diff options
Diffstat (limited to 'pkg/tcpip')
31 files changed, 2234 insertions, 224 deletions
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/header/icmpv6.go b/pkg/tcpip/header/icmpv6.go index c2bfd8c79..b4037b6c8 100644 --- a/pkg/tcpip/header/icmpv6.go +++ b/pkg/tcpip/header/icmpv6.go @@ -132,7 +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) } @@ -197,7 +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/ndp_options.go b/pkg/tcpip/header/ndp_options.go index b28bde15b..a2b9d7435 100644 --- a/pkg/tcpip/header/ndp_options.go +++ b/pkg/tcpip/header/ndp_options.go @@ -15,6 +15,10 @@ package header import ( + "encoding/binary" + "errors" + "time" + "gvisor.dev/gvisor/pkg/tcpip" ) @@ -27,15 +31,187 @@ 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 + + // NDPPrefixInformationInfiniteLifetime is a value that represents + // infinity for the Valid and Preferred Lifetime fields in a NDP Prefix + // Information option. Its value is (2^32 - 1)s = 4294967295s + NDPPrefixInformationInfiniteLifetime = time.Second * 4294967295 + // lengthByteUnits is the multiplier factor for the Length field of an // NDP option. That is, the length field for NDP options is in units of // 8 octets, as per RFC 4861 section 4.6. lengthByteUnits = 8 ) +// NDPOptionIterator is an iterator of NDPOption. +// +// Note, between when an NDPOptionIterator is obtained and last used, no changes +// to the NDPOptions may happen. Doing so may cause undefined and unexpected +// behaviour. It is fine to obtain an NDPOptionIterator, iterate over the first +// few NDPOption then modify the backing NDPOptions so long as the +// NDPOptionIterator obtained before modification is no longer used. +type NDPOptionIterator struct { + // The NDPOptions this NDPOptionIterator is iterating over. + opts NDPOptions +} + +// Potential errors when iterating over an NDPOptions. +var ( + ErrNDPOptBufExhausted = errors.New("Buffer unexpectedly exhausted") + ErrNDPOptZeroLength = errors.New("NDP option has zero-valued Length field") + ErrNDPOptMalformedBody = errors.New("NDP option has a malformed body") +) + +// Next returns the next element in the backing NDPOptions, or true if we are +// done, or false if an error occured. +// +// The return can be read as option, done, error. Note, option should only be +// used if done is false and error is nil. +func (i *NDPOptionIterator) Next() (NDPOption, bool, error) { + for { + // Do we still have elements to look at? + if len(i.opts) == 0 { + return nil, true, nil + } + + // Do we have enough bytes for an NDP option that has a Length + // field of at least 1? Note, 0 in the Length field is invalid. + if len(i.opts) < lengthByteUnits { + return nil, true, ErrNDPOptBufExhausted + } + + // Get the Type field. + t := i.opts[0] + + // Get the Length field. + l := i.opts[1] + + // This would indicate an erroneous NDP option as the Length + // field should never be 0. + if l == 0 { + return nil, true, ErrNDPOptZeroLength + } + + // How many bytes are in the option body? + numBytes := int(l) * lengthByteUnits + numBodyBytes := numBytes - 2 + + potentialBody := i.opts[2:] + + // This would indicate an erroenous NDPOptions buffer as we ran + // out of the buffer in the middle of an NDP option. + if left := len(potentialBody); left < numBodyBytes { + return nil, true, ErrNDPOptBufExhausted + } + + // Get only the options body, leaving the rest of the options + // buffer alone. + body := potentialBody[:numBodyBytes] + + // Update opts with the remaining options body. + i.opts = i.opts[numBytes:] + + switch t { + case NDPTargetLinkLayerAddressOptionType: + return NDPTargetLinkLayerAddressOption(body), false, nil + + case NDPPrefixInformationType: + // Make sure the length of a Prefix Information option + // body is ndpPrefixInformationLength, as per RFC 4861 + // section 4.6.2. + if numBodyBytes != ndpPrefixInformationLength { + return nil, true, ErrNDPOptMalformedBody + } + + return NDPPrefixInformation(body), false, nil + default: + // We do not yet recognize the option, just skip for + // now. This is okay because RFC 4861 allows us to + // skip/ignore any unrecognized options. However, + // we MUST recognized all the options in RFC 4861. + // + // TODO(b/141487990): Handle all NDP options as defined + // by RFC 4861. + } + } +} + // NDPOptions is a buffer of NDP options as defined by RFC 4861 section 4.6. type NDPOptions []byte +// Iter returns an iterator of NDPOption. +// +// If check is true, Iter will do an integrity check on the options by iterating +// over it and returning an error if detected. +// +// See NDPOptionIterator for more information. +func (b NDPOptions) Iter(check bool) (NDPOptionIterator, error) { + it := NDPOptionIterator{opts: b} + + if check { + for it2 := it; true; { + if _, done, err := it2.Next(); err != nil || done { + return it, err + } + } + } + + return it, nil +} + // Serialize serializes the provided list of NDP options into o. // // Note, b must be of sufficient size to hold all the options in s. See @@ -75,15 +251,15 @@ 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 +268,15 @@ 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 +315,7 @@ 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 +330,134 @@ 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 +// NDPPrefixInformationInfiniteLifetime. +func (o NDPPrefixInformation) ValidLifetime() time.Duration { + // The field is the time in seconds, as per RFC 4861 section 4.6.2. + return time.Second * time.Duration(binary.BigEndian.Uint32(o[ndpPrefixInformationValidLifetimeOffset:])) +} + +// PreferredLifetime returns the length of time that an address generated from +// the prefix via Stateless Address Auto-Configuration remains preferred. This +// value is relative to the send time of the packet that the Prefix Information +// option was present in. +// +// Note, a value of 0 implies that addresses generated from the prefix should +// no longer remain preferred, and a value of infinity is represented by +// NDPPrefixInformationInfiniteLifetime. +// +// Also note that the value of this field MUST NOT exceed the Valid Lifetime +// field to avoid preferring addresses that are no longer valid, for the +// purpose of Stateless Address Auto-Configuration. +func (o NDPPrefixInformation) PreferredLifetime() time.Duration { + // The field is the time in seconds, as per RFC 4861 section 4.6.2. + return time.Second * time.Duration(binary.BigEndian.Uint32(o[ndpPrefixInformationPreferredLifetimeOffset:])) +} + +// Prefix returns an IPv6 address or a prefix of an IPv6 address. The Prefix +// Length field (see NDPPrefixInformation.PrefixLength) contains the number +// of valid leading bits in the prefix. +// +// Hosts SHOULD ignore an NDP Prefix Information option where the Prefix field +// holds the link-local prefix (fe80::). +func (o NDPPrefixInformation) Prefix() tcpip.Address { + return tcpip.Address(o[ndpPrefixInformationPrefixOffset:][:IPv6AddressSize]) +} diff --git a/pkg/tcpip/header/ndp_test.go b/pkg/tcpip/header/ndp_test.go index 0aac14f43..ad6daafcd 100644 --- a/pkg/tcpip/header/ndp_test.go +++ b/pkg/tcpip/header/ndp_test.go @@ -36,18 +36,18 @@ 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) } } @@ -65,56 +65,56 @@ 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") } } @@ -128,30 +128,66 @@ func TestNDPRouterAdvert(t *testing.T) { ra := NDPRouterAdvert(b) if got := ra.CurrHopLimit(); got != 64 { - t.Fatalf("got ra.CurrHopLimit = %d, want = 64", got) + t.Errorf("got ra.CurrHopLimit = %d, want = 64", got) } if got := ra.ManagedAddrConfFlag(); !got { - t.Fatalf("got ManagedAddrConfFlag = false, want = true") + t.Errorf("got ManagedAddrConfFlag = false, want = true") } if got := ra.OtherConfFlag(); got { - t.Fatalf("got OtherConfFlag = true, want = false") + t.Errorf("got OtherConfFlag = true, want = false") } if got, want := ra.RouterLifetime(), time.Second*258; got != want { - t.Fatalf("got ra.RouterLifetime = %d, want = %d", got, want) + t.Errorf("got ra.RouterLifetime = %d, want = %d", got, want) } if got, want := ra.ReachableTime(), time.Millisecond*50595078; got != want { - t.Fatalf("got ra.ReachableTime = %d, want = %d", got, want) + t.Errorf("got ra.ReachableTime = %d, want = %d", got, want) } if got, want := ra.RetransTimer(), time.Millisecond*117967114; got != want { - t.Fatalf("got ra.RetransTimer = %d, want = %d", 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) { @@ -194,6 +230,341 @@ 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) + } +} + +// TestNDPOptionsIterCheck tests that Iter will return false if the NDPOptions +// the iterator was returned for is malformed. +func TestNDPOptionsIterCheck(t *testing.T) { + tests := []struct { + name string + buf []byte + expected error + }{ + { + "ZeroLengthField", + []byte{0, 0, 0, 0, 0, 0, 0, 0}, + ErrNDPOptZeroLength, + }, + { + "ValidTargetLinkLayerAddressOption", + []byte{2, 1, 1, 2, 3, 4, 5, 6}, + nil, + }, + { + "TooSmallTargetLinkLayerAddressOption", + []byte{2, 1, 1, 2, 3, 4, 5}, + ErrNDPOptBufExhausted, + }, + { + "ValidPrefixInformation", + []byte{ + 3, 4, 43, 64, + 1, 2, 3, 4, + 5, 6, 7, 8, + 0, 0, 0, 0, + 9, 10, 11, 12, + 13, 14, 15, 16, + 17, 18, 19, 20, + 21, 22, 23, 24, + }, + nil, + }, + { + "TooSmallPrefixInformation", + []byte{ + 3, 4, 43, 64, + 1, 2, 3, 4, + 5, 6, 7, 8, + 0, 0, 0, 0, + 9, 10, 11, 12, + 13, 14, 15, 16, + 17, 18, 19, 20, + 21, 22, 23, + }, + ErrNDPOptBufExhausted, + }, + { + "InvalidPrefixInformationLength", + []byte{ + 3, 3, 43, 64, + 1, 2, 3, 4, + 5, 6, 7, 8, + 0, 0, 0, 0, + 9, 10, 11, 12, + 13, 14, 15, 16, + }, + ErrNDPOptMalformedBody, + }, + { + "ValidTargetLinkLayerAddressWithPrefixInformation", + []byte{ + // Target Link-Layer Address. + 2, 1, 1, 2, 3, 4, 5, 6, + + // Prefix information. + 3, 4, 43, 64, + 1, 2, 3, 4, + 5, 6, 7, 8, + 0, 0, 0, 0, + 9, 10, 11, 12, + 13, 14, 15, 16, + 17, 18, 19, 20, + 21, 22, 23, 24, + }, + nil, + }, + { + "ValidTargetLinkLayerAddressWithPrefixInformationWithUnrecognized", + []byte{ + // Target Link-Layer Address. + 2, 1, 1, 2, 3, 4, 5, 6, + + // 255 is an unrecognized type. If 255 ends up + // being the type for some recognized type, + // update 255 to some other unrecognized value. + 255, 2, 1, 2, 3, 4, 5, 6, 1, 2, 3, 4, 5, 6, 7, 8, + + // Prefix information. + 3, 4, 43, 64, + 1, 2, 3, 4, + 5, 6, 7, 8, + 0, 0, 0, 0, + 9, 10, 11, 12, + 13, 14, 15, 16, + 17, 18, 19, 20, + 21, 22, 23, 24, + }, + nil, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + opts := NDPOptions(test.buf) + + if _, err := opts.Iter(true); err != test.expected { + t.Fatalf("got Iter(true) = (_, %v), want = (_, %v)", err, test.expected) + } + + // test.buf may be malformed but we chose not to check + // the iterator so it must return true. + if _, err := opts.Iter(false); err != nil { + t.Fatalf("got Iter(false) = (_, %s), want = (_, nil)", err) + } }) } } + +// TestNDPOptionsIter tests that we can iterator over a valid NDPOptions. Note, +// this test does not actually check any of the option's getters, it simply +// checks the option Type and Body. We have other tests that