(eps, id); ep != nil { + destEps = append(destEps, ep) + } + + case header.TCPProtocolNumber: + if !(isUnicast(r.LocalAddress) && isUnicast(r.RemoteAddress)) { + // TCP can only be used to communicate between a single + // source and a single destination; the addresses must + // be unicast. + eps.mu.RUnlock() + r.Stats().TCP.InvalidSegmentsReceived.Increment() + return true + } + + fallthrough + + default: + if ep := d.findEndpointLocked(eps, id); ep != nil { + destEps = append(destEps, ep) + } } eps.mu.RUnlock() @@ -361,17 +445,19 @@ 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 @@ -385,7 +471,7 @@ 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() @@ -395,7 +481,7 @@ 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(n *NIC, 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 @@ -403,7 +489,7 @@ func (d *transportDemuxer) deliverControlPacket(n *NIC, net tcpip.NetworkProtoco // 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. @@ -412,12 +498,12 @@ func (d *transportDemuxer) deliverControlPacket(n *NIC, net tcpip.NetworkProtoco } // Deliver the packet. - ep.handleControlPacket(n, id, typ, extra, vv) + ep.handleControlPacket(n, id, typ, extra, pkt) return true } -func (d *transportDemuxer) findAllEndpointsLocked(eps *transportEndpoints, vv buffer.VectorisedView, id TransportEndpointID) []*endpointsByNic { +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 { @@ -445,14 +531,44 @@ func (d *transportDemuxer) findAllEndpointsLocked(eps *transportEndpoints, vv bu if ep, ok := eps.endpoints[nid]; ok { 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, vv buffer.VectorisedView, id TransportEndpointID) *endpointsByNic { - if matchedEPs := d.findAllEndpointsLocked(eps, vv, id); len(matchedEPs) > 0 { +func (d *transportDemuxer) findEndpointLocked(eps *transportEndpoints, id TransportEndpointID) *endpointsByNic { + if matchedEPs := d.findAllEndpointsLocked(eps, id); len(matchedEPs) > 0 { return matchedEPs[0] } return nil @@ -465,7 +581,7 @@ 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() @@ -496,3 +612,7 @@ func (d *transportDemuxer) unregisterRawEndpoint(netProto tcpip.NetworkProtocolN func isMulticastOrBroadcast(addr tcpip.Address) bool { return addr == header.IPv4Broadcast || header.IsV4MulticastAddress(addr) || header.IsV6MulticastAddress(addr) } + +func isUnicast(addr tcpip.Address) bool { + return addr != header.IPv4Any && addr != header.IPv6Any && !isMulticastOrBroadcast(addr) +} diff --git a/pkg/tcpip/stack/transport_demuxer_test.go b/pkg/tcpip/stack/transport_demuxer_test.go index 210233dc0..3b28b06d0 100644 --- a/pkg/tcpip/stack/transport_demuxer_test.go +++ b/pkg/tcpip/stack/transport_demuxer_test.go @@ -79,17 +79,17 @@ func newDualTestContextMultiNic(t *testing.T, mtu uint32, linkEpNames []string) 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 { + 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 { + 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 { + if err := s.AddAddress(nicID, ipv6.ProtocolNumber, stackV6Addr); err != nil { t.Fatalf("AddAddress IPv6 failed: %v", err) } } @@ -156,7 +156,9 @@ func (c *testContext) sendV6Packet(payload []byte, h *headers, linkEpName string u.SetChecksum(^u.CalculateChecksum(xsum)) // Inject packet. - c.linkEPs[linkEpName].Inject(ipv6.ProtocolNumber, buf.ToVectorisedView()) + c.linkEPs[linkEpName].InjectInbound(ipv6.ProtocolNumber, tcpip.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) } func TestTransportDemuxerRegister(t *testing.T) { diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index 86c62be25..748ce4ea5 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -43,6 +43,7 @@ type fakeTransportEndpoint struct { proto *fakeTransportProtocol peerAddr tcpip.Address route stack.Route + uniqueID uint64 // acceptQueue is non-nil iff bound. acceptQueue []fakeTransportEndpoint @@ -56,8 +57,8 @@ func (f *fakeTransportEndpoint) Stats() tcpip.EndpointStats { return nil } -func newFakeTransportEndpoint(s *stack.Stack, proto *fakeTransportProtocol, netProto tcpip.NetworkProtocolNumber) tcpip.Endpoint { - return &fakeTransportEndpoint{stack: s, TransportEndpointInfo: stack.TransportEndpointInfo{NetProto: netProto}, proto: proto} +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() { @@ -82,7 +83,10 @@ func (f *fakeTransportEndpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions if err != nil { return 0, nil, err } - if err := f.route.WritePacket(nil /* gso */, hdr, buffer.View(v).ToVectorisedView(), stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}); err != nil { + if err := f.route.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + Data: buffer.View(v).ToVectorisedView(), + }); err != nil { return 0, nil, err } @@ -144,6 +148,10 @@ 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 } @@ -192,7 +200,7 @@ 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 { @@ -209,7 +217,7 @@ 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++ } @@ -218,15 +226,15 @@ 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 @@ -251,7 +259,7 @@ 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) { @@ -266,7 +274,7 @@ 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 } @@ -337,7 +345,9 @@ 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) } @@ -346,7 +356,9 @@ 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) } @@ -355,7 +367,9 @@ 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) } @@ -408,7 +422,9 @@ 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) } @@ -417,7 +433,9 @@ 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) } @@ -426,7 +444,9 @@ 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) } @@ -579,7 +599,9 @@ 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 { @@ -598,10 +620,10 @@ func TestTransportForwarding(t *testing.T) { t.Fatal("Response packet not forwarded") } - if dst := p.Header[0]; dst != 3 { + if dst := p.Pkt.Header.View()[0]; dst != 3 { t.Errorf("Response packet has incorrect destination addresss: got = %d, want = 3", dst) } - if src := p.Header[1]; src != 1 { + if src := p.Pkt.Header.View()[1]; src != 1 { t.Errorf("Response packet has incorrect source addresss: got = %d, want = 3", src) } } diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 678a94616..f62fd729f 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -231,6 +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 @@ -255,7 +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. @@ -301,7 +308,7 @@ type ControlMessages struct { // HasTimestamp indicates whether Timestamp is valid/set. HasTimestamp bool - // Timestamp is the time (in ns) that the last packed used to create + // Timestamp is the time (in ns) that the last packet used to create // the read data was received. Timestamp int64 @@ -310,6 +317,18 @@ type ControlMessages struct { // Inq is the number of bytes ready to be received. Inq int32 + + // HasTOS indicates whether Tos is valid/set. + HasTOS bool + + // TOS is the IPv4 type of service of the associated packet. + TOS int8 + + // HasTClass indicates whether Tclass is valid/set. + HasTClass bool + + // Tclass is the IPv6 traffic class of the associated packet. + TClass int32 } // Endpoint is the interface implemented by transport protocols (e.g., tcp, udp) @@ -489,6 +508,11 @@ 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. ) @@ -501,11 +525,6 @@ 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 @@ -557,6 +576,11 @@ type KeepaliveIntervalOption time.Duration // closed. type KeepaliveCountOption int +// TCPUserTimeoutOption is used by SetSockOpt/GetSockOpt to specify a user +// specified timeout for a given TCP connection. +// See: RFC5482 for details. +type TCPUserTimeoutOption time.Duration + // CongestionControlOption is used by SetSockOpt/GetSockOpt to set/get // the current congestion control algorithm. type CongestionControlOption string @@ -579,6 +603,16 @@ type MaxSegOption int // 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 @@ -673,6 +707,11 @@ 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) @@ -881,6 +920,23 @@ 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 + + // EstablishedClosed is the number of times established TCP connections + // made a transition to CLOSED state. + EstablishedClosed *StatCounter + + // EstablishedTimedout is the number of times an established connection + // was reset because of keep-alive time out. + EstablishedTimedout *StatCounter + // ListenOverflowSynDrop is the number of times the listen queue overflowed // and a SYN was dropped. ListenOverflowSynDrop *StatCounter @@ -1308,8 +1364,8 @@ 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/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 9254c3dea..d8c5b5058 100644 --- a/pkg/tcpip/transport/icmp/BUILD +++ b/pkg/tcpip/transport/icmp/BUILD @@ -38,11 +38,3 @@ go_library( "//pkg/waiter", ], ) - -filegroup( - name = "autogen", - srcs = [ - "icmp_packet_list.go", - ], - visibility = ["//:sandbox"], -) diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index 3187b336b..9c40931b5 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 @@ -58,6 +55,7 @@ type endpoint struct { // immutable. stack *stack.Stack `state:"manual"` waiterQueue *waiter.Queue + uniqueID uint64 // The following fields are used to manage the receive queue, and are // protected by rcvMu. @@ -90,9 +88,15 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProt 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() { @@ -274,13 +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 + nicID := to.NIC if e.BindNICID != 0 { - if nicid != 0 && nicid != e.BindNICID { + if nicID != 0 && nicID != e.BindNICID { return 0, nil, tcpip.ErrNoRoute } - nicid = e.BindNICID + nicID = e.BindNICID } toCopy := *to @@ -291,7 +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 } @@ -425,7 +429,11 @@ func send4(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Err if ttl == 0 { ttl = r.DefaultTTL() } - return r.WritePacket(nil /* gso */, hdr, data.ToVectorisedView(), stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}) + return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + Data: data.ToVectorisedView(), + TransportHeader: buffer.View(icmpv4), + }) } func send6(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Error { @@ -445,13 +453,17 @@ func send6(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Err 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)) if ttl == 0 { ttl = r.DefaultTTL() } - return r.WritePacket(nil /* gso */, hdr, data.ToVectorisedView(), stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}) + return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + Data: dataVV, + TransportHeader: buffer.View(icmpv6), + }) } func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress, allowMismatch bool) (tcpip.NetworkProtocolNumber, *tcpip.Error) { @@ -479,7 +491,7 @@ 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: @@ -488,11 +500,11 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { 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 } @@ -503,7 +515,7 @@ 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 } @@ -520,14 +532,14 @@ 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.route = r.Clone() - e.RegisterNICID = nicid + e.RegisterNICID = nicID e.state = stateConnected @@ -578,18 +590,18 @@ 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 /* reuse */, 0 /* bindToDevice */) + 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 /* reuse */, 0 /* bindtodevice */) + err := e.stack.RegisterTransportEndpoint(nicID, netProtos, e.TransProto, id, e, false /* reuse */, 0 /* bindtodevice */) switch err { case nil: return true, nil @@ -714,18 +726,18 @@ 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 { 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() @@ -753,19 +765,19 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, vv 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() @@ -776,7 +788,7 @@ 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 @@ -798,3 +810,6 @@ func (e *endpoint) Info() tcpip.EndpointInfo { func (e *endpoint) Stats() tcpip.EndpointStats { return &e.stats } + +// Wait implements stack.TransportEndpoint.Wait. +func (*endpoint) Wait() {} 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..44b58ff6b --- /dev/null +++ b/pkg/tcpip/transport/packet/BUILD @@ -0,0 +1,38 @@ +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", + ], +) 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 fba598d51..00991ac8e 100644 --- a/pkg/tcpip/transport/raw/BUILD +++ b/pkg/tcpip/transport/raw/BUILD @@ -4,14 +4,14 @@ 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"], @@ -34,14 +34,7 @@ go_library( "//pkg/tcpip/header", "//pkg/tcpip/iptables", "//pkg/tcpip/stack", + "//pkg/tcpip/transport/packet", "//pkg/waiter", ], ) - -filegroup( - name = "autogen", - srcs = [ - "packet_list.go", - ], - visibility = ["//:sandbox"], -) diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index b4c660859..5aafe2615 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. @@ -72,7 +67,7 @@ type endpoint struct { // 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 @@ -90,7 +85,6 @@ type endpoint struct { } // 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 */) } @@ -187,17 +181,17 @@ func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMess return buffer.View{}, tcpip.ControlMessages{}, err } - packet := e.rcvList.Front() - e.rcvList.Remove(packet) - e.rcvBufSize -= packet.data.Size() + pkt := e.rcvList.Front() + e.rcvList.Remove(pkt) + e.rcvBufSize -= pkt.data.Size() 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. @@ -344,13 +338,18 @@ func (e *endpoint) finishWrite(payloadBytes []byte, route *stack.Route) (int64, switch e.NetProto { case header.IPv4ProtocolNumber: if !e.associated { - if err := route.WriteHeaderIncludedPacket(buffer.View(payloadBytes).ToVectorisedView()); err != nil { + if err := route.WriteHeaderIncludedPacket(tcpip.PacketBuffer{ + Data: buffer.View(payloadBytes).ToVectorisedView(), + }); err != nil { return 0, nil, err } break } hdr := buffer.NewPrependable(len(payloadBytes) + int(route.MaxHeaderLength())) - if err := route.WritePacket(nil /* gso */, hdr, buffer.View(payloadBytes).ToVectorisedView(), stack.NetworkHeaderParams{Protocol: e.TransProto, TTL: route.DefaultTTL(), TOS: stack.DefaultTOS}); err != nil { + if err := route.