Merge remote-tracking branch 'upstream/master' into HEAD
Change-Id: I04ae80c5f0d2e616b2a5771d8e6f6dda30abc90d
diff --git a/tcpip/buffer/prependable.go b/tcpip/buffer/prependable.go
index 4287464..48a2a27 100644
--- a/tcpip/buffer/prependable.go
+++ b/tcpip/buffer/prependable.go
@@ -41,6 +41,11 @@
return Prependable{buf: v, usedIdx: 0}
}
+// NewEmptyPrependableFromView creates a new prependable buffer from a View.
+func NewEmptyPrependableFromView(v View) Prependable {
+ return Prependable{buf: v, usedIdx: len(v)}
+}
+
// View returns a View of the backing buffer that contains all prepended
// data so far.
func (p Prependable) View() View {
diff --git a/tcpip/header/checksum.go b/tcpip/header/checksum.go
index 9698763..b67ceca 100644
--- a/tcpip/header/checksum.go
+++ b/tcpip/header/checksum.go
@@ -23,11 +23,17 @@
"github.com/google/netstack/tcpip/buffer"
)
-func calculateChecksum(buf []byte, initial uint32) uint16 {
+func calculateChecksum(buf []byte, odd bool, initial uint32) (uint16, bool) {
v := initial
+ if odd {
+ v += uint32(buf[0])
+ buf = buf[1:]
+ }
+
l := len(buf)
- if l&1 != 0 {
+ odd = l&1 != 0
+ if odd {
l--
v += uint32(buf[l]) << 8
}
@@ -36,7 +42,7 @@
v += (uint32(buf[i]) << 8) + uint32(buf[i+1])
}
- return ChecksumCombine(uint16(v), uint16(v>>16))
+ return ChecksumCombine(uint16(v), uint16(v>>16)), odd
}
// Checksum calculates the checksum (as defined in RFC 1071) of the bytes in the
@@ -44,7 +50,8 @@
//
// The initial checksum must have been computed on an even number of bytes.
func Checksum(buf []byte, initial uint16) uint16 {
- return calculateChecksum(buf, uint32(initial))
+ s, _ := calculateChecksum(buf, false, uint32(initial))
+ return s
}
// ChecksumVV calculates the checksum (as defined in RFC 1071) of the bytes in
@@ -52,19 +59,40 @@
//
// The initial checksum must have been computed on an even number of bytes.
func ChecksumVV(vv buffer.VectorisedView, initial uint16) uint16 {
- var odd bool
+ return ChecksumVVWithOffset(vv, initial, 0, vv.Size())
+}
+
+// ChecksumVVWithOffset calculates the checksum (as defined in RFC 1071) of the
+// bytes in the given VectorizedView.
+//
+// The initial checksum must have been computed on an even number of bytes.
+func ChecksumVVWithOffset(vv buffer.VectorisedView, initial uint16, off int, size int) uint16 {
+ odd := false
sum := initial
for _, v := range vv.Views() {
if len(v) == 0 {
continue
}
- s := uint32(sum)
- if odd {
- s += uint32(v[0])
- v = v[1:]
+
+ if off >= len(v) {
+ off -= len(v)
+ continue
}
- odd = len(v)&1 != 0
- sum = calculateChecksum(v, s)
+ v = v[off:]
+
+ l := len(v)
+ if l > size {
+ l = size
+ }
+ v = v[:l]
+
+ sum, odd = calculateChecksum(v, odd, uint32(sum))
+
+ size -= len(v)
+ if size == 0 {
+ break
+ }
+ off = 0
}
return sum
}
diff --git a/tcpip/header/checksum_test.go b/tcpip/header/checksum_test.go
new file mode 100644
index 0000000..43f0447
--- /dev/null
+++ b/tcpip/header/checksum_test.go
@@ -0,0 +1,109 @@
+// Copyright 2019 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Package header provides the implementation of the encoding and decoding of
+// network protocol headers.
+package header_test
+
+import (
+ "testing"
+
+ "github.com/google/netstack/tcpip/buffer"
+ "github.com/google/netstack/tcpip/header"
+)
+
+func TestChecksumVVWithOffset(t *testing.T) {
+ testCases := []struct {
+ name string
+ vv buffer.VectorisedView
+ off, size int
+ initial uint16
+ want uint16
+ }{
+ {
+ name: "empty",
+ vv: buffer.NewVectorisedView(0, []buffer.View{
+ buffer.NewViewFromBytes([]byte{1, 9, 0, 5, 4}),
+ }),
+ off: 0,
+ size: 0,
+ want: 0,
+ },
+ {
+ name: "OneView",
+ vv: buffer.NewVectorisedView(0, []buffer.View{
+ buffer.NewViewFromBytes([]byte{1, 9, 0, 5, 4}),
+ }),
+ off: 0,
+ size: 5,
+ want: 1294,
+ },
+ {
+ name: "TwoViews",
+ vv: buffer.NewVectorisedView(0, []buffer.View{
+ buffer.NewViewFromBytes([]byte{1, 9, 0, 5, 4}),
+ buffer.NewViewFromBytes([]byte{4, 3, 7, 1, 2, 123}),
+ }),
+ off: 0,
+ size: 11,
+ want: 33819,
+ },
+ {
+ name: "TwoViewsWithOffset",
+ vv: buffer.NewVectorisedView(0, []buffer.View{
+ buffer.NewViewFromBytes([]byte{98, 1, 9, 0, 5, 4}),
+ buffer.NewViewFromBytes([]byte{4, 3, 7, 1, 2, 123}),
+ }),
+ off: 1,
+ size: 11,
+ want: 33819,
+ },
+ {
+ name: "ThreeViewsWithOffset",
+ vv: buffer.NewVectorisedView(0, []buffer.View{
+ buffer.NewViewFromBytes([]byte{98, 1, 9, 0, 5, 4}),
+ buffer.NewViewFromBytes([]byte{98, 1, 9, 0, 5, 4}),
+ buffer.NewViewFromBytes([]byte{4, 3, 7, 1, 2, 123}),
+ }),
+ off: 7,
+ size: 11,
+ want: 33819,
+ },
+ {
+ name: "ThreeViewsWithInitial",
+ vv: buffer.NewVectorisedView(0, []buffer.View{
+ buffer.NewViewFromBytes([]byte{77, 11, 33, 0, 55, 44}),
+ buffer.NewViewFromBytes([]byte{98, 1, 9, 0, 5, 4}),
+ buffer.NewViewFromBytes([]byte{4, 3, 7, 1, 2, 123, 99}),
+ }),
+ initial: 77,
+ off: 7,
+ size: 11,
+ want: 33896,
+ },
+ }
+ for _, tc := range testCases {
+ t.Run(tc.name, func(t *testing.T) {
+ if got, want := header.ChecksumVVWithOffset(tc.vv, tc.initial, tc.off, tc.size), tc.want; got != want {
+ t.Errorf("header.ChecksumVVWithOffset(%v) = %v, want: %v", tc, got, tc.want)
+ }
+ v := tc.vv.ToView()
+ v.TrimFront(tc.off)
+ v.CapLength(tc.size)
+ if got, want := header.Checksum(v, tc.initial), tc.want; got != want {
+ t.Errorf("header.Checksum(%v) = %v, want: %v", tc, got, tc.want)
+ }
+ })
+ }
+}
diff --git a/tcpip/header/eth.go b/tcpip/header/eth.go
index 3c182fa..8c31a0f 100644
--- a/tcpip/header/eth.go
+++ b/tcpip/header/eth.go
@@ -48,8 +48,48 @@
// EthernetAddressSize is the size, in bytes, of an ethernet address.
EthernetAddressSize = 6
+
+ // unspecifiedEthernetAddress is the unspecified ethernet address
+ // (all bits set to 0).
+ unspecifiedEthernetAddress = tcpip.LinkAddress("\x00\x00\x00\x00\x00\x00")
+
+ // unicastMulticastFlagMask is the mask of the least significant bit in
+ // the first octet (in network byte order) of an ethernet address that
+ // determines whether the ethernet address is a unicast or multicast. If
+ // the masked bit is a 1, then the address is a multicast, unicast
+ // otherwise.
+ //
+ // See the IEEE Std 802-2001 document for more details. Specifically,
+ // section 9.2.1 of http://ieee802.org/secmail/pdfocSP2xXA6d.pdf:
+ // "A 48-bit universal address consists of two parts. The first 24 bits
+ // correspond to the OUI as assigned by the IEEE, expect that the
+ // assignee may set the LSB of the first octet to 1 for group addresses
+ // or set it to 0 for individual addresses."
+ unicastMulticastFlagMask = 1
+
+ // unicastMulticastFlagByteIdx is the byte that holds the
+ // unicast/multicast flag. See unicastMulticastFlagMask.
+ unicastMulticastFlagByteIdx = 0
)
+const (
+ // EthernetProtocolAll is a catch-all for all protocols carried inside
+ // an ethernet frame. It is mainly used to create packet sockets that
+ // capture all traffic.
+ EthernetProtocolAll tcpip.NetworkProtocolNumber = 0x0003
+
+ // EthernetProtocolPUP is the PARC Universial Packet protocol ethertype.
+ EthernetProtocolPUP tcpip.NetworkProtocolNumber = 0x0200
+)
+
+// Ethertypes holds the protocol numbers describing the payload of an ethernet
+// frame. These types aren't necessarily supported by netstack, but can be used
+// to catch all traffic of a type via packet endpoints.
+var Ethertypes = []tcpip.NetworkProtocolNumber{
+ EthernetProtocolAll,
+ EthernetProtocolPUP,
+}
+
// SourceAddress returns the "MAC source" field of the ethernet frame header.
func (b Ethernet) SourceAddress() tcpip.LinkAddress {
return tcpip.LinkAddress(b[srcMAC:][:EthernetAddressSize])
@@ -72,3 +112,25 @@
copy(b[srcMAC:][:EthernetAddressSize], e.SrcAddr)
copy(b[dstMAC:][:EthernetAddressSize], e.DstAddr)
}
+
+// IsValidUnicastEthernetAddress returns true if addr is a valid unicast
+// ethernet address.
+func IsValidUnicastEthernetAddress(addr tcpip.LinkAddress) bool {
+ // Must be of the right length.
+ if len(addr) != EthernetAddressSize {
+ return false
+ }
+
+ // Must not be unspecified.
+ if addr == unspecifiedEthernetAddress {
+ return false
+ }
+
+ // Must not be a multicast.
+ if addr[unicastMulticastFlagByteIdx]&unicastMulticastFlagMask != 0 {
+ return false
+ }
+
+ // addr is a valid unicast ethernet address.
+ return true
+}
diff --git a/tcpip/header/eth_test.go b/tcpip/header/eth_test.go
new file mode 100644
index 0000000..c97fb5c
--- /dev/null
+++ b/tcpip/header/eth_test.go
@@ -0,0 +1,68 @@
+// Copyright 2018 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package header
+
+import (
+ "testing"
+
+ "github.com/google/netstack/tcpip"
+)
+
+func TestIsValidUnicastEthernetAddress(t *testing.T) {
+ tests := []struct {
+ name string
+ addr tcpip.LinkAddress
+ expected bool
+ }{
+ {
+ "Nil",
+ tcpip.LinkAddress([]byte(nil)),
+ false,
+ },
+ {
+ "Empty",
+ tcpip.LinkAddress(""),
+ false,
+ },
+ {
+ "InvalidLength",
+ tcpip.LinkAddress("\x01\x02\x03"),
+ false,
+ },
+ {
+ "Unspecified",
+ unspecifiedEthernetAddress,
+ false,
+ },
+ {
+ "Multicast",
+ tcpip.LinkAddress("\x01\x02\x03\x04\x05\x06"),
+ false,
+ },
+ {
+ "Valid",
+ tcpip.LinkAddress("\x02\x02\x03\x04\x05\x06"),
+ true,
+ },
+ }
+
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ if got := IsValidUnicastEthernetAddress(test.addr); got != test.expected {
+ t.Fatalf("got IsValidUnicastEthernetAddress = %t, want = %t", got, test.expected)
+ }
+ })
+ }
+}
diff --git a/tcpip/header/ipv6.go b/tcpip/header/ipv6.go
index a496402..c2223bb 100644
--- a/tcpip/header/ipv6.go
+++ b/tcpip/header/ipv6.go
@@ -101,6 +101,15 @@
return subnet
}()
+// IPv6LinkLocalPrefix is the prefix for IPv6 link-local addresses, as defined
+// by RFC 4291 section 2.5.6.
+//
+// The prefix is fe80::/64
+var IPv6LinkLocalPrefix = tcpip.AddressWithPrefix{
+ Address: "\xfe\x80\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00",
+ PrefixLen: 64,
+}
+
// PayloadLength returns the value of the "payload length" field of the ipv6
// header.
func (b IPv6) PayloadLength() uint16 {
diff --git a/tcpip/header/ndp_neighbor_advert.go b/tcpip/header/ndp_neighbor_advert.go
index 3abc5e1..6fbbf77 100644
--- a/tcpip/header/ndp_neighbor_advert.go
+++ b/tcpip/header/ndp_neighbor_advert.go
@@ -18,6 +18,8 @@
// NDPNeighborAdvert is an NDP Neighbor Advertisement message. It will
// only contain the body of an ICMPv6 packet.
+//
+// See RFC 4861 section 4.4 for more details.
type NDPNeighborAdvert []byte
const (
diff --git a/tcpip/header/ndp_neighbor_solicit.go b/tcpip/header/ndp_neighbor_solicit.go
index 2637759..d5a3208 100644
--- a/tcpip/header/ndp_neighbor_solicit.go
+++ b/tcpip/header/ndp_neighbor_solicit.go
@@ -18,6 +18,8 @@
// NDPNeighborSolicit is an NDP Neighbor Solicitation message. It will only
// contain the body of an ICMPv6 packet.
+//
+// See RFC 4861 section 4.3 for more details.
type NDPNeighborSolicit []byte
const (
diff --git a/tcpip/header/ndp_options.go b/tcpip/header/ndp_options.go
index dd02a04..cff265d 100644
--- a/tcpip/header/ndp_options.go
+++ b/tcpip/header/ndp_options.go
@@ -15,6 +15,9 @@
package header
import (
+ "encoding/binary"
+ "time"
+
"github.com/google/netstack/tcpip"
)
@@ -27,6 +30,65 @@
// 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.
@@ -154,6 +216,9 @@
// 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.
@@ -170,3 +235,102 @@
func (o NDPTargetLinkLayerAddressOption) serializeInto(b []byte) int {
return copy(b, o)
}
+
+// 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/tcpip/header/ndp_router_advert.go b/tcpip/header/ndp_router_advert.go
new file mode 100644
index 0000000..bf76108
--- /dev/null
+++ b/tcpip/header/ndp_router_advert.go
@@ -0,0 +1,112 @@
+// Copyright 2019 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package header
+
+import (
+ "encoding/binary"
+ "time"
+)
+
+// NDPRouterAdvert is an NDP Router Advertisement message. It will only contain
+// the body of an ICMPv6 packet.
+//
+// See RFC 4861 section 4.2 for more details.
+type NDPRouterAdvert []byte
+
+const (
+ // NDPRAMinimumSize is the minimum size of a valid NDP Router
+ // Advertisement message (body of an ICMPv6 packet).
+ NDPRAMinimumSize = 12
+
+ // ndpRACurrHopLimitOffset is the byte of the Curr Hop Limit field
+ // within an NDPRouterAdvert.
+ ndpRACurrHopLimitOffset = 0
+
+ // ndpRAFlagsOffset is the byte with the NDP RA bit-fields/flags
+ // within an NDPRouterAdvert.
+ ndpRAFlagsOffset = 1
+
+ // ndpRAManagedAddrConfFlagMask is the mask of the Managed Address
+ // Configuration flag within the bit-field/flags byte of an
+ // NDPRouterAdvert.
+ ndpRAManagedAddrConfFlagMask = (1 << 7)
+
+ // ndpRAOtherConfFlagMask is the mask of the Other Configuration flag
+ // within the bit-field/flags byte of an NDPRouterAdvert.
+ ndpRAOtherConfFlagMask = (1 << 6)
+
+ // ndpRARouterLifetimeOffset is the start of the 2-byte Router Lifetime
+ // field within an NDPRouterAdvert.
+ ndpRARouterLifetimeOffset = 2
+
+ // ndpRAReachableTimeOffset is the start of the 4-byte Reachable Time
+ // field within an NDPRouterAdvert.
+ ndpRAReachableTimeOffset = 4
+
+ // ndpRARetransTimerOffset is the start of the 4-byte Retrans Timer
+ // field within an NDPRouterAdvert.
+ ndpRARetransTimerOffset = 8
+
+ // ndpRAOptionsOffset is the start of the NDP options in an
+ // NDPRouterAdvert.
+ ndpRAOptionsOffset = 12
+)
+
+// CurrHopLimit returns the value of the Curr Hop Limit field.
+func (b NDPRouterAdvert) CurrHopLimit() uint8 {
+ return b[ndpRACurrHopLimitOffset]
+}
+
+// ManagedAddrConfFlag returns the value of the Managed Address Configuration
+// flag.
+func (b NDPRouterAdvert) ManagedAddrConfFlag() bool {
+ return b[ndpRAFlagsOffset]&ndpRAManagedAddrConfFlagMask != 0
+}
+
+// OtherConfFlag returns the value of the Other Configuration flag.
+func (b NDPRouterAdvert) OtherConfFlag() bool {
+ return b[ndpRAFlagsOffset]&ndpRAOtherConfFlagMask != 0
+}
+
+// RouterLifetime returns the lifetime associated with the default router. A
+// value of 0 means the source of the Router Advertisement is not a default
+// router and SHOULD NOT appear on the default router list. Note, a value of 0
+// only means that the router should not be used as a default router, it does
+// not apply to other information contained in the Router Advertisement.