tests the option +// field gettings given an option body and don't need to duplicate those tests +// here. +func TestNDPOptionsIter(t *testing.T) { + buf := []byte{ + // Target Link-Layer Address. + 2, 1, 1, 2, 3, 4, 5, 6, + + // 255 is an unrecognized type. If 255 ends up being the type + // for some recognized type, update 255 to some other + // unrecognized value. Note, this option should be skipped when + // iterating. + 255, 2, 1, 2, 3, 4, 5, 6, 1, 2, 3, 4, 5, 6, 7, 8, + + // Prefix information. + 3, 4, 43, 64, + 1, 2, 3, 4, + 5, 6, 7, 8, + 0, 0, 0, 0, + 9, 10, 11, 12, + 13, 14, 15, 16, + 17, 18, 19, 20, + 21, 22, 23, 24, + } + + opts := NDPOptions(buf) + it, err := opts.Iter(true) + if err != nil { + t.Fatalf("got Iter = (_, %s), want = (_, nil)", err) + } + + // Test the first (Taret Link-Layer) option. + next, done, err := it.Next() + if err != nil { + t.Fatalf("got Next = (_, _, %s), want = (_, _, nil)", err) + } + if done { + t.Fatal("got Next = (_, true, _), want = (_, false, _)") + } + if got, want := []byte(next.(NDPTargetLinkLayerAddressOption)), buf[2:][:6]; !bytes.Equal(got, want) { + t.Errorf("got Next = (%x, _, _), want = (%x, _, _)", got, want) + } + if got := next.Type(); got != NDPTargetLinkLayerAddressOptionType { + t.Errorf("got Type = %d, want = %d", got, NDPTargetLinkLayerAddressOptionType) + } + + // Test the next (Prefix Information) option. + // Note, the unrecognized option should be skipped. + next, done, err = it.Next() + if err != nil { + t.Fatalf("got Next = (_, _, %s), want = (_, _, nil)", err) + } + if done { + t.Fatal("got Next = (_, true, _), want = (_, false, _)") + } + if got, want := next.(NDPPrefixInformation), buf[26:][:30]; !bytes.Equal(got, want) { + t.Errorf("got Next = (%x, _, _), want = (%x, _, _)", got, want) + } + if got := next.Type(); got != NDPPrefixInformationType { + t.Errorf("got Type = %d, want = %d", got, NDPPrefixInformationType) + } + + // Iterator should not return anything else. + next, done, err = it.Next() + if err != nil { + t.Errorf("got Next = (_, _, %s), want = (_, _, nil)", err) + } + if !done { + t.Error("got Next = (_, false, _), want = (_, true, _)") + } + if next != nil { + t.Errorf("got Next = (%x, _, _), want = (nil, _, _)", next) + } +} diff --git a/pkg/tcpip/link/fdbased/endpoint_test.go b/pkg/tcpip/link/fdbased/endpoint_test.go index 59378b96c..e7c05ca4f 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 buffer.VectorisedView linkHeader buffer.View } @@ -94,7 +94,7 @@ func (c *context) cleanup() { } func (c *context) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote tcpip.LinkAddress, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView, linkHeader buffer.View) { - c.ch <- packetInfo{remote, protocol, vv.ToView(), linkHeader} + c.ch <- packetInfo{remote, protocol, vv, linkHeader} } func TestNoEthernetProperties(t *testing.T) { @@ -319,13 +319,17 @@ func TestDeliverPacket(t *testing.T) { want := packetInfo{ raddr: raddr, proto: proto, - contents: b, + contents: buffer.View(b).ToVectorisedView(), linkHeader: buffer.View(hdr), } if !eth { want.proto = header.IPv4ProtocolNumber want.raddr = "" } + // want.contents will be a single view, + // so make pi do the same for the + // DeepEqual check. + pi.contents = pi.contents.ToView().ToVectorisedView() if !reflect.DeepEqual(want, pi) { t.Fatalf("Unexpected received packet: %+v, want %+v", pi, want) } diff --git a/pkg/tcpip/link/fdbased/packet_dispatchers.go b/pkg/tcpip/link/fdbased/packet_dispatchers.go index 12168a1dc..3331b6453 100644 --- a/pkg/tcpip/link/fdbased/packet_dispatchers.go +++ b/pkg/tcpip/link/fdbased/packet_dispatchers.go @@ -139,7 +139,7 @@ func (d *readVDispatcher) dispatch() (bool, *tcpip.Error) { } used := d.capViews(n, BufConfig) - vv := buffer.NewVectorisedView(n, d.views[:used]) + vv := buffer.NewVectorisedView(n, append([]buffer.View(nil), d.views[:used]...)) vv.TrimFront(d.e.hdrSize) d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, vv, buffer.View(eth)) @@ -293,7 +293,7 @@ func (d *recvMMsgDispatcher) dispatch() (bool, *tcpip.Error) { } used := d.capViews(k, int(n), BufConfig) - vv := buffer.NewVectorisedView(int(n), d.views[k][:used]) + vv := buffer.NewVectorisedView(int(n), append([]buffer.View(nil), d.views[k][:used]...)) vv.TrimFront(d.e.hdrSize) d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, vv, buffer.View(eth)) 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/network/ip_test.go b/pkg/tcpip/network/ip_test.go index 8d74497ba..666d8b92a 100644 --- a/pkg/tcpip/network/ip_test.go +++ b/pkg/tcpip/network/ip_test.go @@ -519,6 +519,7 @@ 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 @@ -570,7 +571,7 @@ 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, }) @@ -618,6 +619,10 @@ func TestIPv6ReceiveControl(t *testing.T) { o.extra = c.expectedExtra vv := view[:len(view)-c.trunc].ToVectorisedView() + + // Set ICMPv6 checksum. + icmp.SetChecksum(header.ICMPv6Checksum(icmp, outerSrcAddr, localIpv6Addr, buffer.VectorisedView{})) + ep.HandlePacket(&r, 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) diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index 6c14b4aae..c3f1dd488 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -72,6 +72,18 @@ 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 := vv + 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. @@ -246,15 +258,15 @@ 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) if targetAddr != r.RemoteAddress { diff --git a/pkg/tcpip/network/ipv6/icmp_test.go b/pkg/tcpip/network/ipv6/icmp_test.go