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: e.TransProto, TTL: route.DefaultTTL(), TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + Data: buffer.View(payloadBytes).ToVectorisedView(), + }); err != nil { return 0, nil, err } @@ -561,7 +560,7 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { } // HandlePacket implements stack.RawTransportEndpoint.HandlePacket. -func (e *endpoint) HandlePacket(route *stack.Route, netHeader buffer.View, vv buffer.VectorisedView) { +func (e *endpoint) HandlePacket(route *stack.Route, pkt tcpip.PacketBuffer) { e.rcvMu.Lock() // Drop the packet if our buffer is currently full. @@ -602,16 +601,17 @@ func (e *endpoint) HandlePacket(route *stack.Route, netHeader buffer.View, vv bu 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[:]) + networkHeader := append(buffer.View(nil), pkt.NetworkHeader...) + combinedVV := networkHeader.ToVectorisedView() + combinedVV.Append(pkt.Data) + packet.data = combinedVV packet.timestampNS = e.stack.NowNanoseconds() e.rcvList.PushBack(packet) @@ -643,3 +643,6 @@ func (e *endpoint) Info() tcpip.EndpointInfo { 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 a6c7cc43a..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 @@ -86,7 +86,9 @@ func (ep *endpoint) Resume(s *stack.Stack) { } } - if err := ep.stack.RegisterRawTransportEndpoint(ep.RegisterNICID, 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 aed70e06f..3b353d56c 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -28,6 +28,7 @@ go_library( "forwarder.go", "protocol.go", "rcv.go", + "rcv_state.go", "reno.go", "sack.go", "sack_scoreboard.go", @@ -44,6 +45,7 @@ go_library( imports = ["gvisor.dev/gvisor/pkg/tcpip/buffer"], visibility = ["//visibility:public"], deps = [ + "//pkg/log", "//pkg/rand", "//pkg/sleep", "//pkg/tcpip", @@ -51,6 +53,7 @@ go_library( "//pkg/tcpip/hash/jenkins", "//pkg/tcpip/header", "//pkg/tcpip/iptables", + "//pkg/tcpip/ports", "//pkg/tcpip/seqnum", "//pkg/tcpip/stack", "//pkg/tcpip/transport/raw", @@ -60,17 +63,9 @@ go_library( ], ) -filegroup( - name = "autogen", - srcs = [ - "tcp_segment_list.go", - ], - visibility = ["//:sandbox"], -) - 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 844959fa0..5422ae80c 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" @@ -241,8 +242,15 @@ func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, i n.initGSO() + // Now inherit any socket options that should be inherited from the + // listening endpoint. + // In case of Forwarder listenEP will be nil and hence this check. + if l.listenEP != nil { + l.listenEP.propagateInheritableOptions(n) + } + // Register new endpoint so that packets are routed to it. - if err := n.stack.RegisterTransportEndpoint(n.boundNICID, n.effectiveNetProtos, ProtocolNumber, n.ID, n, n.reusePort, n.bindToDevice); err != nil { + if err := n.stack.RegisterTransportEndpoint(n.boundNICID, n.effectiveNetProtos, ProtocolNumber, n.ID, n, n.reusePort, n.boundBindToDevice); err != nil { n.Close() return nil, err } @@ -268,8 +276,8 @@ 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 } @@ -288,7 +296,7 @@ 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.Close() if l.listenEP != nil { @@ -297,7 +305,7 @@ func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *head return nil, err } ep.mu.Lock() - ep.state = StateEstablished + ep.isConnectNotified = true ep.mu.Unlock() // Update the receive window scaling. We can't do it before the @@ -349,6 +357,14 @@ func (e *endpoint) deliverAccepted(n *endpoint) { } } +// propagateInheritableOptions propagates any options set on the listening +// endpoint to the newly created endpoint. +func (e *endpoint) propagateInheritableOptions(n *endpoint) { + e.mu.Lock() + n.userTimeout = e.userTimeout + e.mu.Unlock() +} + // handleSynSegment is called in its own goroutine once the listening endpoint // receives a SYN segment. It is responsible for completing the handshake and // queueing the new endpoint for acceptance. @@ -359,6 +375,7 @@ 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() @@ -366,6 +383,11 @@ func (e *endpoint) handleSynSegment(ctx *listenContext, s *segment, opts *header return } ctx.removePendingEndpoint(n) + // Start the protocol goroutine. + wq := &waiter.Queue{} + n.startAcceptedLoop(wq) + e.stack.Stats().TCP.PassiveConnectionOpenings.Increment() + e.deliverAccepted(n) } @@ -399,8 +421,19 @@ 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) { - switch s.flags { - case header.TCPFlagSyn: + 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 { + case s.flags == header.TCPFlagSyn: opts := parseSynSegmentOptions(s) if incSynRcvdCount() { // Only handle the syn if the following conditions hold @@ -433,19 +466,18 @@ 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), } 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() } - case header.TCPFlagAck: + case (s.flags & header.TCPFlagAck) != 0: if e.acceptQueueIsFull() { // Silently drop the ack as the application can't accept // the connection at this point. The ack will be @@ -459,6 +491,14 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { } if !synCookiesInUse() { + // When not using SYN cookies, as per RFC 793, section 3.9, page 64: + // Any acknowledgment is bad if it arrives on a connection still in + // the LISTEN state. An acceptable reset segment should be formed + // for any arriving ACK-bearing segment. The RST should be + // formatted as follows: + // + // <SEQ=SEG.ACK><CTL=RST> + // // Send a reset as this is an ACK for which there is no // half open connections and we are not using cookies // yet. @@ -519,7 +559,11 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { n.tsOffset = 0 // Switch state to connected. + // We do not use transitionToStateEstablishedLocked here as there is + // no handshake state available when doing a SYN cookie based accept. + 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 @@ -530,6 +574,11 @@ 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) } } diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 5ea036bea..cdd69f360 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. @@ -194,6 +218,14 @@ func (h *handshake) synSentState(s *segment) *tcpip.Error { // acceptable if the ack field acknowledges the SYN. if s.flagIsSet(header.TCPFlagRst) { if s.flagIsSet(header.TCPFlagAck) && s.ackNumber == h.iss+1 { + // RFC 793, page 67, states that "If the RST bit is set [and] If the ACK + // was acceptable then signal the user "error: connection reset", drop + // the segment, enter CLOSED state, delete TCB, and return." + h.ep.mu.Lock() + h.ep.workerCleanup = true + h.ep.mu.Unlock() + // Although the RFC above calls out ECONNRESET, Linux actually returns + // ECONNREFUSED here so we do as well. return tcpip.ErrConnectionRefused } return nil @@ -228,6 +260,11 @@ func (h *handshake) synSentState(s *segment) *tcpip.Error { // and the handshake is completed. if s.flagIsSet(header.TCPFlagAck) { h.state = handshakeCompleted + + h.ep.mu.Lock() + h.ep.transitionToStateEstablishedLocked(h) + h.ep.mu.Unlock() + h.ep.sendRaw(buffer.VectorisedView{}, header.TCPFlagAck, h.iss+1, h.ackNum, h.rcvWnd>>h.effectiveRcvWndScale()) return nil } @@ -275,6 +312,15 @@ func (h *handshake) synRcvdState(s *segment) *tcpip.Error { return nil } + // RFC 793, Section 3.9, page 69, states that in the SYN-RCVD state, a + // sequence number outside of the window causes an ACK with the proper seq + // number and "After sending the acknowledgment, drop the unacceptable + // segment and return." + if !s.sequenceNumber.InWindow(h.ackNum, h.rcvWnd) { + h.ep.sendRaw(buffer.VectorisedView{}, header.TCPFlagAck, h.iss+1, h.ackNum, h.rcvWnd) + return nil + } + if s.flagIsSet(header.TCPFlagSyn) && s.sequenceNumber != h.ackNum-1 { // We received two SYN segments with different sequence // numbers, so we reset this and restart the whole @@ -319,6 +365,10 @@ func (h *handshake) synRcvdState(s *segment) *tcpip.Error { h.ep.updateRecentTimestamp(s.parsedOptions.TSVal, h.ackNum, s.sequenceNumber) } h.state = handshakeCompleted + h.ep.mu.Lock() + h.ep.transitionToStateEstablishedLocked(h) + h.ep.mu.Unlock() + return nil } @@ -445,7 +495,7 @@ 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, @@ -607,19 +657,14 @@ func (e *endpoint) sendTCP(r *stack.Route, id stack.TransportEndpointID, data bu 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, tos uint8, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) *tcpip.Error { +func buildTCPHdr(r *stack.Route, id stack.TransportEndpointID, pkt *tcpip.PacketBuffer, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) { optLen := len(opts) - // Allocate a buffer for the TCP header. - hdr := buffer.NewPrependable(header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen) - - if rcvWnd > 0xffff { - rcvWnd = 0xffff - } - + hdr := &pkt.Header + packetSize := pkt.DataSize + off := pkt.DataOffset // Initialize the header. tcp := header.TCP(hdr.Prepend(header.TCPMinimumSize + optLen)) + pkt.TransportHeader = buffer.View(tcp) tcp.Encode(&header.TCPFields{ SrcPort: id.LocalPort, DstPort: id.RemotePort, @@ -631,7 +676,7 @@ 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 { @@ -641,14 +686,79 @@ 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(pkt.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 + + // Allocate one big slice for all the headers. + hdrSize := header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen + buf := make([]byte, n*hdrSize) + pkts := make([]tcpip.PacketBuffer, n) + for i := range pkts { + pkts[i].Header = buffer.NewEmptyPrependableFromView(buf[i*hdrSize:][:hdrSize]) + } + + size := data.Size() + off := 0 + for i := 0; i < n; i++ { + packetSize := mss + if packetSize > size { + packetSize = size + } + size -= packetSize + pkts[i].DataOffset = off + pkts[i].DataSize = packetSize + pkts[i].Data = data + buildTCPHdr(r, id, &pkts[i], flags, seq, ack, rcvWnd, opts, gso) + off += packetSize + seq = seq.Add(seqnum.Size(packetSize)) + } + if ttl == 0 { + ttl = r.DefaultTTL() + } + sent, err := r.WritePackets(gso, pkts, 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) + } + + pkt := tcpip.PacketBuffer{ + Header: buffer.NewPrependable(header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen), + DataOffset: 0, + DataSize: data.Size(), + Data: data, + } + buildTCPHdr(r, id, &pkt, flags, seq, ack, rcvWnd, opts, gso) + if ttl == 0 { ttl = r.DefaultTTL() } - if err := r.WritePacket(gso, hdr, data, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}); err != nil { + if err := r.WritePacket(gso, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}, pkt); err != nil { r.Stats().TCP.SegmentSendErrors.Increment() return err } @@ -754,10 +864,26 @@ 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 - if err != tcpip.ErrConnectionReset { - e.sendRaw(buffer.VectorisedView{}, header.TCPFlagAck|header.TCPFlagRst, e.snd.sndUna, e.rcv.rcvNxt, 0) + if err != tcpip.ErrConnectionReset && err != tcpip.ErrTimeout { + // 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) } } @@ -771,6 +897,99 @@ func (e *endpoint) completeWorkerLocked() { } } +// transitionToStateEstablisedLocked transitions a given endpoint +// to an established state using the handshake parameters provided. +// It also initializes sender/receiver if required. +func (e *endpoint) transitionToStateEstablishedLocked(h *handshake) { + if e.snd == nil { + // Transfer handshake state to TCP connection. We disable + // receive window scaling if the peer doesn't support it + // (indicated by a negative send window scale). + e.snd = newSender(e, h.iss, h.ackNum-1, h.sndWnd, h.mss, h.sndWndScale) + } + if e.rcv == nil { + rcvBufSize := seqnum.Size(e.receiveBufferSize()) + e.rcvListMu.Lock() + e.rcv = newReceiver(e, h.ackNum-1, h.rcvWnd, h.effectiveRcvWndScale(), rcvBufSize) + // Bootstrap the auto tuning algorithm. Starting at zero will + // result in a really large receive window after the first auto + // tuning adjustment. + e.rcvAutoParams.prevCopied = int(h.rcvWnd) + e.rcvListMu.Unlock() + } + h.ep.stack.Stats().TCP.CurrentEstablished.Increment() + e.state = StateEstablished +} + +// transitionToStateCloseLocked ensures that the endpoint is +// cleaned up from the transport demuxer, "before" moving to +// StateClose. This will ensure that no packet will be +// delivered to this endpoint from the demuxer when the endpoint +// is transitioned to StateClose. +func (e *endpoint) transitionToStateCloseLocked() { + if e.state == StateClose { + return + } + e.cleanupLocked() + e.state = StateClose + e.stack.Stats().TCP.EstablishedClosed.Increment() +} + +// tryDeliverSegmentFromClosedEndpoint attempts to deliver the parsed +// segment to any other endpoint other than the current one. This is called +// only when the endpoint is in StateClose and we want to deliver the segment +// to any other listening endpoint. We reply with RST if we cannot find one. +func (e *endpoint) tryDeliverSegmentFromClosedEndpoint(s *segment) { + ep := e.stack.FindTransportEndpoint(e.NetProto, e.TransProto, e.ID, &s.route) + if ep == nil { + replyWithReset(s) + s.decRef() + return + } + ep.