+func (b NDPRouterAdvert) RouterLifetime() time.Duration {
+ // The field is the time in seconds, as per RFC 4861 section 4.2.
+ return time.Second * time.Duration(binary.BigEndian.Uint16(b[ndpRARouterLifetimeOffset:]))
+}
+
+// ReachableTime returns the time that a node assumes a neighbor is reachable
+// after having received a reachability confirmation. A value of 0 means
+// that it is unspecified by the source of the Router Advertisement message.
+func (b NDPRouterAdvert) ReachableTime() time.Duration {
+ // The field is the time in milliseconds, as per RFC 4861 section 4.2.
+ return time.Millisecond * time.Duration(binary.BigEndian.Uint32(b[ndpRAReachableTimeOffset:]))
+}
+
+// RetransTimer returns the time between retransmitted Neighbor Solicitation
+// messages. A value of 0 means that it is unspecified by the source of the
+// Router Advertisement message.
+func (b NDPRouterAdvert) RetransTimer() time.Duration {
+ // The field is the time in milliseconds, as per RFC 4861 section 4.2.
+ return time.Millisecond * time.Duration(binary.BigEndian.Uint32(b[ndpRARetransTimerOffset:]))
+}
+
+// Options returns an NDPOptions of the the options body.
+func (b NDPRouterAdvert) Options() NDPOptions {
+ return NDPOptions(b[ndpRAOptionsOffset:])
+}
diff --git a/tcpip/header/ndp_test.go b/tcpip/header/ndp_test.go
index 7d66e70..7297a61 100644
--- a/tcpip/header/ndp_test.go
+++ b/tcpip/header/ndp_test.go
@@ -17,6 +17,7 @@
import (
"bytes"
"testing"
+ "time"
"github.com/google/netstack/tcpip"
)
@@ -117,6 +118,40 @@
}
}
+func TestNDPRouterAdvert(t *testing.T) {
+ b := []byte{
+ 64, 128, 1, 2,
+ 3, 4, 5, 6,
+ 7, 8, 9, 10,
+ }
+
+ ra := NDPRouterAdvert(b)
+
+ if got := ra.CurrHopLimit(); got != 64 {
+ t.Fatalf("got ra.CurrHopLimit = %d, want = 64", got)
+ }
+
+ if got := ra.ManagedAddrConfFlag(); !got {
+ t.Fatalf("got ManagedAddrConfFlag = false, want = true")
+ }
+
+ if got := ra.OtherConfFlag(); got {
+ t.Fatalf("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)
+ }
+
+ if got, want := ra.ReachableTime(), time.Millisecond*50595078; got != want {
+ t.Fatalf("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)
+ }
+}
+
// TestNDPTargetLinkLayerAddressOptionSerialize tests serializing a
// NDPTargetLinkLayerAddressOption.
func TestNDPTargetLinkLayerAddressOptionSerialize(t *testing.T) {
@@ -162,3 +197,74 @@
})
}
}
+
+// 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)
+ }
+
+ // First two bytes are the Type and Length fields, which are not part of
+ // the option body.
+ pi := NDPPrefixInformation(targetBuf[2:])
+
+ if got := pi.Type(); got != 3 {
+ t.Fatalf("got Type = %d, want = 3", got)
+ }
+
+ if got := pi.Length(); got != 30 {
+ t.Fatalf("got Length = %d, want = 30", got)
+ }
+
+ if got := pi.PrefixLength(); got != 43 {
+ t.Fatalf("got PrefixLength = %d, want = 43", got)
+ }
+
+ if pi.OnLinkFlag() {
+ t.Fatalf("got OnLinkFlag = true, want = false")
+ }
+
+ if !pi.AutonomousAddressConfigurationFlag() {
+ t.Fatalf("got AutonomousAddressConfigurationFlag = false, want = true")
+ }
+
+ if got, want := pi.ValidLifetime(), 16909060*time.Second; got != want {
+ t.Fatalf("got ValidLifetime = %d, want = %d", got, want)
+ }
+
+ if got, want := pi.PreferredLifetime(), 84281096*time.Second; got != want {
+ t.Fatalf("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.Fatalf("got Prefix = %s, want = %s", got, want)
+ }
+}
diff --git a/tcpip/link/channel/channel.go b/tcpip/link/channel/channel.go
index e13061c..fdaa8be 100644
--- a/tcpip/link/channel/channel.go
+++ b/tcpip/link/channel/channel.go
@@ -72,7 +72,7 @@
// InjectLinkAddr injects an inbound packet with a remote link address.
func (e *Endpoint) InjectLinkAddr(protocol tcpip.NetworkProtocolNumber, remote tcpip.LinkAddress, vv buffer.VectorisedView) {
- e.dispatcher.DeliverNetworkPacket(e, remote, "" /* local */, protocol, vv.Clone(nil))
+ e.dispatcher.DeliverNetworkPacket(e, remote, "" /* local */, protocol, vv.Clone(nil), nil /* linkHeader */)
}
// Attach saves the stack network-layer dispatcher for use later when packets
@@ -96,7 +96,7 @@
func (e *Endpoint) Capabilities() stack.LinkEndpointCapabilities {
caps := stack.LinkEndpointCapabilities(0)
if e.GSO {
- caps |= stack.CapabilityGSO
+ caps |= stack.CapabilityHardwareGSO
}
return caps
}
@@ -134,5 +134,49 @@
return nil
}
+// WritePackets stores outbound packets into the channel.
+func (e *Endpoint) WritePackets(_ *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) {
+ payloadView := payload.ToView()
+ n := 0
+packetLoop:
+ for i := range hdrs {
+ hdr := &hdrs[i].Hdr
+ off := hdrs[i].Off
+ size := hdrs[i].Size
+ p := PacketInfo{
+ Header: hdr.View(),
+ Proto: protocol,
+ Payload: buffer.NewViewFromBytes(payloadView[off : off+size]),
+ GSO: gso,
+ }
+
+ select {
+ case e.C <- p:
+ n++
+ default:
+ break packetLoop
+ }
+ }
+
+ return n, nil
+}
+
+// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket.
+func (e *Endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error {
+ p := PacketInfo{
+ Header: packet.ToView(),
+ Proto: 0,
+ Payload: buffer.View{},
+ GSO: nil,
+ }
+
+ select {
+ case e.C <- p:
+ default:
+ }
+
+ return nil
+}
+
// Wait implements stack.LinkEndpoint.Wait.
func (*Endpoint) Wait() {}
diff --git a/tcpip/link/fdbased/endpoint.go b/tcpip/link/fdbased/endpoint.go
index 9269298..776db51 100644
--- a/tcpip/link/fdbased/endpoint.go
+++ b/tcpip/link/fdbased/endpoint.go
@@ -165,6 +165,9 @@
// disabled.
GSOMaxSize uint32
+ // SoftwareGSOEnabled indicates whether software GSO is enabled or not.
+ SoftwareGSOEnabled bool
+
// PacketDispatchMode specifies the type of inbound dispatcher to be
// used for this endpoint.
PacketDispatchMode PacketDispatchMode
@@ -242,7 +245,11 @@
}
if isSocket {
if opts.GSOMaxSize != 0 {
- e.caps |= stack.CapabilityGSO
+ if opts.SoftwareGSOEnabled {
+ e.caps |= stack.CapabilitySoftwareGSO
+ } else {
+ e.caps |= stack.CapabilityHardwareGSO
+ }
e.gsoMaxSize = opts.GSOMaxSize
}
}
@@ -397,7 +404,7 @@
eth.Encode(ethHdr)
}
- if e.Capabilities()&stack.CapabilityGSO != 0 {
+ if e.Capabilities()&stack.CapabilityHardwareGSO != 0 {
vnetHdr := virtioNetHdr{}
vnetHdrBuf := vnetHdrToByteSlice(&vnetHdr)
if gso != nil {
@@ -430,8 +437,130 @@
return rawfile.NonBlockingWrite3(e.fds[0], hdr.View(), payload.ToView(), nil)
}
-// WriteRawPacket writes a raw packet directly to the file descriptor.
-func (e *endpoint) WriteRawPacket(dest tcpip.Address, packet []byte) *tcpip.Error {
+// WritePackets writes outbound packets to the file descriptor. If it is not
+// currently writable, the packet is dropped.
+func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) {
+ var ethHdrBuf []byte
+ // hdr + data
+ iovLen := 2
+ if e.hdrSize > 0 {
+ // Add ethernet header if needed.
+ ethHdrBuf = make([]byte, header.EthernetMinimumSize)
+ eth := header.Ethernet(ethHdrBuf)
+ ethHdr := &header.EthernetFields{
+ DstAddr: r.RemoteLinkAddress,
+ Type: protocol,
+ }
+
+ // Preserve the src address if it's set in the route.
+ if r.LocalLinkAddress != "" {
+ ethHdr.SrcAddr = r.LocalLinkAddress
+ } else {
+ ethHdr.SrcAddr = e.addr
+ }
+ eth.Encode(ethHdr)
+ iovLen++
+ }
+
+ n := len(hdrs)
+
+ views := payload.Views()
+ /*
+ * Each bondary in views can add one more iovec.
+ *
+ * payload | | | |
+ * -----------------------------
+ * packets | | | | | | |
+ * -----------------------------
+ * iovecs | | | | | | | | |
+ */
+ iovec := make([]syscall.Iovec, n*iovLen+len(views)-1)
+ mmsgHdrs := make([]rawfile.MMsgHdr, n)
+
+ iovecIdx := 0
+ viewIdx := 0
+ viewOff := 0
+ off := 0
+ nextOff := 0
+ for i := range hdrs {
+ prevIovecIdx := iovecIdx
+ mmsgHdr := &mmsgHdrs[i]
+ mmsgHdr.Msg.Iov = &iovec[iovecIdx]
+ packetSize := hdrs[i].Size
+ hdr := &hdrs[i].Hdr
+
+ off = hdrs[i].Off
+ if off != nextOff {
+ // We stop in a different point last time.
+ size := packetSize
+ viewIdx = 0
+ viewOff = 0
+ for size > 0 {
+ if size >= len(views[viewIdx]) {
+ viewIdx++
+ viewOff = 0
+ size -= len(views[viewIdx])
+ } else {
+ viewOff = size
+ size = 0
+ }
+ }
+ }
+ nextOff = off + packetSize
+
+ if ethHdrBuf != nil {
+ v := &iovec[iovecIdx]
+ v.Base = ðHdrBuf[0]
+ v.Len = uint64(len(ethHdrBuf))
+ iovecIdx++
+ }
+
+ v := &iovec[iovecIdx]
+ hdrView := hdr.View()
+ v.Base = &hdrView[0]
+ v.Len = uint64(len(hdrView))
+ iovecIdx++
+
+ for packetSize > 0 {
+ vec := &iovec[iovecIdx]
+ iovecIdx++
+
+ v := views[viewIdx]
+ vec.Base = &v[viewOff]
+ s := len(v) - viewOff
+ if s <= packetSize {
+ viewIdx++
+ viewOff = 0
+ } else {
+ s = packetSize
+ viewOff += s
+ }
+ vec.Len = uint64(s)
+ packetSize -= s
+ }
+
+ mmsgHdr.Msg.Iovlen = uint64(iovecIdx - prevIovecIdx)
+ }
+
+ packets := 0
+ for packets < n {
+ sent, err := rawfile.NonBlockingSendMMsg(e.fds[0], mmsgHdrs)
+ if err != nil {
+ return packets, err
+ }
+ packets += sent
+ mmsgHdrs = mmsgHdrs[sent:]
+ }
+ return packets, nil
+}
+
+// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket.
+func (e *endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error {
+ return rawfile.NonBlockingWrite(e.fds[0], packet.ToView())
+}
+
+// InjectOutobund implements stack.InjectableEndpoint.InjectOutbound.
+func (e *endpoint) InjectOutbound(dest tcpip.Address, packet []byte) *tcpip.Error {
return rawfile.NonBlockingWrite(e.fds[0], packet)
}
@@ -468,9 +597,9 @@
e.dispatcher = dispatcher
}
-// Inject injects an inbound packet.
-func (e *InjectableEndpoint) Inject(protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) {
- e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, vv)
+// InjectInbound injects an inbound packet.
+func (e *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) {
+ e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, vv, nil /* linkHeader */)
}
// NewInjectable creates a new fd-based InjectableEndpoint.
diff --git a/tcpip/link/fdbased/endpoint_test.go b/tcpip/link/fdbased/endpoint_test.go
index 93e9e80..3bf9e29 100644
--- a/tcpip/link/fdbased/endpoint_test.go
+++ b/tcpip/link/fdbased/endpoint_test.go
@@ -43,9 +43,10 @@
)
type packetInfo struct {
- raddr tcpip.LinkAddress
- proto tcpip.NetworkProtocolNumber
- contents buffer.View
+ raddr tcpip.LinkAddress
+ proto tcpip.NetworkProtocolNumber
+ contents buffer.View
+ linkHeader buffer.View
}
type context struct {
@@ -92,8 +93,8 @@
syscall.Close(c.fds[1])
}
-func (c *context) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote tcpip.LinkAddress, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) {
- c.ch <- packetInfo{remote, protocol, vv.ToView()}
+func (c *context) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote tcpip.LinkAddress, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView, linkHeader buffer.View) {
+ c.ch <- packetInfo{remote, protocol, vv.ToView(), linkHeader}
}
func TestNoEthernetProperties(t *testing.T) {
@@ -293,11 +294,12 @@
b[i] = uint8(rand.Intn(256))
}
+ var hdr header.Ethernet
if !eth {
// So that it looks like an IPv4 packet.
b[0] = 0x40
} else {
- hdr := make(header.Ethernet, header.EthernetMinimumSize)
+ hdr = make(header.Ethernet, header.EthernetMinimumSize)
hdr.Encode(&header.EthernetFields{
SrcAddr: raddr,
DstAddr: laddr,
@@ -315,9 +317,10 @@
select {
case pi := <-c.ch:
want := packetInfo{
- raddr: raddr,
- proto: proto,
- contents: b,
+ raddr: raddr,
+ proto: proto,
+ contents: b,
+ linkHeader: buffer.View(hdr),
}
if !eth {
want.proto = header.IPv4ProtocolNumber
diff --git a/tcpip/link/fdbased/mmap.go b/tcpip/link/fdbased/mmap.go
index d1a0a7c..a53383c 100644
--- a/tcpip/link/fdbased/mmap.go
+++ b/tcpip/link/fdbased/mmap.go
@@ -169,9 +169,10 @@
var (
p tcpip.NetworkProtocolNumber
remote, local tcpip.LinkAddress
+ eth header.Ethernet
)
if d.e.hdrSize > 0 {
- eth := header.Ethernet(pkt)
+ eth = header.Ethernet(pkt)
p = eth.Type()
remote = eth.SourceAddress()
local = eth.DestinationAddress()
@@ -189,6 +190,6 @@
}
pkt = pkt[d.e.hdrSize:]
- d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, buffer.NewVectorisedView(len(pkt), []buffer.View{buffer.View(pkt)}))
+ d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, buffer.NewVectorisedView(len(pkt), []buffer.View{buffer.View(pkt)}), buffer.View(eth))
return true, nil
}
diff --git a/tcpip/link/fdbased/packet_dispatchers.go b/tcpip/link/fdbased/packet_dispatchers.go
index 6575ae3..adf9096 100644
--- a/tcpip/link/fdbased/packet_dispatchers.go
+++ b/tcpip/link/fdbased/packet_dispatchers.go
@@ -53,7 +53,7 @@
d := &readVDispatcher{fd: fd, e: e}
d.views = make([]buffer.View, len(BufConfig))
iovLen := len(BufConfig)
- if d.e.Capabilities()&stack.CapabilityGSO != 0 {
+ if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 {
iovLen++
}
d.iovecs = make([]syscall.Iovec, iovLen)
@@ -63,7 +63,7 @@
func (d *readVDispatcher) allocateViews(bufConfig []int) {
var vnetHdr [virtioNetHdrSize]byte
vnetHdrOff := 0
- if d.e.Capabilities()&stack.CapabilityGSO != 0 {
+ if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 {
// The kernel adds virtioNetHdr before each packet, but
// we don't use it, so so we allocate a buffer for it,
// add it in iovecs but don't add it in a view.
@@ -106,7 +106,7 @@
if err != nil {
return false, err
}
- if d.e.Capabilities()&stack.CapabilityGSO != 0 {
+ if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 {
// Skip virtioNetHdr which is added before each packet, it
// isn't used and it isn't in a view.
n -= virtioNetHdrSize
@@ -118,9 +118,10 @@
var (
p tcpip.NetworkProtocolNumber
remote, local tcpip.LinkAddress
+ eth header.Ethernet
)
if d.e.hdrSize > 0 {
- eth := header.Ethernet(d.views[0])
+ eth = header.Ethernet(d.views[0][:header.EthernetMinimumSize])
p = eth.Type()
remote = eth.SourceAddress()
local = eth.DestinationAddress()
@@ -141,7 +142,7 @@
vv := buffer.NewVectorisedView(n, d.views[:used])
vv.TrimFront(d.e.hdrSize)
- d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, vv)
+ d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, vv, buffer.View(eth))
// Prepare e.views for another packet: release used views.
for i := 0; i < used; i++ {
@@ -194,7 +195,7 @@
}
d.iovecs = make([][]syscall.Iovec, MaxMsgsPerRecv)
iovLen := len(BufConfig)
- if d.e.Capabilities()&stack.CapabilityGSO != 0 {
+ if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 {
// virtioNetHdr is prepended before each packet.
iovLen++
}
@@ -225,7 +226,7 @@
for k := 0; k < len(d.views); k++ {
var vnetHdr [virtioNetHdrSize]byte
vnetHdrOff := 0
- if d.e.Capabilities()&stack.CapabilityGSO != 0 {
+ if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 {
// The kernel adds virtioNetHdr before each packet, but
// we don't use it, so so we allocate a buffer for it,
// add it in iovecs but don't add it in a view.
