summaryrefslogtreecommitdiffhomepage
path: root/pkg/tcpip/stack
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/tcpip/stack')
-rw-r--r--pkg/tcpip/stack/BUILD103
-rw-r--r--pkg/tcpip/stack/conntrack.go631
-rw-r--r--pkg/tcpip/stack/dhcpv6configurationfromndpra_string.go40
-rw-r--r--pkg/tcpip/stack/fake_time_test.go209
-rw-r--r--pkg/tcpip/stack/forwarder.go131
-rw-r--r--pkg/tcpip/stack/forwarder_test.go648
-rw-r--r--pkg/tcpip/stack/headertype_string.go39
-rw-r--r--pkg/tcpip/stack/iptables.go423
-rw-r--r--pkg/tcpip/stack/iptables_state.go40
-rw-r--r--pkg/tcpip/stack/iptables_targets.go163
-rw-r--r--pkg/tcpip/stack/iptables_types.go262
-rw-r--r--pkg/tcpip/stack/linkaddrcache.go4
-rw-r--r--pkg/tcpip/stack/linkaddrcache_test.go4
-rw-r--r--pkg/tcpip/stack/ndp.go1910
-rw-r--r--pkg/tcpip/stack/ndp_test.go5204
-rw-r--r--pkg/tcpip/stack/neighbor_cache.go333
-rw-r--r--pkg/tcpip/stack/neighbor_cache_test.go1726
-rw-r--r--pkg/tcpip/stack/neighbor_entry.go482
-rw-r--r--pkg/tcpip/stack/neighbor_entry_test.go2870
-rw-r--r--pkg/tcpip/stack/neighborstate_string.go44
-rw-r--r--pkg/tcpip/stack/nic.go1346
-rw-r--r--pkg/tcpip/stack/nic_test.go316
-rw-r--r--pkg/tcpip/stack/nud.go466
-rw-r--r--pkg/tcpip/stack/nud_test.go795
-rw-r--r--pkg/tcpip/stack/packet_buffer.go299
-rw-r--r--pkg/tcpip/stack/packet_buffer_test.go397
-rw-r--r--pkg/tcpip/stack/rand.go40
-rw-r--r--pkg/tcpip/stack/registration.go197
-rw-r--r--pkg/tcpip/stack/route.go117
-rw-r--r--pkg/tcpip/stack/stack.go847
-rw-r--r--pkg/tcpip/stack/stack_options.go106
-rw-r--r--pkg/tcpip/stack/stack_test.go2219
-rw-r--r--pkg/tcpip/stack/transport_demuxer.go566
-rw-r--r--pkg/tcpip/stack/transport_demuxer_test.go294
-rw-r--r--pkg/tcpip/stack/transport_test.go121
35 files changed, 21641 insertions, 1751 deletions
diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD
index 460db3cf8..900938dd1 100644
--- a/pkg/tcpip/stack/BUILD
+++ b/pkg/tcpip/stack/BUILD
@@ -1,6 +1,5 @@
-load("@io_bazel_rules_go//go:def.bzl", "go_test")
+load("//tools:defs.bzl", "go_library", "go_test")
load("//tools/go_generics:defs.bzl", "go_template_instance")
-load("//tools/go_stateify:defs.bzl", "go_library")
package(licenses = ["notice"])
@@ -16,35 +15,88 @@ go_template_instance(
},
)
+go_template_instance(
+ name = "neighbor_entry_list",
+ out = "neighbor_entry_list.go",
+ package = "stack",
+ prefix = "neighborEntry",
+ template = "//pkg/ilist:generic_list",
+ types = {
+ "Element": "*neighborEntry",
+ "Linker": "*neighborEntry",
+ },
+)
+
+go_template_instance(
+ name = "packet_buffer_list",
+ out = "packet_buffer_list.go",
+ package = "stack",
+ prefix = "PacketBuffer",
+ template = "//pkg/ilist:generic_list",
+ types = {
+ "Element": "*PacketBuffer",
+ "Linker": "*PacketBuffer",
+ },
+)
+
+go_template_instance(
+ name = "tuple_list",
+ out = "tuple_list.go",
+ package = "stack",
+ prefix = "tuple",
+ template = "//pkg/ilist:generic_list",
+ types = {
+ "Element": "*tuple",
+ "Linker": "*tuple",
+ },
+)
+
go_library(
name = "stack",
srcs = [
+ "conntrack.go",
+ "dhcpv6configurationfromndpra_string.go",
+ "forwarder.go",
+ "headertype_string.go",
"icmp_rate_limit.go",
+ "iptables.go",
+ "iptables_state.go",
+ "iptables_targets.go",
+ "iptables_types.go",
"linkaddrcache.go",
"linkaddrentry_list.go",
"ndp.go",
+ "neighbor_cache.go",
+ "neighbor_entry.go",
+ "neighbor_entry_list.go",
+ "neighborstate_string.go",
"nic.go",
+ "nud.go",
+ "packet_buffer.go",
+ "packet_buffer_list.go",
+ "rand.go",
"registration.go",
"route.go",
"stack.go",
"stack_global_state.go",
+ "stack_options.go",
"transport_demuxer.go",
+ "tuple_list.go",
],
- importpath = "gvisor.dev/gvisor/pkg/tcpip/stack",
- visibility = [
- "//visibility:public",
- ],
+ visibility = ["//visibility:public"],
deps = [
"//pkg/ilist",
+ "//pkg/log",
"//pkg/rand",
"//pkg/sleep",
+ "//pkg/sync",
"//pkg/tcpip",
"//pkg/tcpip/buffer",
"//pkg/tcpip/hash/jenkins",
"//pkg/tcpip/header",
- "//pkg/tcpip/iptables",
"//pkg/tcpip/ports",
"//pkg/tcpip/seqnum",
+ "//pkg/tcpip/transport/tcpconntrack",
"//pkg/waiter",
"@org_golang_x_time//rate:go_default_library",
],
@@ -52,46 +104,57 @@ go_library(
go_test(
name = "stack_x_test",
- size = "small",
+ size = "medium",
srcs = [
"ndp_test.go",
+ "nud_test.go",
"stack_test.go",
"transport_demuxer_test.go",
"transport_test.go",
],
+ shard_count = 20,
deps = [
":stack",
+ "//pkg/rand",
"//pkg/tcpip",
"//pkg/tcpip/buffer",
"//pkg/tcpip/checker",
"//pkg/tcpip/header",
- "//pkg/tcpip/iptables",
"//pkg/tcpip/link/channel",
"//pkg/tcpip/link/loopback",
+ "//pkg/tcpip/network/arp",
"//pkg/tcpip/network/ipv4",
"//pkg/tcpip/network/ipv6",
+ "//pkg/tcpip/ports",
"//pkg/tcpip/transport/icmp",
"//pkg/tcpip/transport/udp",
"//pkg/waiter",
- "@com_github_google_go-cmp//cmp:go_default_library",
+ "@com_github_google_go_cmp//cmp:go_default_library",
+ "@com_github_google_go_cmp//cmp/cmpopts:go_default_library",
],
)
go_test(
name = "stack_test",
size = "small",
- srcs = ["linkaddrcache_test.go"],
- embed = [":stack"],
+ srcs = [
+ "fake_time_test.go",
+ "forwarder_test.go",
+ "linkaddrcache_test.go",
+ "neighbor_cache_test.go",
+ "neighbor_entry_test.go",
+ "nic_test.go",
+ "packet_buffer_test.go",
+ ],
+ library = ":stack",
deps = [
"//pkg/sleep",
+ "//pkg/sync",
"//pkg/tcpip",
+ "//pkg/tcpip/buffer",
+ "//pkg/tcpip/header",
+ "@com_github_dpjacques_clockwork//:go_default_library",
+ "@com_github_google_go_cmp//cmp:go_default_library",
+ "@com_github_google_go_cmp//cmp/cmpopts:go_default_library",
],
)
-
-filegroup(
- name = "autogen",
- srcs = [
- "linkaddrentry_list.go",
- ],
- visibility = ["//:sandbox"],
-)
diff --git a/pkg/tcpip/stack/conntrack.go b/pkg/tcpip/stack/conntrack.go
new file mode 100644
index 000000000..7dd344b4f
--- /dev/null
+++ b/pkg/tcpip/stack/conntrack.go
@@ -0,0 +1,631 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package stack
+
+import (
+ "encoding/binary"
+ "sync"
+ "time"
+
+ "gvisor.dev/gvisor/pkg/tcpip"
+ "gvisor.dev/gvisor/pkg/tcpip/hash/jenkins"
+ "gvisor.dev/gvisor/pkg/tcpip/header"
+ "gvisor.dev/gvisor/pkg/tcpip/transport/tcpconntrack"
+)
+
+// Connection tracking is used to track and manipulate packets for NAT rules.
+// The connection is created for a packet if it does not exist. Every
+// connection contains two tuples (original and reply). The tuples are
+// manipulated if there is a matching NAT rule. The packet is modified by
+// looking at the tuples in the Prerouting and Output hooks.
+//
+// Currently, only TCP tracking is supported.
+
+// Our hash table has 16K buckets.
+// TODO(gvisor.dev/issue/170): These should be tunable.
+const numBuckets = 1 << 14
+
+// Direction of the tuple.
+type direction int
+
+const (
+ dirOriginal direction = iota
+ dirReply
+)
+
+// Manipulation type for the connection.
+type manipType int
+
+const (
+ manipNone manipType = iota
+ manipDstPrerouting
+ manipDstOutput
+)
+
+// tuple holds a connection's identifying and manipulating data in one
+// direction. It is immutable.
+//
+// +stateify savable
+type tuple struct {
+ // tupleEntry is used to build an intrusive list of tuples.
+ tupleEntry
+
+ tupleID
+
+ // conn is the connection tracking entry this tuple belongs to.
+ conn *conn
+
+ // direction is the direction of the tuple.
+ direction direction
+}
+
+// tupleID uniquely identifies a connection in one direction. It currently
+// contains enough information to distinguish between any TCP or UDP
+// connection, and will need to be extended to support other protocols.
+//
+// +stateify savable
+type tupleID struct {
+ srcAddr tcpip.Address
+ srcPort uint16
+ dstAddr tcpip.Address
+ dstPort uint16
+ transProto tcpip.TransportProtocolNumber
+ netProto tcpip.NetworkProtocolNumber
+}
+
+// reply creates the reply tupleID.
+func (ti tupleID) reply() tupleID {
+ return tupleID{
+ srcAddr: ti.dstAddr,
+ srcPort: ti.dstPort,
+ dstAddr: ti.srcAddr,
+ dstPort: ti.srcPort,
+ transProto: ti.transProto,
+ netProto: ti.netProto,
+ }
+}
+
+// conn is a tracked connection.
+//
+// +stateify savable
+type conn struct {
+ // original is the tuple in original direction. It is immutable.
+ original tuple
+
+ // reply is the tuple in reply direction. It is immutable.
+ reply tuple
+
+ // manip indicates if the packet should be manipulated. It is immutable.
+ manip manipType
+
+ // tcbHook indicates if the packet is inbound or outbound to
+ // update the state of tcb. It is immutable.
+ tcbHook Hook
+
+ // mu protects all mutable state.
+ mu sync.Mutex `state:"nosave"`
+ // tcb is TCB control block. It is used to keep track of states
+ // of tcp connection and is protected by mu.
+ tcb tcpconntrack.TCB
+ // lastUsed is the last time the connection saw a relevant packet, and
+ // is updated by each packet on the connection. It is protected by mu.
+ lastUsed time.Time `state:".(unixTime)"`
+}
+
+// timedOut returns whether the connection timed out based on its state.
+func (cn *conn) timedOut(now time.Time) bool {
+ const establishedTimeout = 5 * 24 * time.Hour
+ const defaultTimeout = 120 * time.Second
+ cn.mu.Lock()
+ defer cn.mu.Unlock()
+ if cn.tcb.State() == tcpconntrack.ResultAlive {
+ // Use the same default as Linux, which doesn't delete
+ // established connections for 5(!) days.
+ return now.Sub(cn.lastUsed) > establishedTimeout
+ }
+ // Use the same default as Linux, which lets connections in most states
+ // other than established remain for <= 120 seconds.
+ return now.Sub(cn.lastUsed) > defaultTimeout
+}
+
+// update the connection tracking state.
+//
+// Precondition: ct.mu must be held.
+func (ct *conn) updateLocked(tcpHeader header.TCP, hook Hook) {
+ // Update the state of tcb. tcb assumes it's always initialized on the
+ // client. However, we only need to know whether the connection is
+ // established or not, so the client/server distinction isn't important.
+ // TODO(gvisor.dev/issue/170): Add support in tcpconntrack to handle
+ // other tcp states.
+ if ct.tcb.IsEmpty() {
+ ct.tcb.Init(tcpHeader)
+ } else if hook == ct.tcbHook {
+ ct.tcb.UpdateStateOutbound(tcpHeader)
+ } else {
+ ct.tcb.UpdateStateInbound(tcpHeader)
+ }
+}
+
+// ConnTrack tracks all connections created for NAT rules. Most users are
+// expected to only call handlePacket, insertRedirectConn, and maybeInsertNoop.
+//
+// ConnTrack keeps all connections in a slice of buckets, each of which holds a
+// linked list of tuples. This gives us some desirable properties:
+// - Each bucket has its own lock, lessening lock contention.
+// - The slice is large enough that lists stay short (<10 elements on average).
+// Thus traversal is fast.
+// - During linked list traversal we reap expired connections. This amortizes
+// the cost of reaping them and makes reapUnused faster.
+//
+// Locks are ordered by their location in the buckets slice. That is, a
+// goroutine that locks buckets[i] can only lock buckets[j] s.t. i < j.
+//
+// +stateify savable
+type ConnTrack struct {
+ // seed is a one-time random value initialized at stack startup
+ // and is used in the calculation of hash keys for the list of buckets.
+ // It is immutable.
+ seed uint32
+
+ // mu protects the buckets slice, but not buckets' contents. Only take
+ // the write lock if you are modifying the slice or saving for S/R.
+ mu sync.RWMutex `state:"nosave"`
+
+ // buckets is protected by mu.
+ buckets []bucket
+}
+
+// +stateify savable
+type bucket struct {
+ // mu protects tuples.
+ mu sync.Mutex `state:"nosave"`
+ tuples tupleList
+}
+
+// packetToTupleID converts packet to a tuple ID. It fails when pkt lacks a valid
+// TCP header.
+func packetToTupleID(pkt *PacketBuffer) (tupleID, *tcpip.Error) {
+ // TODO(gvisor.dev/issue/170): Need to support for other
+ // protocols as well.
+ netHeader := header.IPv4(pkt.NetworkHeader().View())
+ if len(netHeader) < header.IPv4MinimumSize || netHeader.TransportProtocol() != header.TCPProtocolNumber {
+ return tupleID{}, tcpip.ErrUnknownProtocol
+ }
+ tcpHeader := header.TCP(pkt.TransportHeader().View())
+ if len(tcpHeader) < header.TCPMinimumSize {
+ return tupleID{}, tcpip.ErrUnknownProtocol
+ }
+
+ return tupleID{
+ srcAddr: netHeader.SourceAddress(),
+ srcPort: tcpHeader.SourcePort(),
+ dstAddr: netHeader.DestinationAddress(),
+ dstPort: tcpHeader.DestinationPort(),
+ transProto: netHeader.TransportProtocol(),
+ netProto: header.IPv4ProtocolNumber,
+ }, nil
+}
+
+// newConn creates new connection.
+func newConn(orig, reply tupleID, manip manipType, hook Hook) *conn {
+ conn := conn{
+ manip: manip,
+ tcbHook: hook,
+ lastUsed: time.Now(),
+ }
+ conn.original = tuple{conn: &conn, tupleID: orig}
+ conn.reply = tuple{conn: &conn, tupleID: reply, direction: dirReply}
+ return &conn
+}
+
+// connFor gets the conn for pkt if it exists, or returns nil
+// if it does not. It returns an error when pkt does not contain a valid TCP
+// header.
+// TODO(gvisor.dev/issue/170): Only TCP packets are supported. Need to support
+// other transport protocols.
+func (ct *ConnTrack) connFor(pkt *PacketBuffer) (*conn, direction) {
+ tid, err := packetToTupleID(pkt)
+ if err != nil {
+ return nil, dirOriginal
+ }
+ return ct.connForTID(tid)
+}
+
+func (ct *ConnTrack) connForTID(tid tupleID) (*conn, direction) {
+ bucket := ct.bucket(tid)
+ now := time.Now()
+
+ ct.mu.RLock()
+ defer ct.mu.RUnlock()
+ ct.buckets[bucket].mu.Lock()
+ defer ct.buckets[bucket].mu.Unlock()
+
+ // Iterate over the tuples in a bucket, cleaning up any unused
+ // connections we find.
+ for other := ct.buckets[bucket].tuples.Front(); other != nil; other = other.Next() {
+ // Clean up any timed-out connections we happen to find.
+ if ct.reapTupleLocked(other, bucket, now) {
+ // The tuple expired.
+ continue
+ }
+ if tid == other.tupleID {
+ return other.conn, other.direction
+ }
+ }
+
+ return nil, dirOriginal
+}
+
+func (ct *ConnTrack) insertRedirectConn(pkt *PacketBuffer, hook Hook, rt RedirectTarget) *conn {
+ tid, err := packetToTupleID(pkt)
+ if err != nil {
+ return nil
+ }
+ if hook != Prerouting && hook != Output {
+ return nil
+ }
+
+ // Create a new connection and change the port as per the iptables
+ // rule. This tuple will be used to manipulate the packet in
+ // handlePacket.
+ replyTID := tid.reply()
+ replyTID.srcAddr = rt.MinIP
+ replyTID.srcPort = rt.MinPort
+ var manip manipType
+ switch hook {
+ case Prerouting:
+ manip = manipDstPrerouting
+ case Output:
+ manip = manipDstOutput
+ }
+ conn := newConn(tid, replyTID, manip, hook)
+ ct.insertConn(conn)
+ return conn
+}
+
+// insertConn inserts conn into the appropriate table bucket.
+func (ct *ConnTrack) insertConn(conn *conn) {
+ // Lock the buckets in the correct order.
+ tupleBucket := ct.bucket(conn.original.tupleID)
+ replyBucket := ct.bucket(conn.reply.tupleID)
+ ct.mu.RLock()
+ defer ct.mu.RUnlock()
+ if tupleBucket < replyBucket {
+ ct.buckets[tupleBucket].mu.Lock()
+ ct.buckets[replyBucket].mu.Lock()
+ } else if tupleBucket > replyBucket {
+ ct.buckets[replyBucket].mu.Lock()
+ ct.buckets[tupleBucket].mu.Lock()
+ } else {
+ // Both tuples are in the same bucket.
+ ct.buckets[tupleBucket].mu.Lock()
+ }
+
+ // Now that we hold the locks, ensure the tuple hasn't been inserted by
+ // another thread.
+ alreadyInserted := false
+ for other := ct.buckets[tupleBucket].tuples.Front(); other != nil; other = other.Next() {
+ if other.tupleID == conn.original.tupleID {
+ alreadyInserted = true
+ break
+ }
+ }
+
+ if !alreadyInserted {
+ // Add the tuple to the map.
+ ct.buckets[tupleBucket].tuples.PushFront(&conn.original)
+ ct.buckets[replyBucket].tuples.PushFront(&conn.reply)
+ }
+
+ // Unlocking can happen in any order.
+ ct.buckets[tupleBucket].mu.Unlock()
+ if tupleBucket != replyBucket {
+ ct.buckets[replyBucket].mu.Unlock()
+ }
+}
+
+// handlePacketPrerouting manipulates ports for packets in Prerouting hook.
+// TODO(gvisor.dev/issue/170): Change address for Prerouting hook.
+func handlePacketPrerouting(pkt *PacketBuffer, conn *conn, dir direction) {
+ // If this is a noop entry, don't do anything.
+ if conn.manip == manipNone {
+ return
+ }
+
+ netHeader := header.IPv4(pkt.NetworkHeader().View())
+ tcpHeader := header.TCP(pkt.TransportHeader().View())
+
+ // For prerouting redirection, packets going in the original direction
+ // have their destinations modified and replies have their sources
+ // modified.
+ switch dir {
+ case dirOriginal:
+ port := conn.reply.srcPort
+ tcpHeader.SetDestinationPort(port)
+ netHeader.SetDestinationAddress(conn.reply.srcAddr)
+ case dirReply:
+ port := conn.original.dstPort
+ tcpHeader.SetSourcePort(port)
+ netHeader.SetSourceAddress(conn.original.dstAddr)
+ }
+
+ // TODO(gvisor.dev/issue/170): TCP checksums aren't usually validated
+ // on inbound packets, so we don't recalculate them. However, we should
+ // support cases when they are validated, e.g. when we can't offload
+ // receive checksumming.
+
+ netHeader.SetChecksum(0)
+ netHeader.SetChecksum(^netHeader.CalculateChecksum())
+}
+
+// handlePacketOutput manipulates ports for packets in Output hook.
+func handlePacketOutput(pkt *PacketBuffer, conn *conn, gso *GSO, r *Route, dir direction) {
+ // If this is a noop entry, don't do anything.
+ if conn.manip == manipNone {
+ return
+ }
+
+ netHeader := header.IPv4(pkt.NetworkHeader().View())
+ tcpHeader := header.TCP(pkt.TransportHeader().View())
+
+ // For output redirection, packets going in the original direction
+ // have their destinations modified and replies have their sources
+ // modified. For prerouting redirection, we only reach this point
+ // when replying, so packet sources are modified.
+ if conn.manip == manipDstOutput && dir == dirOriginal {
+ port := conn.reply.srcPort
+ tcpHeader.SetDestinationPort(port)
+ netHeader.SetDestinationAddress(conn.reply.srcAddr)
+ } else {
+ port := conn.original.dstPort
+ tcpHeader.SetSourcePort(port)
+ netHeader.SetSourceAddress(conn.original.dstAddr)
+ }
+
+ // Calculate the TCP checksum and set it.
+ tcpHeader.SetChecksum(0)
+ length := uint16(pkt.Size()) - uint16(netHeader.HeaderLength())
+ xsum := r.PseudoHeaderChecksum(header.TCPProtocolNumber, length)
+ if gso != nil && gso.NeedsCsum {
+ tcpHeader.SetChecksum(xsum)
+ } else if r.Capabilities()&CapabilityTXChecksumOffload == 0 {
+ xsum = header.ChecksumVVWithOffset(pkt.Data, xsum, int(tcpHeader.DataOffset()), pkt.Data.Size())
+ tcpHeader.SetChecksum(^tcpHeader.CalculateChecksum(xsum))
+ }
+
+ netHeader.SetChecksum(0)
+ netHeader.SetChecksum(^netHeader.CalculateChecksum())
+}
+
+// handlePacket will manipulate the port and address of the packet if the
+// connection exists. Returns whether, after the packet traverses the tables,
+// it should create a new entry in the table.
+func (ct *ConnTrack) handlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r *Route) bool {
+ if pkt.NatDone {
+ return false
+ }
+
+ if hook != Prerouting && hook != Output {
+ return false
+ }
+
+ // TODO(gvisor.dev/issue/170): Support other transport protocols.
+ if nh := pkt.NetworkHeader().View(); nh.IsEmpty() || header.IPv4(nh).TransportProtocol() != header.TCPProtocolNumber {
+ return false
+ }
+
+ conn, dir := ct.connFor(pkt)
+ // Connection or Rule not found for the packet.
+ if conn == nil {
+ return true
+ }
+
+ tcpHeader := header.TCP(pkt.TransportHeader().View())
+ if len(tcpHeader) < header.TCPMinimumSize {
+ return false
+ }
+
+ switch hook {
+ case Prerouting:
+ handlePacketPrerouting(pkt, conn, dir)
+ case Output:
+ handlePacketOutput(pkt, conn, gso, r, dir)
+ }
+ pkt.NatDone = true
+
+ // Update the state of tcb.
+ // TODO(gvisor.dev/issue/170): Add support in tcpcontrack to handle
+ // other tcp states.
+ conn.mu.Lock()
+ defer conn.mu.Unlock()
+
+ // Mark the connection as having been used recently so it isn't reaped.
+ conn.lastUsed = time.Now()
+ // Update connection state.
+ conn.updateLocked(header.TCP(pkt.TransportHeader().View()), hook)
+
+ return false
+}
+
+// maybeInsertNoop tries to insert a no-op connection entry to keep connections
+// from getting clobbered when replies arrive. It only inserts if there isn't
+// already a connection for pkt.
+//
+// This should be called after traversing iptables rules only, to ensure that
+// pkt.NatDone is set correctly.
+func (ct *ConnTrack) maybeInsertNoop(pkt *PacketBuffer, hook Hook) {
+ // If there were a rule applying to this packet, it would be marked
+ // with NatDone.
+ if pkt.NatDone {
+ return
+ }
+
+ // We only track TCP connections.
+ if nh := pkt.NetworkHeader().View(); nh.IsEmpty() || header.IPv4(nh).TransportProtocol() != header.TCPProtocolNumber {
+ return
+ }
+
+ // This is the first packet we're seeing for the TCP connection. Insert
+ // the noop entry (an identity mapping) so that the response doesn't
+ // get NATed, breaking the connection.
+ tid, err := packetToTupleID(pkt)
+ if err != nil {
+ return
+ }
+ conn := newConn(tid, tid.reply(), manipNone, hook)
+ conn.updateLocked(header.TCP(pkt.TransportHeader().View()), hook)
+ ct.insertConn(conn)
+}
+
+// bucket gets the conntrack bucket for a tupleID.
+func (ct *ConnTrack) bucket(id tupleID) int {
+ h := jenkins.Sum32(ct.seed)
+ h.Write([]byte(id.srcAddr))
+ h.Write([]byte(id.dstAddr))
+ shortBuf := make([]byte, 2)
+ binary.LittleEndian.PutUint16(shortBuf, id.srcPort)
+ h.Write([]byte(shortBuf))
+ binary.LittleEndian.PutUint16(shortBuf, id.dstPort)
+ h.Write([]byte(shortBuf))
+ binary.LittleEndian.PutUint16(shortBuf, uint16(id.transProto))
+ h.Write([]byte(shortBuf))
+ binary.LittleEndian.PutUint16(shortBuf, uint16(id.netProto))
+ h.Write([]byte(shortBuf))
+ ct.mu.RLock()
+ defer ct.mu.RUnlock()
+ return int(h.Sum32()) % len(ct.buckets)
+}
+
+// reapUnused deletes timed out entries from the conntrack map. The rules for
+// reaping are:
+// - Most reaping occurs in connFor, which is called on each packet. connFor
+// cleans up the bucket the packet's connection maps to. Thus calls to
+// reapUnused should be fast.
+// - Each call to reapUnused traverses a fraction of the conntrack table.
+// Specifically, it traverses len(ct.buckets)/fractionPerReaping.
+// - After reaping, reapUnused decides when it should next run based on the
+// ratio of expired connections to examined connections. If the ratio is
+// greater than maxExpiredPct, it schedules the next run quickly. Otherwise it
+// slightly increases the interval between runs.
+// - maxFullTraversal caps the time it takes to traverse the entire table.
+//
+// reapUnused returns the next bucket that should be checked and the time after
+// which it should be called again.
+func (ct *ConnTrack) reapUnused(start int, prevInterval time.Duration) (int, time.Duration) {
+ // TODO(gvisor.dev/issue/170): This can be more finely controlled, as
+ // it is in Linux via sysctl.
+ const fractionPerReaping = 128
+ const maxExpiredPct = 50
+ const maxFullTraversal = 60 * time.Second
+ const minInterval = 10 * time.Millisecond
+ const maxInterval = maxFullTraversal / fractionPerReaping
+
+ now := time.Now()
+ checked := 0
+ expired := 0
+ var idx int
+ ct.mu.RLock()
+ defer ct.mu.RUnlock()
+ for i := 0; i < len(ct.buckets)/fractionPerReaping; i++ {
+ idx = (i + start) % len(ct.buckets)
+ ct.buckets[idx].mu.Lock()
+ for tuple := ct.buckets[idx].tuples.Front(); tuple != nil; tuple = tuple.Next() {
+ checked++
+ if ct.reapTupleLocked(tuple, idx, now) {
+ expired++
+ }
+ }
+ ct.buckets[idx].mu.Unlock()
+ }
+ // We already checked buckets[idx].
+ idx++
+
+ // If half or more of the connections are expired, the table has gotten
+ // stale. Reschedule quickly.
+ expiredPct := 0
+ if checked != 0 {
+ expiredPct = expired * 100 / checked
+ }
+ if expiredPct > maxExpiredPct {
+ return idx, minInterval
+ }
+ if interval := prevInterval + minInterval; interval <= maxInterval {
+ // Increment the interval between runs.
+ return idx, interval
+ }
+ // We've hit the maximum interval.
+ return idx, maxInterval
+}
+
+// reapTupleLocked tries to remove tuple and its reply from the table. It
+// returns whether the tuple's connection has timed out.
+//
+// Preconditions: ct.mu is locked for reading and bucket is locked.
+func (ct *ConnTrack) reapTupleLocked(tuple *tuple, bucket int, now time.Time) bool {
+ if !tuple.conn.timedOut(now) {
+ return false
+ }
+
+ // To maintain lock order, we can only reap these tuples if the reply
+ // appears later in the table.
+ replyBucket := ct.bucket(tuple.reply())
+ if bucket > replyBucket {
+ return true
+ }
+
+ // Don't re-lock if both tuples are in the same bucket.
+ differentBuckets := bucket != replyBucket
+ if differentBuckets {
+ ct.buckets[replyBucket].mu.Lock()
+ }
+
+ // We have the buckets locked and can remove both tuples.
+ if tuple.direction == dirOriginal {
+ ct.buckets[replyBucket].tuples.Remove(&tuple.conn.reply)
+ } else {
+ ct.buckets[replyBucket].tuples.Remove(&tuple.conn.original)
+ }
+ ct.buckets[bucket].tuples.Remove(tuple)
+
+ // Don't re-unlock if both tuples are in the same bucket.
+ if differentBuckets {
+ ct.buckets[replyBucket].mu.Unlock()
+ }
+
+ return true
+}
+
+func (ct *ConnTrack) originalDst(epID TransportEndpointID) (tcpip.Address, uint16, *tcpip.Error) {
+ // Lookup the connection. The reply's original destination
+ // describes the original address.
+ tid := tupleID{
+ srcAddr: epID.LocalAddress,
+ srcPort: epID.LocalPort,
+ dstAddr: epID.RemoteAddress,
+ dstPort: epID.RemotePort,
+ transProto: header.TCPProtocolNumber,
+ netProto: header.IPv4ProtocolNumber,
+ }
+ conn, _ := ct.connForTID(tid)
+ if conn == nil {
+ // Not a tracked connection.
+ return "", 0, tcpip.ErrNotConnected
+ } else if conn.manip == manipNone {
+ // Unmanipulated connection.
+ return "", 0, tcpip.ErrInvalidOptionValue
+ }
+
+ return conn.original.dstAddr, conn.original.dstPort, nil
+}
diff --git a/pkg/tcpip/stack/dhcpv6configurationfromndpra_string.go b/pkg/tcpip/stack/dhcpv6configurationfromndpra_string.go
new file mode 100644
index 000000000..d199ded6a
--- /dev/null
+++ b/pkg/tcpip/stack/dhcpv6configurationfromndpra_string.go
@@ -0,0 +1,40 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Code generated by "stringer -type DHCPv6ConfigurationFromNDPRA"; DO NOT EDIT.