(*endpoint).enqueueSegment(s) +} + +func (e *endpoint) handleReset(s *segment) (ok bool, err *tcpip.Error) { + if e.rcv.acceptable(s.sequenceNumber, 0) { + // RFC 793, page 37 states that "in all states + // 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.transitionToStateCloseLocked() + e.HardError = tcpip.ErrAborted + 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 { @@ -788,14 +1007,34 @@ 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. @@ -804,7 +1043,33 @@ 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 + } + + // Now check if the received segment has caused us to transition + // to a CLOSED state, if yes then terminate processing and do + // not invoke the sender. + e.mu.RLock() + state := e.state + e.mu.RUnlock() + if state == StateClose { + // When we get into StateClose while processing from the queue, + // return immediately and let the protocolMainloop handle it. + // + // We can reach StateClose only while processing a previous segment + // or a notification from the protocolMainLoop (caller goroutine). + // This means that with this return, the segment dequeue below can + // never occur on a closed endpoint. + s.decRef() + return nil + } e.snd.handleRcvdSegment(s) } s.decRef() @@ -830,14 +1095,27 @@ func (e *endpoint) handleSegments() *tcpip.Error { // keepalive packets periodically when the connection is idle. If we don't hear // from the other side after a number of tries, we terminate the connection. func (e *endpoint) keepaliveTimerExpired() *tcpip.Error { + e.mu.RLock() + userTimeout := e.userTimeout + e.mu.RUnlock() + e.keepalive.Lock() if !e.keepalive.enabled || !e.keepalive.timer.checkExpiration() { e.keepalive.Unlock() return nil } + // If a userTimeout is set then abort the connection if it is + // exceeded. + if userTimeout != 0 && time.Since(e.rcv.lastRcvdAckTime) >= userTimeout && e.keepalive.unacked > 0 { + e.keepalive.Unlock() + e.stack.Stats().TCP.EstablishedTimedout.Increment() + return tcpip.ErrTimeout + } + if e.keepalive.unacked >= e.keepalive.count { e.keepalive.Unlock() + e.stack.Stats().TCP.EstablishedTimedout.Increment() return tcpip.ErrTimeout } @@ -854,7 +1132,6 @@ func (e *endpoint) keepaliveTimerExpired() *tcpip.Error { // whether it is enabled for this endpoint. func (e *endpoint) resetKeepaliveTimer(receivedData bool) { e.keepalive.Lock() - defer e.keepalive.Unlock() if receivedData { e.keepalive.unacked = 0 } @@ -862,6 +1139,7 @@ func (e *endpoint) resetKeepaliveTimer(receivedData bool) { // data to send. if !e.keepalive.enabled || e.snd == nil || e.snd.sndUna != e.snd.sndNxt { e.keepalive.timer.disable() + e.keepalive.Unlock() return } if e.keepalive.unacked > 0 { @@ -869,6 +1147,7 @@ func (e *endpoint) resetKeepaliveTimer(receivedData bool) { } else { e.keepalive.timer.enable(e.keepalive.idle) } + e.keepalive.Unlock() } // disableKeepaliveTimer stops the keepalive timer. @@ -903,7 +1182,6 @@ 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) } @@ -932,21 +1210,6 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { return err } - - // Transfer handshake state to TCP connection. We disable - // receive window scaling if the peer doesn't support it - // (indicated by a negative send window scale). - e.snd = newSender(e, h.iss, h.ackNum-1, h.sndWnd, h.mss, h.sndWndScale) - - rcvBufSize := seqnum.Size(e.receiveBufferSize()) - e.rcvListMu.Lock() - e.rcv = newReceiver(e, h.ackNum-1, h.rcvWnd, h.effectiveRcvWndScale(), rcvBufSize) - // boot strap the auto tuning algorithm. Starting at zero will - // result in a large step function on the first proper causing - // the window to just go to a really large value after the first - // RTT itself. - e.rcvAutoParams.prevCopied = initialRcvWnd - e.rcvListMu.Unlock() } e.keepalive.timer.init(&e.keepalive.waker) @@ -954,7 +1217,6 @@ 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 { @@ -985,13 +1247,20 @@ 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.transitionToStateCloseLocked() + e.mu.Unlock() + return nil }, }, { w: &e.snd.resendWaker, f: func() *tcpip.Error { if !e.snd.retransmitTimerExpired() { + e.stack.Stats().TCP.EstablishedTimedout.Increment() return tcpip.ErrTimeout } return nil @@ -1028,17 +1297,18 @@ 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 { @@ -1054,12 +1324,20 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { return err } } - if e.state != StateError { + if e.state != StateClose && e.state != StateError { + // Only block the worker if the endpoint + // is not in closed state or error state. close(e.drainDone) <-e.undrain } } + if n¬ifyTickleWorker != 0 { + // Just a tickle notification. No need to do + // anything. + return nil + } + return nil }, }, @@ -1086,15 +1364,16 @@ 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() @@ -1110,15 +1389,168 @@ 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.state = StateClose + e.stack.Stats().TCP.CurrentEstablished.Decrement() + e.transitionToStateCloseLocked() } + // Lock released below. epilogue() + // epilogue removes the endpoint from the transport-demuxer and + // unlocks e.mu. Now that no new segments can get enqueued to this + // endpoint, try to re-match the segment to a different endpoint + // as the current endpoint is closed. + for { + s := e.segmentQueue.dequeue() + if s == nil { + break + } + + e.tryDeliverSegmentFromClosedEndpoint(s) + } + + // A new SYN was received during TIME_WAIT and we need to abort + // the timewait and redirect the segment to the listener queue + if reuseTW != nil { + 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/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index a1b784b49..fe629aa40 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -30,6 +30,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/hash/jenkins" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/iptables" + "gvisor.dev/gvisor/pkg/tcpip/ports" "gvisor.dev/gvisor/pkg/tcpip/seqnum" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/tmutex" @@ -121,6 +122,11 @@ 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. @@ -287,6 +293,7 @@ type endpoint struct { // change throughout the lifetime of the endpoint. 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. @@ -319,6 +326,11 @@ type endpoint struct { 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"` @@ -330,6 +342,11 @@ type endpoint struct { // disabling SO_BROADCAST, albeit as a NOOP. broadcast bool + // Values used to reserve a port or register a transport endpoint + // (which ever happens first). + boundBindToDevice tcpip.NICID + boundPortFlags ports.Flags + // effectiveNetProtos contains the network protocols actually in use. In // most cases it will only contain "netProto", but in cases like IPv6 // endpoints with v6only set to false, this could include multiple @@ -411,7 +428,7 @@ 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 @@ -458,6 +475,12 @@ type endpoint struct { // without hearing a response, the connection is closed. keepalive keepalive + // userTimeout if non-zero specifies a user specified timeout for + // a connection w/ pending data to send. A connection that has pending + // unacked data will be forcibily aborted if the timeout is reached + // without any data being acked. + userTimeout time.Duration + // pendingAccepted is a synchronization primitive used to track number // of connections that are queued up to be delivered to the accepted // channel. We use this to ensure that all goroutines blocked on writing @@ -502,6 +525,36 @@ type endpoint struct { // 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. @@ -550,6 +603,7 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue interval: 75 * time.Second, count: 9, }, + uniqueID: s.UniqueID(), } var ss SendBufferSizeOption @@ -572,6 +626,16 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue e.rcvAutoParams.disabled = !bool(mrb) } + 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 } @@ -659,6 +723,13 @@ 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) @@ -671,14 +742,18 @@ 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.bindToDevice) + e.stack.StartTransportEndpointCleanup(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.boundBindToDevice) e.isRegistered = false } - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.bindToDevice) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.boundPortFlags, e.boundBindToDevice) e.isPortReserved = false + e.boundBindToDevice = 0 + e.boundPortFlags = ports.Flags{} } + // 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) @@ -704,9 +779,7 @@ 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() } }() @@ -732,16 +805,19 @@ func (e *endpoint) cleanupLocked() { e.workerCleanup = false if e.isRegistered { - e.stack.UnregisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.bindToDevice) + e.stack.StartTransportEndpointCleanup(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.boundBindToDevice) e.isRegistered = false } if e.isPortReserved { - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.bindToDevice) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.boundPortFlags, e.boundBindToDevice) e.isPortReserved = false } + e.boundBindToDevice = 0 + e.boundPortFlags = ports.Flags{} e.route.Release() + e.stack.CompleteTransportEndpointCleanup(e) tcpip.DeleteDanglingEndpoint(e) } @@ -752,7 +828,9 @@ 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 } @@ -1133,16 +1211,6 @@ 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 { - // Lower 2 bits represents ECN bits. RFC 3168, section 23.1 - const inetECNMask = 3 - switch v := opt.