@@ -261,7 +262,7 @@
// Process each of received packets.
for k := 0; k < nMsgs; k++ {
n := int(d.msgHdrs[k].Len)
- if d.e.Capabilities()&stack.CapabilityGSO != 0 {
+ if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 {
n -= virtioNetHdrSize
}
if n <= d.e.hdrSize {
@@ -271,9 +272,10 @@
var (
p tcpip.NetworkProtocolNumber
remote, local tcpip.LinkAddress
+ eth header.Ethernet
)
if d.e.hdrSize > 0 {
- eth := header.Ethernet(d.views[k][0])
+ eth = header.Ethernet(d.views[k][0])
p = eth.Type()
remote = eth.SourceAddress()
local = eth.DestinationAddress()
@@ -293,7 +295,7 @@
used := d.capViews(k, int(n), BufConfig)
vv := buffer.NewVectorisedView(int(n), d.views[k][:used])
vv.TrimFront(d.e.hdrSize)
- d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, vv)
+ d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, vv, buffer.View(eth))
// Prepare e.views for another packet: release used views.
for i := 0; i < used; i++ {
diff --git a/tcpip/link/loopback/loopback.go b/tcpip/link/loopback/loopback.go
index a631eb8..dc9a9e6 100644
--- a/tcpip/link/loopback/loopback.go
+++ b/tcpip/link/loopback/loopback.go
@@ -23,6 +23,7 @@
import (
"github.com/google/netstack/tcpip"
"github.com/google/netstack/tcpip/buffer"
+ "github.com/google/netstack/tcpip/header"
"github.com/google/netstack/tcpip/stack"
)
@@ -70,6 +71,9 @@
return ""
}
+// Wait implements stack.LinkEndpoint.Wait.
+func (*endpoint) Wait() {}
+
// WritePacket implements stack.LinkEndpoint.WritePacket. It delivers outbound
// packets to the network-layer dispatcher.
func (e *endpoint) WritePacket(_ *stack.Route, _ *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error {
@@ -81,10 +85,27 @@
// Because we're immediately turning around and writing the packet back to the
// rx path, we intentionally don't preserve the remote and local link
// addresses from the stack.Route we're passed.
- e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, vv)
+ e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, vv, nil /* linkHeader */)
return nil
}
-// Wait implements stack.LinkEndpoint.Wait.
-func (*endpoint) Wait() {}
+// WritePackets implements stack.LinkEndpoint.WritePackets.
+func (e *endpoint) WritePackets(_ *stack.Route, _ *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) {
+ panic("not implemented")
+}
+
+// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket.
+func (e *endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error {
+ // Reject the packet if it's shorter than an ethernet header.
+ if packet.Size() < header.EthernetMinimumSize {
+ return tcpip.ErrBadAddress
+ }
+
+ // There should be an ethernet header at the beginning of packet.
+ linkHeader := header.Ethernet(packet.First()[:header.EthernetMinimumSize])
+ packet.TrimFront(len(linkHeader))
+ e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, linkHeader.Type(), packet, buffer.View(linkHeader))
+
+ return nil
+}
diff --git a/tcpip/link/muxed/injectable.go b/tcpip/link/muxed/injectable.go
index 28d82d4..397431c 100644
--- a/tcpip/link/muxed/injectable.go
+++ b/tcpip/link/muxed/injectable.go
@@ -79,29 +79,47 @@
return m.dispatcher != nil
}
-// Inject implements stack.InjectableLinkEndpoint.
-func (m *InjectableEndpoint) Inject(protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) {
- m.dispatcher.DeliverNetworkPacket(m, "" /* remote */, "" /* local */, protocol, vv)
+// InjectInbound implements stack.InjectableLinkEndpoint.
+func (m *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) {
+ m.dispatcher.DeliverNetworkPacket(m, "" /* remote */, "" /* local */, protocol, vv, nil /* linkHeader */)
+}
+
+// WritePackets writes outbound packets to the appropriate
+// LinkInjectableEndpoint based on the RemoteAddress. HandleLocal only works if
+// r.RemoteAddress has a route registered in this endpoint.
+func (m *InjectableEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) {
+ endpoint, ok := m.routes[r.RemoteAddress]
+ if !ok {
+ return 0, tcpip.ErrNoRoute
+ }
+ return endpoint.WritePackets(r, gso, hdrs, payload, protocol)
}
// WritePacket writes outbound packets to the appropriate LinkInjectableEndpoint
// based on the RemoteAddress. HandleLocal only works if r.RemoteAddress has a
// route registered in this endpoint.
-func (m *InjectableEndpoint) WritePacket(r *stack.Route, _ *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error {
+func (m *InjectableEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error {
if endpoint, ok := m.routes[r.RemoteAddress]; ok {
- return endpoint.WritePacket(r, nil /* gso */, hdr, payload, protocol)
+ return endpoint.WritePacket(r, gso, hdr, payload, protocol)
}
return tcpip.ErrNoRoute
}
-// WriteRawPacket writes outbound packets to the appropriate
+// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket.
+func (m *InjectableEndpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error {
+ // WriteRawPacket doesn't get a route or network address, so there's
+ // nowhere to write this.
+ return tcpip.ErrNoRoute
+}
+
+// InjectOutbound writes outbound packets to the appropriate
// LinkInjectableEndpoint based on the dest address.
-func (m *InjectableEndpoint) WriteRawPacket(dest tcpip.Address, packet []byte) *tcpip.Error {
+func (m *InjectableEndpoint) InjectOutbound(dest tcpip.Address, packet []byte) *tcpip.Error {
endpoint, ok := m.routes[dest]
if !ok {
return tcpip.ErrNoRoute
}
- return endpoint.WriteRawPacket(dest, packet)
+ return endpoint.InjectOutbound(dest, packet)
}
// Wait implements stack.LinkEndpoint.Wait.
diff --git a/tcpip/link/muxed/injectable_test.go b/tcpip/link/muxed/injectable_test.go
index 23afef0..84e4eb6 100644
--- a/tcpip/link/muxed/injectable_test.go
+++ b/tcpip/link/muxed/injectable_test.go
@@ -31,7 +31,7 @@
func TestInjectableEndpointRawDispatch(t *testing.T) {
endpoint, sock, dstIP := makeTestInjectableEndpoint(t)
- endpoint.WriteRawPacket(dstIP, []byte{0xFA})
+ endpoint.InjectOutbound(dstIP, []byte{0xFA})
buf := make([]byte, ipv4.MaxTotalSize)
bytesRead, err := sock.Read(buf)
diff --git a/tcpip/link/rawfile/rawfile_unsafe.go b/tcpip/link/rawfile/rawfile_unsafe.go
index c4e122b..762c5c9 100644
--- a/tcpip/link/rawfile/rawfile_unsafe.go
+++ b/tcpip/link/rawfile/rawfile_unsafe.go
@@ -23,6 +23,7 @@
"unsafe"
"github.com/google/netstack/tcpip"
+ "golang.org/x/sys/unix"
)
// GetMTU determines the MTU of a network interface device.
@@ -101,6 +102,16 @@
return nil
}
+// NonBlockingSendMMsg sends multiple messages on a socket.
+func NonBlockingSendMMsg(fd int, msgHdrs []MMsgHdr) (int, *tcpip.Error) {
+ n, _, e := syscall.RawSyscall6(unix.SYS_SENDMMSG, uintptr(fd), uintptr(unsafe.Pointer(&msgHdrs[0])), uintptr(len(msgHdrs)), syscall.MSG_DONTWAIT, 0, 0)
+ if e != 0 {
+ return 0, TranslateErrno(e)
+ }
+
+ return int(n), nil
+}
+
// PollEvent represents the pollfd structure passed to a poll() system call.
type PollEvent struct {
FD int32
diff --git a/tcpip/link/sharedmem/sharedmem.go b/tcpip/link/sharedmem/sharedmem.go
index 38746ef..227dcbc 100644
--- a/tcpip/link/sharedmem/sharedmem.go
+++ b/tcpip/link/sharedmem/sharedmem.go
@@ -212,6 +212,26 @@
return nil
}
+// WritePackets implements stack.LinkEndpoint.WritePackets.
+func (e *endpoint) WritePackets(r *stack.Route, _ *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) {
+ panic("not implemented")
+}
+
+// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket.
+func (e *endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error {
+ v := packet.ToView()
+ // Transmit the packet.
+ e.mu.Lock()
+ ok := e.tx.transmit(v, buffer.View{})
+ e.mu.Unlock()
+
+ if !ok {
+ return tcpip.ErrWouldBlock
+ }
+
+ return nil
+}
+
// dispatchLoop reads packets from the rx queue in a loop and dispatches them
// to the network stack.
func (e *endpoint) dispatchLoop(d stack.NetworkDispatcher) {
@@ -254,7 +274,7 @@
// Send packet up the stack.
eth := header.Ethernet(b)
- d.DeliverNetworkPacket(e, eth.SourceAddress(), eth.DestinationAddress(), eth.Type(), buffer.View(b[header.EthernetMinimumSize:]).ToVectorisedView())
+ d.DeliverNetworkPacket(e, eth.SourceAddress(), eth.DestinationAddress(), eth.Type(), buffer.View(b[header.EthernetMinimumSize:]).ToVectorisedView(), buffer.View(eth))
}
// Clean state.
diff --git a/tcpip/link/sharedmem/sharedmem_test.go b/tcpip/link/sharedmem/sharedmem_test.go
index 00cffa5..5ac2e4a 100644
--- a/tcpip/link/sharedmem/sharedmem_test.go
+++ b/tcpip/link/sharedmem/sharedmem_test.go
@@ -78,9 +78,10 @@
}
type packetInfo struct {
- addr tcpip.LinkAddress
- proto tcpip.NetworkProtocolNumber
- vv buffer.VectorisedView
+ addr tcpip.LinkAddress
+ proto tcpip.NetworkProtocolNumber
+ vv buffer.VectorisedView
+ linkHeader buffer.View
}
type testContext struct {
@@ -130,12 +131,13 @@
return c
}
-func (c *testContext) DeliverNetworkPacket(_ stack.LinkEndpoint, remoteLinkAddr, localLinkAddr tcpip.LinkAddress, proto tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) {
+func (c *testContext) DeliverNetworkPacket(_ stack.LinkEndpoint, remoteLinkAddr, localLinkAddr tcpip.LinkAddress, proto tcpip.NetworkProtocolNumber, vv buffer.VectorisedView, linkHeader buffer.View) {
c.mu.Lock()
c.packets = append(c.packets, packetInfo{
- addr: remoteLinkAddr,
- proto: proto,
- vv: vv.Clone(nil),
+ addr: remoteLinkAddr,
+ proto: proto,
+ vv: vv.Clone(nil),
+ linkHeader: linkHeader,
})
c.mu.Unlock()
diff --git a/tcpip/link/sniffer/sniffer.go b/tcpip/link/sniffer/sniffer.go
index 10ddb8b..6a7d564 100644
--- a/tcpip/link/sniffer/sniffer.go
+++ b/tcpip/link/sniffer/sniffer.go
@@ -116,7 +116,7 @@
// DeliverNetworkPacket implements the stack.NetworkDispatcher interface. It is
// called by the link-layer endpoint being wrapped when a packet arrives, and
// logs the packet before forwarding to the actual dispatcher.
-func (e *endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) {
+func (e *endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView, linkHeader buffer.View) {
if atomic.LoadUint32(&LogPackets) == 1 && e.file == nil {
logPacket("recv", protocol, vv.First(), nil)
}
@@ -147,7 +147,7 @@
panic(err)
}
}
- e.dispatcher.DeliverNetworkPacket(e, remote, local, protocol, vv)
+ e.dispatcher.DeliverNetworkPacket(e, remote, local, protocol, vv, linkHeader)
}
// Attach implements the stack.LinkEndpoint interface. It saves the dispatcher
@@ -193,10 +193,7 @@
return 0
}
-// WritePacket implements the stack.LinkEndpoint interface. It is called by
-// higher-level protocols to write packets; it just logs the packet and forwards
-// the request to the lower endpoint.
-func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error {
+func (e *endpoint) dumpPacket(gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) {
if atomic.LoadUint32(&LogPackets) == 1 && e.file == nil {
logPacket("send", protocol, hdr.View(), gso)
}
@@ -218,28 +215,74 @@
panic(err)
}
length -= len(hdrBuf)
- if length > 0 {
- for _, v := range payload.Views() {
- if len(v) > length {
- v = v[:length]
- }
- n, err := buf.Write(v)
- if err != nil {
- panic(err)
- }
- length -= n
- if length == 0 {
- break
- }
- }
- }
+ logVectorisedView(payload, length, buf)
if _, err := e.file.Write(buf.Bytes()); err != nil {
panic(err)
}
}
+}
+
+// WritePacket implements the stack.LinkEndpoint interface. It is called by
+// higher-level protocols to write packets; it just logs the packet and
+// forwards the request to the lower endpoint.
+func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error {
+ e.dumpPacket(gso, hdr, payload, protocol)
return e.lower.WritePacket(r, gso, hdr, payload, protocol)
}
+// WritePackets implements the stack.LinkEndpoint interface. It is called by
+// higher-level protocols to write packets; it just logs the packet and
+// forwards the request to the lower endpoint.
+func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) {
+ view := payload.ToView()
+ for _, d := range hdrs {
+ e.dumpPacket(gso, d.Hdr, buffer.NewVectorisedView(d.Size, []buffer.View{view[d.Off:][:d.Size]}), protocol)
+ }
+ return e.lower.WritePackets(r, gso, hdrs, payload, protocol)
+}
+
+// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket.
+func (e *endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error {
+ if atomic.LoadUint32(&LogPackets) == 1 && e.file == nil {
+ logPacket("send", 0, buffer.View("[raw packet, no header available]"), nil /* gso */)
+ }
+ if e.file != nil && atomic.LoadUint32(&LogPacketsToFile) == 1 {
+ length := packet.Size()
+ if length > int(e.maxPCAPLen) {
+ length = int(e.maxPCAPLen)
+ }
+
+ buf := bytes.NewBuffer(make([]byte, 0, pcapPacketHeaderLen+length))
+ if err := binary.Write(buf, binary.BigEndian, newPCAPPacketHeader(uint32(length), uint32(packet.Size()))); err != nil {
+ panic(err)
+ }
+ logVectorisedView(packet, length, buf)
+ if _, err := e.file.Write(buf.Bytes()); err != nil {
+ panic(err)
+ }
+ }
+ return e.lower.WriteRawPacket(packet)
+}
+
+func logVectorisedView(vv buffer.VectorisedView, length int, buf *bytes.Buffer) {
+ if length <= 0 {
+ return
+ }
+ for _, v := range vv.Views() {
+ if len(v) > length {
+ v = v[:length]
+ }
+ n, err := buf.Write(v)
+ if err != nil {
+ panic(err)
+ }
+ length -= n
+ if length == 0 {
+ return
+ }
+ }
+}
+
// Wait implements stack.LinkEndpoint.Wait.
func (*endpoint) Wait() {}
diff --git a/tcpip/link/waitable/waitable.go b/tcpip/link/waitable/waitable.go
index f867ea4..ada74d0 100644
--- a/tcpip/link/waitable/waitable.go
+++ b/tcpip/link/waitable/waitable.go
@@ -50,12 +50,12 @@
// It is called by the link-layer endpoint being wrapped when a packet arrives,
// and only forwards to the actual dispatcher if Wait or WaitDispatch haven't
// been called.
-func (e *Endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) {
+func (e *Endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView, linkHeader buffer.View) {
if !e.dispatchGate.Enter() {
return
}
- e.dispatcher.DeliverNetworkPacket(e, remote, local, protocol, vv)
+ e.dispatcher.DeliverNetworkPacket(e, remote, local, protocol, vv, linkHeader)
e.dispatchGate.Leave()
}
@@ -109,6 +109,30 @@
return err
}
+// WritePackets implements stack.LinkEndpoint.WritePackets. It is called by
+// higher-level protocols to write packets. It only forwards packets to the
+// lower endpoint if Wait or WaitWrite haven't been called.
+func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) {
+ if !e.writeGate.Enter() {
+ return len(hdrs), nil
+ }
+
+ n, err := e.lower.WritePackets(r, gso, hdrs, payload, protocol)
+ e.writeGate.Leave()
+ return n, err
+}
+
+// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket.
+func (e *Endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error {
+ if !e.writeGate.Enter() {
+ return nil
+ }
+
+ err := e.lower.WriteRawPacket(packet)
+ e.writeGate.Leave()
+ return err
+}
+
// WaitWrite prevents new calls to WritePacket from reaching the lower endpoint,
// and waits for inflight ones to finish before returning.
func (e *Endpoint) WaitWrite() {
diff --git a/tcpip/link/waitable/waitable_test.go b/tcpip/link/waitable/waitable_test.go
index 2016eaa..7935cb4 100644
--- a/tcpip/link/waitable/waitable_test.go
+++ b/tcpip/link/waitable/waitable_test.go
@@ -35,7 +35,7 @@
dispatcher stack.NetworkDispatcher
}
-func (e *countedEndpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) {
+func (e *countedEndpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView, linkHeader buffer.View) {
e.dispatchCount++
}
@@ -70,6 +70,17 @@
return nil
}
+// WritePackets implements stack.LinkEndpoint.WritePackets.