index 7c11dde55..b112303b6 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") ) @@ -359,3 +359,533 @@ 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.ICMPv6MinimumSize, + 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.Inject(ProtocolNumber, 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.Inject(ProtocolNumber, 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.Inject(ProtocolNumber, + 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/stack/ndp.go b/pkg/tcpip/stack/ndp.go index 921d1c9c7..03ddebdbd 100644 --- a/pkg/tcpip/stack/ndp.go +++ b/pkg/tcpip/stack/ndp.go @@ -51,6 +51,22 @@ const ( minimumRetransmitTimer = time.Millisecond ) +// 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) +} + // NDPConfigurations is the NDP configurations for the netstack. type NDPConfigurations struct { // The number of Neighbor Solicitation messages to send when doing @@ -88,6 +104,12 @@ 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 } @@ -110,8 +132,8 @@ type dadState struct { // 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 +149,13 @@ 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 +168,59 @@ 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 + + // doDadIteration does a single iteration of the DAD loop. + // + // Returns true if the integrator needs to be informed of DAD + // completing. + doDadIteration := func() bool { + ndp.nic.mu.Lock() + defer ndp.nic.mu.Unlock() + + if done { + // If we reach this point, it means that the DAD + // timer fired after another goroutine already + // obtained the NIC lock and stopped DAD before + // this function obtained the NIC lock. Simply + // return here and do nothing further. + return false + } - 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 - } + 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())) + } - 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())) - } + d, err = ndp.doDuplicateAddressDetection(addr, remaining, ref) + if err != nil || d { + delete(ndp.dad, addr) - 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) + 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 } - ndp.stopDuplicateAddressDetection(addr) - return + remaining-- + timer.Reset(ndp.nic.stack.ndpConfigs.RetransmitTimer) + return false } - timer.Reset(n.stack.ndpConfigs.RetransmitTimer) - remaining-- - + if doDadIteration() && ndp.nic.stack.ndpDisp != nil { + ndp.nic.stack.ndpDisp.OnDuplicateAddressDetectionStatus(ndp.nic.ID(), addr, d, err) + } }) ndp.dad[addr] = dadState{ @@ -204,11 +243,11 @@ 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 +258,17 @@ 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)) @@ -275,5 +314,8 @@ 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) + } } diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go index 8995fbfc3..525a25218 100644 --- a/pkg/tcpip/stack/ndp_test.go +++ b/pkg/tcpip/stack/ndp_test.go @@ -31,6 +31,7 @@ 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" + addr3 = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x03" linkAddr1 = "\x02\x02\x03\x04\x05\x06" ) @@ -67,6 +68,35 @@ 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 +} + +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 +} + +// Implements stack.NDPDispatcher.OnDuplicateAddressDetectionStatus. +// +// If the DAD event matches what we are expecting, send signal on n.dadC. +func (n *ndpDispatcher) OnDuplicateAddressDetectionStatus(nicid tcpip.NICID, addr tcpip.Address, resolved bool, err *tcpip.Error) { + n.dadC <- ndpDADEvent{ + nicid, + addr, + resolved, + err, + } +} + // 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 @@ -88,8 +118,12 @@ func TestDADResolve(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { + ndpDisp := ndpDispatcher{ + dadC: make(chan ndpDADEvent), + } opts := stack.Options{ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPDisp: &ndpDisp, } opts.NDPConfigs.RetransmitTimer = test.retransTimer opts.NDPConfigs.DupAddrDetectTransmits = test.dupAddrDetectTransmits @@ -106,8 +140,7 @@ 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) @@ -123,16 +156,10 @@ 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) @@ -141,13 +168,30 @@ 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) @@ -250,9 +294,14 @@ 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 @@ -286,8 +335,28 @@ func TestDADFail(t *testing.T) { 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) @@ -302,11 +371,18 @@ 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) s := stack.New(opts) @@ -332,11 +408,27 @@ 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 +442,168 @@ 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(10, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPDisp: &ndpDisp, + }) + + // This NIC(1)'s NDP configurations will be updated to + // be different from the default. + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + // Created before updating NIC(1)'s NDP configurations + // but updating NIC(1)'s NDP configurations should not + // affect other existing NICs. + if err := s.CreateNIC(2, e); err != nil { + t.Fatalf("CreateNIC(2) = %s", err) + } + + // Update the NDP configurations on NIC(1) to use DAD. + configs := stack.NDPConfigurations{ + DupAddrDetectTransmits: test.dupAddrDetectTransmits, + RetransmitTimer: test.retransmitTimer, + } + if err := s.SetNDPConfigurations(1, configs); err != nil { + t.Fatalf("got SetNDPConfigurations(1, _) = %s", err) + } + + // Created after updating NIC(1)'s NDP configurations + // but the stack's default NDP configurations should not + // have