(type) { case tcpip.DelayOption: if v == 0 { atomic.StoreUint32(&e.delay, 0) @@ -1154,6 +1222,16 @@ 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) @@ -1184,9 +1262,9 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.bindToDevice = 0 return nil } - for nicid, nic := range e.stack.NICInfo() { + for nicID, nic := range e.stack.NICInfo() { if nic.Name == string(v) { - e.bindToDevice = nicid + e.bindToDevice = nicID return nil } } @@ -1206,7 +1284,7 @@ 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 @@ -1262,6 +1340,12 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.notifyProtocolGoroutine(notifyKeepaliveChanged) return nil + case tcpip.TCPUserTimeoutOption: + e.mu.Lock() + e.userTimeout = time.Duration(v) + e.mu.Unlock() + return nil + case tcpip.BroadcastOption: e.mu.Lock() e.broadcast = v != 0 @@ -1319,6 +1403,28 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { 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 } @@ -1345,6 +1451,7 @@ 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 @@ -1357,8 +1464,16 @@ 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. @@ -1379,13 +1494,6 @@ 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 { @@ -1496,6 +1604,12 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { e.keepalive.Unlock() return nil + case *tcpip.TCPUserTimeoutOption: + e.mu.Lock() + *o = tcpip.TCPUserTimeoutOption(e.userTimeout) + e.mu.Unlock() + return nil + case *tcpip.OutOfBandInlineOption: // We don't currently support disabling this option. *o = 1 @@ -1530,6 +1644,12 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { 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 } @@ -1602,7 +1722,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tc 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 @@ -1611,11 +1731,11 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tc 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) @@ -1634,7 +1754,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tc } // 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 } @@ -1649,7 +1769,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tc if e.ID.LocalPort != 0 { // The endpoint is bound to a port, attempt to register it. - err := e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, e.ID, e, e.reusePort, e.bindToDevice) + err := e.stack.RegisterTransportEndpoint(nicID, netProtos, ProtocolNumber, e.ID, e, e.reusePort, e.boundBindToDevice) if err != nil { return err } @@ -1664,7 +1784,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tc // 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.PortSeed()) + h := jenkins.Sum32(e.stack.Seed()) h.Write([]byte(e.ID.LocalAddress)) h.Write([]byte(e.ID.RemoteAddress)) portBuf := make([]byte, 2) @@ -1678,15 +1798,18 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tc } // 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) { + if !e.stack.IsPortAvailable(netProtos, ProtocolNumber, e.ID.LocalAddress, p, ports.Flags{LoadBalanced: false}, e.bindToDevice) { return false, nil } id := e.ID id.LocalPort = p - switch e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, id, e, e.reusePort, e.bindToDevice) { + switch e.stack.RegisterTransportEndpoint(nicID, netProtos, ProtocolNumber, id, e, e.reusePort, e.bindToDevice) { case nil: + // Port picking successful. Save the details of + // the selected port. e.ID = id + e.boundBindToDevice = e.bindToDevice return true, nil case tcpip.ErrPortInUse: return false, nil @@ -1702,14 +1825,14 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tc // before Connect: in such a case we don't want to hold on to // reservations anymore. if e.isPortReserved { - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, origID.LocalAddress, origID.LocalPort, e.bindToDevice) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, origID.LocalAddress, origID.LocalPort, e.boundPortFlags, e.boundBindToDevice) 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 @@ -1729,6 +1852,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tc e.segmentQueue.mu.Unlock() e.snd.updateMaxPayloadSize(int(e.route.MTU()), 0) e.state = StateEstablished + e.stack.Stats().TCP.CurrentEstablished.Increment() } if run { @@ -1749,9 +1873,8 @@ 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. @@ -1766,6 +1889,7 @@ 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 } } @@ -1784,14 +1908,11 @@ func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { 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: @@ -1799,11 +1920,20 @@ 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 } @@ -1844,6 +1974,15 @@ func (e *endpoint) listen(backlog int) *tcpip.Error { return nil } + if e.state == StateInitial { + // The listen is called on an unbound socket, the socket is + // automatically bound to a random free port with the local + // address set to INADDR_ANY. + if err := e.bindLocked(tcpip.FullAddress{}); err != nil { + return err + } + } + // Endpoint must be bound before it can transition to listen mode. if e.state != StateBound { e.stats.ReadErrors.InvalidEndpointState.Increment() @@ -1851,7 +1990,7 @@ func (e *endpoint) listen(backlog int) *tcpip.Error { } // Register the endpoint. - if err := e.stack.RegisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.reusePort, e.bindToDevice); err != nil { + if err := e.stack.RegisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.reusePort, e.boundBindToDevice); err != nil { return err } @@ -1895,12 +2034,7 @@ 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. @@ -1908,6 +2042,10 @@ func (e *endpoint) Bind(addr tcpip.FullAddress) (err *tcpip.Error) { e.mu.Lock() defer e.mu.Unlock() + return e.bindLocked(addr) +} + +func (e *endpoint) bindLocked(addr tcpip.FullAddress) (err *tcpip.Error) { // Don't allow binding once endpoint is not in the initial state // anymore. This is because once the endpoint goes into a connected or // listen state, it is already bound. @@ -1932,26 +2070,33 @@ func (e *endpoint) Bind(addr tcpip.FullAddress) (err *tcpip.Error) { } } - port, err := e.stack.ReservePort(netProtos, ProtocolNumber, addr.Addr, addr.Port, e.reusePort, e.bindToDevice) + flags := ports.Flags{ + LoadBalanced: e.reusePort, + } + port, err := e.stack.ReservePort(netProtos, ProtocolNumber, addr.Addr, addr.Port, flags, e.bindToDevice) if err != nil { return err } + e.boundBindToDevice = e.bindToDevice + e.boundPortFlags = flags e.isPortReserved = true e.effectiveNetProtos = netProtos e.ID.LocalPort = port // Any failures beyond this point must remove the port registration. - defer func(bindToDevice tcpip.NICID) { + defer func(portFlags ports.Flags, bindToDevice tcpip.NICID) { if err != nil { - e.stack.ReleasePort(netProtos, ProtocolNumber, addr.Addr, port, bindToDevice) + e.stack.ReleasePort(netProtos, ProtocolNumber, addr.Addr, port, portFlags, bindToDevice) e.isPortReserved = false e.effectiveNetProtos = nil e.ID.LocalPort = 0 e.ID.LocalAddress = "" e.boundNICID = 0 + e.boundBindToDevice = 0 + e.boundPortFlags = ports.Flags{} } - }(e.bindToDevice) + }(e.boundPortFlags, e.boundBindToDevice) // If an address is specified, we must ensure that it's one of our // local addresses. @@ -2001,8 +2146,8 @@ func (e *endpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Error) { // 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() @@ -2025,6 +2170,10 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, vv 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() @@ -2037,7 +2186,7 @@ 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) { switch typ { case stack.ControlPacketTooBig: e.sndBufMu.Lock() @@ -2327,11 +2476,7 @@ 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: @@ -2349,6 +2494,18 @@ 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 { @@ -2371,6 +2528,23 @@ 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 eae17237e..7aa4c3f0e 100644 --- a/pkg/tcpip/transport/tcp/endpoint_state.go +++ b/pkg/tcpip/transport/tcp/endpoint_state.go @@ -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.BindAddr) == 0 { e.BindAddr = e.ID.LocalAddress } - if err := e.Bind(tcpip.FullAddress{Addr: e.BindAddr, 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)) } } @@ -217,6 +224,16 @@ func (e *endpoint) Resume(s *stack.Stack) { 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) @@ -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) } } 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 db40785d3..bc718064c 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 { @@ -147,13 +162,26 @@ func (*protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Transpo func replyWithReset(s *segment) { // Get the seqnum from the packet if the ack flag is set. seq := seqnum.Value(0) + ack := seqnum.Value(0) + flags := byte(header.TCPFlagRst) + // As per RFC 793 page 35 (Reset Generation) + // 1. If the connection does not exist (CLOSED) then 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. + + // If the incoming segment has an ACK field, the reset takes its + // sequence number from the ACK field of the segment, otherwise the + // reset has sequence number zero and the ACK field is set to the sum + // of the sequence number and segment length of the incoming segment. + // The connection remains in the CLOSED state. if s.flagIsSet(header.TCPFlagAck) { seq = s.ackNumber + } else { + flags |= header.TCPFlagAck + ack = s.sequenceNumber.Add(s.logicalLen()) } - - ack := s.sequenceNumber.Add(s.logicalLen()) - - sendTCP(&s.route, s.id, buffer.VectorisedView{}, s.route.DefaultTTL(), stack.DefaultTOS, header.TCPFlagRst|header.TCPFlagAck, seq, ack, 0 /* rcvWnd */, nil /* options */, nil /* gso */) + sendTCP(&s.route, s.id, buffer.VectorisedView{}, s.route.DefaultTTL(), stack.DefaultTOS, flags, seq, ack, 0 /* rcvWnd */, nil /* options */, nil /* gso */) } // SetOption implements TransportProtocol.SetOption. @@ -165,6 +193,12 @@ 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 +236,24 @@ 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 +268,12 @@ 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 +304,18 @@ 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 +328,7 @@ 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..0a5534959 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" ) @@ -49,16 +50,20 @@ type receiver struct { pendingRcvdSegments segmentHeap pendingBufUsed seqnum.Size pendingBufSize seqnum.Size + + // Time when the last ack was received. + lastRcvdAckTime time.Time `state:".(unixTime)"` } func newReceiver(ep *endpoint, irs seqnum.Value, rcvWnd seqnum.Size, rcvWndScale uint8, pendingBufSize seqnum.Size) *receiver { return &receiver{ - ep: ep, - rcvNxt: irs + 1, - rcvAcc: irs.Add(rcvWnd + 1), - rcvWnd: rcvWnd, - rcvWndScale: rcvWndScale, - pendingBufSize: pendingBufSize, + ep: ep, + rcvNxt: irs + 1, + rcvAcc: irs.Add(rcvWnd + 1), + rcvWnd: rcvWnd, + rcvWndScale: rcvWndScale, + pendingBufSize: pendingBufSize, + lastRcvdAckTime: time.Now(), } } @@ -204,15 +209,20 @@ func (r *receiver) consumeSegment(s *segment, segSeq seqnum.Value, segLen seqnum // Handle ACK (not FIN-ACK, which we handled above) during one of the // shutdown states. - if s.flagIsSet(header.TCPFlagAck) { + if s.flagIsSet(header.TCPFlagAck) && s.ackNumber == r.ep.snd.sndNxt { r.ep.mu.Lock() 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: - r.ep.state = StateClose + r.ep.transitionToStateCloseLocked() } r.ep.mu.Unlock() } @@ -253,25 +263,110 @@ 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 } + // Store the time of the last ack. + r.lastRcvdAckTime = time.Now() + // Defer segment processing if it can't be consumed now. if !r.consumeSegment(s, segSeq, segLen) { if segLen > 0 || s.flagIsSet(header.TCPFlagFin) { @@ -288,7 +383,7 @@ 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 +410,67 @@ 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/rcv_state.go b/pkg/tcpip/transport/tcp/rcv_state.go new file mode 100644 index 000000000..2bf21a2e7 --- /dev/null +++ b/pkg/tcpip/transport/tcp/rcv_state.go @@ -0,0 +1,29 @@ +// 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 tcp + +import ( + "time" +) + +// saveLastRcvdAckTime is invoked by stateify. +func (r *receiver) saveLastRcvdAckTime() unixTime { + return unixTime{r.lastRcvdAckTime.Unix(), r.lastRcvdAckTime.UnixNano()} +} + +// loadLastRcvdAckTime is invoked by stateify. +func (r *receiver) loadLastRcvdAckTime(unix unixTime) { + r.lastRcvdAckTime = time.Unix(unix.second, unix.nano) +} 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 8332a0179..8a947dc66 100644 --- a/pkg/tcpip/transport/tcp/snd.go +++ b/pkg/tcpip/transport/tcp/snd.go @@ -28,8 +28,11 @@ import ( ) const ( - // minRTO is the minimum allowed value for the retransmit timeout. - minRTO = 200 * time.Millisecond + // MinRTO is the minimum allowed value for the retransmit timeout. + MinRTO = 200 * time.Millisecond + + // MaxRTO is the maximum allowed value for the retransmit timeout. + MaxRTO = 120 * time.Second // InitialCwnd is the initial congestion window. InitialCwnd = 10 @@ -134,6 +137,10 @@ type sender struct { // rttMeasureTime is the time when the rttMeasureSeqNum was sent. rttMeasureTime time.Time `state:".(unixTime)"` + // firstRetransmittedSegXmitTime is the original transmit time of + // the first segment that was retransmitted due to RTO expiration. + firstRetransmittedSegXmitTime time.Time `state:".