+func (e *countedEndpoint) WritePackets(r *stack.Route, _ *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) {
+ e.writeCount += len(hdrs)
+ return len(hdrs), nil
+}
+
+func (e *countedEndpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error {
+ e.writeCount++
+ return nil
+}
+
// Wait implements stack.LinkEndpoint.Wait.
func (*countedEndpoint) Wait() {}
@@ -109,21 +120,21 @@
}
// Dispatch and check that it goes through.
- ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, buffer.VectorisedView{})
+ ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, buffer.VectorisedView{}, buffer.View{})
if want := 1; ep.dispatchCount != want {
t.Fatalf("Unexpected dispatchCount: got=%v, want=%v", ep.dispatchCount, want)
}
// Wait on writes, then try to dispatch. It must go through.
wep.WaitWrite()
- ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, buffer.VectorisedView{})
+ ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, buffer.VectorisedView{}, buffer.View{})
if want := 2; ep.dispatchCount != want {
t.Fatalf("Unexpected dispatchCount: got=%v, want=%v", ep.dispatchCount, want)
}
// Wait on dispatches, then try to dispatch. It must not go through.
wep.WaitDispatch()
- ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, buffer.VectorisedView{})
+ ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, buffer.VectorisedView{}, buffer.View{})
if want := 2; ep.dispatchCount != want {
t.Fatalf("Unexpected dispatchCount: got=%v, want=%v", ep.dispatchCount, want)
}
diff --git a/tcpip/network/arp/arp.go b/tcpip/network/arp/arp.go
index d245583..4d23943 100644
--- a/tcpip/network/arp/arp.go
+++ b/tcpip/network/arp/arp.go
@@ -83,6 +83,11 @@
return tcpip.ErrNotSupported
}
+// WritePackets implements stack.NetworkEndpoint.WritePackets.
+func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, []stack.PacketDescriptor, buffer.VectorisedView, stack.NetworkHeaderParams, stack.PacketLooping) (int, *tcpip.Error) {
+ return 0, tcpip.ErrNotSupported
+}
+
func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error {
return tcpip.ErrNotSupported
}
diff --git a/tcpip/network/ip_test.go b/tcpip/network/ip_test.go
index f1ebbf6..bd5be08 100644
--- a/tcpip/network/ip_test.go
+++ b/tcpip/network/ip_test.go
@@ -171,6 +171,15 @@
return nil
}
+// WritePackets implements stack.LinkEndpoint.WritePackets.
+func (t *testObject) WritePackets(_ *stack.Route, _ *stack.GSO, hdr []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) {
+ panic("not implemented")
+}
+
+func (t *testObject) WriteRawPacket(_ buffer.VectorisedView) *tcpip.Error {
+ return tcpip.ErrNotSupported
+}
+
func buildIPv4Route(local, remote tcpip.Address) (stack.Route, *tcpip.Error) {
s := stack.New(stack.Options{
NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()},
diff --git a/tcpip/network/ipv4/ipv4.go b/tcpip/network/ipv4/ipv4.go
index 30176c4..b781cb6 100644
--- a/tcpip/network/ipv4/ipv4.go
+++ b/tcpip/network/ipv4/ipv4.go
@@ -198,10 +198,9 @@
return nil
}
-// WritePacket writes a packet to the given destination address and protocol.
-func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) *tcpip.Error {
+func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadSize int, params stack.NetworkHeaderParams) {
ip := header.IPv4(hdr.Prepend(header.IPv4MinimumSize))
- length := uint16(hdr.UsedLength() + payload.Size())
+ length := uint16(hdr.UsedLength() + payloadSize)
id := uint32(0)
if length > header.IPv4MaximumHeaderSize+8 {
// Packets of 68 bytes or less are required by RFC 791 to not be
@@ -219,6 +218,11 @@
DstAddr: r.RemoteAddress,
})
ip.SetChecksum(^ip.CalculateChecksum())
+}
+
+// WritePacket writes a packet to the given destination address and protocol.
+func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) *tcpip.Error {
+ e.addIPHeader(r, &hdr, payload.Size(), params)
if loop&stack.PacketLoop != 0 {
views := make([]buffer.View, 1, 1+len(payload.Views()))
@@ -242,6 +246,23 @@
return nil
}
+// WritePackets implements stack.NetworkEndpoint.WritePackets.
+func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) {
+ if loop&stack.PacketLoop != 0 {
+ panic("multiple packets in local loop")
+ }
+ if loop&stack.PacketOut == 0 {
+ return len(hdrs), nil
+ }
+
+ for i := range hdrs {
+ e.addIPHeader(r, &hdrs[i].Hdr, hdrs[i].Size, params)
+ }
+ n, err := e.linkEP.WritePackets(r, gso, hdrs, payload, ProtocolNumber)
+ r.Stats().IP.PacketsSent.IncrementBy(uint64(n))
+ return n, err
+}
+
// WriteHeaderIncludedPacket writes a packet already containing a network
// header through the given route.
func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error {
diff --git a/tcpip/network/ipv6/icmp.go b/tcpip/network/ipv6/icmp.go
index cfec146..8208527 100644
--- a/tcpip/network/ipv6/icmp.go
+++ b/tcpip/network/ipv6/icmp.go
@@ -246,15 +246,15 @@
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/tcpip/network/ipv6/ipv6.go b/tcpip/network/ipv6/ipv6.go
index 10729c8..b0eb6b0 100644
--- a/tcpip/network/ipv6/ipv6.go
+++ b/tcpip/network/ipv6/ipv6.go
@@ -97,9 +97,8 @@
return 0
}
-// WritePacket writes a packet to the given destination address and protocol.
-func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) *tcpip.Error {
- length := uint16(hdr.UsedLength() + payload.Size())
+func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadSize int, params stack.NetworkHeaderParams) {
+ length := uint16(hdr.UsedLength() + payloadSize)
ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize))
ip.Encode(&header.IPv6Fields{
PayloadLength: length,
@@ -109,6 +108,11 @@
SrcAddr: r.LocalAddress,
DstAddr: r.RemoteAddress,
})
+}
+
+// WritePacket writes a packet to the given destination address and protocol.
+func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) *tcpip.Error {
+ e.addIPHeader(r, &hdr, payload.Size(), params)
if loop&stack.PacketLoop != 0 {
views := make([]buffer.View, 1, 1+len(payload.Views()))
@@ -127,6 +131,26 @@
return e.linkEP.WritePacket(r, gso, hdr, payload, ProtocolNumber)
}
+// WritePackets implements stack.LinkEndpoint.WritePackets.
+func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) {
+ if loop&stack.PacketLoop != 0 {
+ panic("not implemented")
+ }
+ if loop&stack.PacketOut == 0 {
+ return len(hdrs), nil
+ }
+
+ for i := range hdrs {
+ hdr := &hdrs[i].Hdr
+ size := hdrs[i].Size
+ e.addIPHeader(r, hdr, size, params)
+ }
+
+ n, err := e.linkEP.WritePackets(r, gso, hdrs, payload, ProtocolNumber)
+ r.Stats().IP.PacketsSent.IncrementBy(uint64(n))
+ return n, err
+}
+
// WriteHeaderIncludedPacker implements stack.NetworkEndpoint. It is not yet
// supported by IPv6.
func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error {
diff --git a/tcpip/stack/ndp.go b/tcpip/stack/ndp.go
index 05a4561..483ae69 100644
--- a/tcpip/stack/ndp.go
+++ b/tcpip/stack/ndp.go
@@ -51,6 +51,22 @@
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 @@
// 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 @@
// 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 @@
// 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 @@
var done bool
var timer *time.Timer
- timer = time.AfterFunc(n.stack.ndpConfigs.RetransmitTimer, func() {
- n.mu.Lock()
- defer n.mu.Unlock()
+ timer = time.AfterFunc(ndp.configs.RetransmitTimer, func() {
+ var d bool
+ var err *tcpip.Error
- if done {
- // If we reach this point, it means that the DAD timer
- // fired after another goroutine already obtained the
- // NIC lock and stopped DAD before it this function
- // obtained the NIC lock. Simply return here and do
- // nothing further.
- return
- }
+ // doDadIteration does a single iteration of the DAD loop.
+ //
+ // Returns true if the integrator needs to be informed of DAD
+ // completing.
+ doDadIteration := func() bool {
+ ndp.nic.mu.Lock()
+ defer ndp.nic.mu.Unlock()
- ref, ok := n.endpoints[NetworkEndpointID{addr}]
- if !ok {
- // This should never happen.
- // We should have an endpoint for addr since we are
- // still performing DAD on it. If the endpoint does not
- // exist, but we are doing DAD on it, then we started
- // DAD at some point, but forgot to stop it when the
- // endpoint was deleted.
- panic(fmt.Sprintf("ndpdad: unrecognized addr %s for NIC(%d)", addr, n.ID()))
- }
-
- if done, 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 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
}
- ndp.stopDuplicateAddressDetection(addr)
- 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()))
+ }
+
+ d, err = ndp.doDuplicateAddressDetection(addr, remaining, ref)
+ if err != nil || d {
+ delete(ndp.dad, addr)
+
+ if err != nil {
+ log.Printf("ndpdad: Error occured during DAD iteration for addr (%s) on NIC(%d); err = %s", addr, ndp.nic.ID(), err)
+ }
+
+ // Let the integrator know DAD has completed.
+ return true
+ }
+
+ remaining--
+ timer.Reset(ndp.nic.stack.ndpConfigs.RetransmitTimer)
+ return false
}
- 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 @@
// 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 @@
// 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 @@
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/tcpip/stack/ndp_test.go b/tcpip/stack/ndp_test.go
index 40f26b2..e89d1bd 100644
--- a/tcpip/stack/ndp_test.go
+++ b/tcpip/stack/ndp_test.go
@@ -31,8 +31,8 @@
const (
addr1 = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01"
addr2 = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02"
- linkAddr1 = "\x01\x02\x03\x04\x05\x06"
- linkAddr2 = "\x01\x02\x03\x04\x05\x07"
+ addr3 = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x03"
+ linkAddr1 = "\x02\x02\x03\x04\x05\x06"
)
// TestDADDisabled tests that an address successfully resolves immediately
@@ -68,6 +68,35 @@
}
}
+// 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
@@ -89,8 +118,12 @@
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
@@ -107,8 +140,7 @@
stat := s.Stats().ICMP.V6PacketsSent.NeighborSolicit
- // Should have sent an NDP NS almost immediately.
- time.Sleep(100 * time.Millisecond)
+ // Should have sent an NDP NS immediately.
if got := stat.Value(); got != 1 {
t.Fatalf("got NeighborSolicit = %d, want = 1", got)
@@ -124,16 +156,10 @@
t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want)
}
- // Wait for the remaining time - 500ms, to make sure
- // the address is still not resolved. Note, we subtract
- // 600ms because we already waited for 100ms earlier,
- // so our remaining time is 100ms less than the expected
- // time.
- // (X - 100ms) - 500ms = X - 600ms
- //
- // TODO(b/140896005): Use events from the netstack to
- // be signalled before DAD resolves.
- time.Sleep(test.expectedRetransmitTimer*time.Duration(test.dupAddrDetectTransmits) - 600*time.Millisecond)
+ // Wait for the remaining time - some delta (500ms), to
+ // make sure the address is still not resolved.
+ const delta = 500 * time.Millisecond
+ time.Sleep(test.expectedRetransmitTimer*time.Duration(test.dupAddrDetectTransmits) - delta)
addr, err = s.GetMainNICAddress(1, header.IPv6ProtocolNumber)
if err != nil {
t.Fatalf("got stack.GetMainNICAddress(_, _) = (_, %v), want = (_, nil)", err)
@@ -142,13 +168,30 @@
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)
@@ -251,9 +294,14 @@
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
@@ -287,8 +335,28 @@
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)
@@ -303,11 +371,18 @@
// 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)
@@ -333,11 +408,27 @@
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)
@@ -351,3 +442,168 @@
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/tcpip/stack/nic.go b/tcpip/stack/nic.go
index 3fc6dc4..cad6b91 100644
--- a/tcpip/stack/nic.go
+++ b/tcpip/stack/nic.go
@@ -19,7 +19,6 @@
"sync"
"sync/atomic"
- "github.com/google/netstack/ilist"
"github.com/google/netstack/tcpip"
"github.com/google/netstack/tcpip/buffer"
"github.com/google/netstack/tcpip/header"
@@ -37,13 +36,20 @@
mu sync.RWMutex
spoofing bool
promiscuous bool
- primary map[tcpip.NetworkProtocolNumber]*ilist.List
+ primary map[tcpip.NetworkProtocolNumber][]*referencedNetworkEndpoint
endpoints map[NetworkEndpointID]*referencedNetworkEndpoint
addressRanges []tcpip.Subnet
mcastJoins map[NetworkEndpointID]int32
+ // packetEPs is protected by mu, but the contained PacketEndpoint
+ // values are not.
+ packetEPs map[tcpip.NetworkProtocolNumber][]PacketEndpoint
stats NICStats
+ // ndp is the NDP related state for NIC.
+ //
+ // Note, read and write operations on ndp require that the NIC is
+ // appropriately locked.
ndp ndpState
}
@@ -78,16 +84,21 @@
NeverPrimaryEndpoint
)
+// newNIC returns a new NIC using the default NDP configurations from stack.
func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, loopback bool) *NIC {
- return &NIC{
+ // TODO(b/141011931): Validate a LinkEndpoint (ep) is valid. For
+ // example, make sure that the link address it provides is a valid
+ // unicast ethernet address.
+ nic := &NIC{
stack: stack,
id: id,
name: name,
linkEP: ep,
loopback: loopback,
- primary: make(map[tcpip.NetworkProtocolNumber]*ilist.List),
+ primary: make(map[tcpip.NetworkProtocolNumber][]*referencedNetworkEndpoint),
endpoints: make(map[NetworkEndpointID]*referencedNetworkEndpoint),
mcastJoins: make(map[NetworkEndpointID]int32),
+ packetEPs: make(map[tcpip.NetworkProtocolNumber][]PacketEndpoint),
stats: NICStats{
Tx: DirectionStats{
Packets: &tcpip.StatCounter{},
@@ -99,9 +110,21 @@
},
},
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 {
+ nic.packetEPs[netProto] = []PacketEndpoint{}
+ }
+ for _, netProto := range stack.networkProtocols {
+ nic.packetEPs[netProto.Number()] = []PacketEndpoint{}
+ }
+
+ return nic
}
// enable enables the NIC. enable will attach the link to its LinkEndpoint and
@@ -126,11 +149,50 @@
// when we perform Duplicate Address Detection, or Router Advertisement
// when we do Router Discovery. See RFC 4862, section 5.4.2 and RFC 4861
// section 4.2 for more information.
- if _, ok := n.stack.networkProtocols[header.IPv6ProtocolNumber]; ok {
- return n.joinGroup(header.IPv6ProtocolNumber, header.IPv6AllNodesMulticastAddress)
+ //
+ // Also auto-generate an IPv6 link-local address based on the NIC's
+ // link address if it is configured to do so. Note, each interface is
+ // required to have IPv6 link-local unicast address, as per RFC 4291
+ // section 2.1.
+ _, ok := n.stack.networkProtocols[header.IPv6ProtocolNumber]
+ if !ok {
+ return nil
}
- return nil
+ n.mu.Lock()
+ defer n.mu.Unlock()
+
+ if err := n.joinGroupLocked(header.IPv6ProtocolNumber, header.IPv6AllNodesMulticastAddress); err != nil {
+ return err
+ }
+
+ if !n.stack.autoGenIPv6LinkLocal {
+ return nil
+ }
+
+ l2addr := n.linkEP.LinkAddress()
+
+ // Only attempt to generate the link-local address if we have a
+ // valid MAC address.
+ //
+ // TODO(b/141011931): Validate a LinkEndpoint's link address
+ // (provided by LinkEndpoint.LinkAddress) before reaching this
+ // point.
+ if !header.IsValidUnicastEthernetAddress(l2addr) {
+ return nil
+ }
+
+ addr := header.LinkLocalAddr(l2addr)
+
+ _, err := n.addPermanentAddressLocked(tcpip.ProtocolAddress{
+ Protocol: header.IPv6ProtocolNumber,
+ AddressWithPrefix: tcpip.AddressWithPrefix{
+ Address: addr,
+ PrefixLen: header.IPv6LinkLocalPrefix.PrefixLen,
+ },
+ }, CanBePrimaryEndpoint)
+
+ return err
}
// attachLinkEndpoint attaches the NIC to the endpoint, which will enable it
@@ -166,13 +228,7 @@
n.mu.RLock()
defer n.mu.RUnlock()
- list := n.primary[protocol]
- if list == nil {
- return nil
- }
-
- for e := list.Front(); e != nil; e = e.Next() {
- r := e.(*referencedNetworkEndpoint)
+ for _, r := range n.primary[protocol] {
if r.isValidForOutgoing() && r.tryIncRef() {
return r
}
@@ -286,9 +342,31 @@
// The NIC already have a permanent endpoint with that address.
return nil, tcpip.ErrDuplicateAddress
case permanentExpired, temporary:
- // Promote the endpoint to become permanent.
+ // Promote the endpoint to become permanent and respect
+ // the new peb.
if ref.tryIncRef() {
ref.setKind(permanent)
+
+ refs := n.primary[ref.protocol]
+ for i, r := range refs {
+ if r == ref {
+ switch peb {
+ case CanBePrimaryEndpoint:
+ return ref, nil
+ case FirstPrimaryEndpoint:
+ if i == 0 {
+ return ref, nil
+ }
+ n.primary[r.protocol] = append(refs[:i], refs[i+1:]...)
+ case NeverPrimaryEndpoint:
+ n.primary[r.protocol] = append(refs[:i], refs[i+1:]...)