been updated. + if err := s.CreateNIC(3, e); err != nil { + t.Fatalf("CreateNIC(3) = %s", err) + } + + // Add addresses for each NIC. + if err := s.AddAddress(1, header.IPv6ProtocolNumber, addr1); err != nil { + t.Fatalf("AddAddress(1, %d, %s) = %s", header.IPv6ProtocolNumber, addr1, err) + } + if err := s.AddAddress(2, header.IPv6ProtocolNumber, addr2); err != nil { + t.Fatalf("AddAddress(2, %d, %s) = %s", header.IPv6ProtocolNumber, addr2, err) + } + if err := s.AddAddress(3, header.IPv6ProtocolNumber, addr3); err != nil { + t.Fatalf("AddAddress(3, %d, %s) = %s", header.IPv6ProtocolNumber, addr3, err) + } + + // Address should not be considered bound to NIC(1) yet + // (DAD ongoing). + addr, err := s.GetMainNICAddress(1, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (_, %v), want = (_, nil)", err) + } + if want := (tcpip.AddressWithPrefix{}); addr != want { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want) + } + + // Should get the address on NIC(2) and NIC(3) + // immediately since we should not have performed DAD on + // it as the stack was configured to not do DAD by + // default and we only updated the NDP configurations on + // NIC(1). + addr, err = s.GetMainNICAddress(2, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("stack.GetMainNICAddress(2, _) err = %s", err) + } + if addr.Address != addr2 { + t.Fatalf("got stack.GetMainNICAddress(2, _) = %s, want = %s", addr, addr2) + } + addr, err = s.GetMainNICAddress(3, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("stack.GetMainNICAddress(3, _) err = %s", err) + } + if addr.Address != addr3 { + t.Fatalf("got stack.GetMainNICAddress(3, _) = %s, want = %s", addr, addr3) + } + + // Sleep until right (500ms before) before resolution to + // make sure the address didn't resolve on NIC(1) yet. + const delta = 500 * time.Millisecond + time.Sleep(time.Duration(test.dupAddrDetectTransmits)*test.expectedRetransmitTimer - delta) + addr, err = s.GetMainNICAddress(1, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (_, %v), want = (_, nil)", err) + } + if want := (tcpip.AddressWithPrefix{}); addr != want { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want) + } + + // Wait for DAD to resolve. + select { + case <-time.After(2 * delta): + // We should get a resolution event after 500ms + // (delta) since we wait for 500ms less than the + // expected resolution time above to make sure + // that the address did not yet resolve. Waiting + // for 1s (2x delta) without a resolution event + // means something is wrong. + t.Fatal("timed out waiting for DAD resolution") + case e := <-ndpDisp.dadC: + if e.err != nil { + t.Fatal("got DAD error: ", e.err) + } + if e.nicid != 1 { + t.Fatalf("got DAD event w/ nicid = %d, want = 1", e.nicid) + } + if e.addr != addr1 { + t.Fatalf("got DAD event w/ addr = %s, want = %s", addr, addr1) + } + if !e.resolved { + t.Fatal("got DAD event w/ resolved = false, want = true") + } + } + addr, err = s.GetMainNICAddress(1, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("stack.GetMainNICAddress(1, _) err = %s", err) + } + if addr.Address != addr1 { + t.Fatalf("got stack.GetMainNICAddress(1, _) = %s, want = %s", addr, addr1) + } + }) + } +} diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index e456e05f4..fe8f83d58 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -46,6 +46,10 @@ type NIC struct { 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 } @@ -80,10 +84,16 @@ 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 { // 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, @@ -105,9 +115,11 @@ 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), }, } + nic.ndp.nic = nic // Register supported packet endpoint protocols. for _, netProto := range header.Ethertypes { @@ -432,7 +444,7 @@ func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb Primar // 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 } } @@ -750,7 +762,7 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link } n.mu.RUnlock() for _, ep := range packetEPs { - ep.HandlePacket(n.id, local, protocol, vv, linkHeader) + ep.HandlePacket(n.id, local, protocol, vv.Clone(nil), linkHeader) } if netProto.Number() == header.IPv4ProtocolNumber || netProto.Number() == header.IPv6ProtocolNumber { @@ -936,6 +948,18 @@ 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() +} + type networkEndpointKind int32 const ( diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go index 0869fb084..d7c124e81 100644 --- a/pkg/tcpip/stack/registration.go +++ b/pkg/tcpip/stack/registration.go @@ -60,13 +60,34 @@ 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 takes ownership of vv. HandlePacket(r *Route, id TransportEndpointID, vv buffer.VectorisedView) // HandleControlPacket is called by the stack when new control (e.g., // ICMP) packets arrive to this transport endpoint. + // + // HandleControlPacket takes ownership of vv. HandleControlPacket(id TransportEndpointID, typ ControlType, extra uint32, vv buffer.VectorisedView) + + // 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 @@ -77,6 +98,8 @@ 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 takes ownership of packet and netHeader. HandlePacket(r *Route, netHeader buffer.View, packet buffer.VectorisedView) } @@ -93,6 +116,8 @@ type PacketEndpoint interface { // // linkHeader may have a length of 0, in which case the PacketEndpoint // should construct its own ethernet header for applications. + // + // HandlePacket takes ownership of packet and linkHeader. HandlePacket(nicid tcpip.NICID, addr tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, packet buffer.VectorisedView, linkHeader buffer.View) } @@ -143,10 +168,14 @@ 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 takes ownership of vv and netHeader. DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, netHeader buffer.View, vv buffer.VectorisedView) // DeliverTransportControlPacket delivers control packets to the // appropriate transport protocol endpoint. + // + // DeliverTransportControlPacket takes ownership of vv. DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, vv buffer.VectorisedView) } @@ -220,6 +249,8 @@ type NetworkEndpoint interface { // HandlePacket is called by the link layer when new packets arrive to // this network endpoint. + // + // HandlePacket takes ownership of vv. HandlePacket(r *Route, vv buffer.VectorisedView) // Close is called when the endpoint is reomved from a stack. @@ -265,6 +296,8 @@ type NetworkDispatcher interface { // DeliverNetworkPacket finds the appropriate network protocol endpoint // and hands the packet over for further processing. linkHeader may have // length 0 when the caller does not have ethernet data. + // + // DeliverNetworkPacket takes ownership of vv and linkHeader. DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView, linkHeader buffer.View) } diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 284280917..115a6fcb8 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" @@ -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 { @@ -361,9 +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 @@ -399,13 +408,25 @@ type Stack struct { // TODO(gvisor.dev/issue/940): S/R this field. portSeed 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. @@ -429,7 +450,10 @@ type Options struct { // stack (false). HandleLocal bool - // NDPConfigs is the NDP configurations used by interfaces. + // UniqueID is an optional generator of unique identifiers. + UniqueID UniqueID + + // NDPConfigs is the default NDP configurations used by interfaces. // // By default, NDPConfigs will have a zero value for its // DupAddrDetectTransmits field, implying that DAD will not be performed @@ -448,6 +472,10 @@ type Options struct { // 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 @@ -497,6 +525,10 @@ 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() @@ -505,6 +537,7 @@ func New(opts Options) *Stack { 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, @@ -514,6 +547,8 @@ func New(opts Options) *Stack { portSeed: generateRandUint32(), ndpConfigs: opts.NDPConfigs, autoGenIPv6LinkLocal: opts.AutoGenIPv6LinkLocal, + uniqueIDGenerator: opts.UniqueID, + ndpDisp: opts.NDPDisp, } // Add specified network protocols. @@ -540,6 +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 @@ -1127,6 +1167,25 @@ 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() +} + // 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. @@ -1148,6 +1207,69 @@ 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() { @@ -1416,6 +1538,25 @@ func (s *Stack) DupTentativeAddrDetected(id tcpip.NICID, addr tcpip.Address) *tc return nic.dupTentativeAddrDetected(addr) } +// SetNDPConfigurations sets the per-interface NDP configurations on the NIC +// with ID id to c. +// +// Note, if c contains invalid NDP configuration values, it will be fixed to +// use default values for the erroneous values. +func (s *Stack) SetNDPConfigurations(id tcpip.NICID, c NDPConfigurations) *tcpip.Error { + s.mu.Lock() + defer s.mu.Unlock() + + nic, ok := s.nics[id] + if !ok { + return tcpip.ErrUnknownNICID + } + + nic.setNDPConfigs(c) + + return nil +} + // PortSeed returns a 32 bit value that can be used as a seed value for port // picking. // diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index 9a8906a0d..9dae853d0 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -1971,13 +1971,15 @@ func TestNICAutoGenAddr(t *testing.T) { // 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: stack.NDPConfigurations{ - RetransmitTimer: time.Second, - DupAddrDetectTransmits: 1, - }, + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: ndpConfigs, AutoGenIPv6LinkLocal: true, + NDPDisp: &ndpDisp, } e := channel.New(10, 1280, linkAddr1) @@ -1996,21 +1998,35 @@ func TestNICAutoGenAddrDoesDAD(t *testing.T) { t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want) } - // Wait for the address to resolve (an extra - // 250ms to make sure the address resolves). - // - // TODO(b/140896005): Use events from the - // netstack to know immediately when DAD - // completes. - time.Sleep(time.Second + 250*time.Millisecond) + linkLocalAddr := header.LinkLocalAddr(linkAddr1) - // Should have auto-generated an address and - // resolved (if DAD). + // 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: header.LinkLocalAddr(linkAddr1), PrefixLen: header.IPv6LinkLocalPrefix.PrefixLen}); addr != want { + if want := (tcpip.AddressWithPrefix{Address: linkLocalAddr, PrefixLen: header.IPv6LinkLocalPrefix.PrefixLen}); addr != want { t.Fatalf("got stack.GetMainNICAddress(_, _) = %s, want = %s", addr, want) } } diff --git a/pkg/tcpip/stack/transport_demuxer.go b/pkg/tcpip/stack/transport_demuxer.go index 97a1aec4b..ccd3d030e 100644 --- a/pkg/tcpip/stack/transport_demuxer.go +++ b/pkg/tcpip/stack/transport_demuxer.go @@ -17,6 +17,7 @@ package stack import ( "fmt" "math/rand" + "sort" "sync" "gvisor.dev/gvisor/pkg/tcpip" @@ -41,6 +42,31 @@ 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 @@ -48,6 +74,16 @@ 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) { @@ -127,21 +163,6 @@ 