+
+package stack
+
+import "strconv"
+
+func _() {
+ // An "invalid array index" compiler error signifies that the constant values have changed.
+ // Re-run the stringer command to generate them again.
+ var x [1]struct{}
+ _ = x[DHCPv6NoConfiguration-1]
+ _ = x[DHCPv6ManagedAddress-2]
+ _ = x[DHCPv6OtherConfigurations-3]
+}
+
+const _DHCPv6ConfigurationFromNDPRA_name = "DHCPv6NoConfigurationDHCPv6ManagedAddressDHCPv6OtherConfigurations"
+
+var _DHCPv6ConfigurationFromNDPRA_index = [...]uint8{0, 21, 41, 66}
+
+func (i DHCPv6ConfigurationFromNDPRA) String() string {
+ i -= 1
+ if i < 0 || i >= DHCPv6ConfigurationFromNDPRA(len(_DHCPv6ConfigurationFromNDPRA_index)-1) {
+ return "DHCPv6ConfigurationFromNDPRA(" + strconv.FormatInt(int64(i+1), 10) + ")"
+ }
+ return _DHCPv6ConfigurationFromNDPRA_name[_DHCPv6ConfigurationFromNDPRA_index[i]:_DHCPv6ConfigurationFromNDPRA_index[i+1]]
+}
diff --git a/pkg/tcpip/stack/fake_time_test.go b/pkg/tcpip/stack/fake_time_test.go
new file mode 100644
index 000000000..92c8cb534
--- /dev/null
+++ b/pkg/tcpip/stack/fake_time_test.go
@@ -0,0 +1,209 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package stack
+
+import (
+ "container/heap"
+ "sync"
+ "time"
+
+ "github.com/dpjacques/clockwork"
+ "gvisor.dev/gvisor/pkg/tcpip"
+)
+
+type fakeClock struct {
+ clock clockwork.FakeClock
+
+ // mu protects the fields below.
+ mu sync.RWMutex
+
+ // times is min-heap of times. A heap is used for quick retrieval of the next
+ // upcoming time of scheduled work.
+ times *timeHeap
+
+ // waitGroups stores one WaitGroup for all work scheduled to execute at the
+ // same time via AfterFunc. This allows parallel execution of all functions
+ // passed to AfterFunc scheduled for the same time.
+ waitGroups map[time.Time]*sync.WaitGroup
+}
+
+func newFakeClock() *fakeClock {
+ return &fakeClock{
+ clock: clockwork.NewFakeClock(),
+ times: &timeHeap{},
+ waitGroups: make(map[time.Time]*sync.WaitGroup),
+ }
+}
+
+var _ tcpip.Clock = (*fakeClock)(nil)
+
+// NowNanoseconds implements tcpip.Clock.NowNanoseconds.
+func (fc *fakeClock) NowNanoseconds() int64 {
+ return fc.clock.Now().UnixNano()
+}
+
+// NowMonotonic implements tcpip.Clock.NowMonotonic.
+func (fc *fakeClock) NowMonotonic() int64 {
+ return fc.NowNanoseconds()
+}
+
+// AfterFunc implements tcpip.Clock.AfterFunc.
+func (fc *fakeClock) AfterFunc(d time.Duration, f func()) tcpip.Timer {
+ until := fc.clock.Now().Add(d)
+ wg := fc.addWait(until)
+ return &fakeTimer{
+ clock: fc,
+ until: until,
+ timer: fc.clock.AfterFunc(d, func() {
+ defer wg.Done()
+ f()
+ }),
+ }
+}
+
+// addWait adds an additional wait to the WaitGroup for parallel execution of
+// all work scheduled for t. Returns a reference to the WaitGroup modified.
+func (fc *fakeClock) addWait(t time.Time) *sync.WaitGroup {
+ fc.mu.RLock()
+ wg, ok := fc.waitGroups[t]
+ fc.mu.RUnlock()
+
+ if ok {
+ wg.Add(1)
+ return wg
+ }
+
+ fc.mu.Lock()
+ heap.Push(fc.times, t)
+ fc.mu.Unlock()
+
+ wg = &sync.WaitGroup{}
+ wg.Add(1)
+
+ fc.mu.Lock()
+ fc.waitGroups[t] = wg
+ fc.mu.Unlock()
+
+ return wg
+}
+
+// removeWait removes a wait from the WaitGroup for parallel execution of all
+// work scheduled for t.
+func (fc *fakeClock) removeWait(t time.Time) {
+ fc.mu.RLock()
+ defer fc.mu.RUnlock()
+
+ wg := fc.waitGroups[t]
+ wg.Done()
+}
+
+// advance executes all work that have been scheduled to execute within d from
+// the current fake time. Blocks until all work has completed execution.
+func (fc *fakeClock) advance(d time.Duration) {
+ // Block until all the work is done
+ until := fc.clock.Now().Add(d)
+ for {
+ fc.mu.Lock()
+ if fc.times.Len() == 0 {
+ fc.mu.Unlock()
+ return
+ }
+
+ t := heap.Pop(fc.times).(time.Time)
+ if t.After(until) {
+ // No work to do
+ heap.Push(fc.times, t)
+ fc.mu.Unlock()
+ return
+ }
+ fc.mu.Unlock()
+
+ diff := t.Sub(fc.clock.Now())
+ fc.clock.Advance(diff)
+
+ fc.mu.RLock()
+ wg := fc.waitGroups[t]
+ fc.mu.RUnlock()
+
+ wg.Wait()
+
+ fc.mu.Lock()
+ delete(fc.waitGroups, t)
+ fc.mu.Unlock()
+ }
+}
+
+type fakeTimer struct {
+ clock *fakeClock
+ timer clockwork.Timer
+
+ mu sync.RWMutex
+ until time.Time
+}
+
+var _ tcpip.Timer = (*fakeTimer)(nil)
+
+// Reset implements tcpip.Timer.Reset.
+func (ft *fakeTimer) Reset(d time.Duration) {
+ if !ft.timer.Reset(d) {
+ return
+ }
+
+ ft.mu.Lock()
+ defer ft.mu.Unlock()
+
+ ft.clock.removeWait(ft.until)
+ ft.until = ft.clock.clock.Now().Add(d)
+ ft.clock.addWait(ft.until)
+}
+
+// Stop implements tcpip.Timer.Stop.
+func (ft *fakeTimer) Stop() bool {
+ if !ft.timer.Stop() {
+ return false
+ }
+
+ ft.mu.RLock()
+ defer ft.mu.RUnlock()
+
+ ft.clock.removeWait(ft.until)
+ return true
+}
+
+type timeHeap []time.Time
+
+var _ heap.Interface = (*timeHeap)(nil)
+
+func (h timeHeap) Len() int {
+ return len(h)
+}
+
+func (h timeHeap) Less(i, j int) bool {
+ return h[i].Before(h[j])
+}
+
+func (h timeHeap) Swap(i, j int) {
+ h[i], h[j] = h[j], h[i]
+}
+
+func (h *timeHeap) Push(x interface{}) {
+ *h = append(*h, x.(time.Time))
+}
+
+func (h *timeHeap) Pop() interface{} {
+ last := (*h)[len(*h)-1]
+ *h = (*h)[:len(*h)-1]
+ return last
+}
diff --git a/pkg/tcpip/stack/forwarder.go b/pkg/tcpip/stack/forwarder.go
new file mode 100644
index 000000000..3eff141e6
--- /dev/null
+++ b/pkg/tcpip/stack/forwarder.go
@@ -0,0 +1,131 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package stack
+
+import (
+ "fmt"
+
+ "gvisor.dev/gvisor/pkg/sync"
+ "gvisor.dev/gvisor/pkg/tcpip"
+)
+
+const (
+ // maxPendingResolutions is the maximum number of pending link-address
+ // resolutions.
+ maxPendingResolutions = 64
+ maxPendingPacketsPerResolution = 256
+)
+
+type pendingPacket struct {
+ nic *NIC
+ route *Route
+ proto tcpip.NetworkProtocolNumber
+ pkt *PacketBuffer
+}
+
+type forwardQueue struct {
+ sync.Mutex
+
+ // The packets to send once the resolver completes.
+ packets map[<-chan struct{}][]*pendingPacket
+
+ // FIFO of channels used to cancel the oldest goroutine waiting for
+ // link-address resolution.
+ cancelChans []chan struct{}
+}
+
+func newForwardQueue() *forwardQueue {
+ return &forwardQueue{packets: make(map[<-chan struct{}][]*pendingPacket)}
+}
+
+func (f *forwardQueue) enqueue(ch <-chan struct{}, n *NIC, r *Route, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) {
+ shouldWait := false
+
+ f.Lock()
+ packets, ok := f.packets[ch]
+ if !ok {
+ shouldWait = true
+ }
+ for len(packets) == maxPendingPacketsPerResolution {
+ p := packets[0]
+ packets = packets[1:]
+ p.nic.stack.stats.IP.OutgoingPacketErrors.Increment()
+ p.route.Release()
+ }
+ if l := len(packets); l >= maxPendingPacketsPerResolution {
+ panic(fmt.Sprintf("max pending packets for resolution reached; got %d packets, max = %d", l, maxPendingPacketsPerResolution))
+ }
+ f.packets[ch] = append(packets, &pendingPacket{
+ nic: n,
+ route: r,
+ proto: protocol,
+ pkt: pkt,
+ })
+ f.Unlock()
+
+ if !shouldWait {
+ return
+ }
+
+ // Wait for the link-address resolution to complete.
+ // Start a goroutine with a forwarding-cancel channel so that we can
+ // limit the maximum number of goroutines running concurrently.
+ cancel := f.newCancelChannel()
+ go func() {
+ cancelled := false
+ select {
+ case <-ch:
+ case <-cancel:
+ cancelled = true
+ }
+
+ f.Lock()
+ packets := f.packets[ch]
+ delete(f.packets, ch)
+ f.Unlock()
+
+ for _, p := range packets {
+ if cancelled {
+ p.nic.stack.stats.IP.OutgoingPacketErrors.Increment()
+ } else if _, err := p.route.Resolve(nil); err != nil {
+ p.nic.stack.stats.IP.OutgoingPacketErrors.Increment()
+ } else {
+ p.nic.forwardPacket(p.route, p.proto, p.pkt)
+ }
+ p.route.Release()
+ }
+ }()
+}
+
+// newCancelChannel creates a channel that can cancel a pending forwarding
+// activity. The oldest channel is closed if the number of open channels would
+// exceed maxPendingResolutions.
+func (f *forwardQueue) newCancelChannel() chan struct{} {
+ f.Lock()
+ defer f.Unlock()
+
+ if len(f.cancelChans) == maxPendingResolutions {
+ ch := f.cancelChans[0]
+ f.cancelChans = f.cancelChans[1:]
+ close(ch)
+ }
+ if l := len(f.cancelChans); l >= maxPendingResolutions {
+ panic(fmt.Sprintf("max pending resolutions reached; got %d active resolutions, max = %d", l, maxPendingResolutions))
+ }
+
+ ch := make(chan struct{})
+ f.cancelChans = append(f.cancelChans, ch)
+ return ch
+}
diff --git a/pkg/tcpip/stack/forwarder_test.go b/pkg/tcpip/stack/forwarder_test.go
new file mode 100644
index 000000000..9dff23623
--- /dev/null
+++ b/pkg/tcpip/stack/forwarder_test.go
@@ -0,0 +1,648 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package stack
+
+import (
+ "encoding/binary"
+ "testing"
+ "time"
+
+ "gvisor.dev/gvisor/pkg/tcpip"
+ "gvisor.dev/gvisor/pkg/tcpip/buffer"
+ "gvisor.dev/gvisor/pkg/tcpip/header"
+)
+
+const (
+ fwdTestNetHeaderLen = 12
+ fwdTestNetDefaultPrefixLen = 8
+
+ // fwdTestNetDefaultMTU is the MTU, in bytes, used throughout the tests,
+ // except where another value is explicitly used. It is chosen to match
+ // the MTU of loopback interfaces on linux systems.
+ fwdTestNetDefaultMTU = 65536
+
+ dstAddrOffset = 0
+ srcAddrOffset = 1
+ protocolNumberOffset = 2
+)
+
+// fwdTestNetworkEndpoint is a network-layer protocol endpoint.
+// Headers of this protocol are fwdTestNetHeaderLen bytes, but we currently only
+// use the first three: destination address, source address, and transport
+// protocol. They're all one byte fields to simplify parsing.
+type fwdTestNetworkEndpoint struct {
+ nicID tcpip.NICID
+ proto *fwdTestNetworkProtocol
+ dispatcher TransportDispatcher
+ ep LinkEndpoint
+}
+
+func (f *fwdTestNetworkEndpoint) MTU() uint32 {
+ return f.ep.MTU() - uint32(f.MaxHeaderLength())
+}
+
+func (f *fwdTestNetworkEndpoint) NICID() tcpip.NICID {
+ return f.nicID
+}
+
+func (*fwdTestNetworkEndpoint) DefaultTTL() uint8 {
+ return 123
+}
+
+func (f *fwdTestNetworkEndpoint) HandlePacket(r *Route, pkt *PacketBuffer) {
+ // Dispatch the packet to the transport protocol.
+ f.dispatcher.DeliverTransportPacket(r, tcpip.TransportProtocolNumber(pkt.NetworkHeader().View()[protocolNumberOffset]), pkt)
+}
+
+func (f *fwdTestNetworkEndpoint) MaxHeaderLength() uint16 {
+ return f.ep.MaxHeaderLength() + fwdTestNetHeaderLen
+}
+
+func (f *fwdTestNetworkEndpoint) PseudoHeaderChecksum(protocol tcpip.TransportProtocolNumber, dstAddr tcpip.Address) uint16 {
+ return 0
+}
+
+func (f *fwdTestNetworkEndpoint) Capabilities() LinkEndpointCapabilities {
+ return f.ep.Capabilities()
+}
+
+func (f *fwdTestNetworkEndpoint) NetworkProtocolNumber() tcpip.NetworkProtocolNumber {
+ return f.proto.Number()
+}
+
+func (f *fwdTestNetworkEndpoint) WritePacket(r *Route, gso *GSO, params NetworkHeaderParams, pkt *PacketBuffer) *tcpip.Error {
+ // Add the protocol's header to the packet and send it to the link
+ // endpoint.
+ b := pkt.NetworkHeader().Push(fwdTestNetHeaderLen)
+ b[dstAddrOffset] = r.RemoteAddress[0]
+ b[srcAddrOffset] = r.LocalAddress[0]
+ b[protocolNumberOffset] = byte(params.Protocol)
+
+ return f.ep.WritePacket(r, gso, fakeNetNumber, pkt)
+}
+
+// WritePackets implements LinkEndpoint.WritePackets.
+func (f *fwdTestNetworkEndpoint) WritePackets(r *Route, gso *GSO, pkts PacketBufferList, params NetworkHeaderParams) (int, *tcpip.Error) {
+ panic("not implemented")
+}
+
+func (*fwdTestNetworkEndpoint) WriteHeaderIncludedPacket(r *Route, pkt *PacketBuffer) *tcpip.Error {
+ return tcpip.ErrNotSupported
+}
+
+func (*fwdTestNetworkEndpoint) Close() {}
+
+// fwdTestNetworkProtocol is a network-layer protocol that implements Address
+// resolution.
+type fwdTestNetworkProtocol struct {
+ addrCache *linkAddrCache
+ addrResolveDelay time.Duration
+ onLinkAddressResolved func(cache *linkAddrCache, addr tcpip.Address, _ tcpip.LinkAddress)
+ onResolveStaticAddress func(tcpip.Address) (tcpip.LinkAddress, bool)
+}
+
+var _ LinkAddressResolver = (*fwdTestNetworkProtocol)(nil)
+
+func (f *fwdTestNetworkProtocol) Number() tcpip.NetworkProtocolNumber {
+ return fakeNetNumber
+}
+
+func (f *fwdTestNetworkProtocol) MinimumPacketSize() int {
+ return fwdTestNetHeaderLen
+}
+
+func (f *fwdTestNetworkProtocol) DefaultPrefixLen() int {
+ return fwdTestNetDefaultPrefixLen
+}
+
+func (*fwdTestNetworkProtocol) ParseAddresses(v buffer.View) (src, dst tcpip.Address) {
+ return tcpip.Address(v[srcAddrOffset : srcAddrOffset+1]), tcpip.Address(v[dstAddrOffset : dstAddrOffset+1])
+}
+
+func (*fwdTestNetworkProtocol) Parse(pkt *PacketBuffer) (tcpip.TransportProtocolNumber, bool, bool) {
+ netHeader, ok := pkt.NetworkHeader().Consume(fwdTestNetHeaderLen)
+ if !ok {
+ return 0, false, false
+ }
+ return tcpip.TransportProtocolNumber(netHeader[protocolNumberOffset]), true, true
+}
+
+func (f *fwdTestNetworkProtocol) NewEndpoint(nicID tcpip.NICID, linkAddrCache LinkAddressCache, dispatcher TransportDispatcher, ep LinkEndpoint, _ *Stack) NetworkEndpoint {
+ return &fwdTestNetworkEndpoint{
+ nicID: nicID,
+ proto: f,
+ dispatcher: dispatcher,
+ ep: ep,
+ }
+}
+
+func (f *fwdTestNetworkProtocol) SetOption(option interface{}) *tcpip.Error {
+ return tcpip.ErrUnknownProtocolOption
+}
+
+func (f *fwdTestNetworkProtocol) Option(option interface{}) *tcpip.Error {
+ return tcpip.ErrUnknownProtocolOption
+}
+
+func (f *fwdTestNetworkProtocol) Close() {}
+
+func (f *fwdTestNetworkProtocol) Wait() {}
+
+func (f *fwdTestNetworkProtocol) LinkAddressRequest(addr, localAddr tcpip.Address, remoteLinkAddr tcpip.LinkAddress, linkEP LinkEndpoint) *tcpip.Error {
+ if f.addrCache != nil && f.onLinkAddressResolved != nil {
+ time.AfterFunc(f.addrResolveDelay, func() {
+ f.onLinkAddressResolved(f.addrCache, addr, remoteLinkAddr)
+ })
+ }
+ return nil
+}
+
+func (f *fwdTestNetworkProtocol) ResolveStaticAddress(addr tcpip.Address) (tcpip.LinkAddress, bool) {
+ if f.onResolveStaticAddress != nil {
+ return f.onResolveStaticAddress(addr)
+ }
+ return "", false
+}
+
+func (f *fwdTestNetworkProtocol) LinkAddressProtocol() tcpip.NetworkProtocolNumber {
+ return fakeNetNumber
+}
+
+// fwdTestPacketInfo holds all the information about an outbound packet.
+type fwdTestPacketInfo struct {
+ RemoteLinkAddress tcpip.LinkAddress
+ LocalLinkAddress tcpip.LinkAddress
+ Pkt *PacketBuffer
+}
+
+type fwdTestLinkEndpoint struct {
+ dispatcher NetworkDispatcher
+ mtu uint32
+ linkAddr tcpip.LinkAddress
+
+ // C is where outbound packets are queued.
+ C chan fwdTestPacketInfo
+}
+
+// InjectInbound injects an inbound packet.
+func (e *fwdTestLinkEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) {
+ e.InjectLinkAddr(protocol, "", pkt)
+}
+
+// InjectLinkAddr injects an inbound packet with a remote link address.
+func (e *fwdTestLinkEndpoint) InjectLinkAddr(protocol tcpip.NetworkProtocolNumber, remote tcpip.LinkAddress, pkt *PacketBuffer) {
+ e.dispatcher.DeliverNetworkPacket(remote, "" /* local */, protocol, pkt)
+}
+
+// Attach saves the stack network-layer dispatcher for use later when packets
+// are injected.
+func (e *fwdTestLinkEndpoint) Attach(dispatcher NetworkDispatcher) {
+ e.dispatcher = dispatcher
+}
+
+// IsAttached implements stack.LinkEndpoint.IsAttached.
+func (e *fwdTestLinkEndpoint) IsAttached() bool {
+ return e.dispatcher != nil
+}
+
+// MTU implements stack.LinkEndpoint.MTU. It returns the value initialized
+// during construction.
+func (e *fwdTestLinkEndpoint) MTU() uint32 {
+ return e.mtu
+}
+
+// Capabilities implements stack.LinkEndpoint.Capabilities.
+func (e fwdTestLinkEndpoint) Capabilities() LinkEndpointCapabilities {
+ caps := LinkEndpointCapabilities(0)
+ return caps | CapabilityResolutionRequired
+}
+
+// GSOMaxSize returns the maximum GSO packet size.
+func (*fwdTestLinkEndpoint) GSOMaxSize() uint32 {
+ return 1 << 15
+}
+
+// MaxHeaderLength returns the maximum size of the link layer header. Given it
+// doesn't have a header, it just returns 0.
+func (*fwdTestLinkEndpoint) MaxHeaderLength() uint16 {
+ return 0
+}
+
+// LinkAddress returns the link address of this endpoint.
+func (e *fwdTestLinkEndpoint) LinkAddress() tcpip.LinkAddress {
+ return e.linkAddr
+}
+
+func (e fwdTestLinkEndpoint) WritePacket(r *Route, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) *tcpip.Error {
+ p := fwdTestPacketInfo{
+ RemoteLinkAddress: r.RemoteLinkAddress,
+ LocalLinkAddress: r.LocalLinkAddress,
+ Pkt: pkt,
+ }
+
+ select {
+ case e.C <- p:
+ default:
+ }
+
+ return nil
+}
+
+// WritePackets stores outbound packets into the channel.
+func (e *fwdTestLinkEndpoint) WritePackets(r *Route, gso *GSO, pkts PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) {
+ n := 0
+ for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() {
+ e.WritePacket(r, gso, protocol, pkt)
+ n++
+ }
+
+ return n, nil
+}
+
+// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket.
+func (e *fwdTestLinkEndpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error {
+ p := fwdTestPacketInfo{
+ Pkt: NewPacketBuffer(PacketBufferOptions{Data: vv}),
+ }
+
+ select {
+ case e.C <- p:
+ default:
+ }
+
+ return nil
+}
+
+// Wait implements stack.LinkEndpoint.Wait.
+func (*fwdTestLinkEndpoint) Wait() {}
+
+// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType.
+func (*fwdTestLinkEndpoint) ARPHardwareType() header.ARPHardwareType {
+ panic("not implemented")
+}
+
+// AddHeader implements stack.LinkEndpoint.AddHeader.
+func (e *fwdTestLinkEndpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) {
+ panic("not implemented")
+}
+
+func fwdTestNetFactory(t *testing.T, proto *fwdTestNetworkProtocol) (ep1, ep2 *fwdTestLinkEndpoint) {
+ // Create a stack with the network protocol and two NICs.
+ s := New(Options{
+ NetworkProtocols: []NetworkProtocol{proto},
+ })
+
+ proto.addrCache = s.linkAddrCache
+
+ // Enable forwarding.
+ s.SetForwarding(proto.Number(), true)
+
+ // NIC 1 has the link address "a", and added the network address 1.
+ ep1 = &fwdTestLinkEndpoint{
+ C: make(chan fwdTestPacketInfo, 300),
+ mtu: fwdTestNetDefaultMTU,
+ linkAddr: "a",
+ }
+ if err := s.CreateNIC(1, ep1); err != nil {
+ t.Fatal("CreateNIC #1 failed:", err)
+ }
+ if err := s.AddAddress(1, fakeNetNumber, "\x01"); err != nil {
+ t.Fatal("AddAddress #1 failed:", err)
+ }
+
+ // NIC 2 has the link address "b", and added the network address 2.
+ ep2 = &fwdTestLinkEndpoint{
+ C: make(chan fwdTestPacketInfo, 300),
+ mtu: fwdTestNetDefaultMTU,
+ linkAddr: "b",
+ }
+ if err := s.CreateNIC(2, ep2); err != nil {
+ t.Fatal("CreateNIC #2 failed:", err)
+ }
+ if err := s.AddAddress(2, fakeNetNumber, "\x02"); err != nil {
+ t.Fatal("AddAddress #2 failed:", err)
+ }
+
+ // Route all packets to NIC 2.
+ {
+ subnet, err := tcpip.NewSubnet("\x00", "\x00")
+ if err != nil {
+ t.Fatal(err)
+ }
+ s.SetRouteTable([]tcpip.Route{{Destination: subnet, NIC: 2}})
+ }
+
+ return ep1, ep2
+}
+
+func TestForwardingWithStaticResolver(t *testing.T) {
+ // Create a network protocol with a static resolver.
+ proto := &fwdTestNetworkProtocol{
+ onResolveStaticAddress:
+ // The network address 3 is resolved to the link address "c".
+ func(addr tcpip.Address) (tcpip.LinkAddress, bool) {
+ if addr == "\x03" {
+ return "c", true
+ }
+ return "", false
+ },
+ }
+
+ ep1, ep2 := fwdTestNetFactory(t, proto)
+
+ // Inject an inbound packet to address 3 on NIC 1, and see if it is
+ // forwarded to NIC 2.
+ buf := buffer.NewView(30)
+ buf[dstAddrOffset] = 3
+ ep1.InjectInbound(fakeNetNumber, NewPacketBuffer(PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ }))
+
+ var p fwdTestPacketInfo
+
+ select {
+ case p = <-ep2.C:
+ default:
+ t.Fatal("packet not forwarded")
+ }
+
+ // Test that the static address resolution happened correctly.
+ if p.RemoteLinkAddress != "c" {
+ t.Fatalf("got p.RemoteLinkAddress = %s, want = c", p.RemoteLinkAddress)
+ }
+ if p.LocalLinkAddress != "b" {
+ t.Fatalf("got p.LocalLinkAddress = %s, want = b", p.LocalLinkAddress)
+ }
+}
+
+func TestForwardingWithFakeResolver(t *testing.T) {
+ // Create a network protocol with a fake resolver.
+ proto := &fwdTestNetworkProtocol{
+ addrResolveDelay: 500 * time.Millisecond,
+ onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address, _ tcpip.LinkAddress) {
+ // Any address will be resolved to the link address "c".
+ cache.add(tcpip.FullAddress{NIC: 2, Addr: addr}, "c")
+ },
+ }
+
+ ep1, ep2 := fwdTestNetFactory(t, proto)
+
+ // Inject an inbound packet to address 3 on NIC 1, and see if it is
+ // forwarded to NIC 2.
+ buf := buffer.NewView(30)
+ buf[dstAddrOffset] = 3
+ ep1.InjectInbound(fakeNetNumber, NewPacketBuffer(PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ }))
+
+ var p fwdTestPacketInfo
+
+ select {
+ case p = <-ep2.C:
+ case <-time.After(time.Second):
+ t.Fatal("packet not forwarded")
+ }
+
+ // Test that the address resolution happened correctly.
+ if p.RemoteLinkAddress != "c" {
+ t.Fatalf("got p.RemoteLinkAddress = %s, want = c", p.RemoteLinkAddress)
+ }
+ if p.LocalLinkAddress != "b" {
+ t.Fatalf("got p.LocalLinkAddress = %s, want = b", p.LocalLinkAddress)
+ }
+}
+
+func TestForwardingWithNoResolver(t *testing.T) {
+ // Create a network protocol without a resolver.
+ proto := &fwdTestNetworkProtocol{}
+
+ ep1, ep2 := fwdTestNetFactory(t, proto)
+
+ // inject an inbound packet to address 3 on NIC 1, and see if it is
+ // forwarded to NIC 2.
+ buf := buffer.NewView(30)
+ buf[dstAddrOffset] = 3
+ ep1.InjectInbound(fakeNetNumber, NewPacketBuffer(PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ }))
+
+ select {
+ case <-ep2.C:
+ t.Fatal("Packet should not be forwarded")
+ case <-time.After(time.Second):
+ }
+}
+
+func TestForwardingWithFakeResolverPartialTimeout(t *testing.T) {
+ // Create a network protocol with a fake resolver.
+ proto := &fwdTestNetworkProtocol{
+ addrResolveDelay: 500 * time.Millisecond,
+ onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address, _ tcpip.LinkAddress) {
+ // Only packets to address 3 will be resolved to the
+ // link address "c".
+ if addr == "\x03" {
+ cache.add(tcpip.FullAddress{NIC: 2, Addr: addr}, "c")
+ }
+ },
+ }
+
+ ep1, ep2 := fwdTestNetFactory(t, proto)
+
+ // Inject an inbound packet to address 4 on NIC 1. This packet should
+ // not be forwarded.
+ buf := buffer.NewView(30)
+ buf[dstAddrOffset] = 4
+ ep1.InjectInbound(fakeNetNumber, NewPacketBuffer(PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ }))
+
+ // Inject an inbound packet to address 3 on NIC 1, and see if it is
+ // forwarded to NIC 2.
+ buf = buffer.NewView(30)
+ buf[dstAddrOffset] = 3
+ ep1.InjectInbound(fakeNetNumber, NewPacketBuffer(PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ }))
+
+ var p fwdTestPacketInfo
+
+ select {
+ case p = <-ep2.C:
+ case <-time.After(time.Second):
+ t.Fatal("packet not forwarded")
+ }
+
+ if nh := PayloadSince(p.Pkt.NetworkHeader()); nh[dstAddrOffset] != 3 {
+ t.Fatalf("got p.Pkt.NetworkHeader[dstAddrOffset] = %d, want = 3", nh[dstAddrOffset])
+ }
+
+ // Test that the address resolution happened correctly.
+ if p.RemoteLinkAddress != "c" {
+ t.Fatalf("got p.RemoteLinkAddress = %s, want = c", p.RemoteLinkAddress)
+ }
+ if p.LocalLinkAddress != "b" {
+ t.Fatalf("got p.LocalLinkAddress = %s, want = b", p.LocalLinkAddress)
+ }
+}
+
+func TestForwardingWithFakeResolverTwoPackets(t *testing.T) {
+ // Create a network protocol with a fake resolver.
+ proto := &fwdTestNetworkProtocol{
+ addrResolveDelay: 500 * time.Millisecond,
+ onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address, _ tcpip.LinkAddress) {
+ // Any packets will be resolved to the link address "c".
+ cache.add(tcpip.FullAddress{NIC: 2, Addr: addr}, "c")
+ },
+ }
+
+ ep1, ep2 := fwdTestNetFactory(t, proto)
+
+ // Inject two inbound packets to address 3 on NIC 1.
+ for i := 0; i < 2; i++ {
+ buf := buffer.NewView(30)
+ buf[dstAddrOffset] = 3
+ ep1.InjectInbound(fakeNetNumber, NewPacketBuffer(PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ }))
+ }
+
+ for i := 0; i < 2; i++ {
+ var p fwdTestPacketInfo
+
+ select {
+ case p = <-ep2.C:
+ case <-time.After(time.Second):
+ t.Fatal("packet not forwarded")
+ }
+
+ if nh := PayloadSince(p.Pkt.NetworkHeader()); nh[dstAddrOffset] != 3 {
+ t.Fatalf("got p.Pkt.NetworkHeader[dstAddrOffset] = %d, want = 3", nh[dstAddrOffset])
+ }
+
+ // Test that the address resolution happened correctly.
+ if p.RemoteLinkAddress != "c" {
+ t.Fatalf("got p.RemoteLinkAddress = %s, want = c", p.RemoteLinkAddress)
+ }
+ if p.LocalLinkAddress != "b" {
+ t.Fatalf("got p.LocalLinkAddress = %s, want = b", p.LocalLinkAddress)
+ }
+ }
+}
+
+func TestForwardingWithFakeResolverManyPackets(t *testing.T) {
+ // Create a network protocol with a fake resolver.