(unixTime)"` + closed bool writeNext *segment writeList segmentList @@ -392,8 +399,8 @@ func (s *sender) updateRTO(rtt time.Duration) { s.rto = s.rtt.srtt + 4*s.rtt.rttvar s.rtt.Unlock() - if s.rto < minRTO { - s.rto = minRTO + if s.rto < MinRTO { + s.rto = MinRTO } } @@ -438,8 +445,30 @@ 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 { + // Give up if we've waited more than a minute since the last resend or + // if a user time out is set and we have exceeded the user specified + // timeout since the first retransmission. + s.ep.mu.RLock() + uto := s.ep.userTimeout + s.ep.mu.RUnlock() + + if s.firstRetransmittedSegXmitTime.IsZero() { + // We store the original xmitTime of the segment that we are + // about to retransmit as the retransmission time. This is + // required as by the time the retransmitTimer has expired the + // segment has already been sent and unacked for the RTO at the + // time the segment was sent. + s.firstRetransmittedSegXmitTime = s.writeList.Front().xmitTime + } + + elapsed := time.Since(s.firstRetransmittedSegXmitTime) + remaining := MaxRTO + if uto != 0 { + // Cap to the user specified timeout if one is specified. + remaining = uto - elapsed + } + + if remaining <= 0 || s.rto >= MaxRTO { return false } @@ -447,6 +476,11 @@ func (s *sender) retransmitTimerExpired() bool { // below. s.rto *= 2 + // Cap RTO to remaining time. + if s.rto > remaining { + s.rto = remaining + } + // See: https://tools.ietf.org/html/rfc6582#section-3.2 Step 4. // // Retransmit timeouts: @@ -1168,6 +1202,8 @@ func (s *sender) handleRcvdSegment(seg *segment) { // RFC 6298 Rule 5.3 if s.sndUna == s.sndNxt { s.outstanding = 0 + // Reset firstRetransmittedSegXmitTime to the zero value. + s.firstRetransmittedSegXmitTime = time.Time{} s.resendTimer.disable() } } diff --git a/pkg/tcpip/transport/tcp/snd_state.go b/pkg/tcpip/transport/tcp/snd_state.go index 12eff8afc..8b20c3455 100644 --- a/pkg/tcpip/transport/tcp/snd_state.go +++ b/pkg/tcpip/transport/tcp/snd_state.go @@ -48,3 +48,13 @@ func (s *sender) loadRttMeasureTime(unix unixTime) { func (s *sender) afterLoad() { s.resendTimer.init(&s.resendWaker) } + +// saveFirstRetransmittedSegXmitTime is invoked by stateify. +func (s *sender) saveFirstRetransmittedSegXmitTime() unixTime { + return unixTime{s.firstRetransmittedSegXmitTime.Unix(), s.firstRetransmittedSegXmitTime.UnixNano()} +} + +// loadFirstRetransmittedSegXmitTime is invoked by stateify. +func (s *sender) loadFirstRetransmittedSegXmitTime(unix unixTime) { + s.firstRetransmittedSegXmitTime = time.Unix(unix.second, unix.nano) +} diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index 6d022a266..e8fe4dab5 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -75,6 +75,20 @@ func TestGiveUpConnect(t *testing.T) { if err := ep.GetSockOpt(tcpip.ErrorOption{}); err != tcpip.ErrAborted { t.Fatalf("got ep.GetSockOpt(tcpip.ErrorOption{}) = %v, want = %v", err, tcpip.ErrAborted) } + + // Call Connect again to retreive the handshake failure status + // and stats updates. + if err := ep.Connect(tcpip.FullAddress{Addr: context.TestAddr, Port: context.TestPort}); err != tcpip.ErrAborted { + t.Fatalf("got ep.Connect(...) = %v, want = %v", err, tcpip.ErrAborted) + } + + if got := c.Stack().Stats().TCP.FailedConnectionAttempts.Value(); got != 1 { + t.Errorf("got stats.TCP.FailedConnectionAttempts.Value() = %v, want = 1", got) + } + + if got := c.Stack().Stats().TCP.CurrentEstablished.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentEstablished.Value() = %v, want = 0", got) + } } func TestConnectIncrementActiveConnection(t *testing.T) { @@ -206,17 +220,18 @@ 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. @@ -256,7 +271,7 @@ 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): @@ -264,6 +279,13 @@ 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), @@ -285,6 +307,11 @@ 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 @@ -296,8 +323,8 @@ func TestTCPResetSentForACKWhenNotUsingSynCookies(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.SeqNum(uint32(c.IRS+1)), - checker.AckNum(uint32(iss)+1), - checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck))) + checker.AckNum(0), + checker.TCPFlags(header.TCPFlagRst))) } func TestTCPResetsReceivedIncrement(t *testing.T) { @@ -376,6 +403,13 @@ 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 @@ -396,12 +430,24 @@ 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()) @@ -413,15 +459,128 @@ func TestConnectResetAfterClose(t *testing.T) { checker.IPv4(t, b, checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagRst), + checker.SeqNum(uint32(c.IRS)+2), + checker.AckNum(0), + checker.TCPFlags(header.TCPFlagRst), ), ) break } } +// TestClosingWithEnqueuedSegments tests handling of still enqueued segments +// when the endpoint transitions to StateClose. The in-flight segments would be +// re-enqueued to a any listening endpoint. +func TestClosingWithEnqueuedSegments(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) + ep := c.EP + c.EP = nil + + if got, want := tcp.EndpointState(ep.State()), tcp.StateEstablished; got != want { + t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) + } + + // Send a FIN for ESTABLISHED --> CLOSED-WAIT + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + Flags: header.TCPFlagFin | header.TCPFlagAck, + SeqNum: 790, + AckNum: c.IRS.Add(1), + RcvWnd: 30000, + }) + + // Get the ACK for the FIN we sent. + checker.IPv4(t, c.GetPacket(), + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS)+1), + checker.AckNum(791), + checker.TCPFlags(header.TCPFlagAck), + ), + ) + + if got, want := tcp.EndpointState(ep.State()), tcp.StateCloseWait; got != want { + t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) + } + + // Close the application endpoint for CLOSE_WAIT --> LAST_ACK + ep.Close() + + // Get the FIN + checker.IPv4(t, c.GetPacket(), + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS)+1), + checker.AckNum(791), + checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), + ), + ) + + if got, want := tcp.EndpointState(ep.State()), tcp.StateLastAck; got != want { + t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) + } + + // Pause the endpoint`s protocolMainLoop. + ep.(interface{ StopWork() }).StopWork() + + // Enqueue last ACK followed by an ACK matching the endpoint + // + // Send Last ACK for LAST_ACK --> CLOSED + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + Flags: header.TCPFlagAck, + SeqNum: 791, + AckNum: c.IRS.Add(2), + RcvWnd: 30000, + }) + + // Send a packet with ACK set, this would generate RST when + // not using SYN cookies as in this test. + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + Flags: header.TCPFlagAck | header.TCPFlagFin, + SeqNum: 792, + AckNum: c.IRS.Add(2), + RcvWnd: 30000, + }) + + // Unpause endpoint`s protocolMainLoop. + ep.(interface{ ResumeWork() }).ResumeWork() + + // Wait for the protocolMainLoop to resume and update state. + time.Sleep(10 * time.Millisecond) + + // Expect the endpoint to be closed. + if got, want := tcp.EndpointState(ep.State()), tcp.StateClose; got != want { + t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) + } + + if got := c.Stack().Stats().TCP.EstablishedClosed.Value(); got != 1 { + t.Errorf("got c.Stack().Stats().TCP.EstablishedClosed = %v, want = 1", got) + } + + if got := c.Stack().Stats().TCP.CurrentEstablished.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentEstablished.Value() = %v, want = 0", got) + } + + // Check if the endpoint was moved to CLOSED and netstack a reset in + // response to the ACK packet that we sent after last-ACK. + checker.IPv4(t, c.GetPacket(), + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS)+2), + checker.AckNum(0), + checker.TCPFlags(header.TCPFlagRst), + ), + ) +} + func TestSimpleReceive(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() @@ -474,6 +633,352 @@ 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))) +} + +// TestTCPAckBeforeAcceptV4 tests that once the 3-way handshake is complete, +// peers can send data and expect a response within a reasonable ammount of time +// without calling Accept on the listening endpoint first. +// +// This test uses IPv4. +func TestTCPAckBeforeAcceptV4(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) + } + + irs, iss := executeHandshake(t, c, context.TestPort, false /* synCookiesInUse */) + + // Send data before accepting the connection. + c.SendPacket([]byte{1, 2, 3, 4}, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: irs + 1, + AckNum: iss + 1, + }) + + // Receive ACK for the data we sent. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagAck), + checker.SeqNum(uint32(iss+1)), + checker.AckNum(uint32(irs+5)))) +} + +// TestTCPAckBeforeAcceptV6 tests that once the 3-way handshake is complete, +// peers can send data and expect a response within a reasonable ammount of time +// without calling Accept on the listening endpoint first. +// +// This test uses IPv6. +func TestTCPAckBeforeAcceptV6(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) + } + + irs, iss := executeV6Handshake(t, c, context.TestPort, false /* synCookiesInUse */) + + // Send data before accepting the connection. + c.SendV6Packet([]byte{1, 2, 3, 4}, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: irs + 1, + AckNum: iss + 1, + }) + + // Receive ACK for the data we sent. + checker.IPv6(t, c.GetV6Packet(), checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagAck), + checker.SeqNum(uint32(iss+1)), + checker.AckNum(uint32(irs+5)))) +} + +func TestSendRstOnListenerRxAckV4(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.Create(-1 /* epRcvBuf */) + + if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatal("Bind failed:", err) + } + + if err := c.EP.Listen(10 /* backlog */); err != nil { + t.Fatal("Listen failed:", err) + } + + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagFin | header.TCPFlagAck, + SeqNum: 100, + AckNum: 200, + }) + + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagRst), + checker.SeqNum(200))) +} + +func TestSendRstOnListenerRxAckV6(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.CreateV6Endpoint(true /* v6Only */) + + if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatal("Bind failed:", err) + } + + if err := c.EP.Listen(10 /* backlog */); err != nil { + t.Fatal("Listen failed:", err) + } + + c.SendV6Packet(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagFin | header.TCPFlagAck, + SeqNum: 100, + AckNum: 200, + }) + + checker.IPv6(t, c.GetV6Packet(), checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagRst), + checker.SeqNum(200))) +} + +// TestListenShutdown tests for the listening endpoint not processing +// any receive when it is on read shutdown. +func TestListenShutdown(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.Create(-1 /* epRcvBuf */) + + if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatal("Bind failed:", err) + } + + if err := c.EP.Listen(10 /* backlog */); err != nil { + t.Fatal("Listen failed:", err) + } + + if err := c.EP.Shutdown(tcpip.ShutdownRead); err != nil { + t.Fatal("Shutdown failed:", err) + } + + // Wait for the endpoint state to be propagated. + time.Sleep(10 * time.Millisecond) + + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: 100, + AckNum: 200, + }) + + c.CheckNoPacket("Packet received when listening socket was shutdown") +} + func TestTOSV4(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() @@ -635,6 +1140,71 @@ func TestConnectBindToDevice(t *testing.T) { } } +func TestRstOnSynSent(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + // Create an endpoint, don't handshake because we want to interfere with the + // handshake process. + c.Create(-1) + + // Start connection attempt. + waitEntry, ch := waiter.NewChannelEntry(nil) + c.WQ.EventRegister(&waitEntry, waiter.EventOut) + defer c.WQ.EventUnregister(&waitEntry) + + addr := tcpip.FullAddress{Addr: context.TestAddr, Port: context.TestPort} + if err := c.EP.Connect(addr); err != tcpip.ErrConnectStarted { + t.Fatalf("got Connect(%+v) = %v, want %s", addr, err, tcpip.ErrConnectStarted) + } + + // Receive SYN packet. + b := c.GetPacket() + checker.IPv4(t, b, + checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagSyn), + ), + ) + + // Ensure that we've reached SynSent state + if got, want := tcp.EndpointState(c.EP.State()), tcp.StateSynSent; got != want { + t.Fatalf("got State() = %s, want %s", got, want) + } + tcpHdr := header.TCP(header.IPv4(b).Payload()) + c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) + + // Send a packet with a proper ACK and a RST flag to cause the socket + // to Error and close out + iss := seqnum.Value(789) + rcvWnd := seqnum.Size(30000) + c.SendPacket(nil, &context.Headers{ + SrcPort: tcpHdr.DestinationPort(), + DstPort: tcpHdr.SourcePort(), + Flags: header.TCPFlagRst | header.TCPFlagAck, + SeqNum: iss, + AckNum: c.IRS.Add(1), + RcvWnd: rcvWnd, + TCPOpts: nil, + }) + + // Wait for receive to be notified. + select { + case <-ch: + case <-time.After(3 * time.Second): + t.Fatal("timed out waiting for packet to arrive") + } + + if _, _, err := c.EP.Read(nil); err != tcpip.ErrConnectionRefused { + t.Fatalf("got c.EP.Read(nil) = %v, want = %s", err, tcpip.ErrConnectionRefused) + } + + // Due to the RST the endpoint should be in an error state. + if got, want := tcp.EndpointState(c.EP.State()), tcp.StateError; got != want { + t.Fatalf("got State() = %s, want %s", got, want) + } +} + func TestOutOfOrderReceive(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() @@ -930,8 +1500,7 @@ 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 { @@ -1623,7 +2192,7 @@ 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}} { @@ -1671,7 +2240,7 @@ 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 { @@ -1704,7 +2273,7 @@ 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() @@ -1741,7 +2310,7 @@ 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)) }}, } @@ -2211,6 +2780,13 @@ loop: if tcp.EndpointState(c.EP.State()) != tcp.StateError { t.Fatalf("got EP state is not StateError") } + + if got := c.Stack().Stats().TCP.EstablishedResets.Value(); got != 1 { + t.Errorf("got stats.TCP.EstablishedResets.Value() = %v, want = 1", got) + } + if got := c.Stack().Stats().TCP.CurrentEstablished.