+ return ref, nil
+ }
+ }
+ }
+
+ n.insertPrimaryEndpointLocked(ref, peb)
+
return ref, nil
}
// tryIncRef failing means the endpoint is scheduled to be removed once
@@ -357,22 +435,11 @@
n.endpoints[id] = ref
- l, ok := n.primary[protocolAddress.Protocol]
- if !ok {
- l = &ilist.List{}
- n.primary[protocolAddress.Protocol] = l
- }
-
- switch peb {
- case CanBePrimaryEndpoint:
- l.PushBack(ref)
- case FirstPrimaryEndpoint:
- l.PushFront(ref)
- }
+ n.insertPrimaryEndpointLocked(ref, peb)
// If we are adding a tentative IPv6 address, start DAD.
if isIPv6Unicast && kind == permanentTentative {
- if err := n.ndp.startDuplicateAddressDetection(n, protocolAddress.AddressWithPrefix.Address, ref); err != nil {
+ if err := n.ndp.startDuplicateAddressDetection(protocolAddress.AddressWithPrefix.Address, ref); err != nil {
return nil, err
}
}
@@ -425,8 +492,7 @@
var addrs []tcpip.ProtocolAddress
for proto, list := range n.primary {
- for e := list.Front(); e != nil; e = e.Next() {
- ref := e.(*referencedNetworkEndpoint)
+ for _, ref := range list {
// Don't include tentative, expired or tempory endpoints
// to avoid confusion and prevent the caller from using
// those.
@@ -491,6 +557,19 @@
return append(sns, n.addressRanges...)
}
+// insertPrimaryEndpointLocked adds r to n's primary endpoint list as required
+// by peb.
+//
+// n MUST be locked.
+func (n *NIC) insertPrimaryEndpointLocked(r *referencedNetworkEndpoint, peb PrimaryEndpointBehavior) {
+ switch peb {
+ case CanBePrimaryEndpoint:
+ n.primary[r.protocol] = append(n.primary[r.protocol], r)
+ case FirstPrimaryEndpoint:
+ n.primary[r.protocol] = append([]*referencedNetworkEndpoint{r}, n.primary[r.protocol]...)
+ }
+}
+
func (n *NIC) removeEndpointLocked(r *referencedNetworkEndpoint) {
id := *r.ep.ID()
@@ -508,9 +587,12 @@
}
delete(n.endpoints, id)
- wasInList := r.Next() != nil || r.Prev() != nil || r == n.primary[r.protocol].Front()
- if wasInList {
- n.primary[r.protocol].Remove(r)
+ refs := n.primary[r.protocol]
+ for i, ref := range refs {
+ if ref == r {
+ n.primary[r.protocol] = append(refs[:i], refs[i+1:]...)
+ break
+ }
}
r.ep.Close()
@@ -580,6 +662,11 @@
// exists yet. Otherwise it just increments its count. n MUST be locked before
// joinGroupLocked is called.
func (n *NIC) joinGroupLocked(protocol tcpip.NetworkProtocolNumber, addr tcpip.Address) *tcpip.Error {
+ // TODO(b/143102137): When implementing MLD, make sure MLD packets are
+ // not sent unless a valid link-local address is available for use on n
+ // as an MLD packet's source address must be a link-local address as
+ // outlined in RFC 3810 section 5.
+
id := NetworkEndpointID{addr}
joins := n.mcastJoins[id]
if joins == 0 {
@@ -643,7 +730,7 @@
// Note that the ownership of the slice backing vv is retained by the caller.
// This rule applies only to the slice itself, not to the items of the slice;
// the ownership of the items is not retained by the caller.
-func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, _ tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) {
+func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView, linkHeader buffer.View) {
n.stats.Rx.Packets.Increment()
n.stats.Rx.Bytes.IncrementBy(uint64(vv.Size()))
@@ -653,6 +740,26 @@
return
}
+ // If no local link layer address is provided, assume it was sent
+ // directly to this NIC.
+ if local == "" {
+ local = n.linkEP.LinkAddress()
+ }
+
+ // Are any packet sockets listening for this network protocol?
+ n.mu.RLock()
+ packetEPs := n.packetEPs[protocol]
+ // Check whether there are packet sockets listening for every protocol.
+ // If we received a packet with protocol EthernetProtocolAll, then the
+ // previous for loop will have handled it.
+ if protocol != header.EthernetProtocolAll {
+ packetEPs = append(packetEPs, n.packetEPs[header.EthernetProtocolAll]...)
+ }
+ n.mu.RUnlock()
+ for _, ep := range packetEPs {
+ ep.HandlePacket(n.id, local, protocol, vv, linkHeader)
+ }
+
if netProto.Number() == header.IPv4ProtocolNumber || netProto.Number() == header.IPv6ProtocolNumber {
n.stack.stats.IP.PacketsReceived.Increment()
}
@@ -712,7 +819,10 @@
return
}
- n.stack.stats.IP.InvalidAddressesReceived.Increment()
+ // If a packet socket handled the packet, don't treat it as invalid.
+ if len(packetEPs) == 0 {
+ n.stack.stats.IP.InvalidAddressesReceived.Increment()
+ }
}
// DeliverTransportPacket delivers the packets to the appropriate transport
@@ -833,6 +943,18 @@
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 (
@@ -868,8 +990,37 @@
temporary
)
+func (n *NIC) registerPacketEndpoint(netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) *tcpip.Error {
+ n.mu.Lock()
+ defer n.mu.Unlock()
+
+ eps, ok := n.packetEPs[netProto]
+ if !ok {
+ return tcpip.ErrNotSupported
+ }
+ n.packetEPs[netProto] = append(eps, ep)
+
+ return nil
+}
+
+func (n *NIC) unregisterPacketEndpoint(netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) {
+ n.mu.Lock()
+ defer n.mu.Unlock()
+
+ eps, ok := n.packetEPs[netProto]
+ if !ok {
+ return
+ }
+
+ for i, epOther := range eps {
+ if epOther == ep {
+ n.packetEPs[netProto] = append(eps[:i], eps[i+1:]...)
+ return
+ }
+ }
+}
+
type referencedNetworkEndpoint struct {
- ilist.Entry
ep NetworkEndpoint
nic *NIC
protocol tcpip.NetworkProtocolNumber
diff --git a/tcpip/stack/registration.go b/tcpip/stack/registration.go
index d406eca..5d9adc6 100644
--- a/tcpip/stack/registration.go
+++ b/tcpip/stack/registration.go
@@ -71,8 +71,8 @@
// RawTransportEndpoint is the interface that needs to be implemented by raw
// transport protocol endpoints. RawTransportEndpoints receive the entire
-// packet - including the link, network, and transport headers - as delivered
-// to netstack.
+// packet - including the network and transport headers - as delivered to
+// netstack.
type RawTransportEndpoint interface {
// HandlePacket is called by the stack when new packets arrive to
// this transport endpoint. The packet contains all data from the link
@@ -80,6 +80,22 @@
HandlePacket(r *Route, netHeader buffer.View, packet buffer.VectorisedView)
}
+// PacketEndpoint is the interface that needs to be implemented by packet
+// transport protocol endpoints. These endpoints receive link layer headers in
+// addition to whatever they contain (usually network and transport layer
+// headers and a payload).
+type PacketEndpoint interface {
+ // HandlePacket is called by the stack when new packets arrive that
+ // match the endpoint.
+ //
+ // Implementers should treat packet as immutable and should copy it
+ // before before modification.
+ //
+ // linkHeader may have a length of 0, in which case the PacketEndpoint
+ // should construct its own ethernet header for applications.
+ HandlePacket(nicid tcpip.NICID, addr tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, packet buffer.VectorisedView, linkHeader buffer.View)
+}
+
// TransportProtocol is the interface that needs to be implemented by transport
// protocols (e.g., tcp, udp) that want to be part of the networking stack.
type TransportProtocol interface {
@@ -185,6 +201,10 @@
// protocol.
WritePacket(r *Route, gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params NetworkHeaderParams, loop PacketLooping) *tcpip.Error
+ // WritePackets writes packets to the given destination address and
+ // protocol.
+ WritePackets(r *Route, gso *GSO, hdrs []PacketDescriptor, payload buffer.VectorisedView, params NetworkHeaderParams, loop PacketLooping) (int, *tcpip.Error)
+
// WriteHeaderIncludedPacket writes a packet that includes a network
// header to the given destination address.
WriteHeaderIncludedPacket(r *Route, payload buffer.VectorisedView, loop PacketLooping) *tcpip.Error
@@ -242,9 +262,10 @@
// packets to the appropriate network endpoint after it has been handled by
// the data link layer.
type NetworkDispatcher interface {
- // DeliverNetworkPacket finds the appropriate network protocol
- // endpoint and hands the packet over for further processing.
- DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView)
+ // DeliverNetworkPacket finds the appropriate network protocol endpoint
+ // and hands the packet over for further processing. linkHeader may have
+ // length 0 when the caller does not have ethernet data.
+ DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView, linkHeader buffer.View)
}
// LinkEndpointCapabilities is the type associated with the capabilities
@@ -266,7 +287,11 @@
CapabilitySaveRestore
CapabilityDisconnectOk
CapabilityLoopback
- CapabilityGSO
+ CapabilityHardwareGSO
+
+ // CapabilitySoftwareGSO indicates the link endpoint supports of sending
+ // multiple packets using a single call (LinkEndpoint.WritePackets).
+ CapabilitySoftwareGSO
)
// LinkEndpoint is the interface implemented by data link layer protocols (e.g.,
@@ -301,6 +326,18 @@
// r.LocalLinkAddress if it is provided.
WritePacket(r *Route, gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error
+ // WritePackets writes packets with the given protocol through the
+ // given route.
+ //
+ // Right now, WritePackets is used only when the software segmentation
+ // offload is enabled. If it will be used for something else, it may
+ // require to change syscall filters.
+ WritePackets(r *Route, gso *GSO, hdrs []PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error)
+
+ // WriteRawPacket writes a packet directly to the link. The packet
+ // should already have an ethernet header.
+ WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error
+
// Attach attaches the data link layer endpoint to the network-layer
// dispatcher of the stack.
Attach(dispatcher NetworkDispatcher)
@@ -324,13 +361,14 @@
type InjectableLinkEndpoint interface {
LinkEndpoint
- // Inject injects an inbound packet.
- Inject(protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView)
+ // InjectInbound injects an inbound packet.
+ InjectInbound(protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView)
- // WriteRawPacket writes a fully formed outbound packet directly to the link.
+ // InjectOutbound writes a fully formed outbound packet directly to the
+ // link.
//
// dest is used by endpoints with multiple raw destinations.
- WriteRawPacket(dest tcpip.Address, packet []byte) *tcpip.Error
+ InjectOutbound(dest tcpip.Address, packet []byte) *tcpip.Error
}
// A LinkAddressResolver is an extension to a NetworkProtocol that
@@ -379,11 +417,16 @@
RemoveWaker(nicid tcpip.NICID, addr tcpip.Address, waker *sleep.Waker)
}
-// UnassociatedEndpointFactory produces endpoints for writing packets not
-// associated with a particular transport protocol. Such endpoints can be used
-// to write arbitrary packets that include the IP header.
-type UnassociatedEndpointFactory interface {
- NewUnassociatedRawEndpoint(stack *Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error)
+// RawFactory produces endpoints for writing various types of raw packets.
+type RawFactory interface {
+ // NewUnassociatedEndpoint produces endpoints for writing packets not
+ // associated with a particular transport protocol. Such endpoints can
+ // be used to write arbitrary packets that include the network header.
+ NewUnassociatedEndpoint(stack *Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error)
+
+ // NewPacketEndpoint produces endpoints for reading and writing packets
+ // that include network and (when cooked is false) link layer headers.
+ NewPacketEndpoint(stack *Stack, cooked bool, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error)
}
// GSOType is the type of GSO segments.
@@ -394,8 +437,14 @@
// Types of gso segments.
const (
GSONone GSOType = iota
+
+ // Hardware GSO types:
GSOTCPv4
GSOTCPv6
+
+ // GSOSW is used for software GSO segments which have to be sent by
+ // endpoint.WritePackets.
+ GSOSW
)
// GSO contains generic segmentation offload properties.
@@ -423,3 +472,7 @@
// GSOMaxSize returns the maximum GSO packet size.
GSOMaxSize() uint32
}
+
+// SoftwareGSOMaxSize is a maximum allowed size of a software GSO segment.
+// This isn't a hard limit, because it is never set into packet headers.
+const SoftwareGSOMaxSize = (1 << 16)
diff --git a/tcpip/stack/route.go b/tcpip/stack/route.go
index 7a3dd68..d51bde9 100644
--- a/tcpip/stack/route.go
+++ b/tcpip/stack/route.go
@@ -47,8 +47,8 @@
// starts.
ref *referencedNetworkEndpoint
- // loop controls where WritePacket should send packets.
- loop PacketLooping
+ // Loop controls where WritePacket should send packets.
+ Loop PacketLooping
}
// makeRoute initializes a new route. It takes ownership of the provided
@@ -69,7 +69,7 @@
LocalLinkAddress: localLinkAddr,
RemoteAddress: remoteAddr,
ref: ref,
- loop: loop,
+ Loop: loop,
}
}
@@ -159,7 +159,7 @@
return tcpip.ErrInvalidEndpointState
}
- err := r.ref.ep.WritePacket(r, gso, hdr, payload, params, r.loop)
+ err := r.ref.ep.WritePacket(r, gso, hdr, payload, params, r.Loop)
if err != nil {
r.Stats().IP.OutgoingPacketErrors.Increment()
} else {
@@ -169,6 +169,44 @@
return err
}
+// PacketDescriptor is a packet descriptor which contains a packet header and
+// offset and size of packet data in a payload view.
+type PacketDescriptor struct {
+ Hdr buffer.Prependable
+ Off int
+ Size int
+}
+
+// NewPacketDescriptors allocates a set of packet descriptors.
+func NewPacketDescriptors(n int, hdrSize int) []PacketDescriptor {
+ buf := make([]byte, n*hdrSize)
+ hdrs := make([]PacketDescriptor, n)
+ for i := range hdrs {
+ hdrs[i].Hdr = buffer.NewEmptyPrependableFromView(buf[i*hdrSize:][:hdrSize])
+ }
+ return hdrs
+}
+
+// WritePackets writes the set of packets through the given route.
+func (r *Route) WritePackets(gso *GSO, hdrs []PacketDescriptor, payload buffer.VectorisedView, params NetworkHeaderParams) (int, *tcpip.Error) {
+ if !r.ref.isValidForOutgoing() {
+ return 0, tcpip.ErrInvalidEndpointState
+ }
+
+ n, err := r.ref.ep.WritePackets(r, gso, hdrs, payload, params, r.Loop)
+ if err != nil {
+ r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(len(hdrs) - n))
+ }
+ r.ref.nic.stats.Tx.Packets.IncrementBy(uint64(n))
+ payloadSize := 0
+ for i := 0; i < n; i++ {
+ r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(hdrs[i].Hdr.UsedLength()))
+ payloadSize += hdrs[i].Size
+ }
+ r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(payloadSize))
+ return n, err
+}
+
// WriteHeaderIncludedPacket writes a packet already containing a network
// header through the given route.
func (r *Route) WriteHeaderIncludedPacket(payload buffer.VectorisedView) *tcpip.Error {
@@ -176,7 +214,7 @@
return tcpip.ErrInvalidEndpointState
}
- if err := r.ref.ep.WriteHeaderIncludedPacket(r, payload, r.loop); err != nil {
+ if err := r.ref.ep.WriteHeaderIncludedPacket(r, payload, r.Loop); err != nil {
r.Stats().IP.OutgoingPacketErrors.Increment()
return err
}
diff --git a/tcpip/stack/stack.go b/tcpip/stack/stack.go
index d6a0471..54a0a6c 100644
--- a/tcpip/stack/stack.go
+++ b/tcpip/stack/stack.go
@@ -351,10 +351,9 @@
networkProtocols map[tcpip.NetworkProtocolNumber]NetworkProtocol
linkAddrResolvers map[tcpip.NetworkProtocolNumber]LinkAddressResolver
- // unassociatedFactory creates unassociated endpoints. If nil, raw
- // endpoints are disabled. It is set during Stack creation and is
- // immutable.
- unassociatedFactory UnassociatedEndpointFactory
+ // rawFactory creates raw endpoints. If nil, raw endpoints are
+ // disabled. It is set during Stack creation and is immutable.
+ rawFactory RawFactory
demux *transportDemuxer
@@ -400,8 +399,17 @@
// 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
}
// Options contains optional Stack configuration.
@@ -425,16 +433,32 @@
// stack (false).
HandleLocal bool
- // UnassociatedFactory produces unassociated endpoints raw endpoints.
- // Raw endpoints are enabled only if this is non-nil.
- UnassociatedFactory UnassociatedEndpointFactory
-
- // NDPConfigs is the NDP configurations used by interfaces.
+ // NDPConfigs is the default NDP configurations used by interfaces.
//
// By default, NDPConfigs will have a zero value for its
// DupAddrDetectTransmits field, implying that DAD will not be performed
// before assigning an address to a NIC.
NDPConfigs NDPConfigurations
+
+ // AutoGenIPv6LinkLocal determins whether or not the stack will attempt
+ // to auto-generate an IPv6 link-local address for newly enabled NICs.
+ // Note, setting this to true does not mean that a link-local address
+ // will be assigned right away, or at all. If Duplicate Address
+ // Detection is enabled, an address will only be assigned if it
+ // successfully resolves. If it fails, no further attempt will be made
+ // to auto-generate an IPv6 link-local address.
+ //
+ // The generated link-local address will follow RFC 4291 Appendix A
+ // guidelines.