+ proto := &fwdTestNetworkProtocol{
+ addrResolveDelay: 500 * time.Millisecond,
+ onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address, _ tcpip.LinkAddress) {
+ // Any packets will be resolved to the link address "c".
+ cache.add(tcpip.FullAddress{NIC: 2, Addr: addr}, "c")
+ },
+ }
+
+ ep1, ep2 := fwdTestNetFactory(t, proto)
+
+ for i := 0; i < maxPendingPacketsPerResolution+5; i++ {
+ // Inject inbound 'maxPendingPacketsPerResolution + 5' packets on NIC 1.
+ buf := buffer.NewView(30)
+ buf[dstAddrOffset] = 3
+ // Set the packet sequence number.
+ binary.BigEndian.PutUint16(buf[fwdTestNetHeaderLen:], uint16(i))
+ ep1.InjectInbound(fakeNetNumber, NewPacketBuffer(PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ }))
+ }
+
+ for i := 0; i < maxPendingPacketsPerResolution; i++ {
+ var p fwdTestPacketInfo
+
+ select {
+ case p = <-ep2.C:
+ case <-time.After(time.Second):
+ t.Fatal("packet not forwarded")
+ }
+
+ b := PayloadSince(p.Pkt.NetworkHeader())
+ if b[dstAddrOffset] != 3 {
+ t.Fatalf("got b[dstAddrOffset] = %d, want = 3", b[dstAddrOffset])
+ }
+ if len(b) < fwdTestNetHeaderLen+2 {
+ t.Fatalf("packet is too short to hold a sequence number: len(b) = %d", b)
+ }
+ seqNumBuf := b[fwdTestNetHeaderLen:]
+
+ // The first 5 packets should not be forwarded so the sequence number should
+ // start with 5.
+ want := uint16(i + 5)
+ if n := binary.BigEndian.Uint16(seqNumBuf); n != want {
+ t.Fatalf("got the packet #%d, want = #%d", n, want)
+ }
+
+ // Test that the address resolution happened correctly.
+ if p.RemoteLinkAddress != "c" {
+ t.Fatalf("got p.RemoteLinkAddress = %s, want = c", p.RemoteLinkAddress)
+ }
+ if p.LocalLinkAddress != "b" {
+ t.Fatalf("got p.LocalLinkAddress = %s, want = b", p.LocalLinkAddress)
+ }
+ }
+}
+
+func TestForwardingWithFakeResolverManyResolutions(t *testing.T) {
+ // Create a network protocol with a fake resolver.
+ proto := &fwdTestNetworkProtocol{
+ addrResolveDelay: 500 * time.Millisecond,
+ onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address, _ tcpip.LinkAddress) {
+ // Any packets will be resolved to the link address "c".
+ cache.add(tcpip.FullAddress{NIC: 2, Addr: addr}, "c")
+ },
+ }
+
+ ep1, ep2 := fwdTestNetFactory(t, proto)
+
+ for i := 0; i < maxPendingResolutions+5; i++ {
+ // Inject inbound 'maxPendingResolutions + 5' packets on NIC 1.
+ // Each packet has a different destination address (3 to
+ // maxPendingResolutions + 7).
+ buf := buffer.NewView(30)
+ buf[dstAddrOffset] = byte(3 + i)
+ ep1.InjectInbound(fakeNetNumber, NewPacketBuffer(PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ }))
+ }
+
+ for i := 0; i < maxPendingResolutions; i++ {
+ var p fwdTestPacketInfo
+
+ select {
+ case p = <-ep2.C:
+ case <-time.After(time.Second):
+ t.Fatal("packet not forwarded")
+ }
+
+ // The first 5 packets (address 3 to 7) should not be forwarded
+ // because their address resolutions are interrupted.
+ if nh := PayloadSince(p.Pkt.NetworkHeader()); nh[dstAddrOffset] < 8 {
+ t.Fatalf("got p.Pkt.NetworkHeader[dstAddrOffset] = %d, want p.Pkt.NetworkHeader[dstAddrOffset] >= 8", nh[dstAddrOffset])
+ }
+
+ // Test that the address resolution happened correctly.
+ if p.RemoteLinkAddress != "c" {
+ t.Fatalf("got p.RemoteLinkAddress = %s, want = c", p.RemoteLinkAddress)
+ }
+ if p.LocalLinkAddress != "b" {
+ t.Fatalf("got p.LocalLinkAddress = %s, want = b", p.LocalLinkAddress)
+ }
+ }
+}
diff --git a/pkg/tcpip/stack/headertype_string.go b/pkg/tcpip/stack/headertype_string.go
new file mode 100644
index 000000000..5efddfaaf
--- /dev/null
+++ b/pkg/tcpip/stack/headertype_string.go
@@ -0,0 +1,39 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at //
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Code generated by "stringer -type headerType ."; DO NOT EDIT.
+
+package stack
+
+import "strconv"
+
+func _() {
+ // An "invalid array index" compiler error signifies that the constant values have changed.
+ // Re-run the stringer command to generate them again.
+ var x [1]struct{}
+ _ = x[linkHeader-0]
+ _ = x[networkHeader-1]
+ _ = x[transportHeader-2]
+ _ = x[numHeaderType-3]
+}
+
+const _headerType_name = "linkHeadernetworkHeadertransportHeadernumHeaderType"
+
+var _headerType_index = [...]uint8{0, 10, 23, 38, 51}
+
+func (i headerType) String() string {
+ if i < 0 || i >= headerType(len(_headerType_index)-1) {
+ return "headerType(" + strconv.FormatInt(int64(i), 10) + ")"
+ }
+ return _headerType_name[_headerType_index[i]:_headerType_index[i+1]]
+}
diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go
new file mode 100644
index 000000000..c37da814f
--- /dev/null
+++ b/pkg/tcpip/stack/iptables.go
@@ -0,0 +1,423 @@
+// Copyright 2019 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package stack
+
+import (
+ "fmt"
+ "time"
+
+ "gvisor.dev/gvisor/pkg/tcpip"
+ "gvisor.dev/gvisor/pkg/tcpip/header"
+)
+
+// tableID is an index into IPTables.tables.
+type tableID int
+
+const (
+ natID tableID = iota
+ mangleID
+ filterID
+ numTables
+)
+
+// Table names.
+const (
+ NATTable = "nat"
+ MangleTable = "mangle"
+ FilterTable = "filter"
+)
+
+// nameToID is immutable.
+var nameToID = map[string]tableID{
+ NATTable: natID,
+ MangleTable: mangleID,
+ FilterTable: filterID,
+}
+
+// HookUnset indicates that there is no hook set for an entrypoint or
+// underflow.
+const HookUnset = -1
+
+// reaperDelay is how long to wait before starting to reap connections.
+const reaperDelay = 5 * time.Second
+
+// DefaultTables returns a default set of tables. Each chain is set to accept
+// all packets.
+func DefaultTables() *IPTables {
+ return &IPTables{
+ tables: [numTables]Table{
+ natID: Table{
+ Rules: []Rule{
+ Rule{Target: AcceptTarget{}},
+ Rule{Target: AcceptTarget{}},
+ Rule{Target: AcceptTarget{}},
+ Rule{Target: AcceptTarget{}},
+ Rule{Target: ErrorTarget{}},
+ },
+ BuiltinChains: [NumHooks]int{
+ Prerouting: 0,
+ Input: 1,
+ Forward: HookUnset,
+ Output: 2,
+ Postrouting: 3,
+ },
+ Underflows: [NumHooks]int{
+ Prerouting: 0,
+ Input: 1,
+ Forward: HookUnset,
+ Output: 2,
+ Postrouting: 3,
+ },
+ },
+ mangleID: Table{
+ Rules: []Rule{
+ Rule{Target: AcceptTarget{}},
+ Rule{Target: AcceptTarget{}},
+ Rule{Target: ErrorTarget{}},
+ },
+ BuiltinChains: [NumHooks]int{
+ Prerouting: 0,
+ Output: 1,
+ },
+ Underflows: [NumHooks]int{
+ Prerouting: 0,
+ Input: HookUnset,
+ Forward: HookUnset,
+ Output: 1,
+ Postrouting: HookUnset,
+ },
+ },
+ filterID: Table{
+ Rules: []Rule{
+ Rule{Target: AcceptTarget{}},
+ Rule{Target: AcceptTarget{}},
+ Rule{Target: AcceptTarget{}},
+ Rule{Target: ErrorTarget{}},
+ },
+ BuiltinChains: [NumHooks]int{
+ Prerouting: HookUnset,
+ Input: 0,
+ Forward: 1,
+ Output: 2,
+ Postrouting: HookUnset,
+ },
+ Underflows: [NumHooks]int{
+ Prerouting: HookUnset,
+ Input: 0,
+ Forward: 1,
+ Output: 2,
+ Postrouting: HookUnset,
+ },
+ },
+ },
+ priorities: [NumHooks][]tableID{
+ Prerouting: []tableID{mangleID, natID},
+ Input: []tableID{natID, filterID},
+ Output: []tableID{mangleID, natID, filterID},
+ },
+ connections: ConnTrack{
+ seed: generateRandUint32(),
+ },
+ reaperDone: make(chan struct{}, 1),
+ }
+}
+
+// EmptyFilterTable returns a Table with no rules and the filter table chains
+// mapped to HookUnset.
+func EmptyFilterTable() Table {
+ return Table{
+ Rules: []Rule{},
+ BuiltinChains: [NumHooks]int{
+ Prerouting: HookUnset,
+ Postrouting: HookUnset,
+ },
+ Underflows: [NumHooks]int{
+ Prerouting: HookUnset,
+ Postrouting: HookUnset,
+ },
+ }
+}
+
+// EmptyNATTable returns a Table with no rules and the filter table chains
+// mapped to HookUnset.
+func EmptyNATTable() Table {
+ return Table{
+ Rules: []Rule{},
+ BuiltinChains: [NumHooks]int{
+ Forward: HookUnset,
+ },
+ Underflows: [NumHooks]int{
+ Forward: HookUnset,
+ },
+ }
+}
+
+// GetTable returns a table by name.
+func (it *IPTables) GetTable(name string) (Table, bool) {
+ id, ok := nameToID[name]
+ if !ok {
+ return Table{}, false
+ }
+ it.mu.RLock()
+ defer it.mu.RUnlock()
+ return it.tables[id], true
+}
+
+// ReplaceTable replaces or inserts table by name.
+func (it *IPTables) ReplaceTable(name string, table Table) *tcpip.Error {
+ id, ok := nameToID[name]
+ if !ok {
+ return tcpip.ErrInvalidOptionValue
+ }
+ it.mu.Lock()
+ defer it.mu.Unlock()
+ // If iptables is being enabled, initialize the conntrack table and
+ // reaper.
+ if !it.modified {
+ it.connections.buckets = make([]bucket, numBuckets)
+ it.startReaper(reaperDelay)
+ }
+ it.modified = true
+ it.tables[id] = table
+ return nil
+}
+
+// A chainVerdict is what a table decides should be done with a packet.
+type chainVerdict int
+
+const (
+ // chainAccept indicates the packet should continue through netstack.
+ chainAccept chainVerdict = iota
+
+ // chainAccept indicates the packet should be dropped.
+ chainDrop
+
+ // chainReturn indicates the packet should return to the calling chain
+ // or the underflow rule of a builtin chain.
+ chainReturn
+)
+
+// Check runs pkt through the rules for hook. It returns true when the packet
+// should continue traversing the network stack and false when it should be
+// dropped.
+//
+// Precondition: pkt.NetworkHeader is set.
+func (it *IPTables) Check(hook Hook, pkt *PacketBuffer, gso *GSO, r *Route, address tcpip.Address, nicName string) bool {
+ // Many users never configure iptables. Spare them the cost of rule
+ // traversal if rules have never been set.
+ it.mu.RLock()
+ defer it.mu.RUnlock()
+ if !it.modified {
+ return true
+ }
+
+ // Packets are manipulated only if connection and matching
+ // NAT rule exists.
+ shouldTrack := it.connections.handlePacket(pkt, hook, gso, r)
+
+ // Go through each table containing the hook.
+ priorities := it.priorities[hook]
+ for _, tableID := range priorities {
+ // If handlePacket already NATed the packet, we don't need to
+ // check the NAT table.
+ if tableID == natID && pkt.NatDone {
+ continue
+ }
+ table := it.tables[tableID]
+ ruleIdx := table.BuiltinChains[hook]
+ switch verdict := it.checkChain(hook, pkt, table, ruleIdx, gso, r, address, nicName); verdict {
+ // If the table returns Accept, move on to the next table.
+ case chainAccept:
+ continue
+ // The Drop verdict is final.
+ case chainDrop:
+ return false
+ case chainReturn:
+ // Any Return from a built-in chain means we have to
+ // call the underflow.
+ underflow := table.Rules[table.Underflows[hook]]
+ switch v, _ := underflow.Target.Action(pkt, &it.connections, hook, gso, r, address); v {
+ case RuleAccept:
+ continue
+ case RuleDrop:
+ return false
+ case RuleJump, RuleReturn:
+ panic("Underflows should only return RuleAccept or RuleDrop.")
+ default:
+ panic(fmt.Sprintf("Unknown verdict: %d", v))
+ }
+
+ default:
+ panic(fmt.Sprintf("Unknown verdict %v.", verdict))
+ }
+ }
+
+ // If this connection should be tracked, try to add an entry for it. If
+ // traversing the nat table didn't end in adding an entry,
+ // maybeInsertNoop will add a no-op entry for the connection. This is
+ // needeed when establishing connections so that the SYN/ACK reply to an
+ // outgoing SYN is delivered to the correct endpoint rather than being
+ // redirected by a prerouting rule.
+ //
+ // From the iptables documentation: "If there is no rule, a `null'
+ // binding is created: this usually does not map the packet, but exists
+ // to ensure we don't map another stream over an existing one."
+ if shouldTrack {
+ it.connections.maybeInsertNoop(pkt, hook)
+ }
+
+ // Every table returned Accept.
+ return true
+}
+
+// beforeSave is invoked by stateify.
+func (it *IPTables) beforeSave() {
+ // Ensure the reaper exits cleanly.
+ it.reaperDone <- struct{}{}
+ // Prevent others from modifying the connection table.
+ it.connections.mu.Lock()
+}
+
+// afterLoad is invoked by stateify.
+func (it *IPTables) afterLoad() {
+ it.startReaper(reaperDelay)
+}
+
+// startReaper starts a goroutine that wakes up periodically to reap timed out
+// connections.
+func (it *IPTables) startReaper(interval time.Duration) {
+ go func() { // S/R-SAFE: reaperDone is signalled when iptables is saved.
+ bucket := 0
+ for {
+ select {
+ case <-it.reaperDone:
+ return
+ case <-time.After(interval):
+ bucket, interval = it.connections.reapUnused(bucket, interval)
+ }
+ }
+ }()
+}
+
+// CheckPackets runs pkts through the rules for hook and returns a map of packets that
+// should not go forward.
+//
+// Preconditions:
+// - pkt is a IPv4 packet of at least length header.IPv4MinimumSize.
+// - pkt.NetworkHeader is not nil.
+//
+// NOTE: unlike the Check API the returned map contains packets that should be
+// dropped.
+func (it *IPTables) CheckPackets(hook Hook, pkts PacketBufferList, gso *GSO, r *Route, nicName string) (drop map[*PacketBuffer]struct{}, natPkts map[*PacketBuffer]struct{}) {
+ for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() {
+ if !pkt.NatDone {
+ if ok := it.Check(hook, pkt, gso, r, "", nicName); !ok {
+ if drop == nil {
+ drop = make(map[*PacketBuffer]struct{})
+ }
+ drop[pkt] = struct{}{}
+ }
+ if pkt.NatDone {
+ if natPkts == nil {
+ natPkts = make(map[*PacketBuffer]struct{})
+ }
+ natPkts[pkt] = struct{}{}
+ }
+ }
+ }
+ return drop, natPkts
+}
+
+// Preconditions:
+// - pkt is a IPv4 packet of at least length header.IPv4MinimumSize.
+// - pkt.NetworkHeader is not nil.
+func (it *IPTables) checkChain(hook Hook, pkt *PacketBuffer, table Table, ruleIdx int, gso *GSO, r *Route, address tcpip.Address, nicName string) chainVerdict {
+ // Start from ruleIdx and walk the list of rules until a rule gives us
+ // a verdict.
+ for ruleIdx < len(table.Rules) {
+ switch verdict, jumpTo := it.checkRule(hook, pkt, table, ruleIdx, gso, r, address, nicName); verdict {
+ case RuleAccept:
+ return chainAccept
+
+ case RuleDrop:
+ return chainDrop
+
+ case RuleReturn:
+ return chainReturn
+
+ case RuleJump:
+ // "Jumping" to the next rule just means we're
+ // continuing on down the list.
+ if jumpTo == ruleIdx+1 {
+ ruleIdx++
+ continue
+ }
+ switch verdict := it.checkChain(hook, pkt, table, jumpTo, gso, r, address, nicName); verdict {
+ case chainAccept:
+ return chainAccept
+ case chainDrop:
+ return chainDrop
+ case chainReturn:
+ ruleIdx++
+ continue
+ default:
+ panic(fmt.Sprintf("Unknown verdict: %d", verdict))
+ }
+
+ default:
+ panic(fmt.Sprintf("Unknown verdict: %d", verdict))
+ }
+
+ }
+
+ // We got through the entire table without a decision. Default to DROP
+ // for safety.
+ return chainDrop
+}
+
+// Preconditions:
+// - pkt is a IPv4 packet of at least length header.IPv4MinimumSize.
+// - pkt.NetworkHeader is not nil.
+func (it *IPTables) checkRule(hook Hook, pkt *PacketBuffer, table Table, ruleIdx int, gso *GSO, r *Route, address tcpip.Address, nicName string) (RuleVerdict, int) {
+ rule := table.Rules[ruleIdx]
+
+ // Check whether the packet matches the IP header filter.
+ if !rule.Filter.match(header.IPv4(pkt.NetworkHeader().View()), hook, nicName) {
+ // Continue on to the next rule.
+ return RuleJump, ruleIdx + 1
+ }
+
+ // Go through each rule matcher. If they all match, run
+ // the rule target.
+ for _, matcher := range rule.Matchers {
+ matches, hotdrop := matcher.Match(hook, pkt, "")
+ if hotdrop {
+ return RuleDrop, 0
+ }
+ if !matches {
+ // Continue on to the next rule.
+ return RuleJump, ruleIdx + 1
+ }
+ }
+
+ // All the matchers matched, so run the target.
+ return rule.Target.Action(pkt, &it.connections, hook, gso, r, address)
+}
+
+// OriginalDst returns the original destination of redirected connections. It
+// returns an error if the connection doesn't exist or isn't redirected.
+func (it *IPTables) OriginalDst(epID TransportEndpointID) (tcpip.Address, uint16, *tcpip.Error) {
+ return it.connections.originalDst(epID)
+}
diff --git a/pkg/tcpip/stack/iptables_state.go b/pkg/tcpip/stack/iptables_state.go
new file mode 100644
index 000000000..529e02a07
--- /dev/null
+++ b/pkg/tcpip/stack/iptables_state.go
@@ -0,0 +1,40 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package stack
+
+import (
+ "time"
+)
+
+// +stateify savable
+type unixTime struct {
+ second int64
+ nano int64
+}
+
+// saveLastUsed is invoked by stateify.
+func (cn *conn) saveLastUsed() unixTime {
+ return unixTime{cn.lastUsed.Unix(), cn.lastUsed.UnixNano()}
+}
+
+// loadLastUsed is invoked by stateify.
+func (cn *conn) loadLastUsed(unix unixTime) {
+ cn.lastUsed = time.Unix(unix.second, unix.nano)
+}
+
+// beforeSave is invoked by stateify.
+func (ct *ConnTrack) beforeSave() {
+ ct.mu.Lock()
+}
diff --git a/pkg/tcpip/stack/iptables_targets.go b/pkg/tcpip/stack/iptables_targets.go
new file mode 100644
index 000000000..5f1b2af64
--- /dev/null
+++ b/pkg/tcpip/stack/iptables_targets.go
@@ -0,0 +1,163 @@
+// Copyright 2019 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package stack
+
+import (
+ "gvisor.dev/gvisor/pkg/log"
+ "gvisor.dev/gvisor/pkg/tcpip"
+ "gvisor.dev/gvisor/pkg/tcpip/header"
+)
+
+// AcceptTarget accepts packets.
+type AcceptTarget struct{}
+
+// Action implements Target.Action.
+func (AcceptTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) {
+ return RuleAccept, 0
+}
+
+// DropTarget drops packets.
+type DropTarget struct{}
+
+// Action implements Target.Action.
+func (DropTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) {
+ return RuleDrop, 0
+}
+
+// ErrorTarget logs an error and drops the packet. It represents a target that
+// should be unreachable.
+type ErrorTarget struct{}
+
+// Action implements Target.Action.
+func (ErrorTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) {
+ log.Debugf("ErrorTarget triggered.")
+ return RuleDrop, 0
+}
+
+// UserChainTarget marks a rule as the beginning of a user chain.
+type UserChainTarget struct {
+ Name string
+}
+
+// Action implements Target.Action.
+func (UserChainTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) {
+ panic("UserChainTarget should never be called.")
+}
+
+// ReturnTarget returns from the current chain. If the chain is a built-in, the
+// hook's underflow should be called.
+type ReturnTarget struct{}
+
+// Action implements Target.Action.
+func (ReturnTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) {
+ return RuleReturn, 0
+}
+
+// RedirectTarget redirects the packet by modifying the destination port/IP.
+// Min and Max values for IP and Ports in the struct indicate the range of
+// values which can be used to redirect.
+type RedirectTarget struct {
+ // TODO(gvisor.dev/issue/170): Other flags need to be added after
+ // we support them.
+ // RangeProtoSpecified flag indicates single port is specified to
+ // redirect.
+ RangeProtoSpecified bool
+
+ // MinIP indicates address used to redirect.
+ MinIP tcpip.Address
+
+ // MaxIP indicates address used to redirect.
+ MaxIP tcpip.Address
+
+ // MinPort indicates port used to redirect.
+ MinPort uint16
+
+ // MaxPort indicates port used to redirect.
+ MaxPort uint16
+}
+
+// Action implements Target.Action.
+// TODO(gvisor.dev/issue/170): Parse headers without copying. The current
+// implementation only works for PREROUTING and calls pkt.Clone(), neither
+// of which should be the case.
+func (rt RedirectTarget) Action(pkt *PacketBuffer, ct *ConnTrack, hook Hook, gso *GSO, r *Route, address tcpip.Address) (RuleVerdict, int) {
+ // Packet is already manipulated.
+ if pkt.NatDone {
+ return RuleAccept, 0
+ }
+
+ // Drop the packet if network and transport header are not set.
+ if pkt.NetworkHeader().View().IsEmpty() || pkt.TransportHeader().View().IsEmpty() {
+ return RuleDrop, 0
+ }
+
+ // Change the address to localhost (127.0.0.1) in Output and
+ // to primary address of the incoming interface in Prerouting.
+ switch hook {
+ case Output:
+ rt.MinIP = tcpip.Address([]byte{127, 0, 0, 1})
+ rt.MaxIP = tcpip.Address([]byte{127, 0, 0, 1})
+ case Prerouting:
+ rt.MinIP = address
+ rt.MaxIP = address
+ default:
+ panic("redirect target is supported only on output and prerouting hooks")
+ }
+
+ // TODO(gvisor.dev/issue/170): Check Flags in RedirectTarget if
+ // we need to change dest address (for OUTPUT chain) or ports.
+ netHeader := header.IPv4(pkt.NetworkHeader().View())
+ switch protocol := netHeader.TransportProtocol(); protocol {
+ case header.UDPProtocolNumber:
+ udpHeader := header.UDP(pkt.TransportHeader().View())
+ udpHeader.SetDestinationPort(rt.MinPort)
+
+ // Calculate UDP checksum and set it.
+ if hook == Output {
+ udpHeader.SetChecksum(0)
+ length := uint16(pkt.Size()) - uint16(netHeader.HeaderLength())
+
+ // Only calculate the checksum if offloading isn't supported.
+ if r.Capabilities()&CapabilityTXChecksumOffload == 0 {
+ xsum := r.PseudoHeaderChecksum(protocol, length)
+ for _, v := range pkt.Data.Views() {
+ xsum = header.Checksum(v, xsum)
+ }
+ udpHeader.SetChecksum(0)
+ udpHeader.SetChecksum(^udpHeader.CalculateChecksum(xsum))
+ }
+ }
+ // Change destination address.
+ netHeader.SetDestinationAddress(rt.MinIP)
+ netHeader.SetChecksum(0)
+ netHeader.SetChecksum(^netHeader.CalculateChecksum())
+ pkt.NatDone = true
+ case header.TCPProtocolNumber:
+ if ct == nil {
+ return RuleAccept, 0
+ }
+
+ // Set up conection for matching NAT rule. Only the first
+ // packet of the connection comes here. Other packets will be
+ // manipulated in connection tracking.
+ if conn := ct.insertRedirectConn(pkt, hook, rt); conn != nil {
+ ct.handlePacket(pkt, hook, gso, r)
+ }
+ default:
+ return RuleDrop, 0
+ }
+
+ return RuleAccept, 0
+}
diff --git a/pkg/tcpip/stack/iptables_types.go b/pkg/tcpip/stack/iptables_types.go
new file mode 100644
index 000000000..73274ada9
--- /dev/null
+++ b/pkg/tcpip/stack/iptables_types.go
@@ -0,0 +1,262 @@
+// Copyright 2019 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package stack
+
+import (
+ "strings"
+ "sync"
+
+ "gvisor.dev/gvisor/pkg/tcpip"
+ "gvisor.dev/gvisor/pkg/tcpip/header"
+)
+
+// A Hook specifies one of the hooks built into the network stack.
+//
+// Userspace app Userspace app
+// ^ |
+// | v
+// [Input] [Output]
+// ^ |
+// | v
+// | routing
+// | |
+// | v
+// ----->[Prerouting]----->routing----->[Forward]---------[Postrouting]----->
+type Hook uint
+
+// These values correspond to values in include/uapi/linux/netfilter.h.
+const (
+ // Prerouting happens before a packet is routed to applications or to
+ // be forwarded.
+ Prerouting Hook = iota
+
+ // Input happens before a packet reaches an application.
+ Input
+
+ // Forward happens once it's decided that a packet should be forwarded
+ // to another host.
+ Forward
+
+ // Output happens after a packet is written by an application to be
+ // sent out.
+ Output
+
+ // Postrouting happens just before a packet goes out on the wire.
+ Postrouting
+
+ // The total number of hooks.
+ NumHooks
+)
+
+// A RuleVerdict is what a rule decides should be done with a packet.
+type RuleVerdict int
+
+const (
+ // RuleAccept indicates the packet should continue through netstack.
+ RuleAccept RuleVerdict = iota
+
+ // RuleDrop indicates the packet should be dropped.
+ RuleDrop
+
+ // RuleJump indicates the packet should jump to another chain.
+ RuleJump
+
+ // RuleReturn indicates the packet should return to the previous chain.
+ RuleReturn
+)
+
+// IPTables holds all the tables for a netstack.
+//
+// +stateify savable
+type IPTables struct {
+ // mu protects tables, priorities, and modified.
+ mu sync.RWMutex
+
+ // tables maps tableIDs to tables. Holds builtin tables only, not user
+ // tables. mu must be locked for accessing.
+ tables [numTables]Table
+
+ // priorities maps each hook to a list of table names. The order of the
+ // list is the order in which each table should be visited for that
+ // hook. mu needs to be locked for accessing.
+ priorities [NumHooks][]tableID
+
+ // modified is whether tables have been modified at least once. It is
+ // used to elide the iptables performance overhead for workloads that
+ // don't utilize iptables.
+ modified bool
+
+ connections ConnTrack
+
+ // reaperDone can be signalled to stop the reaper goroutine.
+ reaperDone chan struct{}
+}
+
+// A Table defines a set of chains and hooks into the network stack. It is
+// really just a list of rules.
+//
+// +stateify savable
+type Table struct {
+ // Rules holds the rules that make up the table.
+ Rules []Rule
+
+ // BuiltinChains maps builtin chains to their entrypoint rule in Rules.
+ BuiltinChains [NumHooks]int
+
+ // Underflows maps builtin chains to their underflow rule in Rules
+ // (i.e. the rule to execute if the chain returns without a verdict).
+ Underflows [NumHooks]int
+}
+
+// ValidHooks returns a bitmap of the builtin hooks for the given table.
+func (table *Table) ValidHooks() uint32 {
+ hooks := uint32(0)
+ for hook, ruleIdx := range table.BuiltinChains {
+ if ruleIdx != HookUnset {
+ hooks |= 1 << hook
+ }
+ }
+ return hooks
+}
+
+// A Rule is a packet processing rule. It consists of two pieces. First it
+// contains zero or more matchers, each of which is a specification of which
+// packets this rule applies to. If there are no matchers in the rule, it
+// applies to any packet.
+//
+// +stateify savable
+type Rule struct {
+ // Filter holds basic IP filtering fields common to every rule.
+ Filter IPHeaderFilter
+
+ // Matchers is the list of matchers for this rule.
+ Matchers []Matcher
+
+ // Target is the action to invoke if all the matchers match the packet.
+ Target Target
+}
+
+// IPHeaderFilter holds basic IP filtering data common to every rule.
+//
+// +stateify savable
+type IPHeaderFilter struct {
+ // Protocol matches the transport protocol.
+ Protocol tcpip.TransportProtocolNumber
+
+ // Dst matches the destination IP address.
+ Dst tcpip.Address
+
+ // DstMask masks bits of the destination IP address when comparing with
+ // Dst.
+ DstMask tcpip.Address
+
+ // DstInvert inverts the meaning of the destination IP check, i.e. when
+ // true the filter will match packets that fail the destination
+ // comparison.
+ DstInvert bool
+
+ // Src matches the source IP address.
+ Src tcpip.Address
+
+ // SrcMask masks bits of the source IP address when comparing with Src.
+ SrcMask tcpip.Address
+
+ // SrcInvert inverts the meaning of the source IP check, i.e. when true the
+ // filter will match packets that fail the source comparison.
+ SrcInvert bool
+
+ // OutputInterface matches the name of the outgoing interface for the
+ // packet.
+ OutputInterface string
+
+ // OutputInterfaceMask masks the characters of the interface name when
+ // comparing with OutputInterface.
+ OutputInterfaceMask string
+
+ // OutputInterfaceInvert inverts the meaning of outgoing interface check,
+ // i.e. when true the filter will match packets that fail the outgoing
+ // interface comparison.
+ OutputInterfaceInvert bool
+}
+
+// match returns whether hdr matches the filter.
+func (fl IPHeaderFilter) match(hdr header.IPv4, hook Hook, nicName string) bool {
+ // TODO(gvisor.dev/issue/170): Support other fields of the filter.
+ // Check the transport protocol.
+ if fl.Protocol != 0 && fl.Protocol != hdr.TransportProtocol() {
+ return false
+ }
+
+ // Check the source and destination IPs.
+ if !filterAddress(hdr.DestinationAddress(), fl.DstMask, fl.Dst, fl.DstInvert) || !filterAddress(hdr.SourceAddress(), fl.SrcMask, fl.Src, fl.SrcInvert) {
+ return false
+ }
+
+ // Check the output interface.