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentEstablished.Value() = %v, want = 0", got) + } } func TestSendOnResetConnection(t *testing.T) { @@ -2905,6 +3481,13 @@ 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) @@ -2912,12 +3495,12 @@ 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(), @@ -2955,10 +3538,9 @@ 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) @@ -2975,7 +3557,7 @@ 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] @@ -2986,7 +3568,7 @@ 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) { @@ -2996,11 +3578,11 @@ 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) } } @@ -3773,8 +4355,9 @@ func TestKeepalive(t *testing.T) { c.CreateConnected(789, 30000, -1 /* epRcvBuf */) + const keepAliveInterval = 10 * time.Millisecond c.EP.SetSockOpt(tcpip.KeepaliveIdleOption(10 * time.Millisecond)) - c.EP.SetSockOpt(tcpip.KeepaliveIntervalOption(10 * time.Millisecond)) + c.EP.SetSockOpt(tcpip.KeepaliveIntervalOption(keepAliveInterval)) c.EP.SetSockOpt(tcpip.KeepaliveCountOption(5)) c.EP.SetSockOpt(tcpip.KeepaliveEnabledOption(1)) @@ -3864,19 +4447,43 @@ func TestKeepalive(t *testing.T) { ) } + // Sleep for a litte over the KeepAlive interval to make sure + // the timer has time to fire after the last ACK and close the + // close the socket. + time.Sleep(keepAliveInterval + 5*time.Millisecond) + // The connection should be terminated after 5 unacked keepalives. + // Send an ACK to trigger a RST from the stack as the endpoint should + // be dead. + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + Flags: header.TCPFlagAck, + SeqNum: 790, + AckNum: seqnum.Value(next), + RcvWnd: 30000, + }) + checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), checker.SeqNum(uint32(next)), - checker.AckNum(uint32(790)), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagRst), + checker.AckNum(uint32(0)), + checker.TCPFlags(header.TCPFlagRst), ), ) + if got := c.Stack().Stats().TCP.EstablishedTimedout.Value(); got != 1 { + t.Errorf("got c.Stack().Stats().TCP.EstablishedTimedout.Value() = %v, want = 1", got) + } + if _, _, err := c.EP.Read(nil); err != tcpip.ErrTimeout { t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrTimeout) } + + if got := c.Stack().Stats().TCP.CurrentEstablished.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentEstablished.Value() = %v, want = 0", got) + } } func executeHandshake(t *testing.T, c *context.Context, srcPort uint16, synCookieInUse bool) (irs, iss seqnum.Value) { @@ -3890,7 +4497,7 @@ func executeHandshake(t *testing.T, c *context.Context, srcPort uint16, synCooki RcvWnd: 30000, }) - // Receive the SYN-ACK reply.w + // Receive the SYN-ACK reply. b := c.GetPacket() tcp := header.TCP(header.IPv4(b).Payload()) iss = seqnum.Value(tcp.SequenceNumber()) @@ -3923,6 +4530,50 @@ func executeHandshake(t *testing.T, c *context.Context, srcPort uint16, synCooki return irs, iss } +func executeV6Handshake(t *testing.T, c *context.Context, srcPort uint16, synCookieInUse bool) (irs, iss seqnum.Value) { + // Send a SYN request. + irs = seqnum.Value(789) + c.SendV6Packet(nil, &context.Headers{ + SrcPort: srcPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: irs, + RcvWnd: 30000, + }) + + // Receive the SYN-ACK reply. + b := c.GetV6Packet() + tcp := header.TCP(header.IPv6(b).Payload()) + iss = seqnum.Value(tcp.SequenceNumber()) + tcpCheckers := []checker.TransportChecker{ + checker.SrcPort(context.StackPort), + checker.DstPort(srcPort), + checker.TCPFlags(header.TCPFlagAck | header.TCPFlagSyn), + checker.AckNum(uint32(irs) + 1), + } + + if synCookieInUse { + // When cookies are in use window scaling is disabled. + tcpCheckers = append(tcpCheckers, checker.TCPSynOptions(header.TCPSynOptions{ + WS: -1, + MSS: c.MSSWithoutOptionsV6(), + })) + } + + checker.IPv6(t, b, checker.TCP(tcpCheckers...)) + + // Send ACK. + c.SendV6Packet(nil, &context.Headers{ + SrcPort: srcPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: irs + 1, + AckNum: iss + 1, + RcvWnd: 30000, + }) + return irs, iss +} + // TestListenBacklogFull tests that netstack does not complete handshakes if the // listen backlog for the endpoint is full. func TestListenBacklogFull(t *testing.T) { @@ -4025,6 +4676,210 @@ func TestListenBacklogFull(t *testing.T) { } } +// TestListenNoAcceptMulticastBroadcastV4 makes sure that TCP segments with a +// non unicast IPv4 address are not accepted. +func TestListenNoAcceptNonUnicastV4(t *testing.T) { + multicastAddr := tcpip.Address("\xe0\x00\x01\x02") + otherMulticastAddr := tcpip.Address("\xe0\x00\x01\x03") + + tests := []struct { + name string + srcAddr tcpip.Address + dstAddr tcpip.Address + }{ + { + "SourceUnspecified", + header.IPv4Any, + context.StackAddr, + }, + { + "SourceBroadcast", + header.IPv4Broadcast, + context.StackAddr, + }, + { + "SourceOurMulticast", + multicastAddr, + context.StackAddr, + }, + { + "SourceOtherMulticast", + otherMulticastAddr, + context.StackAddr, + }, + { + "DestUnspecified", + context.TestAddr, + header.IPv4Any, + }, + { + "DestBroadcast", + context.TestAddr, + header.IPv4Broadcast, + }, + { + "DestOurMulticast", + context.TestAddr, + multicastAddr, + }, + { + "DestOtherMulticast", + context.TestAddr, + otherMulticastAddr, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + t.Parallel() + + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.Create(-1) + + if err := c.Stack().JoinGroup(header.IPv4ProtocolNumber, 1, multicastAddr); err != nil { + t.Fatalf("JoinGroup failed: %s", err) + } + + if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatalf("Bind failed: %s", err) + } + + if err := c.EP.Listen(1); err != nil { + t.Fatalf("Listen failed: %s", err) + } + + irs := seqnum.Value(789) + c.SendPacketWithAddrs(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: irs, + RcvWnd: 30000, + }, test.srcAddr, test.dstAddr) + c.CheckNoPacket("Should not have received a response") + + // Handle normal packet. + c.SendPacketWithAddrs(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: irs, + RcvWnd: 30000, + }, context.TestAddr, context.StackAddr) + checker.IPv4(t, c.GetPacket(), + checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagAck|header.TCPFlagSyn), + checker.AckNum(uint32(irs)+1))) + }) + } +} + +// TestListenNoAcceptMulticastBroadcastV6 makes sure that TCP segments with a +// non unicast IPv6 address are not accepted. +func TestListenNoAcceptNonUnicastV6(t *testing.T) { + multicastAddr := tcpip.Address("\xff\x0e\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\x01") + otherMulticastAddr := tcpip.Address("\xff\x0e\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\x02") + + tests := []struct { + name string + srcAddr tcpip.Address + dstAddr tcpip.Address + }{ + { + "SourceUnspecified", + header.IPv6Any, + context.StackV6Addr, + }, + { + "SourceAllNodes", + header.IPv6AllNodesMulticastAddress, + context.StackV6Addr, + }, + { + "SourceOurMulticast", + multicastAddr, + context.StackV6Addr, + }, + { + "SourceOtherMulticast", + otherMulticastAddr, + context.StackV6Addr, + }, + { + "DestUnspecified", + context.TestV6Addr, + header.IPv6Any, + }, + { + "DestAllNodes", + context.TestV6Addr, + header.IPv6AllNodesMulticastAddress, + }, + { + "DestOurMulticast", + context.TestV6Addr, + multicastAddr, + }, + { + "DestOtherMulticast", + context.TestV6Addr, + otherMulticastAddr, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + t.Parallel() + + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.CreateV6Endpoint(true) + + if err := c.Stack().JoinGroup(header.IPv6ProtocolNumber, 1, multicastAddr); err != nil { + t.Fatalf("JoinGroup failed: %s", err) + } + + if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatalf("Bind failed: %s", err) + } + + if err := c.EP.Listen(1); err != nil { + t.Fatalf("Listen failed: %s", err) + } + + irs := seqnum.Value(789) + c.SendV6PacketWithAddrs(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: irs, + RcvWnd: 30000, + }, test.srcAddr, test.dstAddr) + c.CheckNoPacket("Should not have received a response") + + // Handle normal packet. + c.SendV6PacketWithAddrs(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: irs, + RcvWnd: 30000, + }, context.TestV6Addr, context.StackV6Addr) + checker.IPv6(t, c.GetV6Packet(), + checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagAck|header.TCPFlagSyn), + checker.AckNum(uint32(irs)+1))) + }) + } +} + func TestListenSynRcvdQueueFull(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() @@ -4056,7 +4911,7 @@ func TestListenSynRcvdQueueFull(t *testing.T) { SrcPort: context.TestPort, DstPort: context.StackPort, Flags: header.TCPFlagSyn, - SeqNum: seqnum.Value(789), + SeqNum: irs, RcvWnd: 30000, }) @@ -4167,7 +5022,8 @@ 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, @@ -4207,6 +5063,125 @@ func TestListenBacklogFullSynCookieInUse(t *testing.T) { } } +func TestSynRcvdBadSeqNumber(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + // Create TCP endpoint. + var err *tcpip.Error + c.EP, err = c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &c.WQ) + if err != nil { + t.Fatalf("NewEndpoint failed: %s", err) + } + + // Bind to wildcard. + if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatalf("Bind failed: %s", err) + } + + // Start listening. + if err := c.EP.Listen(10); err != nil { + t.Fatalf("Listen failed: %s", err) + } + + // Send a SYN to get a SYN-ACK. This should put the ep into SYN-RCVD state + irs := seqnum.Value(789) + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: irs, + RcvWnd: 30000, + }) + + // Receive the SYN-ACK reply. + b := c.GetPacket() + tcpHdr := header.TCP(header.IPv4(b).Payload()) + iss := seqnum.Value(tcpHdr.SequenceNumber()) + tcpCheckers := []checker.TransportChecker{ + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagAck | header.TCPFlagSyn), + checker.AckNum(uint32(irs) + 1), + } + checker.IPv4(t, b, checker.TCP(tcpCheckers...)) + + // Now send a packet with an out-of-window sequence number + largeSeqnum := irs + seqnum.Value(tcpHdr.WindowSize()) + 1 + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: largeSeqnum, + AckNum: iss + 1, + RcvWnd: 30000, + }) + + // Should receive an ACK with the expected SEQ number + b = c.GetPacket() + tcpCheckers = []checker.TransportChecker{ + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagAck), + checker.AckNum(uint32(irs) + 1), + checker.SeqNum(uint32(iss + 1)), + } + checker.IPv4(t, b, checker.TCP(tcpCheckers...)) + + // Now that the socket replied appropriately with the ACK, + // complete the connection to test that the large SEQ num + // did not change the state from SYN-RCVD. + + // Send ACK to move to ESTABLISHED state. + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: irs + 1, + AckNum: iss + 1, + RcvWnd: 30000, + }) + + newEP, _, err := c.EP.Accept() + + if err != nil && err != tcpip.ErrWouldBlock { + t.Fatalf("Accept failed: %s", err) + } + + if err == tcpip.ErrWouldBlock { + // Try to accept the connections in the backlog. + we, ch := waiter.NewChannelEntry(nil) + c.WQ.EventRegister(&we, waiter.EventIn) + defer c.WQ.EventUnregister(&we) + + // Wait for connection to be established. + select { + case <-ch: + newEP, _, err = c.EP.Accept() + if err != nil { + t.Fatalf("Accept failed: %s", err) + } + + case <-time.After(1 * time.Second): + t.Fatalf("Timed out waiting for accept") + } + } + + // Now verify that the TCP socket is usable and in a connected state. + data := "Don't panic" + _, _, err = newEP.Write(tcpip.SlicePayload(buffer.NewViewFromBytes([]byte(data))), tcpip.WriteOptions{}) + + if err != nil { + t.Fatalf("Write failed: %s", err) + } + + pkt := c.GetPacket() + tcpHdr = header.TCP(header.IPv4(pkt).Payload()) + if string(tcpHdr.Payload()) != data { + t.Fatalf("Unexpected data: got %s, want %s", string(tcpHdr.Payload()), data) + } +} + func TestPassiveConnectionAttemptIncrement(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() @@ -4381,6 +5356,9 @@ 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) @@ -4668,3 +5646,918 @@ 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, + RcvWnd: 30000, + }) + + // 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, + RcvWnd: 30000, + }) + + // 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, + RcvWnd: 30000, + }) + + // 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, + RcvWnd: 30000, + }) + + 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, + RcvWnd: 30000, + }) + + // 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) + } + + want := c.Stack().Stats().TCP.EstablishedClosed.Value() + 1 + + 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, + RcvWnd: 30000, + }) + + // 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(0), + checker.TCPFlags(header.TCPFlagRst))) + + if got := c.Stack().Stats().TCP.EstablishedClosed.Value(); got != want { + t.Errorf("got c.Stack().Stats().TCP.EstablishedClosed = %v, want = %v", got, want) + } + if got := c.Stack().Stats().TCP.CurrentEstablished.