+ AutoGenIPv6LinkLocal bool
+
+ // NDPDisp is the NDP event dispatcher that an integrator can provide to
+ // receive NDP related events.
+ NDPDisp NDPDispatcher
+
+ // RawFactory produces raw endpoints. Raw endpoints are enabled only if
+ // this is non-nil.
+ RawFactory RawFactory
}
// TransportEndpointInfo holds useful information about a transport endpoint
@@ -485,18 +509,20 @@
opts.NDPConfigs.validate()
s := &Stack{
- transportProtocols: make(map[tcpip.TransportProtocolNumber]*transportProtocolState),
- networkProtocols: make(map[tcpip.NetworkProtocolNumber]NetworkProtocol),
- linkAddrResolvers: make(map[tcpip.NetworkProtocolNumber]LinkAddressResolver),
- nics: make(map[tcpip.NICID]*NIC),
- linkAddrCache: newLinkAddrCache(ageLimit, resolutionTimeout, resolutionAttempts),
- PortManager: ports.NewPortManager(),
- clock: clock,
- stats: opts.Stats.FillIn(),
- handleLocal: opts.HandleLocal,
- icmpRateLimiter: NewICMPRateLimiter(),
- portSeed: generateRandUint32(),
- ndpConfigs: opts.NDPConfigs,
+ transportProtocols: make(map[tcpip.TransportProtocolNumber]*transportProtocolState),
+ networkProtocols: make(map[tcpip.NetworkProtocolNumber]NetworkProtocol),
+ linkAddrResolvers: make(map[tcpip.NetworkProtocolNumber]LinkAddressResolver),
+ nics: make(map[tcpip.NICID]*NIC),
+ linkAddrCache: newLinkAddrCache(ageLimit, resolutionTimeout, resolutionAttempts),
+ PortManager: ports.NewPortManager(),
+ clock: clock,
+ stats: opts.Stats.FillIn(),
+ handleLocal: opts.HandleLocal,
+ icmpRateLimiter: NewICMPRateLimiter(),
+ portSeed: generateRandUint32(),
+ ndpConfigs: opts.NDPConfigs,
+ autoGenIPv6LinkLocal: opts.AutoGenIPv6LinkLocal,
+ ndpDisp: opts.NDPDisp,
}
// Add specified network protocols.
@@ -514,8 +540,8 @@
}
}
- // Add the factory for unassociated endpoints, if present.
- s.unassociatedFactory = opts.UnassociatedFactory
+ // Add the factory for raw endpoints, if present.
+ s.rawFactory = opts.RawFactory
// Create the global transport demuxer.
s.demux = newTransportDemuxer(s)
@@ -650,12 +676,12 @@
// protocol. Raw endpoints receive all traffic for a given protocol regardless
// of address.
func (s *Stack) NewRawEndpoint(transport tcpip.TransportProtocolNumber, network tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue, associated bool) (tcpip.Endpoint, *tcpip.Error) {
- if s.unassociatedFactory == nil {
+ if s.rawFactory == nil {
return nil, tcpip.ErrNotPermitted
}
if !associated {
- return s.unassociatedFactory.NewUnassociatedRawEndpoint(s, network, transport, waiterQueue)
+ return s.rawFactory.NewUnassociatedEndpoint(s, network, transport, waiterQueue)
}
t, ok := s.transportProtocols[transport]
@@ -666,6 +692,16 @@
return t.proto.NewRawEndpoint(s, network, waiterQueue)
}
+// NewPacketEndpoint creates a new packet endpoint listening for the given
+// netProto.
+func (s *Stack) NewPacketEndpoint(cooked bool, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) {
+ if s.rawFactory == nil {
+ return nil, tcpip.ErrNotPermitted
+ }
+
+ return s.rawFactory.NewPacketEndpoint(s, cooked, netProto, waiterQueue)
+}
+
// createNIC creates a NIC with the provided id and link-layer endpoint, and
// optionally enable it.
func (s *Stack) createNIC(id tcpip.NICID, name string, ep LinkEndpoint, enabled, loopback bool) *tcpip.Error {
@@ -1135,6 +1171,109 @@
}
}
+// RegisterPacketEndpoint registers ep with the stack, causing it to receive
+// all traffic of the specified netProto on the given NIC. If nicID is 0, it
+// receives traffic from every NIC.
+func (s *Stack) RegisterPacketEndpoint(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) *tcpip.Error {
+ s.mu.Lock()
+ defer s.mu.Unlock()
+
+ // If no NIC is specified, capture on all devices.
+ if nicID == 0 {
+ // Register with each NIC.
+ for _, nic := range s.nics {
+ if err := nic.registerPacketEndpoint(netProto, ep); err != nil {
+ s.unregisterPacketEndpointLocked(0, netProto, ep)
+ return err
+ }
+ }
+ return nil
+ }
+
+ // Capture on a specific device.
+ nic, ok := s.nics[nicID]
+ if !ok {
+ return tcpip.ErrUnknownNICID
+ }
+ if err := nic.registerPacketEndpoint(netProto, ep); err != nil {
+ return err
+ }
+
+ return nil
+}
+
+// UnregisterPacketEndpoint unregisters ep for packets of the specified
+// netProto from the specified NIC. If nicID is 0, ep is unregistered from all
+// NICs.
+func (s *Stack) UnregisterPacketEndpoint(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) {
+ s.mu.Lock()
+ defer s.mu.Unlock()
+ s.unregisterPacketEndpointLocked(nicID, netProto, ep)
+}
+
+func (s *Stack) unregisterPacketEndpointLocked(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) {
+ // If no NIC is specified, unregister on all devices.
+ if nicID == 0 {
+ // Unregister with each NIC.
+ for _, nic := range s.nics {
+ nic.unregisterPacketEndpoint(netProto, ep)
+ }
+ return
+ }
+
+ // Unregister in a single device.
+ nic, ok := s.nics[nicID]
+ if !ok {
+ return
+ }
+ nic.unregisterPacketEndpoint(netProto, ep)
+}
+
+// WritePacket writes data directly to the specified NIC. It adds an ethernet
+// header based on the arguments.
+func (s *Stack) WritePacket(nicid tcpip.NICID, dst tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, payload buffer.VectorisedView) *tcpip.Error {
+ s.mu.Lock()
+ nic, ok := s.nics[nicid]
+ s.mu.Unlock()
+ if !ok {
+ return tcpip.ErrUnknownDevice
+ }
+
+ // Add our own fake ethernet header.
+ ethFields := header.EthernetFields{
+ SrcAddr: nic.linkEP.LinkAddress(),
+ DstAddr: dst,
+ Type: netProto,
+ }
+ fakeHeader := make(header.Ethernet, header.EthernetMinimumSize)
+ fakeHeader.Encode(ðFields)
+ ethHeader := buffer.View(fakeHeader).ToVectorisedView()
+ ethHeader.Append(payload)
+
+ if err := nic.linkEP.WriteRawPacket(ethHeader); err != nil {
+ return err
+ }
+
+ return nil
+}
+
+// WriteRawPacket writes data directly to the specified NIC without adding any
+// headers.
+func (s *Stack) WriteRawPacket(nicid tcpip.NICID, payload buffer.VectorisedView) *tcpip.Error {
+ s.mu.Lock()
+ nic, ok := s.nics[nicid]
+ s.mu.Unlock()
+ if !ok {
+ return tcpip.ErrUnknownDevice
+ }
+
+ if err := nic.linkEP.WriteRawPacket(payload); err != nil {
+ return err
+ }
+
+ return nil
+}
+
// NetworkProtocolInstance returns the protocol instance in the stack for the
// specified network protocol. This method is public for protocol implementers
// and tests to use.
@@ -1286,6 +1425,25 @@
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/tcpip/stack/stack_test.go b/tcpip/stack/stack_test.go
index d9c307e..f7b3565 100644
--- a/tcpip/stack/stack_test.go
+++ b/tcpip/stack/stack_test.go
@@ -24,11 +24,14 @@
"sort"
"strings"
"testing"
+ "time"
+ "github.com/google/go-cmp/cmp"
"github.com/google/netstack/tcpip"
"github.com/google/netstack/tcpip/buffer"
"github.com/google/netstack/tcpip/header"
"github.com/google/netstack/tcpip/link/channel"
+ "github.com/google/netstack/tcpip/network/ipv6"
"github.com/google/netstack/tcpip/stack"
)
@@ -144,6 +147,11 @@
return f.ep.WritePacket(r, gso, hdr, payload, fakeNetNumber)
}
+// WritePackets implements stack.LinkEndpoint.WritePackets.
+func (f *fakeNetworkEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) {
+ panic("not implemented")
+}
+
func (*fakeNetworkEndpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error {
return tcpip.ErrNotSupported
}
@@ -1864,3 +1872,297 @@
t.Errorf("got Tx.Bytes.Value() = %d, want = %d", got, want)
}
}
+
+// TestNICAutoGenAddr tests the auto-generation of IPv6 link-local addresses
+// (or lack there-of if disabled (default)). Note, DAD will be disabled in
+// these tests.
+func TestNICAutoGenAddr(t *testing.T) {
+ tests := []struct {
+ name string
+ autoGen bool
+ linkAddr tcpip.LinkAddress
+ shouldGen bool
+ }{
+ {
+ "Disabled",
+ false,
+ linkAddr1,
+ false,
+ },
+ {
+ "Enabled",
+ true,
+ linkAddr1,
+ true,
+ },
+ {
+ "Nil MAC",
+ true,
+ tcpip.LinkAddress([]byte(nil)),
+ false,
+ },
+ {
+ "Empty MAC",
+ true,
+ tcpip.LinkAddress(""),
+ false,
+ },
+ {
+ "Invalid MAC",
+ true,
+ tcpip.LinkAddress("\x01\x02\x03"),
+ false,
+ },
+ {
+ "Multicast MAC",
+ true,
+ tcpip.LinkAddress("\x01\x02\x03\x04\x05\x06"),
+ false,
+ },
+ {
+ "Unspecified MAC",
+ true,
+ tcpip.LinkAddress("\x00\x00\x00\x00\x00\x00"),
+ false,
+ },
+ }
+
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ opts := stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ }
+
+ if test.autoGen {
+ // Only set opts.AutoGenIPv6LinkLocal when
+ // test.autoGen is true because
+ // opts.AutoGenIPv6LinkLocal should be false by
+ // default.
+ opts.AutoGenIPv6LinkLocal = true
+ }
+
+ e := channel.New(10, 1280, test.linkAddr)
+ s := stack.New(opts)
+ if err := s.CreateNIC(1, e); err != nil {
+ t.Fatalf("CreateNIC(_) = %s", err)
+ }
+
+ addr, err := s.GetMainNICAddress(1, header.IPv6ProtocolNumber)
+ if err != nil {
+ t.Fatalf("stack.GetMainNICAddress(_, _) err = %s", err)
+ }
+
+ if test.shouldGen {
+ // Should have auto-generated an address and
+ // resolved immediately (DAD is disabled).
+ if want := (tcpip.AddressWithPrefix{Address: header.LinkLocalAddr(test.linkAddr), PrefixLen: header.IPv6LinkLocalPrefix.PrefixLen}); addr != want {
+ t.Fatalf("got stack.GetMainNICAddress(_, _) = %s, want = %s", addr, want)
+ }
+ } else {
+ // Should not have auto-generated an address.
+ if want := (tcpip.AddressWithPrefix{}); addr != want {
+ t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want)
+ }
+ }
+ })
+ }
+}
+
+// TestNICAutoGenAddrDoesDAD tests that the successful auto-generation of IPv6
+// link-local addresses will only be assigned after the DAD process resolves.
+func TestNICAutoGenAddrDoesDAD(t *testing.T) {
+ ndpDisp := ndpDispatcher{
+ dadC: make(chan ndpDADEvent),
+ }
+ ndpConfigs := stack.DefaultNDPConfigurations()
+ opts := stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: ndpConfigs,
+ AutoGenIPv6LinkLocal: true,
+ NDPDisp: &ndpDisp,
+ }
+
+ e := channel.New(10, 1280, linkAddr1)
+ s := stack.New(opts)
+ if err := s.CreateNIC(1, e); err != nil {
+ t.Fatalf("CreateNIC(_) = %s", err)
+ }
+
+ // Address should not be considered bound to the
+ // NIC yet (DAD ongoing).
+ addr, err := s.GetMainNICAddress(1, header.IPv6ProtocolNumber)
+ if err != nil {
+ t.Fatalf("got stack.GetMainNICAddress(_, _) = (_, %v), want = (_, nil)", err)
+ }
+ if want := (tcpip.AddressWithPrefix{}); addr != want {
+ t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want)
+ }
+
+ linkLocalAddr := header.LinkLocalAddr(linkAddr1)
+
+ // Wait for DAD to resolve.
+ select {
+ case <-time.After(time.Duration(ndpConfigs.DupAddrDetectTransmits)*ndpConfigs.RetransmitTimer + time.Second):
+ // We should get a resolution event after 1s (default time to
+ // resolve as per default NDP configurations). Waiting for that
+ // resolution time + an extra 1s without a resolution event
+ // means something is wrong.
+ t.Fatal("timed out waiting for DAD resolution")
+ case e := <-ndpDisp.dadC:
+ if e.err != nil {
+ t.Fatal("got DAD error: ", e.err)
+ }
+ if e.nicid != 1 {
+ t.Fatalf("got DAD event w/ nicid = %d, want = 1", e.nicid)
+ }
+ if e.addr != linkLocalAddr {
+ t.Fatalf("got DAD event w/ addr = %s, want = %s", addr, linkLocalAddr)
+ }
+ if !e.resolved {
+ t.Fatal("got DAD event w/ resolved = false, want = true")
+ }
+ }
+ addr, err = s.GetMainNICAddress(1, header.IPv6ProtocolNumber)
+ if err != nil {
+ t.Fatalf("stack.GetMainNICAddress(_, _) err = %s", err)
+ }
+ if want := (tcpip.AddressWithPrefix{Address: linkLocalAddr, PrefixLen: header.IPv6LinkLocalPrefix.PrefixLen}); addr != want {
+ t.Fatalf("got stack.GetMainNICAddress(_, _) = %s, want = %s", addr, want)
+ }
+}
+
+// TestNewPEB tests that a new PrimaryEndpointBehavior value (peb) is respected
+// when an address's kind gets "promoted" to permanent from permanentExpired.
+func TestNewPEBOnPromotionToPermanent(t *testing.T) {
+ pebs := []stack.PrimaryEndpointBehavior{
+ stack.NeverPrimaryEndpoint,
+ stack.CanBePrimaryEndpoint,
+ stack.FirstPrimaryEndpoint,
+ }
+
+ for _, pi := range pebs {
+ for _, ps := range pebs {
+ t.Run(fmt.Sprintf("%d-to-%d", pi, ps), func(t *testing.T) {
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()},
+ })
+ ep1 := channel.New(10, defaultMTU, "")
+ if err := s.CreateNIC(1, ep1); err != nil {
+ t.Fatal("CreateNIC failed:", err)
+ }
+
+ // Add a permanent address with initial
+ // PrimaryEndpointBehavior (peb), pi. If pi is
+ // NeverPrimaryEndpoint, the address should not
+ // be returned by a call to GetMainNICAddress;
+ // else, it should.
+ if err := s.AddAddressWithOptions(1, fakeNetNumber, "\x01", pi); err != nil {
+ t.Fatal("AddAddressWithOptions failed:", err)
+ }
+ addr, err := s.GetMainNICAddress(1, fakeNetNumber)
+ if err != nil {
+ t.Fatal("s.GetMainNICAddress failed:", err)
+ }
+ if pi == stack.NeverPrimaryEndpoint {
+ if want := (tcpip.AddressWithPrefix{}); addr != want {
+ t.Fatalf("got GetMainNICAddress = %s, want = %s", addr, want)
+
+ }
+ } else if addr.Address != "\x01" {
+ t.Fatalf("got GetMainNICAddress = %s, want = 1", addr.Address)
+ }
+
+ {
+ subnet, err := tcpip.NewSubnet("\x00", "\x00")
+ if err != nil {
+ t.Fatalf("NewSubnet failed:", err)
+ }
+ s.SetRouteTable([]tcpip.Route{{Destination: subnet, Gateway: "\x00", NIC: 1}})
+ }
+
+ // Take a route through the address so its ref
+ // count gets incremented and does not actually
+ // get deleted when RemoveAddress is called
+ // below. This is because we want to test that a
+ // new peb is respected when an address gets
+ // "promoted" to permanent from a
+ // permanentExpired kind.
+ r, err := s.FindRoute(1, "\x01", "\x02", fakeNetNumber, false)
+ if err != nil {
+ t.Fatal("FindRoute failed:", err)
+ }
+ defer r.Release()
+ if err := s.RemoveAddress(1, "\x01"); err != nil {
+ t.Fatalf("RemoveAddress failed:", err)
+ }
+
+ //
+ // At this point, the address should still be
+ // known by the NIC, but have its
+ // kind = permanentExpired.
+ //
+
+ // Add some other address with peb set to
+ // FirstPrimaryEndpoint.
+ if err := s.AddAddressWithOptions(1, fakeNetNumber, "\x03", stack.FirstPrimaryEndpoint); err != nil {
+ t.Fatal("AddAddressWithOptions failed:", err)
+
+ }
+
+ // Add back the address we removed earlier and
+ // make sure the new peb was respected.
+ // (The address should just be promoted now).