+ // TODO(gvisor.dev/issue/170): Add the check for FORWARD and POSTROUTING
+ // hooks after supported.
+ if hook == Output {
+ n := len(fl.OutputInterface)
+ if n == 0 {
+ return true
+ }
+
+ // If the interface name ends with '+', any interface which begins
+ // with the name should be matched.
+ ifName := fl.OutputInterface
+ matches := true
+ if strings.HasSuffix(ifName, "+") {
+ matches = strings.HasPrefix(nicName, ifName[:n-1])
+ } else {
+ matches = nicName == ifName
+ }
+ return fl.OutputInterfaceInvert != matches
+ }
+
+ return true
+}
+
+// filterAddress returns whether addr matches the filter.
+func filterAddress(addr, mask, filterAddr tcpip.Address, invert bool) bool {
+ matches := true
+ for i := range filterAddr {
+ if addr[i]&mask[i] != filterAddr[i] {
+ matches = false
+ break
+ }
+ }
+ return matches != invert
+}
+
+// A Matcher is the interface for matching packets.
+type Matcher interface {
+ // Name returns the name of the Matcher.
+ Name() string
+
+ // Match returns whether the packet matches and whether the packet
+ // should be "hotdropped", i.e. dropped immediately. This is usually
+ // used for suspicious packets.
+ //
+ // Precondition: packet.NetworkHeader is set.
+ Match(hook Hook, packet *PacketBuffer, interfaceName string) (matches bool, hotdrop bool)
+}
+
+// A Target is the interface for taking an action for a packet.
+type Target interface {
+ // Action takes an action on the packet and returns a verdict on how
+ // traversal should (or should not) continue. If the return value is
+ // Jump, it also returns the index of the rule to jump to.
+ Action(packet *PacketBuffer, connections *ConnTrack, hook Hook, gso *GSO, r *Route, address tcpip.Address) (RuleVerdict, int)
+}
diff --git a/pkg/tcpip/stack/linkaddrcache.go b/pkg/tcpip/stack/linkaddrcache.go
index 267df60d1..6f73a0ce4 100644
--- a/pkg/tcpip/stack/linkaddrcache.go
+++ b/pkg/tcpip/stack/linkaddrcache.go
@@ -16,10 +16,10 @@ package stack
import (
"fmt"
- "sync"
"time"
"gvisor.dev/gvisor/pkg/sleep"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
)
@@ -244,7 +244,7 @@ func (c *linkAddrCache) startAddressResolution(k tcpip.FullAddress, linkRes Link
for i := 0; ; i++ {
// Send link request, then wait for the timeout limit and check
// whether the request succeeded.
- linkRes.LinkAddressRequest(k.Addr, localAddr, linkEP)
+ linkRes.LinkAddressRequest(k.Addr, localAddr, "" /* linkAddr */, linkEP)
select {
case now := <-time.After(c.resolutionTimeout):
diff --git a/pkg/tcpip/stack/linkaddrcache_test.go b/pkg/tcpip/stack/linkaddrcache_test.go
index 9946b8fe8..b15b8d1cb 100644
--- a/pkg/tcpip/stack/linkaddrcache_test.go
+++ b/pkg/tcpip/stack/linkaddrcache_test.go
@@ -16,12 +16,12 @@ package stack
import (
"fmt"
- "sync"
"sync/atomic"
"testing"
"time"
"gvisor.dev/gvisor/pkg/sleep"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
)
@@ -48,7 +48,7 @@ type testLinkAddressResolver struct {
onLinkAddressRequest func()
}
-func (r *testLinkAddressResolver) LinkAddressRequest(addr, _ tcpip.Address, _ LinkEndpoint) *tcpip.Error {
+func (r *testLinkAddressResolver) LinkAddressRequest(addr, _ tcpip.Address, _ tcpip.LinkAddress, _ LinkEndpoint) *tcpip.Error {
time.AfterFunc(r.delay, func() { r.fakeRequest(addr) })
if f := r.onLinkAddressRequest; f != nil {
f()
diff --git a/pkg/tcpip/stack/ndp.go b/pkg/tcpip/stack/ndp.go
index 03ddebdbd..97ca00d16 100644
--- a/pkg/tcpip/stack/ndp.go
+++ b/pkg/tcpip/stack/ndp.go
@@ -17,6 +17,7 @@ package stack
import (
"fmt"
"log"
+ "math/rand"
"time"
"gvisor.dev/gvisor/pkg/tcpip"
@@ -32,39 +33,277 @@ const (
// Default = 1 (from RFC 4862 section 5.1)
defaultDupAddrDetectTransmits = 1
- // defaultRetransmitTimer is the default amount of time to wait between
- // sending NDP Neighbor solicitation messages.
+ // defaultMaxRtrSolicitations is the default number of Router
+ // Solicitation messages to send when a NIC becomes enabled.
//
- // Default = 1s (from RFC 4861 section 10).
- defaultRetransmitTimer = time.Second
+ // Default = 3 (from RFC 4861 section 10).
+ defaultMaxRtrSolicitations = 3
- // minimumRetransmitTimer is the minimum amount of time to wait between
- // sending NDP Neighbor solicitation messages. Note, RFC 4861 does
- // not impose a minimum Retransmit Timer, but we do here to make sure
- // the messages are not sent all at once. We also come to this value
- // because in the RetransmitTimer field of a Router Advertisement, a
- // value of 0 means unspecified, so the smallest valid value is 1.
- // Note, the unit of the RetransmitTimer field in the Router
- // Advertisement is milliseconds.
+ // defaultRtrSolicitationInterval is the default amount of time between
+ // sending Router Solicitation messages.
//
- // Min = 1ms.
- minimumRetransmitTimer = time.Millisecond
+ // Default = 4s (from 4861 section 10).
+ defaultRtrSolicitationInterval = 4 * time.Second
+
+ // defaultMaxRtrSolicitationDelay is the default maximum amount of time
+ // to wait before sending the first Router Solicitation message.
+ //
+ // Default = 1s (from 4861 section 10).
+ defaultMaxRtrSolicitationDelay = time.Second
+
+ // defaultHandleRAs is the default configuration for whether or not to
+ // handle incoming Router Advertisements as a host.
+ defaultHandleRAs = true
+
+ // defaultDiscoverDefaultRouters is the default configuration for
+ // whether or not to discover default routers from incoming Router
+ // Advertisements, as a host.
+ defaultDiscoverDefaultRouters = true
+
+ // defaultDiscoverOnLinkPrefixes is the default configuration for
+ // whether or not to discover on-link prefixes from incoming Router
+ // Advertisements' Prefix Information option, as a host.
+ defaultDiscoverOnLinkPrefixes = true
+
+ // defaultAutoGenGlobalAddresses is the default configuration for
+ // whether or not to generate global IPv6 addresses in response to
+ // receiving a new Prefix Information option with its Autonomous
+ // Address AutoConfiguration flag set, as a host.
+ //
+ // Default = true.
+ defaultAutoGenGlobalAddresses = true
+
+ // minimumRtrSolicitationInterval is the minimum amount of time to wait
+ // between sending Router Solicitation messages. This limit is imposed
+ // to make sure that Router Solicitation messages are not sent all at
+ // once, defeating the purpose of sending the initial few messages.
+ minimumRtrSolicitationInterval = 500 * time.Millisecond
+
+ // minimumMaxRtrSolicitationDelay is the minimum amount of time to wait
+ // before sending the first Router Solicitation message. It is 0 because
+ // we cannot have a negative delay.
+ minimumMaxRtrSolicitationDelay = 0
+
+ // MaxDiscoveredDefaultRouters is the maximum number of discovered
+ // default routers. The stack should stop discovering new routers after
+ // discovering MaxDiscoveredDefaultRouters routers.
+ //
+ // This value MUST be at minimum 2 as per RFC 4861 section 6.3.4, and
+ // SHOULD be more.
+ MaxDiscoveredDefaultRouters = 10
+
+ // MaxDiscoveredOnLinkPrefixes is the maximum number of discovered
+ // on-link prefixes. The stack should stop discovering new on-link
+ // prefixes after discovering MaxDiscoveredOnLinkPrefixes on-link
+ // prefixes.
+ MaxDiscoveredOnLinkPrefixes = 10
+
+ // validPrefixLenForAutoGen is the expected prefix length that an
+ // address can be generated for. Must be 64 bits as the interface
+ // identifier (IID) is 64 bits and an IPv6 address is 128 bits, so
+ // 128 - 64 = 64.
+ validPrefixLenForAutoGen = 64
+
+ // defaultAutoGenTempGlobalAddresses is the default configuration for whether
+ // or not to generate temporary SLAAC addresses.
+ defaultAutoGenTempGlobalAddresses = true
+
+ // defaultMaxTempAddrValidLifetime is the default maximum valid lifetime
+ // for temporary SLAAC addresses generated as part of RFC 4941.
+ //
+ // Default = 7 days (from RFC 4941 section 5).
+ defaultMaxTempAddrValidLifetime = 7 * 24 * time.Hour
+
+ // defaultMaxTempAddrPreferredLifetime is the default preferred lifetime
+ // for temporary SLAAC addresses generated as part of RFC 4941.
+ //
+ // Default = 1 day (from RFC 4941 section 5).
+ defaultMaxTempAddrPreferredLifetime = 24 * time.Hour
+
+ // defaultRegenAdvanceDuration is the default duration before the deprecation
+ // of a temporary address when a new address will be generated.
+ //
+ // Default = 5s (from RFC 4941 section 5).
+ defaultRegenAdvanceDuration = 5 * time.Second
+
+ // minRegenAdvanceDuration is the minimum duration before the deprecation
+ // of a temporary address when a new address will be generated.
+ minRegenAdvanceDuration = time.Duration(0)
+
+ // maxSLAACAddrLocalRegenAttempts is the maximum number of times to attempt
+ // SLAAC address regenerations in response to a NIC-local conflict.
+ maxSLAACAddrLocalRegenAttempts = 10
+)
+
+var (
+ // MinPrefixInformationValidLifetimeForUpdate is the minimum Valid
+ // Lifetime to update the valid lifetime of a generated address by
+ // SLAAC.
+ //
+ // This is exported as a variable (instead of a constant) so tests
+ // can update it to a smaller value.
+ //
+ // Min = 2hrs.
+ MinPrefixInformationValidLifetimeForUpdate = 2 * time.Hour
+
+ // MaxDesyncFactor is the upper bound for the preferred lifetime's desync
+ // factor for temporary SLAAC addresses.
+ //
+ // This is exported as a variable (instead of a constant) so tests
+ // can update it to a smaller value.
+ //
+ // Must be greater than 0.
+ //
+ // Max = 10m (from RFC 4941 section 5).
+ MaxDesyncFactor = 10 * time.Minute
+
+ // MinMaxTempAddrPreferredLifetime is the minimum value allowed for the
+ // maximum preferred lifetime for temporary SLAAC addresses.
+ //
+ // This is exported as a variable (instead of a constant) so tests
+ // can update it to a smaller value.
+ //
+ // This value guarantees that a temporary address will be preferred for at
+ // least 1hr if the SLAAC prefix is valid for at least that time.
+ MinMaxTempAddrPreferredLifetime = defaultRegenAdvanceDuration + MaxDesyncFactor + time.Hour
+
+ // MinMaxTempAddrValidLifetime is the minimum value allowed for the
+ // maximum valid lifetime for temporary SLAAC addresses.
+ //
+ // This is exported as a variable (instead of a constant) so tests
+ // can update it to a smaller value.
+ //
+ // This value guarantees that a temporary address will be valid for at least
+ // 2hrs if the SLAAC prefix is valid for at least that time.
+ MinMaxTempAddrValidLifetime = 2 * time.Hour
+)
+
+// DHCPv6ConfigurationFromNDPRA is a configuration available via DHCPv6 that an
+// NDP Router Advertisement informed the Stack about.
+type DHCPv6ConfigurationFromNDPRA int
+
+const (
+ _ DHCPv6ConfigurationFromNDPRA = iota
+
+ // DHCPv6NoConfiguration indicates that no configurations are available via
+ // DHCPv6.
+ DHCPv6NoConfiguration
+
+ // DHCPv6ManagedAddress indicates that addresses are available via DHCPv6.
+ //
+ // DHCPv6ManagedAddress also implies DHCPv6OtherConfigurations because DHCPv6
+ // will return all available configuration information.
+ DHCPv6ManagedAddress
+
+ // DHCPv6OtherConfigurations indicates that other configuration information is
+ // available via DHCPv6.
+ //
+ // Other configurations are configurations other than addresses. Examples of
+ // other configurations are recursive DNS server list, DNS search lists and
+ // default gateway.
+ DHCPv6OtherConfigurations
)
// 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
+ // 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)
+ // This function is not permitted to block indefinitely. This function
+ // is also not permitted to call into the stack.
+ OnDuplicateAddressDetectionStatus(nicID tcpip.NICID, addr tcpip.Address, resolved bool, err *tcpip.Error)
+
+ // OnDefaultRouterDiscovered will be called when a new default router is
+ // discovered. Implementations must return true if the newly discovered
+ // router should be remembered.
+ //
+ // This function is not permitted to block indefinitely. This function
+ // is also not permitted to call into the stack.
+ OnDefaultRouterDiscovered(nicID tcpip.NICID, addr tcpip.Address) bool
+
+ // OnDefaultRouterInvalidated will be called when a discovered default
+ // router that was remembered is invalidated.
+ //
+ // This function is not permitted to block indefinitely. This function
+ // is also not permitted to call into the stack.
+ OnDefaultRouterInvalidated(nicID tcpip.NICID, addr tcpip.Address)
+
+ // OnOnLinkPrefixDiscovered will be called when a new on-link prefix is
+ // discovered. Implementations must return true if the newly discovered
+ // on-link prefix should be remembered.
+ //
+ // This function is not permitted to block indefinitely. This function
+ // is also not permitted to call into the stack.
+ OnOnLinkPrefixDiscovered(nicID tcpip.NICID, prefix tcpip.Subnet) bool
+
+ // OnOnLinkPrefixInvalidated will be called when a discovered on-link
+ // prefix that was remembered is invalidated.
+ //
+ // This function is not permitted to block indefinitely. This function
+ // is also not permitted to call into the stack.
+ OnOnLinkPrefixInvalidated(nicID tcpip.NICID, prefix tcpip.Subnet)
+
+ // OnAutoGenAddress will be called when a new prefix with its
+ // autonomous address-configuration flag set has been received and SLAAC
+ // has been performed. Implementations may prevent the stack from
+ // assigning the address to the NIC by returning false.
+ //
+ // This function is not permitted to block indefinitely. It must not
+ // call functions on the stack itself.
+ OnAutoGenAddress(tcpip.NICID, tcpip.AddressWithPrefix) bool
+
+ // OnAutoGenAddressDeprecated will be called when an auto-generated
+ // address (as part of SLAAC) has been deprecated, but is still
+ // considered valid. Note, if an address is invalidated at the same
+ // time it is deprecated, the deprecation event MAY be omitted.
+ //
+ // This function is not permitted to block indefinitely. It must not
+ // call functions on the stack itself.
+ OnAutoGenAddressDeprecated(tcpip.NICID, tcpip.AddressWithPrefix)
+
+ // OnAutoGenAddressInvalidated will be called when an auto-generated
+ // address (as part of SLAAC) has been invalidated.
+ //
+ // This function is not permitted to block indefinitely. It must not
+ // call functions on the stack itself.
+ OnAutoGenAddressInvalidated(tcpip.NICID, tcpip.AddressWithPrefix)
+
+ // OnRecursiveDNSServerOption will be called when an NDP option with
+ // recursive DNS servers has been received. Note, addrs may contain
+ // link-local addresses.
+ //
+ // It is up to the caller to use the DNS Servers only for their valid
+ // lifetime. OnRecursiveDNSServerOption may be called for new or
+ // already known DNS servers. If called with known DNS servers, their
+ // valid lifetimes must be refreshed to lifetime (it may be increased,
+ // decreased, or completely invalidated when lifetime = 0).
+ //
+ // This function is not permitted to block indefinitely. It must not
+ // call functions on the stack itself.
+ OnRecursiveDNSServerOption(nicID tcpip.NICID, addrs []tcpip.Address, lifetime time.Duration)
+
+ // OnDNSSearchListOption will be called when an NDP option with a DNS
+ // search list has been received.
+ //
+ // It is up to the caller to use the domain names in the search list
+ // for only their valid lifetime. OnDNSSearchListOption may be called
+ // with new or already known domain names. If called with known domain
+ // names, their valid lifetimes must be refreshed to lifetime (it may
+ // be increased, decreased or completely invalidated when lifetime = 0.
+ OnDNSSearchListOption(nicID tcpip.NICID, domainNames []string, lifetime time.Duration)
+
+ // OnDHCPv6Configuration will be called with an updated configuration that is
+ // available via DHCPv6 for a specified NIC.
+ //
+ // This function is not permitted to block indefinitely. It must not
+ // call functions on the stack itself.
+ OnDHCPv6Configuration(tcpip.NICID, DHCPv6ConfigurationFromNDPRA)
}
// NDPConfigurations is the NDP configurations for the netstack.
@@ -78,28 +317,124 @@ type NDPConfigurations struct {
// The amount of time to wait between sending Neighbor solicitation
// messages.
//
- // Must be greater than 0.5s.
+ // Must be greater than or equal to 1ms.
RetransmitTimer time.Duration
+
+ // The number of Router Solicitation messages to send when the NIC
+ // becomes enabled.
+ MaxRtrSolicitations uint8
+
+ // The amount of time between transmitting Router Solicitation messages.
+ //
+ // Must be greater than or equal to 0.5s.
+ RtrSolicitationInterval time.Duration
+
+ // The maximum amount of time before transmitting the first Router
+ // Solicitation message.
+ //
+ // Must be greater than or equal to 0s.
+ MaxRtrSolicitationDelay time.Duration
+
+ // HandleRAs determines whether or not Router Advertisements will be
+ // processed.
+ HandleRAs bool
+
+ // DiscoverDefaultRouters determines whether or not default routers will
+ // be discovered from Router Advertisements. This configuration is
+ // ignored if HandleRAs is false.
+ DiscoverDefaultRouters bool
+
+ // DiscoverOnLinkPrefixes determines whether or not on-link prefixes
+ // will be discovered from Router Advertisements' Prefix Information
+ // option. This configuration is ignored if HandleRAs is false.
+ DiscoverOnLinkPrefixes bool
+
+ // AutoGenGlobalAddresses determines whether or not global IPv6
+ // addresses will be generated for a NIC in response to receiving a new
+ // Prefix Information option with its Autonomous Address
+ // AutoConfiguration flag set, as a host, as per RFC 4862 (SLAAC).
+ //
+ // Note, if an address was already generated for some unique prefix, as
+ // part of SLAAC, this option does not affect whether or not the
+ // lifetime(s) of the generated address changes; this option only
+ // affects the generation of new addresses as part of SLAAC.
+ AutoGenGlobalAddresses bool
+
+ // AutoGenAddressConflictRetries determines how many times to attempt to retry
+ // generation of a permanent auto-generated address in response to DAD
+ // conflicts.
+ //
+ // If the method used to generate the address does not support creating
+ // alternative addresses (e.g. IIDs based on the modified EUI64 of a NIC's
+ // MAC address), then no attempt will be made to resolve the conflict.
+ AutoGenAddressConflictRetries uint8
+
+ // AutoGenTempGlobalAddresses determines whether or not temporary SLAAC
+ // addresses will be generated for a NIC as part of SLAAC privacy extensions,
+ // RFC 4941.
+ //
+ // Ignored if AutoGenGlobalAddresses is false.
+ AutoGenTempGlobalAddresses bool
+
+ // MaxTempAddrValidLifetime is the maximum valid lifetime for temporary
+ // SLAAC addresses.
+ MaxTempAddrValidLifetime time.Duration
+
+ // MaxTempAddrPreferredLifetime is the maximum preferred lifetime for
+ // temporary SLAAC addresses.
+ MaxTempAddrPreferredLifetime time.Duration
+
+ // RegenAdvanceDuration is the duration before the deprecation of a temporary
+ // address when a new address will be generated.
+ RegenAdvanceDuration time.Duration
}
// DefaultNDPConfigurations returns an NDPConfigurations populated with
// default values.
func DefaultNDPConfigurations() NDPConfigurations {
return NDPConfigurations{
- DupAddrDetectTransmits: defaultDupAddrDetectTransmits,
- RetransmitTimer: defaultRetransmitTimer,
+ DupAddrDetectTransmits: defaultDupAddrDetectTransmits,
+ RetransmitTimer: defaultRetransmitTimer,
+ MaxRtrSolicitations: defaultMaxRtrSolicitations,
+ RtrSolicitationInterval: defaultRtrSolicitationInterval,
+ MaxRtrSolicitationDelay: defaultMaxRtrSolicitationDelay,
+ HandleRAs: defaultHandleRAs,
+ DiscoverDefaultRouters: defaultDiscoverDefaultRouters,
+ DiscoverOnLinkPrefixes: defaultDiscoverOnLinkPrefixes,
+ AutoGenGlobalAddresses: defaultAutoGenGlobalAddresses,
+ AutoGenTempGlobalAddresses: defaultAutoGenTempGlobalAddresses,
+ MaxTempAddrValidLifetime: defaultMaxTempAddrValidLifetime,
+ MaxTempAddrPreferredLifetime: defaultMaxTempAddrPreferredLifetime,
+ RegenAdvanceDuration: defaultRegenAdvanceDuration,
}
}
// validate modifies an NDPConfigurations with valid values. If invalid values
// are present in c, the corresponding default values will be used instead.
-//
-// If RetransmitTimer is less than minimumRetransmitTimer, then a value of
-// defaultRetransmitTimer will be used.
func (c *NDPConfigurations) validate() {
if c.RetransmitTimer < minimumRetransmitTimer {
c.RetransmitTimer = defaultRetransmitTimer
}
+
+ if c.RtrSolicitationInterval < minimumRtrSolicitationInterval {
+ c.RtrSolicitationInterval = defaultRtrSolicitationInterval
+ }
+
+ if c.MaxRtrSolicitationDelay < minimumMaxRtrSolicitationDelay {
+ c.MaxRtrSolicitationDelay = defaultMaxRtrSolicitationDelay
+ }
+
+ if c.MaxTempAddrValidLifetime < MinMaxTempAddrValidLifetime {
+ c.MaxTempAddrValidLifetime = MinMaxTempAddrValidLifetime
+ }
+
+ if c.MaxTempAddrPreferredLifetime < MinMaxTempAddrPreferredLifetime || c.MaxTempAddrPreferredLifetime > c.MaxTempAddrValidLifetime {
+ c.MaxTempAddrPreferredLifetime = MinMaxTempAddrPreferredLifetime
+ }
+
+ if c.RegenAdvanceDuration < minRegenAdvanceDuration {
+ c.RegenAdvanceDuration = minRegenAdvanceDuration
+ }
}
// ndpState is the per-interface NDP state.
@@ -112,13 +447,47 @@ type ndpState struct {
// The DAD state to send the next NS message, or resolve the address.
dad map[tcpip.Address]dadState
+
+ // The default routers discovered through Router Advertisements.
+ defaultRouters map[tcpip.Address]defaultRouterState
+
+ rtrSolicit struct {
+ // The timer used to send the next router solicitation message.
+ timer tcpip.Timer
+
+ // Used to let the Router Solicitation timer know that it has been stopped.
+ //
+ // Must only be read from or written to while protected by the lock of
+ // the NIC this ndpState is associated with. MUST be set when the timer is
+ // set.
+ done *bool
+ }
+
+ // The on-link prefixes discovered through Router Advertisements' Prefix
+ // Information option.
+ onLinkPrefixes map[tcpip.Subnet]onLinkPrefixState
+
+ // The SLAAC prefixes discovered through Router Advertisements' Prefix
+ // Information option.
+ slaacPrefixes map[tcpip.Subnet]slaacPrefixState
+
+ // The last learned DHCPv6 configuration from an NDP RA.
+ dhcpv6Configuration DHCPv6ConfigurationFromNDPRA
+
+ // temporaryIIDHistory is the history value used to generate a new temporary
+ // IID.
+ temporaryIIDHistory [header.IIDSize]byte
+
+ // temporaryAddressDesyncFactor is the preferred lifetime's desync factor for
+ // temporary SLAAC addresses.
+ temporaryAddressDesyncFactor time.Duration
}
// dadState holds the Duplicate Address Detection timer and channel to signal
// to the DAD goroutine that DAD should stop.
type dadState struct {
// The DAD timer to send the next NS message, or resolve the address.
- timer *time.Timer
+ timer tcpip.Timer
// Used to let the DAD timer know that it has been stopped.
//
@@ -127,6 +496,102 @@ type dadState struct {
done *bool
}
+// defaultRouterState holds data associated with a default router discovered by
+// a Router Advertisement (RA).
+type defaultRouterState struct {
+ // Job to invalidate the default router.
+ //
+ // Must not be nil.
+ invalidationJob *tcpip.Job
+}
+
+// onLinkPrefixState holds data associated with an on-link prefix discovered by
+// a Router Advertisement's Prefix Information option (PI) when the NDP
+// configurations was configured to do so.
+type onLinkPrefixState struct {
+ // Job to invalidate the on-link prefix.
+ //
+ // Must not be nil.
+ invalidationJob *tcpip.Job
+}
+
+// tempSLAACAddrState holds state associated with a temporary SLAAC address.
+type tempSLAACAddrState struct {
+ // Job to deprecate the temporary SLAAC address.
+ //
+ // Must not be nil.
+ deprecationJob *tcpip.Job
+
+ // Job to invalidate the temporary SLAAC address.
+ //
+ // Must not be nil.
+ invalidationJob *tcpip.Job
+
+ // Job to regenerate the temporary SLAAC address.
+ //
+ // Must not be nil.
+ regenJob *tcpip.Job
+
+ createdAt time.Time
+
+ // The address's endpoint.
+ //
+ // Must not be nil.
+ ref *referencedNetworkEndpoint
+
+ // Has a new temporary SLAAC address already been regenerated?
+ regenerated bool
+}
+
+// slaacPrefixState holds state associated with a SLAAC prefix.
+type slaacPrefixState struct {
+ // Job to deprecate the prefix.
+ //
+ // Must not be nil.
+ deprecationJob *tcpip.Job
+
+ // Job to invalidate the prefix.
+ //
+ // Must not be nil.
+ invalidationJob *tcpip.Job
+
+ // Nonzero only when the address is not valid forever.
+ validUntil time.Time
+
+ // Nonzero only when the address is not preferred forever.
+ preferredUntil time.Time
+
+ // State associated with the stable address generated for the prefix.
+ stableAddr struct {
+ // The address's endpoint.
+ //
+ // May only be nil when the address is being (re-)generated. Otherwise,
+ // must not be nil as all SLAAC prefixes must have a stable address.
+ ref *referencedNetworkEndpoint
+
+ // The number of times an address has been generated locally where the NIC
+ // already had the generated address.
+ localGenerationFailures uint8
+ }
+
+ // The temporary (short-lived) addresses generated for the SLAAC prefix.
+ tempAddrs map[tcpip.Address]tempSLAACAddrState
+
+ // The next two fields are used by both stable and temporary addresses
+ // generated for a SLAAC prefix. This is safe as only 1 address will be
+ // in the generation and DAD process at any time. That is, no two addresses
+ // will be generated at the same time for a given SLAAC prefix.
+
+ // The number of times an address has been generated and added to the NIC.
+ //
+ // Addresses may be regenerated in reseponse to a DAD conflicts.
+ generationAttempts uint8
+
+ // The maximum number of times to attempt regeneration of a SLAAC address
+ // in response to DAD conflicts.
+ maxGenerationAttempts uint8
+}
+
// startDuplicateAddressDetection performs Duplicate Address Detection.
//
// This function must only be called by IPv6 addresses that are currently
@@ -139,87 +604,110 @@ func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, ref *ref
return tcpip.ErrAddressFamilyNotSupported
}
- // Should not attempt to perform DAD on an address that is currently in
- // the DAD process.
+ if ref.getKind() != permanentTentative {
+ // The endpoint should be marked as tentative since we are starting DAD.
+ panic(fmt.Sprintf("ndpdad: addr %s is not tentative on NIC(%d)", addr, ndp.nic.ID()))
+ }
+
+ // Should not attempt to perform DAD on an address that is currently in the
+ // DAD process.
if _, ok := ndp.dad[addr]; ok {
- // Should never happen because we should only ever call this
- // function for newly created addresses. If we attemped to
- // "add" an address that already existed, we would returned an
- // error since we attempted to add a duplicate address, or its
- // reference count would have been increased without doing the
- // work that would have been done for an address that was brand
- // new. See NIC.addPermanentAddressLocked.
+ // Should never happen because we should only ever call this function for
+ // newly created addresses. If we attemped to "add" an address that already
+ // existed, we would get an error since we attempted to add a duplicate
+ // address, or its reference count would have been increased without doing
+ // the work that would have been done for an address that was brand new.
+ // See NIC.addAddressLocked.
panic(fmt.Sprintf("ndpdad: already performing DAD for addr %s on NIC(%d)", addr, ndp.nic.ID()))
}
remaining := ndp.configs.DupAddrDetectTransmits
+ if remaining == 0 {
+ ref.setKind(permanent)
- {
- done, err := ndp.doDuplicateAddressDetection(addr, remaining, ref)
- if err != nil {
- return err
- }
- if done {
- return nil
+ // Consider DAD to have resolved even if no DAD messages were actually
+ // transmitted.
+ if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil {
+ ndpDisp.OnDuplicateAddressDetectionStatus(ndp.nic.ID(), addr, true, nil)
}
- }
- remaining--
+ return nil
+ }
var done bool
- var timer *time.Timer
- timer = time.AfterFunc(ndp.configs.RetransmitTimer, func() {
- var d bool
- var err *tcpip.Error
+ var timer tcpip.Timer
+ // We initially start a timer to fire immediately because some of the DAD work
+ // cannot be done while holding the NIC's lock. This is effectively the same
+ // as starting a goroutine but we use a timer that fires immediately so we can
+ // reset it for the next DAD iteration.
+ timer = ndp.nic.stack.Clock().AfterFunc(0, func() {
+ ndp.nic.mu.Lock()
+ defer ndp.nic.mu.Unlock()
- // doDadIteration does a single iteration of the DAD loop.
- //
- // Returns true if the integrator needs to be informed of DAD
- // completing.
- doDadIteration := func() bool {
- ndp.nic.mu.Lock()
- defer ndp.nic.mu.Unlock()
-
- if done {
- // If we reach this point, it means that the DAD
- // timer fired after another goroutine already
- // obtained the NIC lock and stopped DAD before
- // this function obtained the NIC lock. Simply
- // return here and do nothing further.
- return false
- }
+ if done {
+ // If we reach this point, it means that the DAD timer fired after
+ // another goroutine already obtained the NIC lock and stopped DAD
+ // before this function obtained the NIC lock. Simply return here and do
+ // nothing further.
+ return
+ }
- ref, ok := ndp.nic.endpoints[NetworkEndpointID{addr}]
- if !ok {
- // This should never happen.
- // We should have an endpoint for addr since we
- // are still performing DAD on it. If the
- // endpoint does not exist, but we are doing DAD
- // on it, then we started DAD at some point, but
- // forgot to stop it when the endpoint was
- // deleted.