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 +204,27 @@ 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. @@ -240,6 +274,26 @@ func (ep *multiPortEndpoint) handlePacketAll(r *Route, id TransportEndpointID, v 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 +311,15 @@ 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 } diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index 86c62be25..203e79f56 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() { @@ -144,6 +145,10 @@ func (f *fakeTransportEndpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { return nil } +func (f *fakeTransportEndpoint) UniqueID() uint64 { + return f.uniqueID +} + func (f *fakeTransportEndpoint) ConnectEndpoint(e tcpip.Endpoint) *tcpip.Error { return nil } @@ -218,15 +223,15 @@ func (f *fakeTransportEndpoint) State() uint32 { return 0 } -func (f *fakeTransportEndpoint) ModerateRecvBuf(copied int) { -} +func (f *fakeTransportEndpoint) ModerateRecvBuf(copied int) {} func (f *fakeTransportEndpoint) IPTables() (iptables.IPTables, error) { return iptables.IPTables{}, nil } -func (f *fakeTransportEndpoint) Resume(*stack.Stack) { -} +func (f *fakeTransportEndpoint) Resume(*stack.Stack) {} + +func (f *fakeTransportEndpoint) Wait() {} type fakeTransportGoodOption bool @@ -251,7 +256,7 @@ func (*fakeTransportProtocol) Number() tcpip.TransportProtocolNumber { } func (f *fakeTransportProtocol) NewEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, _ *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { - return newFakeTransportEndpoint(stack, f, netProto), nil + return newFakeTransportEndpoint(stack, f, netProto, stack.UniqueID()), nil } func (f *fakeTransportProtocol) NewRawEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, _ *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 3d10001d8..03be7d3d4 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -489,6 +489,11 @@ 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 +506,6 @@ 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 @@ -673,6 +673,11 @@ 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 +886,15 @@ type TCPStats struct { // successfully via Listen. PassiveConnectionOpenings *StatCounter + // CurrentEstablished is the number of TCP connections for which the + // current state is either ESTABLISHED or CLOSE-WAIT. + CurrentEstablished *StatCounter + + // EstablishedResets is the number of times TCP connections have made + // a direct transition to the CLOSED state from either the + // ESTABLISHED state or the CLOSE-WAIT state. + EstablishedResets *StatCounter + // ListenOverflowSynDrop is the number of times the listen queue overflowed // and a SYN was dropped. ListenOverflowSynDrop *StatCounter 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/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index 3187b336b..33405eb7d 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() { @@ -445,13 +449,13 @@ 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 */, hdr, dataVV, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}) } func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress, allowMismatch bool) (tcpip.NetworkProtocolNumber, *tcpip.Error) { @@ -760,7 +764,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, vv }, } - pkt.data = vv.Clone(pkt.views[:]) + pkt.data = vv e.rcvList.PushBack(pkt) e.rcvBufSize += pkt.data.Size() @@ -798,3 +802,6 @@ 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/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go index 73cdaa265..ead83b83d 100644 --- a/pkg/tcpip/transport/packet/endpoint.go +++ b/pkg/tcpip/transport/packet/endpoint.go @@ -41,10 +41,6 @@ type packet struct { // 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. @@ -310,7 +306,7 @@ func (ep *endpoint) HandlePacket(nicid tcpip.NICID, localAddr tcpip.LinkAddress, if ep.cooked { // Cooked packets can simply be queued. - packet.data = vv.Clone(packet.views[:]) + packet.data = vv } else { // Raw packets need their ethernet headers prepended before // queueing. @@ -328,7 +324,7 @@ func (ep *endpoint) HandlePacket(nicid tcpip.NICID, localAddr tcpip.LinkAddress, } combinedVV := buffer.View(ethHeader).ToVectorisedView() combinedVV.Append(vv) - packet.data = combinedVV.Clone(packet.views[:]) + packet.data = combinedVV } packet.timestampNS = ep.stack.NowNanoseconds() diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index 308f10d24..23922a30e 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -42,10 +42,6 @@ type rawPacket struct { // 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. @@ -609,7 +605,7 @@ func (e *endpoint) HandlePacket(route *stack.Route, netHeader buffer.View, vv bu combinedVV := netHeader.ToVectorisedView() combinedVV.Append(vv) - pkt.data = combinedVV.Clone(pkt.views[:]) + pkt.data = combinedVV pkt.timestampNS = e.stack.NowNanoseconds() e.rcvList.PushBack(pkt) @@ -641,3 +637,6 @@ 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/tcp/accept.go b/pkg/tcpip/transport/tcp/accept.go index 844959fa0..1dd00d026 100644 --- a/pkg/tcpip/transport/tcp/accept.go +++ b/pkg/tcpip/transport/tcp/accept.go @@ -297,6 +297,7 @@ func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *head return nil, err } ep.mu.Lock() + ep.stack.Stats().TCP.CurrentEstablished.Increment() ep.state = StateEstablished ep.mu.Unlock() @@ -399,6 +400,9 @@ 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) { + // TODO(b/143300739): Use the userMSS of the listening socket + // for accepted sockets. + switch s.flags { case header.TCPFlagSyn: opts := parseSynSegmentOptions(s) @@ -433,13 +437,12 @@ 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() @@ -519,6 +522,7 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { n.tsOffset = 0 // Switch state to connected. + n.stack.Stats().TCP.CurrentEstablished.Increment() n.state = StateEstablished // Do the delivery in a separate goroutine so diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 082135374..ca982c451 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -78,9 +78,6 @@ 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 @@ -445,7 +442,7 @@ 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, @@ -805,6 +802,10 @@ 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 { @@ -975,6 +976,8 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { e.lastErrorMu.Unlock() e.mu.Lock() + e.stack.Stats().TCP.EstablishedResets.Increment() + e.stack.Stats().TCP.CurrentEstablished.Decrement() e.state = StateError e.HardError = err @@ -1005,7 +1008,10 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { // Tell waiters that the endpoint is connected and writable. e.mu.Lock() - e.state = StateEstablished + if e.state != StateEstablished { + e.stack.Stats().TCP.CurrentEstablished.Increment() + e.state = StateEstablished + } drained := e.drainDone != nil e.mu.Unlock() if drained { @@ -1166,6 +1172,8 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { // Mark endpoint as closed. e.mu.Lock() if e.state != StateError { + e.stack.Stats().TCP.EstablishedResets.Increment() + e.stack.Stats().TCP.CurrentEstablished.Decrement() e.state = StateClose } // Lock released below. diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 8b9cb4c33..a1efd8d55 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -287,6 +287,7 @@ 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. @@ -411,7 +412,7 @@ 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 @@ -504,6 +505,26 @@ type endpoint struct { stats Stats `state:"nosave"` } +// 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. func (e *endpoint) StopWork() { e.workMu.Lock() @@ -550,6 +571,7 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue interval: 75 * time.Second, count: 9, }, + uniqueID: s.UniqueID(), } var ss SendBufferSizeOption @@ -671,7 +693,7 @@ 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.bindToDevice) e.isRegistered = false } @@ -732,7 +754,7 @@ 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.bindToDevice) e.isRegistered = false } @@ -742,6 +764,7 @@ func (e *endpoint) cleanupLocked() { } e.route.Release() + e.stack.CompleteTransportEndpointCleanup(e) tcpip.DeleteDanglingEndpoint(e) } @@ -752,7 +775,9 @@ 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 +1158,6 @@ 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 +1169,16 @@ 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) @@ -1206,7 +1231,7 @@ 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 @@ -1345,6 +1370,7 @@ 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 +1383,16 @@ 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 +1413,6 @@ 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 { @@ -1729,6 +1756,7 @@ 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 { @@ -2379,6 +2407,23 @@ 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..19f003b6b 100644 --- a/pkg/tcpip/transport/tcp/endpoint_state.go +++ b/pkg/tcpip/transport/tcp/endpoint_state.go @@ -193,8 +193,10 @@ func (e *endpoint) Resume(s *stack.Stack) { 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)) } } @@ -265,6 +267,7 @@ func (e *endpoint) Resume(s *stack.Stack) { } fallthrough case StateError: + e.stack.CompleteTransportEndpointCleanup(e) tcpip.DeleteDanglingEndpoint(e) } } diff --git a/pkg/tcpip/transport/tcp/snd.go b/pkg/tcpip/transport/tcp/snd.go index 8332a0179..d3f7c9125 100644 --- a/pkg/tcpip/transport/tcp/snd.go +++ b/pkg/tcpip/transport/tcp/snd.go @@ -674,6 +674,7 @@ func (s *sender) maybeSendSegment(seg *segment, limit int, end seqnum.Value) (se default: s.ep.state = StateFinWait1 } + s.ep.stack.Stats().TCP.CurrentEstablished.Decrement() s.ep.mu.Unlock() } else { // We're sending a non-FIN segment. diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index 6d022a266..126f26ed3 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -474,6 +474,130 @@ 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 TestTOSV4(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() @@ -1623,7 +1747,7 @@ 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 +1795,7 @@ 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 +1828,7 @@ 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 +1865,7 @@ 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)) }}, } diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 91c8487f3..03bd5c8fd 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -31,9 +31,6 @@ type udpPacket struct { senderAddress tcpip.FullAddress data buffer.VectorisedView `state:".(buffer.VectorisedView)"` timestamp int64 - // views is used as buffer for data when its length is large - // enough to store a VectorisedView. - views [8]buffer.View `state:"nosave"` } // EndpointState represents the state of a UDP endpoint. @@ -80,6 +77,7 @@ 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. @@ -160,9 +158,15 @@ 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() { @@ -1195,7 +1199,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, vv Port: hdr.SourcePort(), }, } - pkt.data = vv.Clone(pkt.views[:]) + pkt.data = vv e.rcvList.PushBack(pkt) e.rcvBufSize += vv.Size() @@ -1234,6 +1238,9 @@ 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) } |