+ if err := s.AddAddressWithOptions(1, fakeNetNumber, "\x01", ps); err != nil {
+ t.Fatal("AddAddressWithOptions failed:", err)
+ }
+ var primaryAddrs []tcpip.Address
+ for _, pa := range s.NICInfo()[1].ProtocolAddresses {
+ primaryAddrs = append(primaryAddrs, pa.AddressWithPrefix.Address)
+ }
+ var expectedList []tcpip.Address
+ switch ps {
+ case stack.FirstPrimaryEndpoint:
+ expectedList = []tcpip.Address{
+ "\x01",
+ "\x03",
+ }
+ case stack.CanBePrimaryEndpoint:
+ expectedList = []tcpip.Address{
+ "\x03",
+ "\x01",
+ }
+ case stack.NeverPrimaryEndpoint:
+ expectedList = []tcpip.Address{
+ "\x03",
+ }
+ }
+ if !cmp.Equal(primaryAddrs, expectedList) {
+ t.Fatalf("got NIC's primary addresses = %v, want = %v", primaryAddrs, expectedList)
+ }
+
+ // Once we remove the other address, if the new
+ // peb, ps, was NeverPrimaryEndpoint, no address
+ // should be returned by a call to
+ // GetMainNICAddress; else, our original address
+ // should be returned.
+ if err := s.RemoveAddress(1, "\x03"); err != nil {
+ t.Fatalf("RemoveAddress failed:", err)
+ }
+ addr, err = s.GetMainNICAddress(1, fakeNetNumber)
+ if err != nil {
+ t.Fatal("s.GetMainNICAddress failed:", err)
+ }
+ if ps == stack.NeverPrimaryEndpoint {
+ if want := (tcpip.AddressWithPrefix{}); addr != want {
+ t.Fatalf("got GetMainNICAddress = %s, want = %s", addr, want)
+
+ }
+ } else {
+ if addr.Address != "\x01" {
+ t.Fatalf("got GetMainNICAddress = %s, want = 1", addr.Address)
+ }
+ }
+ })
+ }
+ }
+}
diff --git a/tcpip/stack/transport_demuxer.go b/tcpip/stack/transport_demuxer.go
index e10f874..cc7a2e0 100644
--- a/tcpip/stack/transport_demuxer.go
+++ b/tcpip/stack/transport_demuxer.go
@@ -465,7 +465,7 @@
func (d *transportDemuxer) registerRawEndpoint(netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, ep RawTransportEndpoint) *tcpip.Error {
eps, ok := d.protocol[protocolIDs{netProto, transProto}]
if !ok {
- return nil
+ return tcpip.ErrNotSupported
}
eps.mu.Lock()
diff --git a/tcpip/tcpip.go b/tcpip/tcpip.go
index 7d47aa8..3a8e56e 100644
--- a/tcpip/tcpip.go
+++ b/tcpip/tcpip.go
@@ -255,7 +255,7 @@
// This may not be used by all endpoint types.
NIC NICID
- // Addr is the network address.
+ // Addr is the network or link layer address.
Addr Address
// Port is the transport port.
@@ -673,6 +673,11 @@
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 @@
// 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/tcpip/transport/packet/endpoint.go b/tcpip/transport/packet/endpoint.go
new file mode 100644
index 0000000..8d3bf3f
--- /dev/null
+++ b/tcpip/transport/packet/endpoint.go
@@ -0,0 +1,363 @@
+// Copyright 2019 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Package packet provides the implementation of packet sockets (see
+// packet(7)). Packet sockets allow applications to:
+//
+// * manually write and inspect link, network, and transport headers
+// * receive all traffic of a given network protocol, or all protocols
+//
+// Packet sockets are similar to raw sockets, but provide even more power to
+// users, letting them effectively talk directly to the network device.
+//
+// Packet sockets skip the input and output iptables chains.
+package packet
+
+import (
+ "sync"
+
+ "github.com/google/netstack/tcpip"
+ "github.com/google/netstack/tcpip/buffer"
+ "github.com/google/netstack/tcpip/header"
+ "github.com/google/netstack/tcpip/iptables"
+ "github.com/google/netstack/tcpip/stack"
+ "github.com/google/netstack/waiter"
+)
+
+// +stateify savable
+type packet struct {
+ packetEntry
+ // data holds the actual packet data, including any headers and
+ // payload.
+ data 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
+ // timestampNS is the unix time at which the packet was received.
+ timestampNS int64
+ // senderAddr is the network address of the sender.
+ senderAddr tcpip.FullAddress
+}
+
+// endpoint is the packet socket implementation of tcpip.Endpoint. It is legal
+// to have goroutines make concurrent calls into the endpoint.
+//
+// Lock order:
+// endpoint.mu
+// endpoint.rcvMu
+//
+// +stateify savable
+type endpoint struct {
+ stack.TransportEndpointInfo
+ // The following fields are initialized at creation time and are
+ // immutable.
+ stack *stack.Stack
+ netProto tcpip.NetworkProtocolNumber
+ waiterQueue *waiter.Queue
+ cooked bool
+
+ // The following fields are used to manage the receive queue and are
+ // protected by rcvMu.
+ rcvMu sync.Mutex
+ rcvList packetList
+ rcvBufSizeMax int
+ rcvBufSize int
+ rcvClosed bool
+
+ // The following fields are protected by mu.
+ mu sync.RWMutex
+ sndBufSize int
+ closed bool
+ stats tcpip.TransportEndpointStats
+}
+
+// NewEndpoint returns a new packet endpoint.
+func NewEndpoint(s *stack.Stack, cooked bool, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) {
+ ep := &endpoint{
+ stack: s,
+ TransportEndpointInfo: stack.TransportEndpointInfo{
+ NetProto: netProto,
+ },
+ cooked: cooked,
+ netProto: netProto,
+ waiterQueue: waiterQueue,
+ rcvBufSizeMax: 32 * 1024,
+ sndBufSize: 32 * 1024,
+ }
+
+ if err := s.RegisterPacketEndpoint(0, netProto, ep); err != nil {
+ return nil, err
+ }
+ return ep, nil
+}
+
+// Close implements tcpip.Endpoint.Close.
+func (ep *endpoint) Close() {
+ ep.mu.Lock()
+ defer ep.mu.Unlock()
+
+ if ep.closed {
+ return
+ }
+
+ ep.stack.UnregisterPacketEndpoint(0, ep.netProto, ep)
+
+ ep.rcvMu.Lock()
+ defer ep.rcvMu.Unlock()
+
+ // Clear the receive list.
+ ep.rcvClosed = true
+ ep.rcvBufSize = 0
+ for !ep.rcvList.Empty() {
+ ep.rcvList.Remove(ep.rcvList.Front())
+ }
+
+ ep.closed = true
+ ep.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut)
+}
+
+// ModerateRecvBuf implements tcpip.Endpoint.ModerateRecvBuf.
+func (ep *endpoint) ModerateRecvBuf(copied int) {}
+
+// IPTables implements tcpip.Endpoint.IPTables.
+func (ep *endpoint) IPTables() (iptables.IPTables, error) {
+ return ep.stack.IPTables(), nil
+}
+
+// Read implements tcpip.Endpoint.Read.
+func (ep *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) {
+ ep.rcvMu.Lock()
+
+ // If there's no data to read, return that read would block or that the
+ // endpoint is closed.
+ if ep.rcvList.Empty() {
+ err := tcpip.ErrWouldBlock
+ if ep.rcvClosed {
+ ep.stats.ReadErrors.ReadClosed.Increment()
+ err = tcpip.ErrClosedForReceive
+ }
+ ep.rcvMu.Unlock()
+ return buffer.View{}, tcpip.ControlMessages{}, err
+ }
+
+ packet := ep.rcvList.Front()
+ ep.rcvList.Remove(packet)
+ ep.rcvBufSize -= packet.data.Size()
+
+ ep.rcvMu.Unlock()
+
+ if addr != nil {
+ *addr = packet.senderAddr
+ }
+
+ return packet.data.ToView(), tcpip.ControlMessages{HasTimestamp: true, Timestamp: packet.timestampNS}, nil
+}
+
+func (ep *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) {
+ // TODO(b/129292371): Implement.
+ return 0, nil, tcpip.ErrInvalidOptionValue
+}
+
+// Peek implements tcpip.Endpoint.Peek.
+func (ep *endpoint) Peek([][]byte) (int64, tcpip.ControlMessages, *tcpip.Error) {
+ return 0, tcpip.ControlMessages{}, nil
+}
+
+// Disconnect implements tcpip.Endpoint.Disconnect. Packet sockets cannot be
+// disconnected, and this function always returns tpcip.ErrNotSupported.
+func (*endpoint) Disconnect() *tcpip.Error {
+ return tcpip.ErrNotSupported
+}
+
+// Connect implements tcpip.Endpoint.Connect. Packet sockets cannot be
+// connected, and this function always returnes tcpip.ErrNotSupported.
+func (ep *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error {
+ return tcpip.ErrNotSupported
+}
+
+// Shutdown implements tcpip.Endpoint.Shutdown. Packet sockets cannot be used
+// with Shutdown, and this function always returns tcpip.ErrNotSupported.
+func (ep *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error {
+ return tcpip.ErrNotSupported
+}
+
+// Listen implements tcpip.Endpoint.Listen. Packet sockets cannot be used with
+// Listen, and this function always returns tcpip.ErrNotSupported.
+func (ep *endpoint) Listen(backlog int) *tcpip.Error {
+ return tcpip.ErrNotSupported
+}
+
+// Accept implements tcpip.Endpoint.Accept. Packet sockets cannot be used with
+// Accept, and this function always returns tcpip.ErrNotSupported.
+func (ep *endpoint) Accept() (tcpip.Endpoint, *waiter.Queue, *tcpip.Error) {
+ return nil, nil, tcpip.ErrNotSupported
+}
+
+// Bind implements tcpip.Endpoint.Bind.
+func (ep *endpoint) Bind(addr tcpip.FullAddress) *tcpip.Error {
+ // TODO(gvisor.dev/issue/173): Add Bind support.
+
+ // "By default, all packets of the specified protocol type are passed
+ // to a packet socket. To get packets only from a specific interface
+ // use bind(2) specifying an address in a struct sockaddr_ll to bind
+ // the packet socket to an interface. Fields used for binding are
+ // sll_family (should be AF_PACKET), sll_protocol, and sll_ifindex."
+ // - packet(7).
+
+ return tcpip.ErrNotSupported
+}
+
+// GetLocalAddress implements tcpip.Endpoint.GetLocalAddress.
+func (ep *endpoint) GetLocalAddress() (tcpip.FullAddress, *tcpip.Error) {
+ return tcpip.FullAddress{}, tcpip.ErrNotSupported
+}
+
+// GetRemoteAddress implements tcpip.Endpoint.GetRemoteAddress.
+func (ep *endpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Error) {
+ // Even a connected socket doesn't return a remote address.
+ return tcpip.FullAddress{}, tcpip.ErrNotConnected
+}
+
+// Readiness implements tcpip.Endpoint.Readiness.
+func (ep *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask {
+ // The endpoint is always writable.
+ result := waiter.EventOut & mask
+
+ // Determine whether the endpoint is readable.
+ if (mask & waiter.EventIn) != 0 {
+ ep.rcvMu.Lock()
+ if !ep.rcvList.Empty() || ep.rcvClosed {
+ result |= waiter.EventIn
+ }
+ ep.rcvMu.Unlock()
+ }
+
+ return result
+}
+
+// SetSockOpt implements tcpip.Endpoint.SetSockOpt. Packet sockets cannot be
+// used with SetSockOpt, and this function always returns
+// tcpip.ErrNotSupported.
+func (ep *endpoint) SetSockOpt(opt interface{}) *tcpip.Error {
+ return tcpip.ErrNotSupported
+}
+
+// SetSockOptInt implements tcpip.Endpoint.SetSockOptInt.
+func (ep *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error {
+ return tcpip.ErrUnknownProtocolOption
+}
+
+// GetSockOptInt implements tcpip.Endpoint.GetSockOptInt.
+func (ep *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) {
+ return 0, tcpip.ErrNotSupported
+}
+
+// GetSockOpt implements tcpip.Endpoint.GetSockOpt.
+func (ep *endpoint) GetSockOpt(opt interface{}) *tcpip.Error {
+ return tcpip.ErrNotSupported
+}
+
+// HandlePacket implements stack.PacketEndpoint.HandlePacket.
+func (ep *endpoint) HandlePacket(nicid tcpip.NICID, localAddr tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, vv buffer.VectorisedView, ethHeader buffer.View) {
+ ep.rcvMu.Lock()
+
+ // Drop the packet if our buffer is currently full.
+ if ep.rcvClosed {
+ ep.rcvMu.Unlock()
+ ep.stack.Stats().DroppedPackets.Increment()
+ ep.stats.ReceiveErrors.ClosedReceiver.Increment()
+ return
+ }
+
+ if ep.rcvBufSize >= ep.rcvBufSizeMax {
+ ep.rcvMu.Unlock()
+ ep.stack.Stats().DroppedPackets.Increment()
+ ep.stats.ReceiveErrors.ReceiveBufferOverflow.Increment()
+ return
+ }
+
+ wasEmpty := ep.rcvBufSize == 0
+
+ // Push new packet into receive list and increment the buffer size.
+ var packet packet
+ // TODO(b/129292371): Return network protocol.
+ if len(ethHeader) > 0 {
+ // Get info directly from the ethernet header.
+ hdr := header.Ethernet(ethHeader)
+ packet.senderAddr = tcpip.FullAddress{
+ NIC: nicid,
+ Addr: tcpip.Address(hdr.SourceAddress()),
+ }
+ } else {
+ // Guess the would-be ethernet header.
+ packet.senderAddr = tcpip.FullAddress{
+ NIC: nicid,
+ Addr: tcpip.Address(localAddr),
+ }
+ }
+
+ if ep.cooked {
+ // Cooked packets can simply be queued.
+ packet.data = vv.Clone(packet.views[:])
+ } else {
+ // Raw packets need their ethernet headers prepended before
+ // queueing.
+ if len(ethHeader) == 0 {
+ // We weren't provided with an actual ethernet header,
+ // so fake one.
+ ethFields := header.EthernetFields{
+ SrcAddr: tcpip.LinkAddress([]byte{0x00, 0x00, 0x00, 0x00, 0x00, 0x00}),
+ DstAddr: localAddr,
+ Type: netProto,
+ }
+ fakeHeader := make(header.Ethernet, header.EthernetMinimumSize)
+ fakeHeader.Encode(ðFields)
+ ethHeader = buffer.View(fakeHeader)
+ }
+ combinedVV := buffer.View(ethHeader).ToVectorisedView()
+ combinedVV.Append(vv)
+ packet.data = combinedVV.Clone(packet.views[:])
+ }
+ packet.timestampNS = ep.stack.NowNanoseconds()
+
+ ep.rcvList.PushBack(&packet)
+ ep.rcvBufSize += packet.data.Size()
+
+ ep.rcvMu.Unlock()
+ ep.stats.PacketsReceived.Increment()
+ // Notify waiters that there's data to be read.
+ if wasEmpty {
+ ep.waiterQueue.Notify(waiter.EventIn)
+ }
+}
+
+// State implements socket.Socket.State.
+func (ep *endpoint) State() uint32 {
+ return 0
+}
+
+// Info returns a copy of the endpoint info.
+func (ep *endpoint) Info() tcpip.EndpointInfo {
+ ep.mu.RLock()
+ // Make a copy of the endpoint info.
+ ret := ep.TransportEndpointInfo
+ ep.mu.RUnlock()
+ return &ret
+}
+
+// Stats returns a pointer to the endpoint stats.
+func (ep *endpoint) Stats() tcpip.EndpointStats {
+ return &ep.stats
+}
diff --git a/tcpip/transport/raw/packet_list.go b/tcpip/transport/packet/packet_list.go
similarity index 99%
rename from tcpip/transport/raw/packet_list.go
rename to tcpip/transport/packet/packet_list.go
index 2e90749..0da0dfc 100644
--- a/tcpip/transport/raw/packet_list.go
+++ b/tcpip/transport/packet/packet_list.go
@@ -1,4 +1,4 @@
-package raw
+package packet
// ElementMapper provides an identity mapping by default.
//
diff --git a/tcpip/transport/raw/endpoint.go b/tcpip/transport/raw/endpoint.go
index 0ff546b..deeae49 100644
--- a/tcpip/transport/raw/endpoint.go
+++ b/tcpip/transport/raw/endpoint.go
@@ -17,8 +17,7 @@
//
// * manually write and inspect transport layer headers and payloads
// * receive all traffic of a given transport protocol (e.g. ICMP or UDP)
-// * optionally write and inspect network layer and link layer headers for
-// packets
+// * optionally write and inspect network layer headers of packets
//
// Raw sockets don't have any notion of ports, and incoming packets are
// demultiplexed solely by protocol number. Thus, a raw UDP endpoint will
@@ -38,8 +37,8 @@
)
// +stateify savable
-type packet struct {
- packetEntry
+type rawPacket struct {
+ rawPacketEntry
// data holds the actual packet data, including any headers and
// payload.
data buffer.VectorisedView
@@ -72,7 +71,7 @@
// The following fields are used to manage the receive queue and are
// protected by rcvMu.
rcvMu sync.Mutex
- rcvList packetList
+ rcvList rawPacketList
rcvBufSizeMax int
rcvBufSize int
rcvClosed bool
@@ -90,7 +89,6 @@
}
// NewEndpoint returns a raw endpoint for the given protocols.