- panic(fmt.Sprintf("ndpdad: unrecognized addr %s for NIC(%d)", addr, ndp.nic.ID()))
- }
+ 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 no longer tentative on NIC(%d)", addr, ndp.nic.ID()))
+ }
- d, err = ndp.doDuplicateAddressDetection(addr, remaining, ref)
- if err != nil || d {
- delete(ndp.dad, addr)
+ dadDone := remaining == 0
- if err != nil {
- log.Printf("ndpdad: Error occured during DAD iteration for addr (%s) on NIC(%d); err = %s", addr, ndp.nic.ID(), err)
- }
+ var err *tcpip.Error
+ if !dadDone {
+ // Use the unspecified address as the source address when performing DAD.
+ ref := ndp.nic.getRefOrCreateTempLocked(header.IPv6ProtocolNumber, header.IPv6Any, NeverPrimaryEndpoint)
- // Let the integrator know DAD has completed.
- return true
- }
+ // Do not hold the lock when sending packets which may be a long running
+ // task or may block link address resolution. We know this is safe
+ // because immediately after obtaining the lock again, we check if DAD
+ // has been stopped before doing any work with the NIC. Note, DAD would be
+ // stopped if the NIC was disabled or removed, or if the address was
+ // removed.
+ ndp.nic.mu.Unlock()
+ err = ndp.sendDADPacket(addr, ref)
+ ndp.nic.mu.Lock()
+ }
+ if done {
+ // If we reach this point, it means that DAD was stopped after we released
+ // the NIC's read lock and before we obtained the write lock.
+ return
+ }
+
+ if dadDone {
+ // DAD has resolved.
+ ref.setKind(permanent)
+ } else if err == nil {
+ // DAD is not done and we had no errors when sending the last NDP NS,
+ // schedule the next DAD timer.
remaining--
timer.Reset(ndp.nic.stack.ndpConfigs.RetransmitTimer)
- return false
+ return
+ }
+
+ // At this point we know that either DAD is done or we hit an error sending
+ // the last NDP NS. Either way, clean up addr's DAD state and let the
+ // integrator know DAD has completed.
+ delete(ndp.dad, addr)
+
+ if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil {
+ ndpDisp.OnDuplicateAddressDetectionStatus(ndp.nic.ID(), addr, dadDone, err)
}
- if doDadIteration() && ndp.nic.stack.ndpDisp != nil {
- ndp.nic.stack.ndpDisp.OnDuplicateAddressDetectionStatus(ndp.nic.ID(), addr, d, err)
+ // If DAD resolved for a stable SLAAC address, attempt generation of a
+ // temporary SLAAC address.
+ if dadDone && ref.configType == slaac {
+ // Reset the generation attempts counter as we are starting the generation
+ // of a new address for the SLAAC prefix.
+ ndp.regenerateTempSLAACAddr(ref.addrWithPrefix().Subnet(), true /* resetGenAttempts */)
}
})
@@ -231,60 +719,58 @@ func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, ref *ref
return nil
}
-// doDuplicateAddressDetection is called on every iteration of the timer, and
-// when DAD starts.
-//
-// It handles resolving the address (if there are no more NS to send), or
-// sending the next NS if there are more NS to send.
-//
-// This function must only be called by IPv6 addresses that are currently
-// tentative.
+// sendDADPacket sends a NS message to see if any nodes on ndp's NIC's link owns
+// addr.
//
-// The NIC that ndp belongs to (n) MUST be locked.
+// addr must be a tentative IPv6 address on ndp's NIC.
//
-// Returns true if DAD has resolved; false if DAD is still ongoing.
-func (ndp *ndpState) doDuplicateAddressDetection(addr tcpip.Address, remaining uint8, ref *referencedNetworkEndpoint) (bool, *tcpip.Error) {
- if ref.getKind() != permanentTentative {
- // The endpoint should still be marked as tentative
- // since we are still performing DAD on it.
- panic(fmt.Sprintf("ndpdad: addr %s is not tentative on NIC(%d)", addr, ndp.nic.ID()))
- }
+// The NIC ndp belongs to MUST NOT be locked.
+func (ndp *ndpState) sendDADPacket(addr tcpip.Address, ref *referencedNetworkEndpoint) *tcpip.Error {
+ snmc := header.SolicitedNodeAddr(addr)
- if remaining == 0 {
- // DAD has resolved.
- ref.setKind(permanent)
- return true, nil
- }
+ r := makeRoute(header.IPv6ProtocolNumber, ref.address(), snmc, ndp.nic.linkEP.LinkAddress(), ref, false, false)
+ defer r.Release()
- // Send a new NS.
- snmc := header.SolicitedNodeAddr(addr)
- snmcRef, ok := ndp.nic.endpoints[NetworkEndpointID{snmc}]
- if !ok {
- // This should never happen as if we have the
- // address, we should have the solicited-node
- // address.
- panic(fmt.Sprintf("ndpdad: NIC(%d) is not in the solicited-node multicast group (%s) but it has addr %s", ndp.nic.ID(), snmc, addr))
- }
+ // Route should resolve immediately since snmc is a multicast address so a
+ // remote link address can be calculated without a resolution process.
+ if c, err := r.Resolve(nil); err != nil {
+ // Do not consider the NIC being unknown or disabled as a fatal error.
+ // Since this method is required to be called when the NIC is not locked,
+ // the NIC could have been disabled or removed by another goroutine.
+ if err == tcpip.ErrUnknownNICID || err != tcpip.ErrInvalidEndpointState {
+ return err
+ }
- // Use the unspecified address as the source address when performing
- // DAD.
- r := makeRoute(header.IPv6ProtocolNumber, header.IPv6Any, snmc, ndp.nic.linkEP.LinkAddress(), snmcRef, false, false)
+ panic(fmt.Sprintf("ndp: error when resolving route to send NDP NS for DAD (%s -> %s on NIC(%d)): %s", header.IPv6Any, snmc, ndp.nic.ID(), err))
+ } else if c != nil {
+ panic(fmt.Sprintf("ndp: route resolution not immediate for route to send NDP NS for DAD (%s -> %s on NIC(%d))", header.IPv6Any, snmc, ndp.nic.ID()))
+ }
- hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv6NeighborSolicitMinimumSize)
- pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborSolicitMinimumSize))
- pkt.SetType(header.ICMPv6NeighborSolicit)
- ns := header.NDPNeighborSolicit(pkt.NDPPayload())
+ icmpData := header.ICMPv6(buffer.NewView(header.ICMPv6NeighborSolicitMinimumSize))
+ icmpData.SetType(header.ICMPv6NeighborSolicit)
+ ns := header.NDPNeighborSolicit(icmpData.NDPPayload())
ns.SetTargetAddress(addr)
- pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{}))
+ icmpData.SetChecksum(header.ICMPv6Checksum(icmpData, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{}))
+
+ pkt := NewPacketBuffer(PacketBufferOptions{
+ ReserveHeaderBytes: int(r.MaxHeaderLength()),
+ Data: buffer.View(icmpData).ToVectorisedView(),
+ })
sent := r.Stats().ICMP.V6PacketsSent
- if err := r.WritePacket(nil, hdr, buffer.VectorisedView{}, NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: DefaultTOS}); err != nil {
+ if err := r.WritePacket(nil,
+ NetworkHeaderParams{
+ Protocol: header.ICMPv6ProtocolNumber,
+ TTL: header.NDPHopLimit,
+ TOS: DefaultTOS,
+ }, pkt,
+ ); err != nil {
sent.Dropped.Increment()
- return false, err
+ return err
}
sent.NeighborSolicit.Increment()
- return false, nil
+ return nil
}
// stopDuplicateAddressDetection ends a running Duplicate Address Detection
@@ -315,7 +801,1173 @@ func (ndp *ndpState) stopDuplicateAddressDetection(addr tcpip.Address) {
delete(ndp.dad, addr)
// Let the integrator know DAD did not resolve.
- if ndp.nic.stack.ndpDisp != nil {
- go ndp.nic.stack.ndpDisp.OnDuplicateAddressDetectionStatus(ndp.nic.ID(), addr, false, nil)
+ if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil {
+ ndpDisp.OnDuplicateAddressDetectionStatus(ndp.nic.ID(), addr, false, nil)
+ }
+}
+
+// handleRA handles a Router Advertisement message that arrived on the NIC
+// this ndp is for. Does nothing if the NIC is configured to not handle RAs.
+//
+// The NIC that ndp belongs to MUST be locked.
+func (ndp *ndpState) handleRA(ip tcpip.Address, ra header.NDPRouterAdvert) {
+ // Is the NIC configured to handle RAs at all?
+ //
+ // Currently, the stack does not determine router interface status on a
+ // per-interface basis; it is a stack-wide configuration, so we check
+ // stack's forwarding flag to determine if the NIC is a routing
+ // interface.
+ if !ndp.configs.HandleRAs || ndp.nic.stack.Forwarding(header.IPv6ProtocolNumber) {
+ return
+ }
+
+ // Only worry about the DHCPv6 configuration if we have an NDPDispatcher as we
+ // only inform the dispatcher on configuration changes. We do nothing else
+ // with the information.
+ if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil {
+ var configuration DHCPv6ConfigurationFromNDPRA
+ switch {
+ case ra.ManagedAddrConfFlag():
+ configuration = DHCPv6ManagedAddress
+
+ case ra.OtherConfFlag():
+ configuration = DHCPv6OtherConfigurations
+
+ default:
+ configuration = DHCPv6NoConfiguration
+ }
+
+ if ndp.dhcpv6Configuration != configuration {
+ ndp.dhcpv6Configuration = configuration
+ ndpDisp.OnDHCPv6Configuration(ndp.nic.ID(), configuration)
+ }
+ }
+
+ // Is the NIC configured to discover default routers?
+ if ndp.configs.DiscoverDefaultRouters {
+ rtr, ok := ndp.defaultRouters[ip]
+ rl := ra.RouterLifetime()
+ switch {
+ case !ok && rl != 0:
+ // This is a new default router we are discovering.
+ //
+ // Only remember it if we currently know about less than
+ // MaxDiscoveredDefaultRouters routers.
+ if len(ndp.defaultRouters) < MaxDiscoveredDefaultRouters {
+ ndp.rememberDefaultRouter(ip, rl)
+ }
+
+ case ok && rl != 0:
+ // This is an already discovered default router. Update
+ // the invalidation job.
+ rtr.invalidationJob.Cancel()
+ rtr.invalidationJob.Schedule(rl)
+ ndp.defaultRouters[ip] = rtr
+
+ case ok && rl == 0:
+ // We know about the router but it is no longer to be
+ // used as a default router so invalidate it.
+ ndp.invalidateDefaultRouter(ip)
+ }
+ }
+
+ // TODO(b/141556115): Do (RetransTimer, ReachableTime)) Parameter
+ // Discovery.
+
+ // We know the options is valid as far as wire format is concerned since
+ // we got the Router Advertisement, as documented by this fn. Given this
+ // we do not check the iterator for errors on calls to Next.
+ it, _ := ra.Options().Iter(false)
+ for opt, done, _ := it.Next(); !done; opt, done, _ = it.Next() {
+ switch opt := opt.(type) {
+ case header.NDPRecursiveDNSServer:
+ if ndp.nic.stack.ndpDisp == nil {
+ continue
+ }
+
+ addrs, _ := opt.Addresses()
+ ndp.nic.stack.ndpDisp.OnRecursiveDNSServerOption(ndp.nic.ID(), addrs, opt.Lifetime())
+
+ case header.NDPDNSSearchList:
+ if ndp.nic.stack.ndpDisp == nil {
+ continue
+ }
+
+ domainNames, _ := opt.DomainNames()
+ ndp.nic.stack.ndpDisp.OnDNSSearchListOption(ndp.nic.ID(), domainNames, opt.Lifetime())
+
+ case header.NDPPrefixInformation:
+ prefix := opt.Subnet()
+
+ // Is the prefix a link-local?
+ if header.IsV6LinkLocalAddress(prefix.ID()) {
+ // ...Yes, skip as per RFC 4861 section 6.3.4,
+ // and RFC 4862 section 5.5.3.b (for SLAAC).
+ continue
+ }
+
+ // Is the Prefix Length 0?
+ if prefix.Prefix() == 0 {
+ // ...Yes, skip as this is an invalid prefix
+ // as all IPv6 addresses cannot be on-link.
+ continue
+ }
+
+ if opt.OnLinkFlag() {
+ ndp.handleOnLinkPrefixInformation(opt)
+ }
+
+ if opt.AutonomousAddressConfigurationFlag() {
+ ndp.handleAutonomousPrefixInformation(opt)
+ }
+ }
+
+ // TODO(b/141556115): Do (MTU) Parameter Discovery.
+ }
+}
+
+// invalidateDefaultRouter invalidates a discovered default router.
+//
+// The NIC that ndp belongs to MUST be locked.
+func (ndp *ndpState) invalidateDefaultRouter(ip tcpip.Address) {
+ rtr, ok := ndp.defaultRouters[ip]
+
+ // Is the router still discovered?
+ if !ok {
+ // ...Nope, do nothing further.
+ return
+ }
+
+ rtr.invalidationJob.Cancel()
+ delete(ndp.defaultRouters, ip)
+
+ // Let the integrator know a discovered default router is invalidated.
+ if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil {
+ ndpDisp.OnDefaultRouterInvalidated(ndp.nic.ID(), ip)
+ }
+}
+
+// rememberDefaultRouter remembers a newly discovered default router with IPv6
+// link-local address ip with lifetime rl.
+//
+// The router identified by ip MUST NOT already be known by the NIC.
+//
+// The NIC that ndp belongs to MUST be locked.
+func (ndp *ndpState) rememberDefaultRouter(ip tcpip.Address, rl time.Duration) {
+ ndpDisp := ndp.nic.stack.ndpDisp
+ if ndpDisp == nil {
+ return
+ }
+
+ // Inform the integrator when we discovered a default router.
+ if !ndpDisp.OnDefaultRouterDiscovered(ndp.nic.ID(), ip) {
+ // Informed by the integrator to not remember the router, do
+ // nothing further.
+ return
+ }
+
+ state := defaultRouterState{
+ invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() {
+ ndp.invalidateDefaultRouter(ip)
+ }),
+ }
+
+ state.invalidationJob.Schedule(rl)
+
+ ndp.defaultRouters[ip] = state
+}
+
+// rememberOnLinkPrefix remembers a newly discovered on-link prefix with IPv6
+// address with prefix prefix with lifetime l.
+//
+// The prefix identified by prefix MUST NOT already be known.
+//
+// The NIC that ndp belongs to MUST be locked.
+func (ndp *ndpState) rememberOnLinkPrefix(prefix tcpip.Subnet, l time.Duration) {
+ ndpDisp := ndp.nic.stack.ndpDisp
+ if ndpDisp == nil {
+ return
+ }
+
+ // Inform the integrator when we discovered an on-link prefix.
+ if !ndpDisp.OnOnLinkPrefixDiscovered(ndp.nic.ID(), prefix) {
+ // Informed by the integrator to not remember the prefix, do
+ // nothing further.
+ return
+ }
+
+ state := onLinkPrefixState{
+ invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() {
+ ndp.invalidateOnLinkPrefix(prefix)
+ }),
+ }
+
+ if l < header.NDPInfiniteLifetime {
+ state.invalidationJob.Schedule(l)
+ }
+
+ ndp.onLinkPrefixes[prefix] = state
+}
+
+// invalidateOnLinkPrefix invalidates a discovered on-link prefix.
+//
+// The NIC that ndp belongs to MUST be locked.
+func (ndp *ndpState) invalidateOnLinkPrefix(prefix tcpip.Subnet) {
+ s, ok := ndp.onLinkPrefixes[prefix]
+
+ // Is the on-link prefix still discovered?
+ if !ok {
+ // ...Nope, do nothing further.
+ return
+ }
+
+ s.invalidationJob.Cancel()
+ delete(ndp.onLinkPrefixes, prefix)
+
+ // Let the integrator know a discovered on-link prefix is invalidated.
+ if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil {
+ ndpDisp.OnOnLinkPrefixInvalidated(ndp.nic.ID(), prefix)
+ }
+}
+
+// handleOnLinkPrefixInformation handles a Prefix Information option with
+// its on-link flag set, as per RFC 4861 section 6.3.4.
+//
+// handleOnLinkPrefixInformation assumes that the prefix this pi is for is
+// not the link-local prefix and the on-link flag is set.
+//
+// The NIC that ndp belongs to MUST be locked.
+func (ndp *ndpState) handleOnLinkPrefixInformation(pi header.NDPPrefixInformation) {
+ prefix := pi.Subnet()
+ prefixState, ok := ndp.onLinkPrefixes[prefix]
+ vl := pi.ValidLifetime()
+
+ if !ok && vl == 0 {
+ // Don't know about this prefix but it has a zero valid
+ // lifetime, so just ignore.
+ return
+ }
+
+ if !ok && vl != 0 {
+ // This is a new on-link prefix we are discovering
+ //
+ // Only remember it if we currently know about less than
+ // MaxDiscoveredOnLinkPrefixes on-link prefixes.
+ if ndp.configs.DiscoverOnLinkPrefixes && len(ndp.onLinkPrefixes) < MaxDiscoveredOnLinkPrefixes {
+ ndp.rememberOnLinkPrefix(prefix, vl)
+ }
+ return
+ }
+
+ if ok && vl == 0 {
+ // We know about the on-link prefix, but it is
+ // no longer to be considered on-link, so
+ // invalidate it.
+ ndp.invalidateOnLinkPrefix(prefix)
+ return
+ }
+
+ // This is an already discovered on-link prefix with a
+ // new non-zero valid lifetime.
+ //
+ // Update the invalidation job.
+
+ prefixState.invalidationJob.Cancel()
+
+ if vl < header.NDPInfiniteLifetime {
+ // Prefix is valid for a finite lifetime, schedule the job to execute after
+ // the new valid lifetime.
+ prefixState.invalidationJob.Schedule(vl)
+ }
+
+ ndp.onLinkPrefixes[prefix] = prefixState
+}
+
+// handleAutonomousPrefixInformation handles a Prefix Information option with
+// its autonomous flag set, as per RFC 4862 section 5.5.3.
+//
+// handleAutonomousPrefixInformation assumes that the prefix this pi is for is
+// not the link-local prefix and the autonomous flag is set.
+//
+// The NIC that ndp belongs to MUST be locked.
+func (ndp *ndpState) handleAutonomousPrefixInformation(pi header.NDPPrefixInformation) {
+ vl := pi.ValidLifetime()
+ pl := pi.PreferredLifetime()
+
+ // If the preferred lifetime is greater than the valid lifetime,
+ // silently ignore the Prefix Information option, as per RFC 4862
+ // section 5.5.3.c.
+ if pl > vl {
+ return
+ }
+
+ prefix := pi.Subnet()
+
+ // Check if we already maintain SLAAC state for prefix.
+ if state, ok := ndp.slaacPrefixes[prefix]; ok {
+ // As per RFC 4862 section 5.5.3.e, refresh prefix's SLAAC lifetimes.
+ ndp.refreshSLAACPrefixLifetimes(prefix, &state, pl, vl)
+ ndp.slaacPrefixes[prefix] = state
+ return
+ }
+
+ // prefix is a new SLAAC prefix. Do the work as outlined by RFC 4862 section
+ // 5.5.3.d if ndp is configured to auto-generate new addresses via SLAAC.
+ if !ndp.configs.AutoGenGlobalAddresses {
+ return
+ }
+
+ ndp.doSLAAC(prefix, pl, vl)
+}
+
+// doSLAAC generates a new SLAAC address with the provided lifetimes
+// for prefix.
+//
+// pl is the new preferred lifetime. vl is the new valid lifetime.
+//
+// The NIC that ndp belongs to MUST be locked.
+func (ndp *ndpState) doSLAAC(prefix tcpip.Subnet, pl, vl time.Duration) {
+ // If we do not already have an address for this prefix and the valid
+ // lifetime is 0, no need to do anything further, as per RFC 4862
+ // section 5.5.3.d.
+ if vl == 0 {
+ return
+ }
+
+ // Make sure the prefix is valid (as far as its length is concerned) to
+ // generate a valid IPv6 address from an interface identifier (IID), as
+ // per RFC 4862 sectiion 5.5.3.d.
+ if prefix.Prefix() != validPrefixLenForAutoGen {
+ return
+ }
+
+ state := slaacPrefixState{
+ deprecationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() {
+ state, ok := ndp.slaacPrefixes[prefix]
+ if !ok {
+ panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for the deprecated SLAAC prefix %s", prefix))
+ }
+
+ ndp.deprecateSLAACAddress(state.stableAddr.ref)
+ }),
+ invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() {
+ state, ok := ndp.slaacPrefixes[prefix]
+ if !ok {
+ panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for the invalidated SLAAC prefix %s", prefix))
+ }
+
+ ndp.invalidateSLAACPrefix(prefix, state)
+ }),
+ tempAddrs: make(map[tcpip.Address]tempSLAACAddrState),
+ maxGenerationAttempts: ndp.configs.AutoGenAddressConflictRetries + 1,
+ }
+
+ now := time.Now()
+
+ // The time an address is preferred until is needed to properly generate the
+ // address.
+ if pl < header.NDPInfiniteLifetime {
+ state.preferredUntil = now.Add(pl)
+ }
+
+ if !ndp.generateSLAACAddr(prefix, &state) {
+ // We were unable to generate an address for the prefix, we do not nothing
+ // further as there is no reason to maintain state or jobs for a prefix we
+ // do not have an address for.
+ return
+ }
+
+ // Setup the initial jobs to deprecate and invalidate prefix.
+
+ if pl < header.NDPInfiniteLifetime && pl != 0 {
+ state.deprecationJob.Schedule(pl)
+ }
+
+ if vl < header.NDPInfiniteLifetime {
+ state.invalidationJob.Schedule(vl)
+ state.validUntil = now.Add(vl)
+ }
+
+ // If the address is assigned (DAD resolved), generate a temporary address.
+ if state.stableAddr.ref.getKind() == permanent {
+ // Reset the generation attempts counter as we are starting the generation
+ // of a new address for the SLAAC prefix.
+ ndp.generateTempSLAACAddr(prefix, &state, true /* resetGenAttempts */)
+ }
+
+ ndp.slaacPrefixes[prefix] = state
+}
+
+// addSLAACAddr adds a SLAAC address to the NIC.
+//
+// The NIC that ndp belongs to MUST be locked.
+func (ndp *ndpState) addSLAACAddr(addr tcpip.AddressWithPrefix, configType networkEndpointConfigType, deprecated bool) *referencedNetworkEndpoint {
+ // Inform the integrator that we have a new SLAAC address.
+ ndpDisp := ndp.nic.stack.ndpDisp
+ if ndpDisp == nil {
+ return nil
+ }
+
+ if !ndpDisp.OnAutoGenAddress(ndp.nic.ID(), addr) {
+ // Informed by the integrator not to add the address.
+ return nil
+ }
+
+ protocolAddr := tcpip.ProtocolAddress{
+ Protocol: header.IPv6ProtocolNumber,
+ AddressWithPrefix: addr,
+ }
+
+ ref, err := ndp.nic.addAddressLocked(protocolAddr, FirstPrimaryEndpoint, permanent, configType, deprecated)
+ if err != nil {
+ panic(fmt.Sprintf("ndp: error when adding SLAAC address %+v: %s", protocolAddr, err))
+ }
+
+ return ref
+}
+
+// generateSLAACAddr generates a SLAAC address for prefix.
+//
+// Returns true if an address was successfully generated.
+//
+// Panics if the prefix is not a SLAAC prefix or it already has an address.
+//
+// The NIC that ndp belongs to MUST be locked.
+func (ndp *ndpState) generateSLAACAddr(prefix tcpip.Subnet, state *slaacPrefixState) bool {
+ if r := state.stableAddr.ref; r != nil {
+ panic(fmt.Sprintf("ndp: SLAAC prefix %s already has a permenant address %s", prefix, r.addrWithPrefix()))
+ }
+
+ // If we have already reached the maximum address generation attempts for the
+ // prefix, do not generate another address.
+ if state.generationAttempts == state.maxGenerationAttempts {
+ return false
+ }
+
+ var generatedAddr tcpip.AddressWithPrefix
+ addrBytes := []byte(prefix.ID())
+
+ for i := 0; ; i++ {
+ // If we were unable to generate an address after the maximum SLAAC address
+ // local regeneration attempts, do nothing further.
+ if i == maxSLAACAddrLocalRegenAttempts {
+ return false
+ }
+
+ dadCounter := state.generationAttempts + state.stableAddr.localGenerationFailures
+ if oIID := ndp.nic.stack.opaqueIIDOpts; oIID.NICNameFromID != nil {
+ addrBytes = header.AppendOpaqueInterfaceIdentifier(
+ addrBytes[:header.IIDOffsetInIPv6Address],
+ prefix,
+ oIID.NICNameFromID(ndp.nic.ID(), ndp.nic.name),
+ dadCounter,
+ oIID.SecretKey,
+ )
+ } else if dadCounter == 0 {
+ // Modified-EUI64 based IIDs have no way to resolve DAD conflicts, so if
+ // the DAD counter is non-zero, we cannot use this method.
+ //
+ // Only attempt to generate an interface-specific IID if we have a valid
+ // link address.
+ //
+ // TODO(b/141011931): Validate a LinkEndpoint's link address (provided by
+ // LinkEndpoint.LinkAddress) before reaching this point.
+ linkAddr := ndp.nic.linkEP.LinkAddress()
+ if !header.IsValidUnicastEthernetAddress(linkAddr) {
+ return false
+ }
+
+ // Generate an address within prefix from the modified EUI-64 of ndp's
+ // NIC's Ethernet MAC address.
+ header.EthernetAdddressToModifiedEUI64IntoBuf(linkAddr, addrBytes[header.IIDOffsetInIPv6Address:])
+ } else {
+ // We have no way to regenerate an address in response to an address
+ // conflict when addresses are not generated with opaque IIDs.
+ return false
+ }
+
+ generatedAddr = tcpip.AddressWithPrefix{
+ Address: tcpip.Address(addrBytes),
+ PrefixLen: validPrefixLenForAutoGen,
+ }
+
+ if !ndp.nic.hasPermanentAddrLocked(generatedAddr.Address) {
+ break
+ }
+
+ state.stableAddr.localGenerationFailures++
+ }
+
+ if ref := ndp.addSLAACAddr(generatedAddr, slaac, time.Since(state.preferredUntil) >= 0 /* deprecated */); ref != nil {
+ state.stableAddr.ref = ref
+ state.generationAttempts++
+ return true
+ }
+
+ return false
+}
+
+// regenerateSLAACAddr regenerates an address for a SLAAC prefix.
+//
+// If generating a new address for the prefix fails, the prefix will be
+// invalidated.
+//
+// The NIC that ndp belongs to MUST be locked.
+func (ndp *ndpState) regenerateSLAACAddr(prefix tcpip.Subnet) {
+ state, ok := ndp.slaacPrefixes[prefix]
+ if !ok {
+ panic(fmt.Sprintf("ndp: SLAAC prefix state not found to regenerate address for %s", prefix))
+ }
+
+ if ndp.generateSLAACAddr(prefix, &state) {
+ ndp.slaacPrefixes[prefix] = state
+ return
+ }
+
+ // We were unable to generate a permanent address for the SLAAC prefix so
+ // invalidate the prefix as there is no reason to maintain state for a
+ // SLAAC prefix we do not have an address for.
+ ndp.invalidateSLAACPrefix(prefix, state)
+}
+
+// generateTempSLAACAddr generates a new temporary SLAAC address.
+//
+// If resetGenAttempts is true, the prefix's generation counter will be reset.
+//
+// Returns true if a new address was generated.
+func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *slaacPrefixState, resetGenAttempts bool) bool {
+ // Are we configured to auto-generate new temporary global addresses for the
+ // prefix?
+ if !ndp.configs.AutoGenTempGlobalAddresses || prefix == header.IPv6LinkLocalPrefix.Subnet() {
+ return false
+ }
+
+ if resetGenAttempts {
+ prefixState.generationAttempts = 0
+ prefixState.maxGenerationAttempts = ndp.configs.AutoGenAddressConflictRetries + 1
+ }
+
+ // If we have already reached the maximum address generation attempts for the
+ // prefix, do not generate another address.
+ if prefixState.generationAttempts == prefixState.maxGenerationAttempts {
+ return false
+ }
+
+ stableAddr := prefixState.stableAddr.ref.address()
+ now := time.Now()
+
+ // As per RFC 4941 section 3.3 step 4, the valid lifetime of a temporary
+ // address is the lower of the valid lifetime of the stable address or the
+ // maximum temporary address valid lifetime.
+ vl := ndp.configs.MaxTempAddrValidLifetime
+ if prefixState.validUntil != (time.Time{}) {
+ if prefixVL := prefixState.validUntil.Sub(now); vl > prefixVL {
+ vl = prefixVL
+ }
+ }
+
+ if vl <= 0 {
+ // Cannot create an address without a valid lifetime.
+ return false
+ }
+
+ // As per RFC 4941 section 3.3 step 4, the preferred lifetime of a temporary
+ // address is the lower of the preferred lifetime of the stable address or the
+ // maximum temporary address preferred lifetime - the temporary address desync
+ // factor.
+ pl := ndp.configs.MaxTempAddrPreferredLifetime - ndp.temporaryAddressDesyncFactor
+ if prefixState.preferredUntil != (time.Time{}) {
+ if prefixPL := prefixState.preferredUntil.Sub(now); pl > prefixPL {
+ // Respect the preferred lifetime of the prefix, as per RFC 4941 section
+ // 3.3 step 4.
+ pl = prefixPL
+ }
+ }
+
+ // As per RFC 4941 section 3.3 step 5, a temporary address is created only if
+ // the calculated preferred lifetime is greater than the advance regeneration
+ // duration. In particular, we MUST NOT create a temporary address with a zero
+ // Preferred Lifetime.
+ if pl <= ndp.configs.RegenAdvanceDuration {
+ return false
+ }
+
+ // Attempt to generate a new address that is not already assigned to the NIC.
+ var generatedAddr tcpip.AddressWithPrefix
+ for i := 0; ; i++ {
+ // If we were unable to generate an address after the maximum SLAAC address
+ // local regeneration attempts, do nothing further.
+ if i == maxSLAACAddrLocalRegenAttempts {
+ return false
+ }
+
+ generatedAddr = header.GenerateTempIPv6SLAACAddr(ndp.temporaryIIDHistory[:], stableAddr)
+ if !ndp.nic.hasPermanentAddrLocked(generatedAddr.Address) {
+ break
+ }
+ }
+
+ // As per RFC RFC 4941 section 3.3 step 5, we MUST NOT create a temporary
+ // address with a zero preferred lifetime. The checks above ensure this
+ // so we know the address is not deprecated.