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentEstablished.Value() = %v, want = 0", got) + } +} + +func TestTCPCloseWithData(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, + RcvWnd: 30000, + }) + + // 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, + RcvWnd: 30000, + } + + // 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") + } + } + + // Now trigger a passive close by sending a FIN. + finHeaders := &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck | header.TCPFlagFin, + SeqNum: iss + 1, + AckNum: c.IRS + 2, + RcvWnd: 30000, + } + + 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+1)), + checker.AckNum(uint32(iss)+2), + checker.TCPFlags(header.TCPFlagAck))) + + // Now write a few bytes and then close the endpoint. + 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(uint32(iss)+2), // Acknum is initial sequence number + 1 + checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), + ), + ) + + if p := b[header.IPv4MinimumSize+header.TCPMinimumSize:]; !bytes.Equal(data, p) { + t.Errorf("got data = %x, want = %x", p, data) + } + + c.EP.Close() + // Check the FIN. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+1)+uint32(len(data))), + checker.AckNum(uint32(iss+2)), + checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) + + // First send a partial ACK. + ackHeaders = &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: iss + 2, + AckNum: c.IRS + 1 + seqnum.Value(len(data)-1), + RcvWnd: 30000, + } + c.SendPacket(nil, ackHeaders) + + // Now send a full ACK. + ackHeaders = &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: iss + 2, + AckNum: c.IRS + 1 + seqnum.Value(len(data)), + RcvWnd: 30000, + } + c.SendPacket(nil, ackHeaders) + + // Now ACK the FIN. + ackHeaders.AckNum++ + c.SendPacket(nil, ackHeaders) + + // Now send an ACK and we should get a RST back as the endpoint should + // be in CLOSED state. + ackHeaders = &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: iss + 2, + AckNum: c.IRS + 1 + seqnum.Value(len(data)), + RcvWnd: 30000, + } + c.SendPacket(nil, ackHeaders) + + // Check the RST. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(ackHeaders.AckNum)), + checker.AckNum(0), + checker.TCPFlags(header.TCPFlagRst))) +} + +func TestTCPUserTimeout(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) + + origEstablishedTimedout := c.Stack().Stats().TCP.EstablishedTimedout.Value() + + userTimeout := 50 * time.Millisecond + c.EP.SetSockOpt(tcpip.TCPUserTimeoutOption(userTimeout)) + + // Send some data and wait before ACKing it. + view := buffer.NewView(3) + if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { + t.Fatalf("Write failed: %v", err) + } + + next := uint32(c.IRS) + 1 + checker.IPv4(t, c.GetPacket(), + checker.PayloadLen(len(view)+header.TCPMinimumSize), + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(next), + checker.AckNum(790), + checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), + ), + ) + + // Wait for a little over the minimum retransmit timeout of 200ms for + // the retransmitTimer to fire and close the connection. + time.Sleep(tcp.MinRTO + 10*time.Millisecond) + + // No packet should be received as the connection should be silently + // closed due to timeout. + c.CheckNoPacket("unexpected packet received after userTimeout has expired") + + next += uint32(len(view)) + + // The connection should be terminated after userTimeout has expired. + // Send an ACK to trigger a RST from the stack as the endpoint should + // be dead. + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + Flags: header.TCPFlagAck, + SeqNum: 790, + AckNum: seqnum.Value(next), + RcvWnd: 30000, + }) + + checker.IPv4(t, c.GetPacket(), + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(next)), + checker.AckNum(uint32(0)), + checker.TCPFlags(header.TCPFlagRst), + ), + ) + + if _, _, err := c.EP.Read(nil); err != tcpip.ErrTimeout { + t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrTimeout) + } + + if got, want := c.Stack().Stats().TCP.EstablishedTimedout.Value(), origEstablishedTimedout+1; got != want { + t.Errorf("got c.Stack().Stats().TCP.EstablishedTimedout = %v, want = %v", got, want) + } +} + +func TestKeepaliveWithUserTimeout(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) + + origEstablishedTimedout := c.Stack().Stats().TCP.EstablishedTimedout.Value() + + const keepAliveInterval = 10 * time.Millisecond + c.EP.SetSockOpt(tcpip.KeepaliveIdleOption(10 * time.Millisecond)) + c.EP.SetSockOpt(tcpip.KeepaliveIntervalOption(keepAliveInterval)) + c.EP.SetSockOpt(tcpip.KeepaliveCountOption(10)) + c.EP.SetSockOpt(tcpip.KeepaliveEnabledOption(1)) + + // Set userTimeout to be the duration for 3 keepalive probes. + userTimeout := 30 * time.Millisecond + c.EP.SetSockOpt(tcpip.TCPUserTimeoutOption(userTimeout)) + + // Check that the connection is still alive. + if _, _, err := c.EP.Read(nil); err != tcpip.ErrWouldBlock { + t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrWouldBlock) + } + + // Now receive 2 keepalives, but don't ACK them. The connection should + // be reset when the 3rd one should be sent due to userTimeout being + // 30ms and each keepalive probe should be sent 10ms apart as set above after + // the connection has been idle for 10ms. + for i := 0; i < 2; i++ { + b := c.GetPacket() + checker.IPv4(t, b, + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS)), + checker.AckNum(uint32(790)), + checker.TCPFlags(header.TCPFlagAck), + ), + ) + } + + // Sleep for a litte over the KeepAlive interval to make sure + // the timer has time to fire after the last ACK and close the + // close the socket. + time.Sleep(keepAliveInterval + 5*time.Millisecond) + + // The connection should be terminated after 30ms. + // Send an ACK to trigger a RST from the stack as the endpoint should + // be dead. + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + Flags: header.TCPFlagAck, + SeqNum: 790, + AckNum: seqnum.Value(c.IRS + 1), + RcvWnd: 30000, + }) + + checker.IPv4(t, c.GetPacket(), + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+1)), + checker.AckNum(uint32(0)), + checker.TCPFlags(header.TCPFlagRst), + ), + ) + + if _, _, err := c.EP.Read(nil); err != tcpip.ErrTimeout { + t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrTimeout) + } + if got, want := c.Stack().Stats().TCP.EstablishedTimedout.Value(), origEstablishedTimedout+1; got != want { + t.Errorf("got c.Stack().Stats().TCP.EstablishedTimedout = %v, want = %v", got, want) + } +} diff --git a/pkg/tcpip/transport/tcp/testing/context/BUILD b/pkg/tcpip/transport/tcp/testing/context/BUILD index 19b0d31c5..b33ec2087 100644 --- a/pkg/tcpip/transport/tcp/testing/context/BUILD +++ b/pkg/tcpip/transport/tcp/testing/context/BUILD @@ -8,7 +8,7 @@ go_library( srcs = ["context.go"], importpath = "gvisor.dev/gvisor/pkg/tcpip/transport/tcp/testing/context", visibility = [ - "//:sandbox", + "//visibility:public", ], deps = [ "//pkg/tcpip", diff --git a/pkg/tcpip/transport/tcp/testing/context/context.go b/pkg/tcpip/transport/tcp/testing/context/context.go index ef823e4ae..b0a376eba 100644 --- a/pkg/tcpip/transport/tcp/testing/context/context.go +++ b/pkg/tcpip/transport/tcp/testing/context/context.go @@ -231,14 +231,15 @@ func (c *Context) CheckNoPacket(errMsg string) { // addresses. It will fail with an error if no packet is received for // 2 seconds. func (c *Context) GetPacket() []byte { + c.t.Helper() select { case p := <-c.linkEP.C: if p.Proto != ipv4.ProtocolNumber { c.t.Fatalf("Bad network protocol: got %v, wanted %v", p.Proto, ipv4.ProtocolNumber) } - b := make([]byte, len(p.Header)+len(p.Payload)) - copy(b, p.Header) - copy(b[len(p.Header):], p.Payload) + + hdr := p.Pkt.Header.View() + b := append(hdr[:len(hdr):len(hdr)], p.Pkt.Data.ToView()...) if p.GSO != nil && p.GSO.L3HdrLen != header.IPv4MinimumSize { c.t.Errorf("L3HdrLen %v (expected %v)", p.GSO.L3HdrLen, header.IPv4MinimumSize) @@ -259,14 +260,15 @@ func (c *Context) GetPacket() []byte { // and destination address. If no packet is available it will return // nil immediately. func (c *Context) GetPacketNonBlocking() []byte { + c.t.Helper() select { case p := <-c.linkEP.C: if p.Proto != ipv4.ProtocolNumber { c.t.Fatalf("Bad network protocol: got %v, wanted %v", p.Proto, ipv4.ProtocolNumber) } - b := make([]byte, len(p.Header)+len(p.Payload)) - copy(b, p.Header) - copy(b[len(p.Header):], p.Payload) + + hdr := p.Pkt.Header.View() + b := append(hdr[:len(hdr):len(hdr)], p.Pkt.Data.ToView()...) checker.IPv4(c.t, b, checker.SrcAddr(StackAddr), checker.DstAddr(TestAddr)) return b @@ -302,11 +304,19 @@ 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. func (c *Context) BuildSegment(payload []byte, h *Headers) buffer.VectorisedView { + return c.BuildSegmentWithAddrs(payload, h, TestAddr, StackAddr) +} + +// BuildSegmentWithAddrs builds a TCP segment based on the given Headers, +// payload and source and destination IPv4 addresses. +func (c *Context) BuildSegmentWithAddrs(payload []byte, h *Headers, src, dst tcpip.Address) buffer.VectorisedView { // Allocate a buffer for data and headers. buf := buffer.NewView(header.TCPMinimumSize + header.IPv4MinimumSize + len(h.TCPOpts) + len(payload)) copy(buf[len(buf)-len(payload):], payload) @@ -319,8 +329,8 @@ func (c *Context) BuildSegment(payload []byte, h *Headers) buffer.VectorisedView TotalLength: uint16(len(buf)), TTL: 65, Protocol: uint8(tcp.ProtocolNumber), - SrcAddr: TestAddr, - DstAddr: StackAddr, + SrcAddr: src, + DstAddr: dst, }) ip.SetChecksum(^ip.CalculateChecksum()) @@ -337,7 +347,7 @@ func (c *Context) BuildSegment(payload []byte, h *Headers) buffer.VectorisedView }) // Calculate the TCP pseudo-header checksum. - xsum := header.PseudoHeaderChecksum(tcp.ProtocolNumber, TestAddr, StackAddr, uint16(len(t))) + xsum := header.PseudoHeaderChecksum(tcp.ProtocolNumber, src, dst, uint16(len(t))) // Calculate the TCP checksum and set it. xsum = header.Checksum(payload, xsum) @@ -350,13 +360,26 @@ 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), + }) +} + +// SendPacketWithAddrs builds and sends a TCP segment(with the provided payload +// & TCPheaders) in an IPv4 packet via the link layer endpoint using the +// provided source and destination IPv4 addresses. +func (c *Context) SendPacketWithAddrs(payload []byte, h *Headers, src, dst tcpip.Address) { + c.linkEP.InjectInbound(ipv4.ProtocolNumber, tcpip.PacketBuffer{ + Data: c.BuildSegmentWithAddrs(payload, h, src, dst), + }) } // SendAck sends an ACK packet. @@ -462,14 +485,15 @@ func (c *Context) CreateV6Endpoint(v6only bool) { // GetV6Packet reads a single packet from the link layer endpoint of the context // and asserts that it is an IPv6 Packet with the expected src/dest addresses. func (c *Context) GetV6Packet() []byte { + c.t.Helper() select { case p := <-c.linkEP.C: if p.Proto != ipv6.ProtocolNumber { c.t.Fatalf("Bad network protocol: got %v, wanted %v", p.Proto, ipv6.ProtocolNumber) } - b := make([]byte, len(p.Header)+len(p.Payload)) - copy(b, p.Header) - copy(b[len(p.Header):], p.Payload) + b := make([]byte, p.Pkt.Header.UsedLength()+p.Pkt.Data.Size()) + copy(b, p.Pkt.Header.View()) + copy(b[p.Pkt.Header.UsedLength():], p.Pkt.Data.ToView()) checker.IPv6(c.t, b, checker.SrcAddr(StackV6Addr), checker.DstAddr(TestV6Addr)) return b @@ -484,6 +508,13 @@ func (c *Context) GetV6Packet() []byte { // SendV6Packet builds and sends an IPv6 Packet via the link layer endpoint of // the context. func (c *Context) SendV6Packet(payload []byte, h *Headers) { + c.SendV6PacketWithAddrs(payload, h, TestV6Addr, StackV6Addr) +} + +// SendV6PacketWithAddrs builds and sends an IPv6 Packet via the link layer +// endpoint of the context using the provided source and destination IPv6 +// addresses. +func (c *Context) SendV6PacketWithAddrs(payload []byte, h *Headers, src, dst tcpip.Address) { // Allocate a buffer for data and headers. buf := buffer.NewView(header.TCPMinimumSize + header.IPv6MinimumSize + len(payload)) copy(buf[len(buf)-len(payload):], payload) @@ -494,8 +525,8 @@ func (c *Context) SendV6Packet(payload []byte, h *Headers) { PayloadLength: uint16(header.TCPMinimumSize + len(payload)), NextHeader: uint8(tcp.ProtocolNumber), HopLimit: 65, - SrcAddr: TestV6Addr, - DstAddr: StackV6Addr, + SrcAddr: src, + DstAddr: dst, }) // Initialize the TCP header. @@ -511,14 +542,16 @@ func (c *Context) SendV6Packet(payload []byte, h *Headers) { }) // Calculate the TCP pseudo-header checksum. - xsum := header.PseudoHeaderChecksum(tcp.ProtocolNumber, TestV6Addr, StackV6Addr, uint16(len(t))) + xsum := header.PseudoHeaderChecksum(tcp.ProtocolNumber, src, dst, uint16(len(t))) // Calculate the TCP checksum and set it. xsum = header.Checksum(payload, xsum) 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. @@ -1059,3 +1092,9 @@ func (c *Context) SetGSOEnabled(enable bool) { func (c *Context) MSSWithoutOptions() uint16 { return uint16(c.linkEP.MTU() - header.IPv4MinimumSize - header.TCPMinimumSize) } + +// MSSWithoutOptionsV6 returns the value for the MSS used by the stack when no +// options are in use for IPv6 packets. +func (c *Context) MSSWithoutOptionsV6() uint16 { + return uint16(c.linkEP.MTU() - header.IPv6MinimumSize - header.TCPMinimumSize) +} diff --git a/pkg/tcpip/transport/udp/BUILD b/pkg/tcpip/transport/udp/BUILD index c9460aa0d..97e4d5825 100644 --- a/pkg/tcpip/transport/udp/BUILD +++ b/pkg/tcpip/transport/udp/BUILD @@ -34,6 +34,7 @@ go_library( "//pkg/tcpip/buffer", "//pkg/tcpip/header", "//pkg/tcpip/iptables", + "//pkg/tcpip/ports", "//pkg/tcpip/stack", "//pkg/tcpip/transport/raw", "//pkg/waiter", @@ -59,11 +60,3 @@ go_test( "//pkg/waiter", ], ) - -filegroup( - name = "autogen", - srcs = [ - "udp_packet_list.go", - ], - visibility = ["//:sandbox"], -) diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 6e87245b7..1ac4705af 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/iptables" + "gvisor.dev/gvisor/pkg/tcpip/ports" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/waiter" ) @@ -31,9 +32,6 @@ 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. @@ -80,6 +78,7 @@ type endpoint struct { // change throughout the lifetime of the endpoint. stack *stack.Stack `state:"manual"` waiterQueue *waiter.Queue + uniqueID uint64 // The following fields are used to manage the receive queue, and are // protected by rcvMu. @@ -106,6 +105,11 @@ type endpoint struct { bindToDevice tcpip.NICID broadcast bool + // Values used to reserve a port or register a transport endpoint. + // (which ever happens first). + boundBindToDevice tcpip.NICID + boundPortFlags ports.Flags + // sendTOS represents IPv4 TOS or IPv6 TrafficClass, // applied while sending packets. Defaults to 0 as on Linux. sendTOS uint8 @@ -140,7 +144,7 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue stack: s, TransportEndpointInfo: stack.TransportEndpointInfo{ NetProto: netProto, - TransProto: header.TCPProtocolNumber, + TransProto: header.UDPProtocolNumber, }, waiterQueue: waiterQueue, // RFC 1075 section 5.4 recommends a TTL of 1 for membership @@ -160,9 +164,15 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue 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() { @@ -171,8 +181,10 @@ func (e *endpoint) Close() { switch e.state { case StateBound, StateConnected: - e.stack.UnregisterTransportEndpoint(e.RegisterNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.bindToDevice) - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.bindToDevice) + e.stack.UnregisterTransportEndpoint(e.RegisterNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.boundBindToDevice) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.boundPortFlags, e.boundBindToDevice) + e.boundBindToDevice = 0 + e.boundPortFlags = ports.Flags{} } for _, mem := range e.multicastMemberships { @@ -278,7 +290,7 @@ 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) { +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). @@ -286,20 +298,20 @@ func (e *endpoint) connectRoute(nicid tcpip.NICID, addr tcpip.FullAddress, netPr } 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 @@ -378,13 +390,13 @@ 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 + nicID := to.NIC if e.BindNICID != 0 { - if nicid != 0 && nicid != e.BindNICID { + if nicID != 0 && nicID != e.BindNICID { return 0, nil, tcpip.ErrNoRoute } - nicid = e.BindNICID + nicID = e.BindNICID } if to.Addr == header.IPv4Broadcast && !e.broadcast { @@ -396,7 +408,7 @@ 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 } @@ -618,9 +630,9 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.bindToDevice = 0 return nil } - for nicid, nic := range e.stack.NICInfo() { + for nicID, nic := range e.stack.NICInfo() { if nic.Name == string(v) { - e.bindToDevice = nicid + e.bindToDevice = nicID return nil } } @@ -728,6 +740,10 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { *o = tcpip.MulticastLoopOption(v) return nil + case *tcpip.ReuseAddressOption: + *o = 0 + return nil + case *tcpip.ReusePortOption: e.mu.RLock() v := e.reusePort @@ -809,7 +825,11 @@ func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort u if useDefaultTTL { ttl = r.DefaultTTL() } - if err := r.WritePacket(nil /* gso */, hdr, data, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}); err != nil { + if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}, tcpip.PacketBuffer{ + Header: hdr, + Data: data, + TransportHeader: buffer.View(udp), + }); err != nil { r.Stats().UDP.PacketSendErrors.Increment() return err } @@ -859,7 +879,10 @@ func (e *endpoint) Disconnect() *tcpip.Error { if e.state != StateConnected { return nil } - id := stack.TransportEndpointID{} + var ( + id stack.TransportEndpointID + btd tcpip.NICID + ) // Exclude ephemerally bound endpoints. if e.BindNICID != 0 || e.ID.LocalAddress == "" { var err *tcpip.Error @@ -867,7 +890,7 @@ func (e *endpoint) Disconnect() *tcpip.Error { LocalPort: e.ID.LocalPort, LocalAddress: e.ID.LocalAddress, } - id, err = e.registerWithStack(e.RegisterNICID, e.effectiveNetProtos, id) + id, btd, err = e.registerWithStack(e.RegisterNICID, e.effectiveNetProtos, id) if err != nil { return err } @@ -875,13 +898,15 @@ func (e *endpoint) Disconnect() *tcpip.Error { } else { if e.ID.LocalPort != 0 { // Release the ephemeral port. - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.bindToDevice) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.boundPortFlags, e.boundBindToDevice) + e.boundPortFlags = ports.Flags{} } e.state = StateInitial } - e.stack.UnregisterTransportEndpoint(e.RegisterNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.bindToDevice) + e.stack.UnregisterTransportEndpoint(e.RegisterNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.boundBindToDevice) e.ID = id + e.boundBindToDevice = btd e.route.Release() e.route = stack.Route{} e.dstPort = 0 @@ -903,7 +928,7 @@ 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: @@ -913,16 +938,16 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { 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 } @@ -950,20 +975,21 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { } } - id, err = e.registerWithStack(nicid, netProtos, id) + id, btd, err := e.registerWithStack(nicID, netProtos, id) if err != nil { return err } // Remove the old registration. if e.ID.LocalPort != 0 { - e.stack.UnregisterTransportEndpoint(e.RegisterNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.bindToDevice) + e.stack.UnregisterTransportEndpoint(e.RegisterNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.boundBindToDevice) } e.ID = id + e.boundBindToDevice = btd e.route = r.Clone() e.dstPort = addr.Port - e.RegisterNICID = nicid + e.RegisterNICID = nicID e.effectiveNetProtos = netProtos e.state = StateConnected @@ -1018,20 +1044,27 @@ func (*endpoint) Accept() (tcpip.Endpoint, *waiter.Queue, *tcpip.Error) { return nil, nil, tcpip.ErrNotSupported } -func (e *endpoint) registerWithStack(nicid tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, id stack.TransportEndpointID) (stack.TransportEndpointID, *tcpip.Error) { +func (e *endpoint) registerWithStack(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, id stack.TransportEndpointID) (stack.TransportEndpointID, tcpip.NICID, *tcpip.Error) { if e.ID.LocalPort == 0 { - port, err := e.stack.ReservePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, e.reusePort, e.bindToDevice) + flags := ports.Flags{ + LoadBalanced: e.reusePort, + // FIXME(b/129164367): Support SO_REUSEADDR. + MostRecent: false, + } + port, err := e.stack.ReservePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, flags, e.bindToDevice) if err != nil { - return id, err + return id, e.bindToDevice, err } + e.boundPortFlags = flags id.LocalPort = port } - err := e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, id, e, e.reusePort, e.bindToDevice) + 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.bindToDevice) + e.stack.ReleasePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, e.boundPortFlags, e.bindToDevice) + e.boundPortFlags = ports.Flags{} } - return id, err + return id, e.bindToDevice, err } func (e *endpoint) bindLocked(addr tcpip.FullAddress) *tcpip.Error { @@ -1057,11 +1090,11 @@ 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 } } @@ -1070,13 +1103,14 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) *tcpip.Error { LocalPort: addr.Port, LocalAddress: addr.Addr, } - id, err = e.registerWithStack(nicid, netProtos, id) + id, btd, err := e.registerWithStack(nicID, netProtos, id) if err != nil { return err } e.ID = id - e.RegisterNICID = nicid + e.boundBindToDevice = btd + e.RegisterNICID = nicID e.effectiveNetProtos = netProtos // Mark endpoint as bound. @@ -1111,9 +1145,14 @@ func (e *endpoint) GetLocalAddress() (tcpip.FullAddress, *tcpip.Error) { e.mu.RLock() defer e.mu.RUnlock() + addr := e.ID.LocalAddress + if e.state == StateConnected { + addr = e.route.LocalAddress + } + return tcpip.FullAddress{ NIC: e.RegisterNICID, - Addr: e.ID.LocalAddress, + Addr: addr, Port: e.ID.LocalPort, }, nil } @@ -1154,17 +1193,17 @@ 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() @@ -1188,18 +1227,18 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, vv 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() @@ -1210,7 +1249,7 @@ 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. @@ -1234,6 +1273,9 @@ 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 b227e353b..43fb047ed 100644 --- a/pkg/tcpip/transport/udp/endpoint_state.go +++ b/pkg/tcpip/transport/udp/endpoint_state.go @@ -109,7 +109,7 @@ func (e *endpoint) Resume(s *stack.Stack) { // pass it to the reservation machinery. id := e.ID e.ID.LocalPort = 0 - e.ID, err = e.registerWithStack(e.RegisterNICID, e.effectiveNetProtos, id) + e.ID, e.boundBindToDevice, err = e.registerWithStack(e.RegisterNICID, e.effectiveNetProtos, id) if err != nil { panic(err) } diff --git a/pkg/tcpip/transport/udp/forwarder.go b/pkg/tcpip/transport/udp/forwarder.go index d399ec722..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 @@ -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 de026880f..259c3072a 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 +135,10 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans pkt.SetType(header.ICMPv4DstUnreachable) pkt.SetCode(header.ICMPv4PortUnreachable) pkt.SetChecksum(header.ICMPv4Checksum(pkt, payload)) - r.WritePacket(nil /* gso */, hdr, payload, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}) + r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + Data: payload, + }) case header.IPv6AddressSize: if !r.Stack().AllowICMPMessage() { @@ -151,12 +159,12 @@ 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 +172,10 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans pkt.SetType(header.ICMPv6DstUnreachable) pkt.SetCode(header.ICMPv6PortUnreachable) pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, payload)) - r.WritePacket(nil /* gso */, hdr, payload, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}) + r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + Header: hdr, + Data: payload, + }) } return true } diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index b724d788c..7051a7a9c 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -356,9 +356,9 @@ func (c *testContext) getPacketAndVerify(flow testFlow, checkers ...checker.Netw if p.Proto != flow.netProto() { c.t.Fatalf("Bad network protocol: got %v, wanted %v", p.Proto, flow.netProto()) } - b := make([]byte, len(p.Header)+len(p.Payload)) - copy(b, p.Header) - copy(b[len(p.Header):], p.Payload) + + hdr := p.Pkt.Header.View() + b := append(hdr[:len(hdr):len(hdr)], p.Pkt.Data.ToView()...) h := flow.header4Tuple(outgoing) checkers := append( @@ -397,7 +397,8 @@ func (c *testContext) injectPacket(flow testFlow, payload []byte) { 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) @@ -431,7 +432,11 @@ func (c *testContext) injectV6Packet(payload []byte, h *header4Tuple, valid bool 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), + }) } // injectV4Packet creates a V4 test packet with the given payload and header @@ -441,7 +446,8 @@ func (c *testContext) injectV6Packet(payload []byte, h *header4Tuple, valid bool 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) @@ -471,7 +477,12 @@ func (c *testContext) injectV4Packet(payload []byte, h *header4Tuple, valid bool 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 { @@ -1442,8 +1453,8 @@ func TestV4UnknownDestination(t *testing.T) { select { case p := <-c.linkEP.C: var pkt []byte - pkt = append(pkt, p.Header...) - pkt = append(pkt, p.Payload...) + pkt = append(pkt, p.Pkt.Header.View()...) + pkt = append(pkt, p.Pkt.Data.ToView()...) if got, want := len(pkt), header.IPv4MinimumProcessableDatagramSize; got > want { t.Fatalf("got an ICMP packet of size: %d, want: sz <= %d", got, want) } @@ -1516,8 +1527,8 @@ func TestV6UnknownDestination(t *testing.T) { select { case p := <-c.linkEP.C: var pkt []byte - pkt = append(pkt, p.Header...) - pkt = append(pkt, p.Payload...) + pkt = append(pkt, p.Pkt.Header.View()...) + pkt = append(pkt, p.Pkt.Data.ToView()...) if got, want := len(pkt), header.IPv6MinimumMTU; got > want { t.Fatalf("got an ICMP packet of size: %d, want: sz <= %d", got, want) } |