-// TODO(b/129292371): IP_HDRINCL and AF_PACKET.
func NewEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) {
return newEndpoint(stack, netProto, transProto, waiterQueue, true /* associated */)
}
@@ -187,17 +185,17 @@
return buffer.View{}, tcpip.ControlMessages{}, err
}
- packet := e.rcvList.Front()
- e.rcvList.Remove(packet)
- e.rcvBufSize -= packet.data.Size()
+ pkt := e.rcvList.Front()
+ e.rcvList.Remove(pkt)
+ e.rcvBufSize -= pkt.data.Size()
e.rcvMu.Unlock()
if addr != nil {
- *addr = packet.senderAddr
+ *addr = pkt.senderAddr
}
- return packet.data.ToView(), tcpip.ControlMessages{HasTimestamp: true, Timestamp: packet.timestampNS}, nil
+ return pkt.data.ToView(), tcpip.ControlMessages{HasTimestamp: true, Timestamp: pkt.timestampNS}, nil
}
// Write implements tcpip.Endpoint.Write.
@@ -602,7 +600,7 @@
wasEmpty := e.rcvBufSize == 0
// Push new packet into receive list and increment the buffer size.
- packet := &packet{
+ pkt := &rawPacket{
senderAddr: tcpip.FullAddress{
NIC: route.NICID(),
Addr: route.RemoteAddress,
@@ -611,11 +609,11 @@
combinedVV := netHeader.ToVectorisedView()
combinedVV.Append(vv)
- packet.data = combinedVV.Clone(packet.views[:])
- packet.timestampNS = e.stack.NowNanoseconds()
+ pkt.data = combinedVV.Clone(pkt.views[:])
+ pkt.timestampNS = e.stack.NowNanoseconds()
- e.rcvList.PushBack(packet)
- e.rcvBufSize += packet.data.Size()
+ e.rcvList.PushBack(pkt)
+ e.rcvBufSize += pkt.data.Size()
e.rcvMu.Unlock()
e.stats.PacketsReceived.Increment()
diff --git a/tcpip/transport/raw/protocol.go b/tcpip/transport/raw/protocol.go
index 619bab8..1117293 100644
--- a/tcpip/transport/raw/protocol.go
+++ b/tcpip/transport/raw/protocol.go
@@ -17,13 +17,19 @@
import (
"github.com/google/netstack/tcpip"
"github.com/google/netstack/tcpip/stack"
+ "github.com/google/netstack/tcpip/transport/packet"
"github.com/google/netstack/waiter"
)
-// EndpointFactory implements stack.UnassociatedEndpointFactory.
+// EndpointFactory implements stack.RawFactory.
type EndpointFactory struct{}
-// NewUnassociatedRawEndpoint implements stack.UnassociatedEndpointFactory.
-func (EndpointFactory) NewUnassociatedRawEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) {
+// NewUnassociatedEndpoint implements stack.RawFactory.NewUnassociatedEndpoint.
+func (EndpointFactory) NewUnassociatedEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) {
return newEndpoint(stack, netProto, transProto, waiterQueue, false /* associated */)
}
+
+// NewPacketEndpoint implements stack.RawFactory.NewPacketEndpoint.
+func (EndpointFactory) NewPacketEndpoint(stack *stack.Stack, cooked bool, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) {
+ return packet.NewEndpoint(stack, cooked, netProto, waiterQueue)
+}
diff --git a/tcpip/transport/raw/raw_packet_list.go b/tcpip/transport/raw/raw_packet_list.go
new file mode 100644
index 0000000..12edb43
--- /dev/null
+++ b/tcpip/transport/raw/raw_packet_list.go
@@ -0,0 +1,173 @@
+package raw
+
+// ElementMapper provides an identity mapping by default.
+//
+// This can be replaced to provide a struct that maps elements to linker
+// objects, if they are not the same. An ElementMapper is not typically
+// required if: Linker is left as is, Element is left as is, or Linker and
+// Element are the same type.
+type rawPacketElementMapper struct{}
+
+// linkerFor maps an Element to a Linker.
+//
+// This default implementation should be inlined.
+//
+//go:nosplit
+func (rawPacketElementMapper) linkerFor(elem *rawPacket) *rawPacket { return elem }
+
+// List is an intrusive list. Entries can be added to or removed from the list
+// in O(1) time and with no additional memory allocations.
+//
+// The zero value for List is an empty list ready to use.
+//
+// To iterate over a list (where l is a List):
+// for e := l.Front(); e != nil; e = e.Next() {
+// // do something with e.
+// }
+//
+// +stateify savable
+type rawPacketList struct {
+ head *rawPacket
+ tail *rawPacket
+}
+
+// Reset resets list l to the empty state.
+func (l *rawPacketList) Reset() {
+ l.head = nil
+ l.tail = nil
+}
+
+// Empty returns true iff the list is empty.
+func (l *rawPacketList) Empty() bool {
+ return l.head == nil
+}
+
+// Front returns the first element of list l or nil.
+func (l *rawPacketList) Front() *rawPacket {
+ return l.head
+}
+
+// Back returns the last element of list l or nil.
+func (l *rawPacketList) Back() *rawPacket {
+ return l.tail
+}
+
+// PushFront inserts the element e at the front of list l.
+func (l *rawPacketList) PushFront(e *rawPacket) {
+ rawPacketElementMapper{}.linkerFor(e).SetNext(l.head)
+ rawPacketElementMapper{}.linkerFor(e).SetPrev(nil)
+
+ if l.head != nil {
+ rawPacketElementMapper{}.linkerFor(l.head).SetPrev(e)
+ } else {
+ l.tail = e
+ }
+
+ l.head = e
+}
+
+// PushBack inserts the element e at the back of list l.
+func (l *rawPacketList) PushBack(e *rawPacket) {
+ rawPacketElementMapper{}.linkerFor(e).SetNext(nil)
+ rawPacketElementMapper{}.linkerFor(e).SetPrev(l.tail)
+
+ if l.tail != nil {
+ rawPacketElementMapper{}.linkerFor(l.tail).SetNext(e)
+ } else {
+ l.head = e
+ }
+
+ l.tail = e
+}
+
+// PushBackList inserts list m at the end of list l, emptying m.
+func (l *rawPacketList) PushBackList(m *rawPacketList) {
+ if l.head == nil {
+ l.head = m.head
+ l.tail = m.tail
+ } else if m.head != nil {
+ rawPacketElementMapper{}.linkerFor(l.tail).SetNext(m.head)
+ rawPacketElementMapper{}.linkerFor(m.head).SetPrev(l.tail)
+
+ l.tail = m.tail
+ }
+
+ m.head = nil
+ m.tail = nil
+}
+
+// InsertAfter inserts e after b.
+func (l *rawPacketList) InsertAfter(b, e *rawPacket) {
+ a := rawPacketElementMapper{}.linkerFor(b).Next()
+ rawPacketElementMapper{}.linkerFor(e).SetNext(a)
+ rawPacketElementMapper{}.linkerFor(e).SetPrev(b)
+ rawPacketElementMapper{}.linkerFor(b).SetNext(e)
+
+ if a != nil {
+ rawPacketElementMapper{}.linkerFor(a).SetPrev(e)
+ } else {
+ l.tail = e
+ }
+}
+
+// InsertBefore inserts e before a.
+func (l *rawPacketList) InsertBefore(a, e *rawPacket) {
+ b := rawPacketElementMapper{}.linkerFor(a).Prev()
+ rawPacketElementMapper{}.linkerFor(e).SetNext(a)
+ rawPacketElementMapper{}.linkerFor(e).SetPrev(b)
+ rawPacketElementMapper{}.linkerFor(a).SetPrev(e)
+
+ if b != nil {
+ rawPacketElementMapper{}.linkerFor(b).SetNext(e)
+ } else {
+ l.head = e
+ }
+}
+
+// Remove removes e from l.
+func (l *rawPacketList) Remove(e *rawPacket) {
+ prev := rawPacketElementMapper{}.linkerFor(e).Prev()
+ next := rawPacketElementMapper{}.linkerFor(e).Next()
+
+ if prev != nil {
+ rawPacketElementMapper{}.linkerFor(prev).SetNext(next)
+ } else {
+ l.head = next
+ }
+
+ if next != nil {
+ rawPacketElementMapper{}.linkerFor(next).SetPrev(prev)
+ } else {
+ l.tail = prev
+ }
+}
+
+// Entry is a default implementation of Linker. Users can add anonymous fields
+// of this type to their structs to make them automatically implement the
+// methods needed by List.
+//
+// +stateify savable
+type rawPacketEntry struct {
+ next *rawPacket
+ prev *rawPacket
+}
+
+// Next returns the entry that follows e in the list.
+func (e *rawPacketEntry) Next() *rawPacket {
+ return e.next
+}
+
+// Prev returns the entry that precedes e in the list.
+func (e *rawPacketEntry) Prev() *rawPacket {
+ return e.prev
+}
+
+// SetNext assigns 'entry' as the entry that follows e in the list.
+func (e *rawPacketEntry) SetNext(elem *rawPacket) {
+ e.next = elem
+}
+
+// SetPrev assigns 'entry' as the entry that precedes e in the list.
+func (e *rawPacketEntry) SetPrev(elem *rawPacket) {
+ e.prev = elem
+}
diff --git a/tcpip/transport/tcp/accept.go b/tcpip/transport/tcp/accept.go
index c287325..72d711b 100644
--- a/tcpip/transport/tcp/accept.go
+++ b/tcpip/transport/tcp/accept.go
@@ -297,6 +297,7 @@
return nil, err
}
ep.mu.Lock()
+ ep.stack.Stats().TCP.CurrentEstablished.Increment()
ep.state = StateEstablished
ep.mu.Unlock()
@@ -519,6 +520,7 @@
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/tcpip/transport/tcp/connect.go b/tcpip/transport/tcp/connect.go
index c3258b2..ee113a4 100644
--- a/tcpip/transport/tcp/connect.go
+++ b/tcpip/transport/tcp/connect.go
@@ -78,9 +78,6 @@
// 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
@@ -607,17 +604,11 @@
return nil
}
-// sendTCP sends a TCP segment with the provided options via the provided
-// network endpoint and under the provided identity.
-func sendTCP(r *stack.Route, id stack.TransportEndpointID, data buffer.VectorisedView, ttl, tos uint8, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) *tcpip.Error {
+func buildTCPHdr(r *stack.Route, id stack.TransportEndpointID, d *stack.PacketDescriptor, data buffer.VectorisedView, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) {
optLen := len(opts)
- // Allocate a buffer for the TCP header.
- hdr := buffer.NewPrependable(header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen)
-
- if rcvWnd > 0xffff {
- rcvWnd = 0xffff
- }
-
+ hdr := &d.Hdr
+ packetSize := d.Size
+ off := d.Off
// Initialize the header.
tcp := header.TCP(hdr.Prepend(header.TCPMinimumSize + optLen))
tcp.Encode(&header.TCPFields{
@@ -631,7 +622,7 @@
})
copy(tcp[header.TCPMinimumSize:], opts)
- length := uint16(hdr.UsedLength() + data.Size())
+ length := uint16(hdr.UsedLength() + packetSize)
xsum := r.PseudoHeaderChecksum(ProtocolNumber, length)
// Only calculate the checksum if offloading isn't supported.
if gso != nil && gso.NeedsCsum {
@@ -641,14 +632,71 @@
// header and data and get the right sum of the TCP packet.
tcp.SetChecksum(xsum)
} else if r.Capabilities()&stack.CapabilityTXChecksumOffload == 0 {
- xsum = header.ChecksumVV(data, xsum)
+ xsum = header.ChecksumVVWithOffset(data, xsum, off, packetSize)
tcp.SetChecksum(^tcp.CalculateChecksum(xsum))
}
+}
+
+func sendTCPBatch(r *stack.Route, id stack.TransportEndpointID, data buffer.VectorisedView, ttl, tos uint8, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) *tcpip.Error {
+ optLen := len(opts)
+ if rcvWnd > 0xffff {
+ rcvWnd = 0xffff
+ }
+
+ mss := int(gso.MSS)
+ n := (data.Size() + mss - 1) / mss
+
+ hdrs := stack.NewPacketDescriptors(n, header.TCPMinimumSize+int(r.MaxHeaderLength())+optLen)
+
+ size := data.Size()
+ off := 0
+ for i := 0; i < n; i++ {
+ packetSize := mss
+ if packetSize > size {
+ packetSize = size
+ }
+ size -= packetSize
+ hdrs[i].Off = off
+ hdrs[i].Size = packetSize
+ buildTCPHdr(r, id, &hdrs[i], data, flags, seq, ack, rcvWnd, opts, gso)
+ off += packetSize
+ seq = seq.Add(seqnum.Size(packetSize))
+ }
+ if ttl == 0 {
+ ttl = r.DefaultTTL()
+ }
+ sent, err := r.WritePackets(gso, hdrs, data, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos})
+ if err != nil {
+ r.Stats().TCP.SegmentSendErrors.IncrementBy(uint64(n - sent))
+ }
+ r.Stats().TCP.SegmentsSent.IncrementBy(uint64(sent))
+ return err
+}
+
+// sendTCP sends a TCP segment with the provided options via the provided
+// network endpoint and under the provided identity.
+func sendTCP(r *stack.Route, id stack.TransportEndpointID, data buffer.VectorisedView, ttl, tos uint8, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) *tcpip.Error {
+ optLen := len(opts)
+ if rcvWnd > 0xffff {
+ rcvWnd = 0xffff
+ }
+
+ if r.Loop&stack.PacketLoop == 0 && gso != nil && gso.Type == stack.GSOSW && int(gso.MSS) < data.Size() {
+ return sendTCPBatch(r, id, data, ttl, tos, flags, seq, ack, rcvWnd, opts, gso)
+ }
+
+ d := &stack.PacketDescriptor{
+ Hdr: buffer.NewPrependable(header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen),
+ Off: 0,
+ Size: data.Size(),
+ }
+ buildTCPHdr(r, id, d, data, flags, seq, ack, rcvWnd, opts, gso)
+
if ttl == 0 {
ttl = r.DefaultTTL()
}
- if err := r.WritePacket(gso, hdr, data, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}); err != nil {
+ if err := r.WritePacket(gso, d.Hdr, data, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}); err != nil {
r.Stats().TCP.SegmentSendErrors.Increment()
return err
}
@@ -754,6 +802,10 @@
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 {
@@ -924,6 +976,8 @@
e.lastErrorMu.Unlock()
e.mu.Lock()
+ e.stack.Stats().TCP.EstablishedResets.Increment()
+ e.stack.Stats().TCP.CurrentEstablished.Decrement()
e.state = StateError
e.HardError = err
@@ -954,7 +1008,10 @@
// 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 {
@@ -1115,6 +1172,8 @@
// 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/tcpip/transport/tcp/endpoint.go b/tcpip/transport/tcp/endpoint.go
index ba897be..e667201 100644
--- a/tcpip/transport/tcp/endpoint.go
+++ b/tcpip/transport/tcp/endpoint.go
@@ -1729,6 +1729,7 @@
e.segmentQueue.mu.Unlock()
e.snd.updateMaxPayloadSize(int(e.route.MTU()), 0)
e.state = StateEstablished
+ e.stack.Stats().TCP.CurrentEstablished.Increment()
}
if run {
@@ -2327,11 +2328,7 @@
return s
}
-func (e *endpoint) initGSO() {
- if e.route.Capabilities()&stack.CapabilityGSO == 0 {
- return
- }
-
+func (e *endpoint) initHardwareGSO() {
gso := &stack.GSO{}
switch e.route.NetProto {
case header.IPv4ProtocolNumber:
@@ -2349,6 +2346,18 @@
e.gso = gso
}
+func (e *endpoint) initGSO() {
+ if e.route.Capabilities()&stack.CapabilityHardwareGSO != 0 {
+ e.initHardwareGSO()
+ } else if e.route.Capabilities()&stack.CapabilitySoftwareGSO != 0 {
+ e.gso = &stack.GSO{
+ MaxSize: e.route.GSOMaxSize(),
+ Type: stack.GSOSW,
+ NeedsCsum: false,
+ }
+ }
+}
+
// State implements tcpip.Endpoint.State. It exports the endpoint's protocol
// state for diagnostics.
func (e *endpoint) State() uint32 {
diff --git a/tcpip/transport/tcp/snd.go b/tcpip/transport/tcp/snd.go
index fa616b9..d4bad63 100644
--- a/tcpip/transport/tcp/snd.go
+++ b/tcpip/transport/tcp/snd.go
@@ -674,6 +674,7 @@
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/tcpip/transport/udp/endpoint.go b/tcpip/transport/udp/endpoint.go
index a4aed3e..c512fc0 100644
--- a/tcpip/transport/udp/endpoint.go
+++ b/tcpip/transport/udp/endpoint.go
@@ -140,7 +140,7 @@
stack: s,
TransportEndpointInfo: stack.TransportEndpointInfo{
NetProto: netProto,
- TransProto: header.TCPProtocolNumber,
+ TransProto: header.UDPProtocolNumber,
},
waiterQueue: waiterQueue,
// RFC 1075 section 5.4 recommends a TTL of 1 for membership
diff --git a/tcpip/transport/udp/protocol.go b/tcpip/transport/udp/protocol.go
index 9969fc3..036b24a 100644
--- a/tcpip/transport/udp/protocol.go
+++ b/tcpip/transport/udp/protocol.go
@@ -121,8 +121,15 @@
payloadLen = available
}
- payload := buffer.NewVectorisedView(len(netHeader), []buffer.View{netHeader})
- payload.Append(vv)
+ // The buffers used by vv and netHeader may be used elsewhere
+ // in the system. For example, a raw or packet socket may use
+ // what UDP considers an unreachable destination. Thus we deep
+ // copy vv and netHeader to prevent multiple ownership and SR
+ // errors.
+ newNetHeader := make(buffer.View, len(netHeader))
+ copy(newNetHeader, netHeader)
+ payload := buffer.NewVectorisedView(len(newNetHeader), []buffer.View{newNetHeader})
+ payload.Append(vv.ToView().ToVectorisedView())
payload.CapLength(payloadLen)
hdr := buffer.NewPrependable(headerLen)