+ ref := ndp.addSLAACAddr(generatedAddr, slaacTemp, false /* deprecated */)
+ if ref == nil {
+ return false
+ }
+
+ state := tempSLAACAddrState{
+ deprecationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() {
+ prefixState, ok := ndp.slaacPrefixes[prefix]
+ if !ok {
+ panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for %s to deprecate temporary address %s", prefix, generatedAddr))
+ }
+
+ tempAddrState, ok := prefixState.tempAddrs[generatedAddr.Address]
+ if !ok {
+ panic(fmt.Sprintf("ndp: must have a tempAddr entry to deprecate temporary address %s", generatedAddr))
+ }
+
+ ndp.deprecateSLAACAddress(tempAddrState.ref)
+ }),
+ invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() {
+ prefixState, ok := ndp.slaacPrefixes[prefix]
+ if !ok {
+ panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for %s to invalidate temporary address %s", prefix, generatedAddr))
+ }
+
+ tempAddrState, ok := prefixState.tempAddrs[generatedAddr.Address]
+ if !ok {
+ panic(fmt.Sprintf("ndp: must have a tempAddr entry to invalidate temporary address %s", generatedAddr))
+ }
+
+ ndp.invalidateTempSLAACAddr(prefixState.tempAddrs, generatedAddr.Address, tempAddrState)
+ }),
+ regenJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() {
+ prefixState, ok := ndp.slaacPrefixes[prefix]
+ if !ok {
+ panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for %s to regenerate temporary address after %s", prefix, generatedAddr))
+ }
+
+ tempAddrState, ok := prefixState.tempAddrs[generatedAddr.Address]
+ if !ok {
+ panic(fmt.Sprintf("ndp: must have a tempAddr entry to regenerate temporary address after %s", generatedAddr))
+ }
+
+ // If an address has already been regenerated for this address, don't
+ // regenerate another address.
+ if tempAddrState.regenerated {
+ return
+ }
+
+ // Reset the generation attempts counter as we are starting the generation
+ // of a new address for the SLAAC prefix.
+ tempAddrState.regenerated = ndp.generateTempSLAACAddr(prefix, &prefixState, true /* resetGenAttempts */)
+ prefixState.tempAddrs[generatedAddr.Address] = tempAddrState
+ ndp.slaacPrefixes[prefix] = prefixState
+ }),
+ createdAt: now,
+ ref: ref,
+ }
+
+ state.deprecationJob.Schedule(pl)
+ state.invalidationJob.Schedule(vl)
+ state.regenJob.Schedule(pl - ndp.configs.RegenAdvanceDuration)
+
+ prefixState.generationAttempts++
+ prefixState.tempAddrs[generatedAddr.Address] = state
+
+ return true
+}
+
+// regenerateTempSLAACAddr regenerates a temporary address for a SLAAC prefix.
+//
+// The NIC that ndp belongs to MUST be locked.
+func (ndp *ndpState) regenerateTempSLAACAddr(prefix tcpip.Subnet, resetGenAttempts bool) {
+ state, ok := ndp.slaacPrefixes[prefix]
+ if !ok {
+ panic(fmt.Sprintf("ndp: SLAAC prefix state not found to regenerate temporary address for %s", prefix))
+ }
+
+ ndp.generateTempSLAACAddr(prefix, &state, resetGenAttempts)
+ ndp.slaacPrefixes[prefix] = state
+}
+
+// refreshSLAACPrefixLifetimes refreshes the lifetimes of a SLAAC prefix.
+//
+// pl is the new preferred lifetime. vl is the new valid lifetime.
+//
+// The NIC that ndp belongs to MUST be locked.
+func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixState *slaacPrefixState, pl, vl time.Duration) {
+ // If the preferred lifetime is zero, then the prefix should be deprecated.
+ deprecated := pl == 0
+ if deprecated {
+ ndp.deprecateSLAACAddress(prefixState.stableAddr.ref)
+ } else {
+ prefixState.stableAddr.ref.deprecated = false
+ }
+
+ // If prefix was preferred for some finite lifetime before, cancel the
+ // deprecation job so it can be reset.
+ prefixState.deprecationJob.Cancel()
+
+ now := time.Now()
+
+ // Schedule the deprecation job if prefix has a finite preferred lifetime.
+ if pl < header.NDPInfiniteLifetime {
+ if !deprecated {
+ prefixState.deprecationJob.Schedule(pl)
+ }
+ prefixState.preferredUntil = now.Add(pl)
+ } else {
+ prefixState.preferredUntil = time.Time{}
+ }
+
+ // As per RFC 4862 section 5.5.3.e, update the valid lifetime for prefix:
+ //
+ // 1) If the received Valid Lifetime is greater than 2 hours or greater than
+ // RemainingLifetime, set the valid lifetime of the prefix to the
+ // advertised Valid Lifetime.
+ //
+ // 2) If RemainingLifetime is less than or equal to 2 hours, ignore the
+ // advertised Valid Lifetime.
+ //
+ // 3) Otherwise, reset the valid lifetime of the prefix to 2 hours.
+
+ if vl >= header.NDPInfiniteLifetime {
+ // Handle the infinite valid lifetime separately as we do not schedule a
+ // job in this case.
+ prefixState.invalidationJob.Cancel()
+ prefixState.validUntil = time.Time{}
+ } else {
+ var effectiveVl time.Duration
+ var rl time.Duration
+
+ // If the prefix was originally set to be valid forever, assume the
+ // remaining time to be the maximum possible value.
+ if prefixState.validUntil == (time.Time{}) {
+ rl = header.NDPInfiniteLifetime
+ } else {
+ rl = time.Until(prefixState.validUntil)
+ }
+
+ if vl > MinPrefixInformationValidLifetimeForUpdate || vl > rl {
+ effectiveVl = vl
+ } else if rl > MinPrefixInformationValidLifetimeForUpdate {
+ effectiveVl = MinPrefixInformationValidLifetimeForUpdate
+ }
+
+ if effectiveVl != 0 {
+ prefixState.invalidationJob.Cancel()
+ prefixState.invalidationJob.Schedule(effectiveVl)
+ prefixState.validUntil = now.Add(effectiveVl)
+ }
+ }
+
+ // If DAD is not yet complete on the stable address, there is no need to do
+ // work with temporary addresses.
+ if prefixState.stableAddr.ref.getKind() != permanent {
+ return
+ }
+
+ // Note, we do not need to update the entries in the temporary address map
+ // after updating the jobs because the jobs are held as pointers.
+ var regenForAddr tcpip.Address
+ allAddressesRegenerated := true
+ for tempAddr, tempAddrState := range prefixState.tempAddrs {
+ // As per RFC 4941 section 3.3 step 4, the valid lifetime of a temporary
+ // address is the lower of the valid lifetime of the stable address or the
+ // maximum temporary address valid lifetime. Note, the valid lifetime of a
+ // temporary address is relative to the address's creation time.
+ validUntil := tempAddrState.createdAt.Add(ndp.configs.MaxTempAddrValidLifetime)
+ if prefixState.validUntil != (time.Time{}) && validUntil.Sub(prefixState.validUntil) > 0 {
+ validUntil = prefixState.validUntil
+ }
+
+ // If the address is no longer valid, invalidate it immediately. Otherwise,
+ // reset the invalidation job.
+ newValidLifetime := validUntil.Sub(now)
+ if newValidLifetime <= 0 {
+ ndp.invalidateTempSLAACAddr(prefixState.tempAddrs, tempAddr, tempAddrState)
+ continue
+ }
+ tempAddrState.invalidationJob.Cancel()
+ tempAddrState.invalidationJob.Schedule(newValidLifetime)
+
+ // As per RFC 4941 section 3.3 step 4, the preferred lifetime of a temporary
+ // address is the lower of the preferred lifetime of the stable address or
+ // the maximum temporary address preferred lifetime - the temporary address
+ // desync factor. Note, the preferred lifetime of a temporary address is
+ // relative to the address's creation time.
+ preferredUntil := tempAddrState.createdAt.Add(ndp.configs.MaxTempAddrPreferredLifetime - ndp.temporaryAddressDesyncFactor)
+ if prefixState.preferredUntil != (time.Time{}) && preferredUntil.Sub(prefixState.preferredUntil) > 0 {
+ preferredUntil = prefixState.preferredUntil
+ }
+
+ // If the address is no longer preferred, deprecate it immediately.
+ // Otherwise, schedule the deprecation job again.
+ newPreferredLifetime := preferredUntil.Sub(now)
+ tempAddrState.deprecationJob.Cancel()
+ if newPreferredLifetime <= 0 {
+ ndp.deprecateSLAACAddress(tempAddrState.ref)
+ } else {
+ tempAddrState.ref.deprecated = false
+ tempAddrState.deprecationJob.Schedule(newPreferredLifetime)
+ }
+
+ tempAddrState.regenJob.Cancel()
+ if tempAddrState.regenerated {
+ } else {
+ allAddressesRegenerated = false
+
+ if newPreferredLifetime <= ndp.configs.RegenAdvanceDuration {
+ // The new preferred lifetime is less than the advance regeneration
+ // duration so regenerate an address for this temporary address
+ // immediately after we finish iterating over the temporary addresses.
+ regenForAddr = tempAddr
+ } else {
+ tempAddrState.regenJob.Schedule(newPreferredLifetime - ndp.configs.RegenAdvanceDuration)
+ }
+ }
+ }
+
+ // Generate a new temporary address if all of the existing temporary addresses
+ // have been regenerated, or we need to immediately regenerate an address
+ // due to an update in preferred lifetime.
+ //
+ // If each temporay address has already been regenerated, no new temporary
+ // address will be generated. To ensure continuation of temporary SLAAC
+ // addresses, we manually try to regenerate an address here.
+ if len(regenForAddr) != 0 || allAddressesRegenerated {
+ // Reset the generation attempts counter as we are starting the generation
+ // of a new address for the SLAAC prefix.
+ if state, ok := prefixState.tempAddrs[regenForAddr]; ndp.generateTempSLAACAddr(prefix, prefixState, true /* resetGenAttempts */) && ok {
+ state.regenerated = true
+ prefixState.tempAddrs[regenForAddr] = state
+ }
+ }
+}
+
+// deprecateSLAACAddress marks ref as deprecated and notifies the stack's NDP
+// dispatcher that ref has been deprecated.
+//
+// deprecateSLAACAddress does nothing if ref is already deprecated.
+//
+// The NIC that ndp belongs to MUST be locked.
+func (ndp *ndpState) deprecateSLAACAddress(ref *referencedNetworkEndpoint) {
+ if ref.deprecated {
+ return
+ }
+
+ ref.deprecated = true
+ if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil {
+ ndpDisp.OnAutoGenAddressDeprecated(ndp.nic.ID(), ref.addrWithPrefix())
+ }
+}
+
+// invalidateSLAACPrefix invalidates a SLAAC prefix.
+//
+// The NIC that ndp belongs to MUST be locked.
+func (ndp *ndpState) invalidateSLAACPrefix(prefix tcpip.Subnet, state slaacPrefixState) {
+ if r := state.stableAddr.ref; r != nil {
+ // Since we are already invalidating the prefix, do not invalidate the
+ // prefix when removing the address.
+ if err := ndp.nic.removePermanentIPv6EndpointLocked(r, false /* allowSLAACInvalidation */); err != nil {
+ panic(fmt.Sprintf("ndp: error removing stable SLAAC address %s: %s", r.addrWithPrefix(), err))
+ }
+ }
+
+ ndp.cleanupSLAACPrefixResources(prefix, state)
+}
+
+// cleanupSLAACAddrResourcesAndNotify cleans up an invalidated SLAAC address's
+// resources.
+//
+// The NIC that ndp belongs to MUST be locked.
+func (ndp *ndpState) cleanupSLAACAddrResourcesAndNotify(addr tcpip.AddressWithPrefix, invalidatePrefix bool) {
+ if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil {
+ ndpDisp.OnAutoGenAddressInvalidated(ndp.nic.ID(), addr)
+ }
+
+ prefix := addr.Subnet()
+ state, ok := ndp.slaacPrefixes[prefix]
+ if !ok || state.stableAddr.ref == nil || addr.Address != state.stableAddr.ref.address() {
+ return
+ }
+
+ if !invalidatePrefix {
+ // If the prefix is not being invalidated, disassociate the address from the
+ // prefix and do nothing further.
+ state.stableAddr.ref = nil
+ ndp.slaacPrefixes[prefix] = state
+ return
+ }
+
+ ndp.cleanupSLAACPrefixResources(prefix, state)
+}
+
+// cleanupSLAACPrefixResources cleans up a SLAAC prefix's jobs and entry.
+//
+// Panics if the SLAAC prefix is not known.
+//
+// The NIC that ndp belongs to MUST be locked.
+func (ndp *ndpState) cleanupSLAACPrefixResources(prefix tcpip.Subnet, state slaacPrefixState) {
+ // Invalidate all temporary addresses.
+ for tempAddr, tempAddrState := range state.tempAddrs {
+ ndp.invalidateTempSLAACAddr(state.tempAddrs, tempAddr, tempAddrState)
+ }
+
+ state.stableAddr.ref = nil
+ state.deprecationJob.Cancel()
+ state.invalidationJob.Cancel()
+ delete(ndp.slaacPrefixes, prefix)
+}
+
+// invalidateTempSLAACAddr invalidates a temporary SLAAC address.
+//
+// The NIC that ndp belongs to MUST be locked.
+func (ndp *ndpState) invalidateTempSLAACAddr(tempAddrs map[tcpip.Address]tempSLAACAddrState, tempAddr tcpip.Address, tempAddrState tempSLAACAddrState) {
+ // Since we are already invalidating the address, do not invalidate the
+ // address when removing the address.
+ if err := ndp.nic.removePermanentIPv6EndpointLocked(tempAddrState.ref, false /* allowSLAACInvalidation */); err != nil {
+ panic(fmt.Sprintf("error removing temporary SLAAC address %s: %s", tempAddrState.ref.addrWithPrefix(), err))
+ }
+
+ ndp.cleanupTempSLAACAddrResources(tempAddrs, tempAddr, tempAddrState)
+}
+
+// cleanupTempSLAACAddrResourcesAndNotify cleans up an invalidated temporary
+// SLAAC address's resources from ndp.
+//
+// The NIC that ndp belongs to MUST be locked.
+func (ndp *ndpState) cleanupTempSLAACAddrResourcesAndNotify(addr tcpip.AddressWithPrefix, invalidateAddr bool) {
+ if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil {
+ ndpDisp.OnAutoGenAddressInvalidated(ndp.nic.ID(), addr)
+ }
+
+ if !invalidateAddr {
+ return
+ }
+
+ prefix := addr.Subnet()
+ state, ok := ndp.slaacPrefixes[prefix]
+ if !ok {
+ panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry to clean up temp addr %s resources", addr))
+ }
+
+ tempAddrState, ok := state.tempAddrs[addr.Address]
+ if !ok {
+ panic(fmt.Sprintf("ndp: must have a tempAddr entry to clean up temp addr %s resources", addr))
+ }
+
+ ndp.cleanupTempSLAACAddrResources(state.tempAddrs, addr.Address, tempAddrState)
+}
+
+// cleanupTempSLAACAddrResourcesAndNotify cleans up a temporary SLAAC address's
+// jobs and entry.
+//
+// The NIC that ndp belongs to MUST be locked.
+func (ndp *ndpState) cleanupTempSLAACAddrResources(tempAddrs map[tcpip.Address]tempSLAACAddrState, tempAddr tcpip.Address, tempAddrState tempSLAACAddrState) {
+ tempAddrState.deprecationJob.Cancel()
+ tempAddrState.invalidationJob.Cancel()
+ tempAddrState.regenJob.Cancel()
+ delete(tempAddrs, tempAddr)
+}
+
+// cleanupState cleans up ndp's state.
+//
+// If hostOnly is true, then only host-specific state will be cleaned up.
+//
+// cleanupState MUST be called with hostOnly set to true when ndp's NIC is
+// transitioning from a host to a router. This function will invalidate all
+// discovered on-link prefixes, discovered routers, and auto-generated
+// addresses.
+//
+// If hostOnly is true, then the link-local auto-generated address will not be
+// invalidated as routers are also expected to generate a link-local address.
+//
+// The NIC that ndp belongs to MUST be locked.
+func (ndp *ndpState) cleanupState(hostOnly bool) {
+ linkLocalSubnet := header.IPv6LinkLocalPrefix.Subnet()
+ linkLocalPrefixes := 0
+ for prefix, state := range ndp.slaacPrefixes {
+ // RFC 4862 section 5 states that routers are also expected to generate a
+ // link-local address so we do not invalidate them if we are cleaning up
+ // host-only state.
+ if hostOnly && prefix == linkLocalSubnet {
+ linkLocalPrefixes++
+ continue
+ }
+
+ ndp.invalidateSLAACPrefix(prefix, state)
+ }
+
+ if got := len(ndp.slaacPrefixes); got != linkLocalPrefixes {
+ panic(fmt.Sprintf("ndp: still have non-linklocal SLAAC prefixes after cleaning up; found = %d prefixes, of which %d are link-local", got, linkLocalPrefixes))
+ }
+
+ for prefix := range ndp.onLinkPrefixes {
+ ndp.invalidateOnLinkPrefix(prefix)
+ }
+
+ if got := len(ndp.onLinkPrefixes); got != 0 {
+ panic(fmt.Sprintf("ndp: still have discovered on-link prefixes after cleaning up; found = %d", got))
+ }
+
+ for router := range ndp.defaultRouters {
+ ndp.invalidateDefaultRouter(router)
+ }
+
+ if got := len(ndp.defaultRouters); got != 0 {
+ panic(fmt.Sprintf("ndp: still have discovered default routers after cleaning up; found = %d", got))
+ }
+
+ ndp.dhcpv6Configuration = 0
+}
+
+// startSolicitingRouters starts soliciting routers, as per RFC 4861 section
+// 6.3.7. If routers are already being solicited, this function does nothing.
+//
+// The NIC ndp belongs to MUST be locked.
+func (ndp *ndpState) startSolicitingRouters() {
+ if ndp.rtrSolicit.timer != nil {
+ // We are already soliciting routers.
+ return
+ }
+
+ remaining := ndp.configs.MaxRtrSolicitations
+ if remaining == 0 {
+ return
+ }
+
+ // Calculate the random delay before sending our first RS, as per RFC
+ // 4861 section 6.3.7.
+ var delay time.Duration
+ if ndp.configs.MaxRtrSolicitationDelay > 0 {
+ delay = time.Duration(rand.Int63n(int64(ndp.configs.MaxRtrSolicitationDelay)))
+ }
+
+ var done bool
+ ndp.rtrSolicit.done = &done
+ ndp.rtrSolicit.timer = ndp.nic.stack.Clock().AfterFunc(delay, func() {
+ ndp.nic.mu.Lock()
+ if done {
+ // If we reach this point, it means that the RS timer fired after another
+ // goroutine already obtained the NIC lock and stopped solicitations.
+ // Simply return here and do nothing further.
+ ndp.nic.mu.Unlock()
+ return
+ }
+
+ // As per RFC 4861 section 4.1, the source of the RS is an address assigned
+ // to the sending interface, or the unspecified address if no address is
+ // assigned to the sending interface.
+ ref := ndp.nic.primaryIPv6EndpointRLocked(header.IPv6AllRoutersMulticastAddress)
+ if ref == nil {
+ ref = ndp.nic.getRefOrCreateTempLocked(header.IPv6ProtocolNumber, header.IPv6Any, NeverPrimaryEndpoint)
+ }
+ ndp.nic.mu.Unlock()
+
+ localAddr := ref.address()
+ r := makeRoute(header.IPv6ProtocolNumber, localAddr, header.IPv6AllRoutersMulticastAddress, ndp.nic.linkEP.LinkAddress(), ref, false, false)
+ defer r.Release()
+
+ // Route should resolve immediately since
+ // header.IPv6AllRoutersMulticastAddress is a multicast address so a
+ // remote link address can be calculated without a resolution process.
+ if c, err := r.Resolve(nil); err != nil {
+ // Do not consider the NIC being unknown or disabled as a fatal error.
+ // Since this method is required to be called when the NIC is not locked,
+ // the NIC could have been disabled or removed by another goroutine.
+ if err == tcpip.ErrUnknownNICID || err == tcpip.ErrInvalidEndpointState {
+ return
+ }
+
+ panic(fmt.Sprintf("ndp: error when resolving route to send NDP RS (%s -> %s on NIC(%d)): %s", header.IPv6Any, header.IPv6AllRoutersMulticastAddress, ndp.nic.ID(), err))
+ } else if c != nil {
+ panic(fmt.Sprintf("ndp: route resolution not immediate for route to send NDP RS (%s -> %s on NIC(%d))", header.IPv6Any, header.IPv6AllRoutersMulticastAddress, ndp.nic.ID()))
+ }
+
+ // As per RFC 4861 section 4.1, an NDP RS SHOULD include the source
+ // link-layer address option if the source address of the NDP RS is
+ // specified. This option MUST NOT be included if the source address is
+ // unspecified.
+ //
+ // TODO(b/141011931): Validate a LinkEndpoint's link address (provided by
+ // LinkEndpoint.LinkAddress) before reaching this point.
+ var optsSerializer header.NDPOptionsSerializer
+ if localAddr != header.IPv6Any && header.IsValidUnicastEthernetAddress(r.LocalLinkAddress) {
+ optsSerializer = header.NDPOptionsSerializer{
+ header.NDPSourceLinkLayerAddressOption(r.LocalLinkAddress),
+ }
+ }
+ payloadSize := header.ICMPv6HeaderSize + header.NDPRSMinimumSize + int(optsSerializer.Length())
+ icmpData := header.ICMPv6(buffer.NewView(payloadSize))
+ icmpData.SetType(header.ICMPv6RouterSolicit)
+ rs := header.NDPRouterSolicit(icmpData.NDPPayload())
+ rs.Options().Serialize(optsSerializer)
+ icmpData.SetChecksum(header.ICMPv6Checksum(icmpData, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{}))
+
+ pkt := NewPacketBuffer(PacketBufferOptions{
+ ReserveHeaderBytes: int(r.MaxHeaderLength()),
+ Data: buffer.View(icmpData).ToVectorisedView(),
+ })
+
+ sent := r.Stats().ICMP.V6PacketsSent
+ if err := r.WritePacket(nil,
+ NetworkHeaderParams{
+ Protocol: header.ICMPv6ProtocolNumber,
+ TTL: header.NDPHopLimit,
+ TOS: DefaultTOS,
+ }, pkt,
+ ); err != nil {
+ sent.Dropped.Increment()
+ log.Printf("startSolicitingRouters: error writing NDP router solicit message on NIC(%d); err = %s", ndp.nic.ID(), err)
+ // Don't send any more messages if we had an error.
+ remaining = 0
+ } else {
+ sent.RouterSolicit.Increment()
+ remaining--
+ }
+
+ ndp.nic.mu.Lock()
+ if done || remaining == 0 {
+ ndp.rtrSolicit.timer = nil
+ ndp.rtrSolicit.done = nil
+ } else if ndp.rtrSolicit.timer != nil {
+ // Note, we need to explicitly check to make sure that
+ // the timer field is not nil because if it was nil but
+ // we still reached this point, then we know the NIC
+ // was requested to stop soliciting routers so we don't
+ // need to send the next Router Solicitation message.
+ ndp.rtrSolicit.timer.Reset(ndp.configs.RtrSolicitationInterval)
+ }
+ ndp.nic.mu.Unlock()
+ })
+
+}
+
+// stopSolicitingRouters stops soliciting routers. If routers are not currently
+// being solicited, this function does nothing.
+//
+// The NIC ndp belongs to MUST be locked.
+func (ndp *ndpState) stopSolicitingRouters() {
+ if ndp.rtrSolicit.timer == nil {
+ // Nothing to do.
+ return
+ }
+
+ *ndp.rtrSolicit.done = true
+ ndp.rtrSolicit.timer.Stop()
+ ndp.rtrSolicit.timer = nil
+ ndp.rtrSolicit.done = nil
+}
+
+// initializeTempAddrState initializes state related to temporary SLAAC
+// addresses.
+func (ndp *ndpState) initializeTempAddrState() {
+ header.InitialTempIID(ndp.temporaryIIDHistory[:], ndp.nic.stack.tempIIDSeed, ndp.nic.ID())
+
+ if MaxDesyncFactor != 0 {
+ ndp.temporaryAddressDesyncFactor = time.Duration(rand.Int63n(int64(MaxDesyncFactor)))
}
}
diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go
index 525a25218..1a6724c31 100644
--- a/pkg/tcpip/stack/ndp_test.go
+++ b/pkg/tcpip/stack/ndp_test.go
@@ -15,9 +15,14 @@
package stack_test
import (
+ "context"
+ "encoding/binary"
+ "fmt"
"testing"
"time"
+ "github.com/google/go-cmp/cmp"
+ "gvisor.dev/gvisor/pkg/rand"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/tcpip/checker"
@@ -26,74 +31,330 @@ import (
"gvisor.dev/gvisor/pkg/tcpip/network/ipv6"
"gvisor.dev/gvisor/pkg/tcpip/stack"
"gvisor.dev/gvisor/pkg/tcpip/transport/icmp"
+ "gvisor.dev/gvisor/pkg/tcpip/transport/udp"
+ "gvisor.dev/gvisor/pkg/waiter"
)
const (
- addr1 = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01"
- addr2 = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02"
- addr3 = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x03"
- linkAddr1 = "\x02\x02\x03\x04\x05\x06"
+ addr1 = tcpip.Address("\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01")
+ addr2 = tcpip.Address("\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02")
+ addr3 = tcpip.Address("\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x03")
+ linkAddr1 = tcpip.LinkAddress("\x02\x02\x03\x04\x05\x06")
+ linkAddr2 = tcpip.LinkAddress("\x02\x02\x03\x04\x05\x07")
+ linkAddr3 = tcpip.LinkAddress("\x02\x02\x03\x04\x05\x08")
+ linkAddr4 = tcpip.LinkAddress("\x02\x02\x03\x04\x05\x09")
+
+ // Extra time to use when waiting for an async event to occur.
+ defaultAsyncPositiveEventTimeout = 10 * time.Second
+
+ // Extra time to use when waiting for an async event to not occur.
+ //
+ // Since a negative check is used to make sure an event did not happen, it is
+ // okay to use a smaller timeout compared to the positive case since execution
+ // stall in regards to the monotonic clock will not affect the expected
+ // outcome.
+ defaultAsyncNegativeEventTimeout = time.Second
)
-// TestDADDisabled tests that an address successfully resolves immediately
-// when DAD is not enabled (the default for an empty stack.Options).
-func TestDADDisabled(t *testing.T) {
- opts := stack.Options{
- NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+var (
+ llAddr1 = header.LinkLocalAddr(linkAddr1)
+ llAddr2 = header.LinkLocalAddr(linkAddr2)
+ llAddr3 = header.LinkLocalAddr(linkAddr3)
+ llAddr4 = header.LinkLocalAddr(linkAddr4)
+ dstAddr = tcpip.FullAddress{
+ Addr: "\x0a\x0b\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01",
+ Port: 25,
}
+)
- e := channel.New(10, 1280, linkAddr1)
- s := stack.New(opts)
- if err := s.CreateNIC(1, e); err != nil {
- t.Fatalf("CreateNIC(_) = %s", err)
+func addrForSubnet(subnet tcpip.Subnet, linkAddr tcpip.LinkAddress) tcpip.AddressWithPrefix {
+ if !header.IsValidUnicastEthernetAddress(linkAddr) {
+ return tcpip.AddressWithPrefix{}
}
- if err := s.AddAddress(1, header.IPv6ProtocolNumber, addr1); err != nil {
- t.Fatalf("AddAddress(_, %d, %s) = %s", header.IPv6ProtocolNumber, addr1, err)
+ addrBytes := []byte(subnet.ID())
+ header.EthernetAdddressToModifiedEUI64IntoBuf(linkAddr, addrBytes[header.IIDOffsetInIPv6Address:])
+ return tcpip.AddressWithPrefix{
+ Address: tcpip.Address(addrBytes),
+ PrefixLen: 64,
}
+}
- // Should get the address immediately since we should not have performed
- // DAD on it.
- addr, err := s.GetMainNICAddress(1, header.IPv6ProtocolNumber)
- if err != nil {
- t.Fatalf("stack.GetMainNICAddress(_, _) err = %s", err)
- }
- if addr.Address != addr1 {
- t.Fatalf("got stack.GetMainNICAddress(_, _) = %s, want = %s", addr, addr1)
+// prefixSubnetAddr returns a prefix (Address + Length), the prefix's equivalent
+// tcpip.Subnet, and an address where the lower half of the address is composed
+// of the EUI-64 of linkAddr if it is a valid unicast ethernet address.
+func prefixSubnetAddr(offset uint8, linkAddr tcpip.LinkAddress) (tcpip.AddressWithPrefix, tcpip.Subnet, tcpip.AddressWithPrefix) {
+ prefixBytes := []byte{1, 2, 3, 4, 5, 6, 7, 8 + offset, 0, 0, 0, 0, 0, 0, 0, 0}
+ prefix := tcpip.AddressWithPrefix{
+ Address: tcpip.Address(prefixBytes),
+ PrefixLen: 64,
}
- // We should not have sent any NDP NS messages.
- if got := s.Stats().ICMP.V6PacketsSent.NeighborSolicit.Value(); got != 0 {
- t.Fatalf("got NeighborSolicit = %d, want = 0", got)
- }
+ subnet := prefix.Subnet()
+
+ return prefix, subnet, addrForSubnet(subnet, linkAddr)
}
// ndpDADEvent is a set of parameters that was passed to
// ndpDispatcher.OnDuplicateAddressDetectionStatus.
type ndpDADEvent struct {
- nicid tcpip.NICID
+ nicID tcpip.NICID
addr tcpip.Address
resolved bool
err *tcpip.Error
}
+type ndpRouterEvent struct {
+ nicID tcpip.NICID
+ addr tcpip.Address
+ // true if router was discovered, false if invalidated.
+ discovered bool
+}
+
+type ndpPrefixEvent struct {
+ nicID tcpip.NICID
+ prefix tcpip.Subnet
+ // true if prefix was discovered, false if invalidated.
+ discovered bool
+}
+
+type ndpAutoGenAddrEventType int
+
+const (
+ newAddr ndpAutoGenAddrEventType = iota
+ deprecatedAddr
+ invalidatedAddr
+)
+
+type ndpAutoGenAddrEvent struct {
+ nicID tcpip.NICID
+ addr tcpip.AddressWithPrefix
+ eventType ndpAutoGenAddrEventType
+}
+
+type ndpRDNSS struct {
+ addrs []tcpip.Address
+ lifetime time.Duration
+}
+
+type ndpRDNSSEvent struct {
+ nicID tcpip.NICID
+ rdnss ndpRDNSS
+}
+
+type ndpDNSSLEvent struct {
+ nicID tcpip.NICID
+ domainNames []string
+ lifetime time.Duration
+}
+
+type ndpDHCPv6Event struct {
+ nicID tcpip.NICID
+ configuration stack.DHCPv6ConfigurationFromNDPRA
+}
+
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
+ dadC chan ndpDADEvent
+ routerC chan ndpRouterEvent
+ rememberRouter bool
+ prefixC chan ndpPrefixEvent
+ rememberPrefix bool
+ autoGenAddrC chan ndpAutoGenAddrEvent
+ rdnssC chan ndpRDNSSEvent
+ dnsslC chan ndpDNSSLEvent
+ routeTable []tcpip.Route
+ dhcpv6ConfigurationC chan ndpDHCPv6Event
}
// 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,
+func (n *ndpDispatcher) OnDuplicateAddressDetectionStatus(nicID tcpip.NICID, addr tcpip.Address, resolved bool, err *tcpip.Error) {
+ if n.dadC != nil {
+ n.dadC <- ndpDADEvent{
+ nicID,
+ addr,
+ resolved,
+ err,
+ }
+ }
+}
+
+// Implements stack.NDPDispatcher.OnDefaultRouterDiscovered.
+func (n *ndpDispatcher) OnDefaultRouterDiscovered(nicID tcpip.NICID, addr tcpip.Address) bool {
+ if c := n.routerC; c != nil {
+ c <- ndpRouterEvent{
+ nicID,
+ addr,
+ true,
+ }
+ }
+
+ return n.rememberRouter
+}
+
+// Implements stack.NDPDispatcher.OnDefaultRouterInvalidated.
+func (n *ndpDispatcher) OnDefaultRouterInvalidated(nicID tcpip.NICID, addr tcpip.Address) {
+ if c := n.routerC; c != nil {
+ c <- ndpRouterEvent{
+ nicID,
+ addr,
+ false,
+ }
+ }
+}
+
+// Implements stack.NDPDispatcher.OnOnLinkPrefixDiscovered.
+func (n *ndpDispatcher) OnOnLinkPrefixDiscovered(nicID tcpip.NICID, prefix tcpip.Subnet) bool {
+ if c := n.prefixC; c != nil {
+ c <- ndpPrefixEvent{
+ nicID,
+ prefix,
+ true,
+ }
+ }
+
+ return n.rememberPrefix
+}
+
+// Implements stack.NDPDispatcher.OnOnLinkPrefixInvalidated.
+func (n *ndpDispatcher) OnOnLinkPrefixInvalidated(nicID tcpip.NICID, prefix tcpip.Subnet) {
+ if c := n.prefixC; c != nil {
+ c <- ndpPrefixEvent{
+ nicID,
+ prefix,
+ false,
+ }
+ }
+}
+
+func (n *ndpDispatcher) OnAutoGenAddress(nicID tcpip.NICID, addr tcpip.AddressWithPrefix) bool {
+ if c := n.autoGenAddrC; c != nil {
+ c <- ndpAutoGenAddrEvent{
+ nicID,
+ addr,
+ newAddr,
+ }
+ }
+ return true
+}
+
+func (n *ndpDispatcher) OnAutoGenAddressDeprecated(nicID tcpip.NICID, addr tcpip.AddressWithPrefix) {
+ if c := n.autoGenAddrC; c != nil {
+ c <- ndpAutoGenAddrEvent{
+ nicID,
+ addr,
+ deprecatedAddr,
+ }
+ }
+}
+
+func (n *ndpDispatcher) OnAutoGenAddressInvalidated(nicID tcpip.NICID, addr tcpip.AddressWithPrefix) {
+ if c := n.autoGenAddrC; c != nil {
+ c <- ndpAutoGenAddrEvent{
+ nicID,
+ addr,
+ invalidatedAddr,
+ }
+ }
+}
+
+// Implements stack.NDPDispatcher.OnRecursiveDNSServerOption.
+func (n *ndpDispatcher) OnRecursiveDNSServerOption(nicID tcpip.NICID, addrs []tcpip.Address, lifetime time.Duration) {
+ if c := n.rdnssC; c != nil {
+ c <- ndpRDNSSEvent{
+ nicID,
+ ndpRDNSS{
+ addrs,
+ lifetime,
+ },
+ }
+ }
+}
+
+// Implements stack.NDPDispatcher.OnDNSSearchListOption.
+func (n *ndpDispatcher) OnDNSSearchListOption(nicID tcpip.NICID, domainNames []string, lifetime time.Duration) {
+ if n.dnsslC != nil {
+ n.dnsslC <- ndpDNSSLEvent{
+ nicID,
+ domainNames,
+ lifetime,
+ }
+ }
+}
+
+// Implements stack.NDPDispatcher.OnDHCPv6Configuration.
+func (n *ndpDispatcher) OnDHCPv6Configuration(nicID tcpip.NICID, configuration stack.DHCPv6ConfigurationFromNDPRA) {
+ if c := n.dhcpv6ConfigurationC; c != nil {
+ c <- ndpDHCPv6Event{
+ nicID,
+ configuration,
+ }
+ }
+}
+
+// channelLinkWithHeaderLength is a channel.Endpoint with a configurable
+// header length.
+type channelLinkWithHeaderLength struct {
+ *channel.Endpoint
+ headerLength uint16
+}
+
+func (l *channelLinkWithHeaderLength) MaxHeaderLength() uint16 {
+ return l.headerLength
+}
+
+// Check e to make sure that the event is for addr on nic with ID 1, and the
+// resolved flag set to resolved with the specified err.
+func checkDADEvent(e ndpDADEvent, nicID tcpip.NICID, addr tcpip.Address, resolved bool, err *tcpip.Error) string {
+ return cmp.Diff(ndpDADEvent{nicID: nicID, addr: addr, resolved: resolved, err: err}, e, cmp.AllowUnexported(e))
+}
+
+// TestDADDisabled tests that an address successfully resolves immediately
+// when DAD is not enabled (the default for an empty stack.Options).
+func TestDADDisabled(t *testing.T) {
+ const nicID = 1
+ ndpDisp := ndpDispatcher{
+ dadC: make(chan ndpDADEvent, 1),
+ }
+ opts := stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPDisp: &ndpDisp,
+ }
+
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(opts)
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+
+ if err := s.AddAddress(nicID, header.IPv6ProtocolNumber, addr1); err != nil {
+ t.Fatalf("AddAddress(%d, %d, %s) = %s", nicID, header.IPv6ProtocolNumber, addr1, err)
+ }
+
+ // Should get the address immediately since we should not have performed
+ // DAD on it.
+ select {
+ case e := <-ndpDisp.dadC:
+ if diff := checkDADEvent(e, nicID, addr1, true, nil); diff != "" {
+ t.Errorf("dad event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected DAD event")
+ }
+ addr, err := s.GetMainNICAddress(nicID, header.IPv6ProtocolNumber)
+ if err != nil {
+ t.Fatalf("stack.GetMainNICAddress(%d, %d) err = %s", nicID, header.IPv6ProtocolNumber, err)
+ }
+ if addr.Address != addr1 {
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = %s, want = %s", nicID, header.IPv6ProtocolNumber, addr, addr1)
+ }
+
+ // We should not have sent any NDP NS messages.
+ if got := s.Stats().ICMP.V6PacketsSent.NeighborSolicit.Value(); got != 0 {
+ t.Fatalf("got NeighborSolicit = %d, want = 0", got)
}
}
@@ -101,23 +362,54 @@ func (n *ndpDispatcher) OnDuplicateAddressDetectionStatus(nicid tcpip.NICID, add
// DAD for various values of DupAddrDetectTransmits and RetransmitTimer.
// Included in the subtests is a test to make sure that an invalid
// RetransmitTimer (<1ms) values get fixed to the default RetransmitTimer of 1s.
+// This tests also validates the NDP NS packet that is transmitted.
func TestDADResolve(t *testing.T) {
+ const nicID = 1
+
tests := []struct {
name string
+ linkHeaderLen uint16
dupAddrDetectTransmits uint8
retransTimer time.Duration
expectedRetransmitTimer time.Duration
}{
- {"1:1s:1s", 1, time.Second, time.Second},
- {"2:1s:1s", 2, time.Second, time.Second},
- {"1:2s:2s", 1, 2 * time.Second, 2 * time.Second},
+ {
+ name: "1:1s:1s",
+ dupAddrDetectTransmits: 1,
+ retransTimer: time.Second,
+ expectedRetransmitTimer: time.Second,
+ },
+ {
+ name: "2:1s:1s",
+ linkHeaderLen: 1,
+ dupAddrDetectTransmits: 2,
+ retransTimer: time.Second,
+ expectedRetransmitTimer: time.Second,
+ },
+ {
+ name: "1:2s:2s",
+ linkHeaderLen: 2,
+ dupAddrDetectTransmits: 1,
+ retransTimer: 2 * time.Second,
+ expectedRetransmitTimer: 2 * time.Second,
+ },
// 0s is an invalid RetransmitTimer timer and will be fixed to
// the default RetransmitTimer value of 1s.
- {"1:0s:1s", 1, 0, time.Second},
+ {
+ name: "1:0s:1s",
+ linkHeaderLen: 3,
+ dupAddrDetectTransmits: 1,
+ retransTimer: 0,
+ expectedRetransmitTimer: time.Second,
+ },
}
for _, test := range tests {
+ test := test
+
t.Run(test.name, func(t *testing.T) {
+ t.Parallel()
+
ndpDisp := ndpDispatcher{
dadC: make(chan ndpDADEvent),
}
@@ -128,101 +420,142 @@ func TestDADResolve(t *testing.T) {
opts.NDPConfigs.RetransmitTimer = test.retransTimer
opts.NDPConfigs.DupAddrDetectTransmits = test.dupAddrDetectTransmits
- e := channel.New(10, 1280, linkAddr1)
- s := stack.New(opts)
- if err := s.CreateNIC(1, e); err != nil {
- t.Fatalf("CreateNIC(_) = %s", err)
+ e := channelLinkWithHeaderLength{
+ Endpoint: channel.New(int(test.dupAddrDetectTransmits), 1280, linkAddr1),
+ headerLength: test.linkHeaderLen,
}
-
- if err := s.AddAddress(1, header.IPv6ProtocolNumber, addr1); err != nil {
- t.Fatalf("AddAddress(_, %d, %s) = %s", header.IPv6ProtocolNumber, addr1, err)
+ e.Endpoint.LinkEPCapabilities |= stack.CapabilityResolutionRequired
+ s := stack.New(opts)
+ if err := s.CreateNIC(nicID, &e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
}
- stat := s.Stats().ICMP.V6PacketsSent.NeighborSolicit
+ // We add a default route so the call to FindRoute below will succeed
+ // once we have an assigned address.
+ s.SetRouteTable([]tcpip.Route{{
+ Destination: header.IPv6EmptySubnet,
+ Gateway: addr3,
+ NIC: nicID,
+ }})
- // Should have sent an NDP NS immediately.
- if got := stat.Value(); got != 1 {
- t.Fatalf("got NeighborSolicit = %d, want = 1", got)
+ if err := s.AddAddress(nicID, header.IPv6ProtocolNumber, addr1); err != nil {
+ t.Fatalf("AddAddress(%d, %d, %s) = %s", nicID, header.IPv6ProtocolNumber, addr1, err)
+ }
+ // Address should not be considered bound to the NIC yet (DAD ongoing).
+ if addr, err := s.GetMainNICAddress(nicID, header.IPv6ProtocolNumber); err != nil {
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (_, %s), want = (_, nil)", nicID, header.IPv6ProtocolNumber, err)
+ } else if want := (tcpip.AddressWithPrefix{}); addr != want {
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (%s, nil), want = (%s, nil)", nicID, header.IPv6ProtocolNumber, addr, want)
}
- // 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)
+ // Make sure the address does not resolve before the resolution time has
+ // passed.
+ time.Sleep(test.expectedRetransmitTimer*time.Duration(test.dupAddrDetectTransmits) - defaultAsyncNegativeEventTimeout)
+ if addr, err := s.GetMainNICAddress(nicID, header.IPv6ProtocolNumber); err != nil {
+ t.Errorf("got stack.GetMainNICAddress(%d, %d) = (_, %s), want = (_, nil)", nicID, header.IPv6ProtocolNumber, err)
+ } else if want := (tcpip.AddressWithPrefix{}); addr != want {
+ t.Errorf("got stack.GetMainNICAddress(%d, %d) = (%s, nil), want = (%s, nil)", nicID, header.IPv6ProtocolNumber, addr, want)
}
- if want := (tcpip.AddressWithPrefix{}); addr != want {
- t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want)
+ // Should not get a route even if we specify the local address as the
+ // tentative address.
+ {
+ r, err := s.FindRoute(nicID, "", addr2, header.IPv6ProtocolNumber, false)
+ if err != tcpip.ErrNoRoute {
+ t.Errorf("got FindRoute(%d, '', %s, %d, false) = (%+v, %v), want = (_, %s)", nicID, addr2, header.IPv6ProtocolNumber, r, err, tcpip.ErrNoRoute)
+ }
+ r.Release()
}
-
- // 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)
+ {
+ r, err := s.FindRoute(nicID, addr1, addr2, header.IPv6ProtocolNumber, false)
+ if err != tcpip.ErrNoRoute {
+ t.Errorf("got FindRoute(%d, %s, %s, %d, false) = (%+v, %v), want = (_, %s)", nicID, addr1, addr2, header.IPv6ProtocolNumber, r, err, tcpip.ErrNoRoute)
+ }
+ r.Release()
}
- if want := (tcpip.AddressWithPrefix{}); addr != want {
- t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want)
+
+ if t.Failed() {
+ t.FailNow()
}
// 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.
+ case <-time.After(defaultAsyncPositiveEventTimeout):
t.Fatal("timed out waiting for DAD resolution")
case e := <-ndpDisp.dadC:
- if e.err != nil {
- t.Fatal("got DAD error: ", e.err)
+ if diff := checkDADEvent(e, nicID, addr1, true, nil); diff != "" {
+ t.Errorf("dad event mismatch (-want +got):\n%s", diff)
}
- 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")
+ }
+ if addr, err := s.GetMainNICAddress(nicID, header.IPv6ProtocolNumber); err != nil {
+ t.Errorf("got stack.GetMainNICAddress(%d, %d) = (_, %s), want = (_, nil)", nicID, header.IPv6ProtocolNumber, err)
+ } else if addr.Address != addr1 {
+ t.Errorf("got stack.GetMainNICAddress(%d, %d) = %s, want = %s", nicID, header.IPv6ProtocolNumber, addr, addr1)
+ }
+ // Should get a route using the address now that it is resolved.
+ {
+ r, err := s.FindRoute(nicID, "", addr2, header.IPv6ProtocolNumber, false)
+ if err != nil {
+ t.Errorf("got FindRoute(%d, '', %s, %d, false): %s", nicID, addr2, header.IPv6ProtocolNumber, err)
+ } else if r.LocalAddress != addr1 {
+ t.Errorf("got r.LocalAddress = %s, want = %s", r.LocalAddress, addr1)
}
+ r.Release()
}
- addr, err = s.GetMainNICAddress(1, header.IPv6ProtocolNumber)
- if err != nil {
- t.Fatalf("stack.GetMainNICAddress(_, _) err = %s", err)
+ {
+ r, err := s.FindRoute(nicID, addr1, addr2, header.IPv6ProtocolNumber, false)
+ if err != nil {
+ t.Errorf("got FindRoute(%d, %s, %s, %d, false): %s", nicID, addr1, addr2, header.IPv6ProtocolNumber, err)
+ } else if r.LocalAddress != addr1 {
+ t.Errorf("got r.LocalAddress = %s, want = %s", r.LocalAddress, addr1)
+ }
+ r.Release()
}
- if addr.Address != addr1 {
- t.Fatalf("got stack.GetMainNICAddress(_, _) = %s, want = %s", addr, addr1)
+
+ if t.Failed() {
+ t.FailNow()
}
// Should not have sent any more NS messages.
- if got := stat.Value(); got != uint64(test.dupAddrDetectTransmits) {
+ if got := s.Stats().ICMP.V6PacketsSent.NeighborSolicit.Value(); got != uint64(test.dupAddrDetectTransmits) {
t.Fatalf("got NeighborSolicit = %d, want = %d", got, test.dupAddrDetectTransmits)
}
// Validate the sent Neighbor Solicitation messages.
for i := uint8(0); i < test.dupAddrDetectTransmits; i++ {
- p := <-e.C
+ p, _ := e.ReadContext(context.Background())
// Make sure its an IPv6 packet.
if p.Proto != header.IPv6ProtocolNumber {
t.Fatalf("got Proto = %d, want = %d", p.Proto, header.IPv6ProtocolNumber)
}
- // Check NDP packet.
- checker.IPv6(t, p.Header.ToVectorisedView().First(),
+ // Make sure the right remote link address is used.
+ snmc := header.SolicitedNodeAddr(addr1)
+ if want := header.EthernetAddressFromMulticastIPv6Address(snmc); p.Route.RemoteLinkAddress != want {
+ t.Errorf("got remote link address = %s, want = %s", p.Route.RemoteLinkAddress, want)
+ }
+
+ // Check NDP NS packet.
+ //
+ // As per RFC 4861 section 4.3, a possible option is the Source Link
+ // Layer option, but this option MUST NOT be included when the source
+ // address of the packet is the unspecified address.
+ checker.IPv6(t, stack.PayloadSince(p.Pkt.NetworkHeader()),
+ checker.SrcAddr(header.IPv6Any),
+ checker.DstAddr(snmc),
checker.TTL(header.NDPHopLimit),
checker.NDPNS(
- checker.NDPNSTargetAddress(addr1)))
+ checker.NDPNSTargetAddress(addr1),
+ checker.NDPNSOptions(nil),
+ ))
+
+ if l, want := p.Pkt.AvailableHeaderBytes(), int(test.linkHeaderLen); l != want {
+ t.Errorf("got p.Pkt.AvailableHeaderBytes() = %d; want = %d", l, want)
+ }
}
})
}
-
}
// TestDADFail tests to make sure that the DAD process fails if another node is
@@ -230,6 +563,8 @@ func TestDADResolve(t *testing.T) {
// a node doing DAD for the same address), or if another node is detected to own
// the address already (receive an NA message for the tentative address).
func TestDADFail(t *testing.T) {
+ const nicID = 1
+
tests := []struct {
name string
makeBuf func(tgt tcpip.Address) buffer.Prependable
@@ -265,13 +600,17 @@ func TestDADFail(t *testing.T) {
{
"RxAdvert",
func(tgt tcpip.Address) buffer.Prependable {
- hdr := buffer.NewPrependable(header.IPv6MinimumSize + header.ICMPv6NeighborAdvertSize)
- pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborAdvertSize))
+ naSize := header.ICMPv6NeighborAdvertMinimumSize + header.NDPLinkLayerAddressSize
+ hdr := buffer.NewPrependable(header.IPv6MinimumSize + naSize)
+ pkt := header.ICMPv6(hdr.Prepend(naSize))
pkt.SetType(header.ICMPv6NeighborAdvert)
na := header.NDPNeighborAdvert(pkt.NDPPayload())
na.SetSolicitedFlag(true)
na.SetOverrideFlag(true)
na.SetTargetAddress(tgt)
+ na.Options().Serialize(header.NDPOptionsSerializer{
+ header.NDPTargetLinkLayerAddressOption(linkAddr1),
+ })
pkt.SetChecksum(header.ICMPv6Checksum(pkt, tgt, header.IPv6AllNodesMulticastAddress, buffer.VectorisedView{}))
payloadLength := hdr.UsedLength()
ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize))
@@ -295,7 +634,7 @@ func TestDADFail(t *testing.T) {
for _, test := range tests {
t.Run(test.name, func(t *testing.T) {
ndpDisp := ndpDispatcher{
- dadC: make(chan ndpDADEvent),
+ dadC: make(chan ndpDADEvent, 1),
}
ndpConfigs := stack.DefaultNDPConfigurations()
opts := stack.Options{
@@ -305,30 +644,33 @@ func TestDADFail(t *testing.T) {
}
opts.NDPConfigs.RetransmitTimer = time.Second * 2
- e := channel.New(10, 1280, linkAddr1)
+ e := channel.New(0, 1280, linkAddr1)
s := stack.New(opts)
- if err := s.CreateNIC(1, e); err != nil {
- t.Fatalf("CreateNIC(_) = %s", err)
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
}
- if err := s.AddAddress(1, header.IPv6ProtocolNumber, addr1); err != nil {
- t.Fatalf("AddAddress(_, %d, %s) = %s", header.IPv6ProtocolNumber, addr1, err)
+ if err := s.AddAddress(nicID, header.IPv6ProtocolNumber, addr1); err != nil {
+ t.Fatalf("AddAddress(%d, %d, %s) = %s", nicID, header.IPv6ProtocolNumber, addr1, err)
}
// Address should not be considered bound to the NIC yet
// (DAD ongoing).
- addr, err := s.GetMainNICAddress(1, header.IPv6ProtocolNumber)
+ addr, err := s.GetMainNICAddress(nicID, header.IPv6ProtocolNumber)
if err != nil {
- t.Fatalf("got stack.GetMainNICAddress(_, _) = (_, %v), want = (_, nil)", err)
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (_, %v), want = (_, nil)", nicID, header.IPv6ProtocolNumber, err)
}
if want := (tcpip.AddressWithPrefix{}); addr != want {
- t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want)
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (%s, nil), want = (%s, nil)", nicID, header.IPv6ProtocolNumber, addr, want)
}
// Receive a packet to simulate multiple nodes owning or
// attempting to own the same address.
hdr := test.makeBuf(addr1)
- e.Inject(header.IPv6ProtocolNumber, hdr.View().ToVectorisedView())
+ pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{
+ Data: hdr.View().ToVectorisedView(),
+ })
+ e.InjectInbound(header.IPv6ProtocolNumber, pkt)
stat := test.getStat(s.Stats().ICMP.V6PacketsReceived)
if got := stat.Value(); got != 1 {
@@ -344,102 +686,132 @@ func TestDADFail(t *testing.T) {
// 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")
+ if diff := checkDADEvent(e, nicID, addr1, false, nil); diff != "" {
+ t.Errorf("dad event mismatch (-want +got):\n%s", diff)
}
}
- addr, err = s.GetMainNICAddress(1, header.IPv6ProtocolNumber)
+ addr, err = s.GetMainNICAddress(nicID, header.IPv6ProtocolNumber)
if err != nil {
- t.Fatalf("got stack.GetMainNICAddress(_, _) = (_, %v), want = (_, nil)", err)
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (_, %v), want = (_, nil)", nicID, header.IPv6ProtocolNumber, err)
}
if want := (tcpip.AddressWithPrefix{}); addr != want {
- t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want)
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (%s, nil), want = (%s, nil)", nicID, header.IPv6ProtocolNumber, addr, want)
+ }
+
+ // Attempting to add the address again should not fail if the address's
+ // state was cleaned up when DAD failed.
+ if err := s.AddAddress(nicID, header.IPv6ProtocolNumber, addr1); err != nil {
+ t.Fatalf("AddAddress(%d, %d, %s) = %s", nicID, header.IPv6ProtocolNumber, addr1, err)
}
})
}
}
-// 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,
- }
+ const nicID = 1
- e := channel.New(10, 1280, linkAddr1)
- s := stack.New(opts)
- if err := s.CreateNIC(1, e); err != nil {
- t.Fatalf("CreateNIC(_) = %s", err)
- }
+ tests := []struct {
+ name string
+ stopFn func(t *testing.T, s *stack.Stack)
+ skipFinalAddrCheck bool
+ }{
+ // Tests to make sure that DAD stops when an address is removed.
+ {
+ name: "Remove address",
+ stopFn: func(t *testing.T, s *stack.Stack) {
+ if err := s.RemoveAddress(nicID, addr1); err != nil {
+ t.Fatalf("RemoveAddress(%d, %s): %s", nicID, addr1, err)
+ }
+ },
+ },
- if err := s.AddAddress(1, header.IPv6ProtocolNumber, addr1); err != nil {
- t.Fatalf("AddAddress(_, %d, %s) = %s", header.IPv6ProtocolNumber, addr1, err)
- }
+ // Tests to make sure that DAD stops when the NIC is disabled.
+ {
+ name: "Disable NIC",
+ stopFn: func(t *testing.T, s *stack.Stack) {
+ if err := s.DisableNIC(nicID); err != nil {
+ t.Fatalf("DisableNIC(%d): %s", nicID, 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)
+ // Tests to make sure that DAD stops when the NIC is removed.
+ {
+ name: "Remove NIC",
+ stopFn: func(t *testing.T, s *stack.Stack) {
+ if err := s.RemoveNIC(nicID); err != nil {
+ t.Fatalf("RemoveNIC(%d): %s", nicID, err)
+ }
+ },
+ // The NIC is removed so we can't check its addresses after calling
+ // stopFn.
+ skipFinalAddrCheck: true,
+ },
}
- // Remove the address. This should stop DAD.
- if err := s.RemoveAddress(1, addr1); err != nil {
- t.Fatalf("RemoveAddress(_, %s) = %s", addr1, err)
- }
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ ndpDisp := ndpDispatcher{
+ dadC: make(chan ndpDADEvent, 1),
+ }
+ ndpConfigs := stack.NDPConfigurations{
+ RetransmitTimer: time.Second,
+ DupAddrDetectTransmits: 2,
+ }
+ opts := stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPDisp: &ndpDisp,
+ NDPConfigs: ndpConfigs,
+ }
- // 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")
- }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(opts)
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _): %s", nicID, err)
+ }
- }
- 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)
- }
+ if err := s.AddAddress(nicID, header.IPv6ProtocolNumber, addr1); err != nil {
+ t.Fatalf("AddAddress(%d, %d, %s): %s", nicID, header.IPv6ProtocolNumber, addr1, err)
+ }
- // Should not have sent more than 1 NS message.
- if got := s.Stats().ICMP.V6PacketsSent.NeighborSolicit.Value(); got > 1 {
- t.Fatalf("got NeighborSolicit = %d, want <= 1", got)
+ // Address should not be considered bound to the NIC yet (DAD ongoing).
+ addr, err := s.GetMainNICAddress(nicID, header.IPv6ProtocolNumber)
+ if err != nil {
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (_, %v), want = (_, nil)", nicID, header.IPv6ProtocolNumber, err)
+ }
+ if want := (tcpip.AddressWithPrefix{}); addr != want {
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (%s, nil), want = (%s, nil)", nicID, header.IPv6ProtocolNumber, addr, want)
+ }
+
+ test.stopFn(t, s)
+
+ // 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 diff := checkDADEvent(e, nicID, addr1, false, nil); diff != "" {
+ t.Errorf("dad event mismatch (-want +got):\n%s", diff)
+ }
+ }
+
+ if !test.skipFinalAddrCheck {
+ addr, err := s.GetMainNICAddress(nicID, header.IPv6ProtocolNumber)
+ if err != nil {
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (_, %v), want = (_, nil)", nicID, header.IPv6ProtocolNumber, err)
+ }
+ if want := (tcpip.AddressWithPrefix{}); addr != want {
+ t.Errorf("got stack.GetMainNICAddress(%d, %d) = (%s, nil), want = (%s, nil)", nicID, header.IPv6ProtocolNumber, addr, want)
+ }
+ }
+
+ // Should not have sent more than 1 NS message.
+ if got := s.Stats().ICMP.V6PacketsSent.NeighborSolicit.Value(); got > 1 {
+ t.Errorf("got NeighborSolicit = %d, want <= 1", got)
+ }
+ })
}
}
@@ -460,6 +832,10 @@ func TestSetNDPConfigurationFailsForBadNICID(t *testing.T) {
// configurations without affecting the default NDP configurations or other
// interfaces' configurations.
func TestSetNDPConfigurations(t *testing.T) {
+ const nicID1 = 1
+ const nicID2 = 2
+ const nicID3 = 3
+
tests := []struct {
name string
dupAddrDetectTransmits uint8
@@ -483,25 +859,36 @@ func TestSetNDPConfigurations(t *testing.T) {
for _, test := range tests {
t.Run(test.name, func(t *testing.T) {
ndpDisp := ndpDispatcher{
- dadC: make(chan ndpDADEvent),
+ dadC: make(chan ndpDADEvent, 1),
}
- e := channel.New(10, 1280, linkAddr1)
+ e := channel.New(0, 1280, linkAddr1)
s := stack.New(stack.Options{
NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
NDPDisp: &ndpDisp,
})
+ expectDADEvent := func(nicID tcpip.NICID, addr tcpip.Address) {
+ select {
+ case e := <-ndpDisp.dadC:
+ if diff := checkDADEvent(e, nicID, addr, true, nil); diff != "" {
+ t.Errorf("dad event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatalf("expected DAD event for %s", addr)
+ }
+ }
+
// 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)
+ if err := s.CreateNIC(nicID1, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID1, 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)
+ if err := s.CreateNIC(nicID2, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID2, err)
}
// Update the NDP configurations on NIC(1) to use DAD.
@@ -509,36 +896,38 @@ func TestSetNDPConfigurations(t *testing.T) {
DupAddrDetectTransmits: test.dupAddrDetectTransmits,
RetransmitTimer: test.retransmitTimer,
}
- if err := s.SetNDPConfigurations(1, configs); err != nil {
- t.Fatalf("got SetNDPConfigurations(1, _) = %s", err)
+ if err := s.SetNDPConfigurations(nicID1, configs); err != nil {
+ t.Fatalf("got SetNDPConfigurations(%d, _) = %s", nicID1, 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)
+ if err := s.CreateNIC(nicID3, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID3, 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(nicID1, header.IPv6ProtocolNumber, addr1); err != nil {
+ t.Fatalf("AddAddress(%d, %d, %s) = %s", nicID1, 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(nicID2, header.IPv6ProtocolNumber, addr2); err != nil {
+ t.Fatalf("AddAddress(%d, %d, %s) = %s", nicID2, 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)
+ expectDADEvent(nicID2, addr2)
+ if err := s.AddAddress(nicID3, header.IPv6ProtocolNumber, addr3); err != nil {
+ t.Fatalf("AddAddress(%d, %d, %s) = %s", nicID3, header.IPv6ProtocolNumber, addr3, err)
}
+ expectDADEvent(nicID3, addr3)
// Address should not be considered bound to NIC(1) yet
// (DAD ongoing).
- addr, err := s.GetMainNICAddress(1, header.IPv6ProtocolNumber)
+ addr, err := s.GetMainNICAddress(nicID1, header.IPv6ProtocolNumber)
if err != nil {
- t.Fatalf("got stack.GetMainNICAddress(_, _) = (_, %v), want = (_, nil)", err)
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (_, %v), want = (_, nil)", nicID1, header.IPv6ProtocolNumber, err)
}
if want := (tcpip.AddressWithPrefix{}); addr != want {
- t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want)
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (%s, nil), want = (%s, nil)", nicID1, header.IPv6ProtocolNumber, addr, want)
}
// Should get the address on NIC(2) and NIC(3)
@@ -546,31 +935,31 @@ func TestSetNDPConfigurations(t *testing.T) {
// 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)
+ addr, err = s.GetMainNICAddress(nicID2, header.IPv6ProtocolNumber)
if err != nil {
- t.Fatalf("stack.GetMainNICAddress(2, _) err = %s", err)
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (_, %v), want = (_, nil)", nicID2, header.IPv6ProtocolNumber, err)
}
if addr.Address != addr2 {
- t.Fatalf("got stack.GetMainNICAddress(2, _) = %s, want = %s", addr, addr2)
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = %s, want = %s", nicID2, header.IPv6ProtocolNumber, addr, addr2)
}
- addr, err = s.GetMainNICAddress(3, header.IPv6ProtocolNumber)
+ addr, err = s.GetMainNICAddress(nicID3, header.IPv6ProtocolNumber)
if err != nil {
- t.Fatalf("stack.GetMainNICAddress(3, _) err = %s", err)
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (_, %v), want = (_, nil)", nicID3, header.IPv6ProtocolNumber, err)
}
if addr.Address != addr3 {
- t.Fatalf("got stack.GetMainNICAddress(3, _) = %s, want = %s", addr, addr3)
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = %s, want = %s", nicID3, header.IPv6ProtocolNumber, 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)
+ addr, err = s.GetMainNICAddress(nicID1, header.IPv6ProtocolNumber)
if err != nil {
- t.Fatalf("got stack.GetMainNICAddress(_, _) = (_, %v), want = (_, nil)", err)
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (_, %v), want = (_, nil)", nicID1, header.IPv6ProtocolNumber, err)
}
if want := (tcpip.AddressWithPrefix{}); addr != want {
- t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want)
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (%s, nil), want = (%s, nil)", nicID1, header.IPv6ProtocolNumber, addr, want)
}
// Wait for DAD to resolve.
@@ -584,25 +973,4386 @@ func TestSetNDPConfigurations(t *testing.T) {
// 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 diff := checkDADEvent(e, nicID1, addr1, true, nil); diff != "" {
+ t.Errorf("dad event mismatch (-want +got):\n%s", diff)
}
- if e.nicid != 1 {
- t.Fatalf("got DAD event w/ nicid = %d, want = 1", e.nicid)
+ }
+ addr, err = s.GetMainNICAddress(nicID1, header.IPv6ProtocolNumber)
+ if err != nil {
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (_, %v), want = (_, nil)", nicID1, header.IPv6ProtocolNumber, err)
+ }
+ if addr.Address != addr1 {
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = %s, want = %s", nicID1, header.IPv6ProtocolNumber, addr, addr1)
+ }
+ })
+ }
+}
+
+// raBufWithOptsAndDHCPv6 returns a valid NDP Router Advertisement with options
+// and DHCPv6 configurations specified.
+func raBufWithOptsAndDHCPv6(ip tcpip.Address, rl uint16, managedAddress, otherConfigurations bool, optSer header.NDPOptionsSerializer) *stack.PacketBuffer {
+ icmpSize := header.ICMPv6HeaderSize + header.NDPRAMinimumSize + int(optSer.Length())
+ hdr := buffer.NewPrependable(header.IPv6MinimumSize + icmpSize)
+ pkt := header.ICMPv6(hdr.Prepend(icmpSize))
+ pkt.SetType(header.ICMPv6RouterAdvert)
+ pkt.SetCode(0)
+ raPayload := pkt.NDPPayload()
+ ra := header.NDPRouterAdvert(raPayload)
+ // Populate the Router Lifetime.
+ binary.BigEndian.PutUint16(raPayload[2:], rl)
+ // Populate the Managed Address flag field.
+ if managedAddress {
+ // The Managed Addresses flag field is the 7th bit of byte #1 (0-indexing)
+ // of the RA payload.
+ raPayload[1] |= (1 << 7)
+ }
+ // Populate the Other Configurations flag field.
+ if otherConfigurations {
+ // The Other Configurations flag field is the 6th bit of byte #1
+ // (0-indexing) of the RA payload.
+ raPayload[1] |= (1 << 6)
+ }
+ opts := ra.Options()
+ opts.Serialize(optSer)
+ pkt.SetChecksum(header.ICMPv6Checksum(pkt, ip, header.IPv6AllNodesMulticastAddress, buffer.VectorisedView{}))
+ payloadLength := hdr.UsedLength()
+ iph := header.IPv6(hdr.Prepend(header.IPv6MinimumSize))
+ iph.Encode(&header.IPv6Fields{
+ PayloadLength: uint16(payloadLength),
+ NextHeader: uint8(icmp.ProtocolNumber6),
+ HopLimit: header.NDPHopLimit,
+ SrcAddr: ip,
+ DstAddr: header.IPv6AllNodesMulticastAddress,
+ })
+
+ return stack.NewPacketBuffer(stack.PacketBufferOptions{
+ Data: hdr.View().ToVectorisedView(),
+ })
+}
+
+// raBufWithOpts returns a valid NDP Router Advertisement with options.
+//
+// Note, raBufWithOpts does not populate any of the RA fields other than the
+// Router Lifetime.
+func raBufWithOpts(ip tcpip.Address, rl uint16, optSer header.NDPOptionsSerializer) *stack.PacketBuffer {
+ return raBufWithOptsAndDHCPv6(ip, rl, false, false, optSer)
+}
+
+// raBufWithDHCPv6 returns a valid NDP Router Advertisement with DHCPv6 related
+// fields set.
+//
+// Note, raBufWithDHCPv6 does not populate any of the RA fields other than the
+// DHCPv6 related ones.
+func raBufWithDHCPv6(ip tcpip.Address, managedAddresses, otherConfiguratiosns bool) *stack.PacketBuffer {
+ return raBufWithOptsAndDHCPv6(ip, 0, managedAddresses, otherConfiguratiosns, header.NDPOptionsSerializer{})
+}
+
+// raBuf returns a valid NDP Router Advertisement.
+//
+// Note, raBuf does not populate any of the RA fields other than the
+// Router Lifetime.
+func raBuf(ip tcpip.Address, rl uint16) *stack.PacketBuffer {
+ return raBufWithOpts(ip, rl, header.NDPOptionsSerializer{})
+}
+
+// raBufWithPI returns a valid NDP Router Advertisement with a single Prefix
+// Information option.
+//
+// Note, raBufWithPI does not populate any of the RA fields other than the
+// Router Lifetime.
+func raBufWithPI(ip tcpip.Address, rl uint16, prefix tcpip.AddressWithPrefix, onLink, auto bool, vl, pl uint32) *stack.PacketBuffer {
+ flags := uint8(0)
+ if onLink {
+ // The OnLink flag is the 7th bit in the flags byte.
+ flags |= 1 << 7
+ }
+ if auto {
+ // The Address Auto-Configuration flag is the 6th bit in the
+ // flags byte.
+ flags |= 1 << 6
+ }
+
+ // A valid header.NDPPrefixInformation must be 30 bytes.
+ buf := [30]byte{}
+ // The first byte in a header.NDPPrefixInformation is the Prefix Length
+ // field.
+ buf[0] = uint8(prefix.PrefixLen)
+ // The 2nd byte within a header.NDPPrefixInformation is the Flags field.
+ buf[1] = flags
+ // The Valid Lifetime field starts after the 2nd byte within a
+ // header.NDPPrefixInformation.
+ binary.BigEndian.PutUint32(buf[2:], vl)
+ // The Preferred Lifetime field starts after the 6th byte within a
+ // header.NDPPrefixInformation.
+ binary.BigEndian.PutUint32(buf[6:], pl)
+ // The Prefix Address field starts after the 14th byte within a
+ // header.NDPPrefixInformation.
+ copy(buf[14:], prefix.Address)
+ return raBufWithOpts(ip, rl, header.NDPOptionsSerializer{
+ header.NDPPrefixInformation(buf[:]),
+ })
+}
+
+// TestNoRouterDiscovery tests that router discovery will not be performed if
+// configured not to.
+func TestNoRouterDiscovery(t *testing.T) {
+ // Being configured to discover routers means handle and
+ // discover are set to true and forwarding is set to false.
+ // This tests all possible combinations of the configurations,
+ // except for the configuration where handle = true, discover =
+ // true and forwarding = false (the required configuration to do
+ // router discovery) - that will done in other tests.
+ for i := 0; i < 7; i++ {
+ handle := i&1 != 0
+ discover := i&2 != 0
+ forwarding := i&4 == 0
+
+ t.Run(fmt.Sprintf("HandleRAs(%t), DiscoverDefaultRouters(%t), Forwarding(%t)", handle, discover, forwarding), func(t *testing.T) {
+ ndpDisp := ndpDispatcher{
+ routerC: make(chan ndpRouterEvent, 1),
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ HandleRAs: handle,
+ DiscoverDefaultRouters: discover,
+ },
+ NDPDisp: &ndpDisp,
+ })
+ s.SetForwarding(ipv6.ProtocolNumber, forwarding)
+
+ if err := s.CreateNIC(1, e); err != nil {
+ t.Fatalf("CreateNIC(1) = %s", err)
+ }
+
+ // Rx an RA with non-zero lifetime.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, 1000))
+ select {
+ case <-ndpDisp.routerC:
+ t.Fatal("unexpectedly discovered a router when configured not to")
+ default:
+ }
+ })
+ }
+}
+
+// Check e to make sure that the event is for addr on nic with ID 1, and the
+// discovered flag set to discovered.
+func checkRouterEvent(e ndpRouterEvent, addr tcpip.Address, discovered bool) string {
+ return cmp.Diff(ndpRouterEvent{nicID: 1, addr: addr, discovered: discovered}, e, cmp.AllowUnexported(e))
+}
+
+// TestRouterDiscoveryDispatcherNoRemember tests that the stack does not
+// remember a discovered router when the dispatcher asks it not to.
+func TestRouterDiscoveryDispatcherNoRemember(t *testing.T) {
+ ndpDisp := ndpDispatcher{
+ routerC: make(chan ndpRouterEvent, 1),
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ HandleRAs: true,
+ DiscoverDefaultRouters: true,
+ },
+ NDPDisp: &ndpDisp,
+ })
+
+ if err := s.CreateNIC(1, e); err != nil {
+ t.Fatalf("CreateNIC(1) = %s", err)
+ }
+
+ // Receive an RA for a router we should not remember.
+ const lifetimeSeconds = 1
+ e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, lifetimeSeconds))
+ select {
+ case e := <-ndpDisp.routerC:
+ if diff := checkRouterEvent(e, llAddr2, true); diff != "" {
+ t.Errorf("router event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected router discovery event")
+ }
+
+ // Wait for the invalidation time plus some buffer to make sure we do
+ // not actually receive any invalidation events as we should not have
+ // remembered the router in the first place.
+ select {
+ case <-ndpDisp.routerC:
+ t.Fatal("should not have received any router events")
+ case <-time.After(lifetimeSeconds*time.Second + defaultAsyncNegativeEventTimeout):
+ }
+}
+
+func TestRouterDiscovery(t *testing.T) {
+ ndpDisp := ndpDispatcher{
+ routerC: make(chan ndpRouterEvent, 1),
+ rememberRouter: true,
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ HandleRAs: true,
+ DiscoverDefaultRouters: true,
+ },
+ NDPDisp: &ndpDisp,
+ })
+
+ expectRouterEvent := func(addr tcpip.Address, discovered bool) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.routerC:
+ if diff := checkRouterEvent(e, addr, discovered); diff != "" {
+ t.Errorf("router event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected router discovery event")
+ }
+ }
+
+ expectAsyncRouterInvalidationEvent := func(addr tcpip.Address, timeout time.Duration) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.routerC:
+ if diff := checkRouterEvent(e, addr, false); diff != "" {
+ t.Errorf("router event mismatch (-want +got):\n%s", diff)
+ }
+ case <-time.After(timeout):
+ t.Fatal("timed out waiting for router discovery event")
+ }
+ }
+
+ if err := s.CreateNIC(1, e); err != nil {
+ t.Fatalf("CreateNIC(1) = %s", err)
+ }
+
+ // Rx an RA from lladdr2 with zero lifetime. It should not be
+ // remembered.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, 0))
+ select {
+ case <-ndpDisp.routerC:
+ t.Fatal("unexpectedly discovered a router with 0 lifetime")
+ default:
+ }
+
+ // Rx an RA from lladdr2 with a huge lifetime.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, 1000))
+ expectRouterEvent(llAddr2, true)
+
+ // Rx an RA from another router (lladdr3) with non-zero lifetime.
+ const l3LifetimeSeconds = 6
+ e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr3, l3LifetimeSeconds))
+ expectRouterEvent(llAddr3, true)
+
+ // Rx an RA from lladdr2 with lesser lifetime.
+ const l2LifetimeSeconds = 2
+ e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, l2LifetimeSeconds))
+ select {
+ case <-ndpDisp.routerC:
+ t.Fatal("Should not receive a router event when updating lifetimes for known routers")
+ default:
+ }
+
+ // Wait for lladdr2's router invalidation job to execute. The lifetime
+ // of the router should have been updated to the most recent (smaller)
+ // lifetime.
+ //
+ // Wait for the normal lifetime plus an extra bit for the
+ // router to get invalidated. If we don't get an invalidation
+ // event after this time, then something is wrong.
+ expectAsyncRouterInvalidationEvent(llAddr2, l2LifetimeSeconds*time.Second+defaultAsyncPositiveEventTimeout)
+
+ // Rx an RA from lladdr2 with huge lifetime.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, 1000))
+ expectRouterEvent(llAddr2, true)
+
+ // Rx an RA from lladdr2 with zero lifetime. It should be invalidated.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, 0))
+ expectRouterEvent(llAddr2, false)
+
+ // Wait for lladdr3's router invalidation job to execute. The lifetime
+ // of the router should have been updated to the most recent (smaller)
+ // lifetime.
+ //
+ // Wait for the normal lifetime plus an extra bit for the
+ // router to get invalidated. If we don't get an invalidation
+ // event after this time, then something is wrong.
+ expectAsyncRouterInvalidationEvent(llAddr3, l3LifetimeSeconds*time.Second+defaultAsyncPositiveEventTimeout)
+}
+
+// TestRouterDiscoveryMaxRouters tests that only
+// stack.MaxDiscoveredDefaultRouters discovered routers are remembered.
+func TestRouterDiscoveryMaxRouters(t *testing.T) {
+ ndpDisp := ndpDispatcher{
+ routerC: make(chan ndpRouterEvent, 1),
+ rememberRouter: true,
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ HandleRAs: true,
+ DiscoverDefaultRouters: true,
+ },
+ NDPDisp: &ndpDisp,
+ })
+
+ if err := s.CreateNIC(1, e); err != nil {
+ t.Fatalf("CreateNIC(1) = %s", err)
+ }
+
+ // Receive an RA from 2 more than the max number of discovered routers.
+ for i := 1; i <= stack.MaxDiscoveredDefaultRouters+2; i++ {
+ linkAddr := []byte{2, 2, 3, 4, 5, 0}
+ linkAddr[5] = byte(i)
+ llAddr := header.LinkLocalAddr(tcpip.LinkAddress(linkAddr))
+
+ e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr, 5))
+
+ if i <= stack.MaxDiscoveredDefaultRouters {
+ select {
+ case e := <-ndpDisp.routerC:
+ if diff := checkRouterEvent(e, llAddr, true); diff != "" {
+ t.Errorf("router event mismatch (-want +got):\n%s", diff)
}
- if e.addr != addr1 {
- t.Fatalf("got DAD event w/ addr = %s, want = %s", addr, addr1)
+ default:
+ t.Fatal("expected router discovery event")
+ }
+
+ } else {
+ select {
+ case <-ndpDisp.routerC:
+ t.Fatal("should not have discovered a new router after we already discovered the max number of routers")
+ default:
+ }
+ }
+ }
+}
+
+// TestNoPrefixDiscovery tests that prefix discovery will not be performed if
+// configured not to.
+func TestNoPrefixDiscovery(t *testing.T) {
+ prefix := tcpip.AddressWithPrefix{
+ Address: tcpip.Address("\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x00"),
+ PrefixLen: 64,
+ }
+
+ // Being configured to discover prefixes means handle and
+ // discover are set to true and forwarding is set to false.
+ // This tests all possible combinations of the configurations,
+ // except for the configuration where handle = true, discover =
+ // true and forwarding = false (the required configuration to do
+ // prefix discovery) - that will done in other tests.
+ for i := 0; i < 7; i++ {
+ handle := i&1 != 0
+ discover := i&2 != 0
+ forwarding := i&4 == 0
+
+ t.Run(fmt.Sprintf("HandleRAs(%t), DiscoverOnLinkPrefixes(%t), Forwarding(%t)", handle, discover, forwarding), func(t *testing.T) {
+ ndpDisp := ndpDispatcher{
+ prefixC: make(chan ndpPrefixEvent, 1),
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ HandleRAs: handle,
+ DiscoverOnLinkPrefixes: discover,
+ },
+ NDPDisp: &ndpDisp,
+ })
+ s.SetForwarding(ipv6.ProtocolNumber, forwarding)
+
+ if err := s.CreateNIC(1, e); err != nil {
+ t.Fatalf("CreateNIC(1) = %s", err)
+ }
+
+ // Rx an RA with prefix with non-zero lifetime.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, 10, 0))
+
+ select {
+ case <-ndpDisp.prefixC:
+ t.Fatal("unexpectedly discovered a prefix when configured not to")
+ default:
+ }
+ })
+ }
+}
+
+// Check e to make sure that the event is for prefix on nic with ID 1, and the
+// discovered flag set to discovered.
+func checkPrefixEvent(e ndpPrefixEvent, prefix tcpip.Subnet, discovered bool) string {
+ return cmp.Diff(ndpPrefixEvent{nicID: 1, prefix: prefix, discovered: discovered}, e, cmp.AllowUnexported(e))
+}
+
+// TestPrefixDiscoveryDispatcherNoRemember tests that the stack does not
+// remember a discovered on-link prefix when the dispatcher asks it not to.
+func TestPrefixDiscoveryDispatcherNoRemember(t *testing.T) {
+ prefix, subnet, _ := prefixSubnetAddr(0, "")
+
+ ndpDisp := ndpDispatcher{
+ prefixC: make(chan ndpPrefixEvent, 1),
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ HandleRAs: true,
+ DiscoverDefaultRouters: false,
+ DiscoverOnLinkPrefixes: true,
+ },
+ NDPDisp: &ndpDisp,
+ })
+
+ if err := s.CreateNIC(1, e); err != nil {
+ t.Fatalf("CreateNIC(1) = %s", err)
+ }
+
+ // Receive an RA with prefix that we should not remember.
+ const lifetimeSeconds = 1
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, lifetimeSeconds, 0))
+ select {
+ case e := <-ndpDisp.prefixC:
+ if diff := checkPrefixEvent(e, subnet, true); diff != "" {
+ t.Errorf("prefix event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected prefix discovery event")
+ }
+
+ // Wait for the invalidation time plus some buffer to make sure we do
+ // not actually receive any invalidation events as we should not have
+ // remembered the prefix in the first place.
+ select {
+ case <-ndpDisp.prefixC:
+ t.Fatal("should not have received any prefix events")
+ case <-time.After(lifetimeSeconds*time.Second + defaultAsyncNegativeEventTimeout):
+ }
+}
+
+func TestPrefixDiscovery(t *testing.T) {
+ prefix1, subnet1, _ := prefixSubnetAddr(0, "")
+ prefix2, subnet2, _ := prefixSubnetAddr(1, "")
+ prefix3, subnet3, _ := prefixSubnetAddr(2, "")
+
+ ndpDisp := ndpDispatcher{
+ prefixC: make(chan ndpPrefixEvent, 1),
+ rememberPrefix: true,
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ HandleRAs: true,
+ DiscoverOnLinkPrefixes: true,
+ },
+ NDPDisp: &ndpDisp,
+ })
+
+ if err := s.CreateNIC(1, e); err != nil {
+ t.Fatalf("CreateNIC(1) = %s", err)
+ }
+
+ expectPrefixEvent := func(prefix tcpip.Subnet, discovered bool) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.prefixC:
+ if diff := checkPrefixEvent(e, prefix, discovered); diff != "" {
+ t.Errorf("prefix event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected prefix discovery event")
+ }
+ }
+
+ // Receive an RA with prefix1 in an NDP Prefix Information option (PI)
+ // with zero valid lifetime.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, false, 0, 0))
+ select {
+ case <-ndpDisp.prefixC:
+ t.Fatal("unexpectedly discovered a prefix with 0 lifetime")
+ default:
+ }
+
+ // Receive an RA with prefix1 in an NDP Prefix Information option (PI)
+ // with non-zero lifetime.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, false, 100, 0))
+ expectPrefixEvent(subnet1, true)
+
+ // Receive an RA with prefix2 in a PI.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix2, true, false, 100, 0))
+ expectPrefixEvent(subnet2, true)
+
+ // Receive an RA with prefix3 in a PI.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix3, true, false, 100, 0))
+ expectPrefixEvent(subnet3, true)
+
+ // Receive an RA with prefix1 in a PI with lifetime = 0.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, false, 0, 0))
+ expectPrefixEvent(subnet1, false)
+
+ // Receive an RA with prefix2 in a PI with lesser lifetime.
+ lifetime := uint32(2)
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix2, true, false, lifetime, 0))
+ select {
+ case <-ndpDisp.prefixC:
+ t.Fatal("unexpectedly received prefix event when updating lifetime")
+ default:
+ }
+
+ // Wait for prefix2's most recent invalidation job plus some buffer to
+ // expire.
+ select {
+ case e := <-ndpDisp.prefixC:
+ if diff := checkPrefixEvent(e, subnet2, false); diff != "" {
+ t.Errorf("prefix event mismatch (-want +got):\n%s", diff)
+ }
+ case <-time.After(time.Duration(lifetime)*time.Second + defaultAsyncPositiveEventTimeout):
+ t.Fatal("timed out waiting for prefix discovery event")
+ }
+
+ // Receive RA to invalidate prefix3.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix3, true, false, 0, 0))
+ expectPrefixEvent(subnet3, false)
+}
+
+func TestPrefixDiscoveryWithInfiniteLifetime(t *testing.T) {
+ // Update the infinite lifetime value to a smaller value so we can test
+ // that when we receive a PI with such a lifetime value, we do not
+ // invalidate the prefix.
+ const testInfiniteLifetimeSeconds = 2
+ const testInfiniteLifetime = testInfiniteLifetimeSeconds * time.Second
+ saved := header.NDPInfiniteLifetime
+ header.NDPInfiniteLifetime = testInfiniteLifetime
+ defer func() {
+ header.NDPInfiniteLifetime = saved
+ }()
+
+ prefix := tcpip.AddressWithPrefix{
+ Address: tcpip.Address("\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x00"),
+ PrefixLen: 64,
+ }
+ subnet := prefix.Subnet()
+
+ ndpDisp := ndpDispatcher{
+ prefixC: make(chan ndpPrefixEvent, 1),
+ rememberPrefix: true,
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ HandleRAs: true,
+ DiscoverOnLinkPrefixes: true,
+ },
+ NDPDisp: &ndpDisp,
+ })
+
+ if err := s.CreateNIC(1, e); err != nil {
+ t.Fatalf("CreateNIC(1) = %s", err)
+ }
+
+ expectPrefixEvent := func(prefix tcpip.Subnet, discovered bool) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.prefixC:
+ if diff := checkPrefixEvent(e, prefix, discovered); diff != "" {
+ t.Errorf("prefix event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected prefix discovery event")
+ }
+ }
+
+ // Receive an RA with prefix in an NDP Prefix Information option (PI)
+ // with infinite valid lifetime which should not get invalidated.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, testInfiniteLifetimeSeconds, 0))
+ expectPrefixEvent(subnet, true)
+ select {
+ case <-ndpDisp.prefixC:
+ t.Fatal("unexpectedly invalidated a prefix with infinite lifetime")
+ case <-time.After(testInfiniteLifetime + defaultAsyncNegativeEventTimeout):
+ }
+
+ // Receive an RA with finite lifetime.
+ // The prefix should get invalidated after 1s.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, testInfiniteLifetimeSeconds-1, 0))
+ select {
+ case e := <-ndpDisp.prefixC:
+ if diff := checkPrefixEvent(e, subnet, false); diff != "" {
+ t.Errorf("prefix event mismatch (-want +got):\n%s", diff)
+ }
+ case <-time.After(testInfiniteLifetime):
+ t.Fatal("timed out waiting for prefix discovery event")
+ }
+
+ // Receive an RA with finite lifetime.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, testInfiniteLifetimeSeconds-1, 0))
+ expectPrefixEvent(subnet, true)
+
+ // Receive an RA with prefix with an infinite lifetime.
+ // The prefix should not be invalidated.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, testInfiniteLifetimeSeconds, 0))
+ select {
+ case <-ndpDisp.prefixC:
+ t.Fatal("unexpectedly invalidated a prefix with infinite lifetime")
+ case <-time.After(testInfiniteLifetime + defaultAsyncNegativeEventTimeout):
+ }
+
+ // Receive an RA with a prefix with a lifetime value greater than the
+ // set infinite lifetime value.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, testInfiniteLifetimeSeconds+1, 0))
+ select {
+ case <-ndpDisp.prefixC:
+ t.Fatal("unexpectedly invalidated a prefix with infinite lifetime")
+ case <-time.After((testInfiniteLifetimeSeconds+1)*time.Second + defaultAsyncNegativeEventTimeout):
+ }
+
+ // Receive an RA with 0 lifetime.
+ // The prefix should get invalidated.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, 0, 0))
+ expectPrefixEvent(subnet, false)
+}
+
+// TestPrefixDiscoveryMaxRouters tests that only
+// stack.MaxDiscoveredOnLinkPrefixes discovered on-link prefixes are remembered.
+func TestPrefixDiscoveryMaxOnLinkPrefixes(t *testing.T) {
+ ndpDisp := ndpDispatcher{
+ prefixC: make(chan ndpPrefixEvent, stack.MaxDiscoveredOnLinkPrefixes+3),
+ rememberPrefix: true,
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ HandleRAs: true,
+ DiscoverDefaultRouters: false,
+ DiscoverOnLinkPrefixes: true,
+ },
+ NDPDisp: &ndpDisp,
+ })
+
+ if err := s.CreateNIC(1, e); err != nil {
+ t.Fatalf("CreateNIC(1) = %s", err)
+ }
+
+ optSer := make(header.NDPOptionsSerializer, stack.MaxDiscoveredOnLinkPrefixes+2)
+ prefixes := [stack.MaxDiscoveredOnLinkPrefixes + 2]tcpip.Subnet{}
+
+ // Receive an RA with 2 more than the max number of discovered on-link
+ // prefixes.
+ for i := 0; i < stack.MaxDiscoveredOnLinkPrefixes+2; i++ {
+ prefixAddr := [16]byte{1, 2, 3, 4, 5, 6, 7, 8, 0, 0, 0, 0, 0, 0, 0, 0}
+ prefixAddr[7] = byte(i)
+ prefix := tcpip.AddressWithPrefix{
+ Address: tcpip.Address(prefixAddr[:]),
+ PrefixLen: 64,
+ }
+ prefixes[i] = prefix.Subnet()
+ buf := [30]byte{}
+ buf[0] = uint8(prefix.PrefixLen)
+ buf[1] = 128
+ binary.BigEndian.PutUint32(buf[2:], 10)
+ copy(buf[14:], prefix.Address)
+
+ optSer[i] = header.NDPPrefixInformation(buf[:])
+ }
+
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithOpts(llAddr1, 0, optSer))
+ for i := 0; i < stack.MaxDiscoveredOnLinkPrefixes+2; i++ {
+ if i < stack.MaxDiscoveredOnLinkPrefixes {
+ select {
+ case e := <-ndpDisp.prefixC:
+ if diff := checkPrefixEvent(e, prefixes[i], true); diff != "" {
+ t.Errorf("prefix event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected prefix discovery event")
+ }
+ } else {
+ select {
+ case <-ndpDisp.prefixC:
+ t.Fatal("should not have discovered a new prefix after we already discovered the max number of prefixes")
+ default:
+ }
+ }
+ }
+}
+
+// Checks to see if list contains an IPv6 address, item.
+func containsV6Addr(list []tcpip.ProtocolAddress, item tcpip.AddressWithPrefix) bool {
+ protocolAddress := tcpip.ProtocolAddress{
+ Protocol: header.IPv6ProtocolNumber,
+ AddressWithPrefix: item,
+ }
+
+ return containsAddr(list, protocolAddress)
+}
+
+// TestNoAutoGenAddr tests that SLAAC is not performed when configured not to.
+func TestNoAutoGenAddr(t *testing.T) {
+ prefix, _, _ := prefixSubnetAddr(0, "")
+
+ // Being configured to auto-generate addresses means handle and
+ // autogen are set to true and forwarding is set to false.
+ // This tests all possible combinations of the configurations,
+ // except for the configuration where handle = true, autogen =
+ // true and forwarding = false (the required configuration to do
+ // SLAAC) - that will done in other tests.
+ for i := 0; i < 7; i++ {
+ handle := i&1 != 0
+ autogen := i&2 != 0
+ forwarding := i&4 == 0
+
+ t.Run(fmt.Sprintf("HandleRAs(%t), AutoGenAddr(%t), Forwarding(%t)", handle, autogen, forwarding), func(t *testing.T) {
+ ndpDisp := ndpDispatcher{
+ autoGenAddrC: make(chan ndpAutoGenAddrEvent, 1),
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ HandleRAs: handle,
+ AutoGenGlobalAddresses: autogen,
+ },
+ NDPDisp: &ndpDisp,
+ })
+ s.SetForwarding(ipv6.ProtocolNumber, forwarding)
+
+ if err := s.CreateNIC(1, e); err != nil {
+ t.Fatalf("CreateNIC(1) = %s", err)
+ }
+
+ // Rx an RA with prefix with non-zero lifetime.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, false, true, 10, 0))
+
+ select {
+ case <-ndpDisp.autoGenAddrC:
+ t.Fatal("unexpectedly auto-generated an address when configured not to")
+ default:
+ }
+ })
+ }
+}
+
+// Check e to make sure that the event is for addr on nic with ID 1, and the
+// event type is set to eventType.
+func checkAutoGenAddrEvent(e ndpAutoGenAddrEvent, addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType) string {
+ return cmp.Diff(ndpAutoGenAddrEvent{nicID: 1, addr: addr, eventType: eventType}, e, cmp.AllowUnexported(e))
+}
+
+// TestAutoGenAddr tests that an address is properly generated and invalidated
+// when configured to do so.
+func TestAutoGenAddr(t *testing.T) {
+ const newMinVL = 2
+ newMinVLDuration := newMinVL * time.Second
+ saved := stack.MinPrefixInformationValidLifetimeForUpdate
+ defer func() {
+ stack.MinPrefixInformationValidLifetimeForUpdate = saved
+ }()
+ stack.MinPrefixInformationValidLifetimeForUpdate = newMinVLDuration
+
+ prefix1, _, addr1 := prefixSubnetAddr(0, linkAddr1)
+ prefix2, _, addr2 := prefixSubnetAddr(1, linkAddr1)
+
+ ndpDisp := ndpDispatcher{
+ autoGenAddrC: make(chan ndpAutoGenAddrEvent, 1),
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ HandleRAs: true,
+ AutoGenGlobalAddresses: true,
+ },
+ NDPDisp: &ndpDisp,
+ })
+
+ if err := s.CreateNIC(1, e); err != nil {
+ t.Fatalf("CreateNIC(1) = %s", err)
+ }
+
+ expectAutoGenAddrEvent := func(addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, eventType); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected addr auto gen event")
+ }
+ }
+
+ // Receive an RA with prefix1 in an NDP Prefix Information option (PI)
+ // with zero valid lifetime.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, 0, 0))
+ select {
+ case <-ndpDisp.autoGenAddrC:
+ t.Fatal("unexpectedly auto-generated an address with 0 lifetime")
+ default:
+ }
+
+ // Receive an RA with prefix1 in an NDP Prefix Information option (PI)
+ // with non-zero lifetime.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, 100, 0))
+ expectAutoGenAddrEvent(addr1, newAddr)
+ if !containsV6Addr(s.NICInfo()[1].ProtocolAddresses, addr1) {
+ t.Fatalf("Should have %s in the list of addresses", addr1)
+ }
+
+ // Receive an RA with prefix2 in an NDP Prefix Information option (PI)
+ // with preferred lifetime > valid lifetime
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, 5, 6))
+ select {
+ case <-ndpDisp.autoGenAddrC:
+ t.Fatal("unexpectedly auto-generated an address with preferred lifetime > valid lifetime")
+ default:
+ }
+
+ // Receive an RA with prefix2 in a PI.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix2, true, true, 100, 0))
+ expectAutoGenAddrEvent(addr2, newAddr)
+ if !containsV6Addr(s.NICInfo()[1].ProtocolAddresses, addr1) {
+ t.Fatalf("Should have %s in the list of addresses", addr1)
+ }
+ if !containsV6Addr(s.NICInfo()[1].ProtocolAddresses, addr2) {
+ t.Fatalf("Should have %s in the list of addresses", addr2)
+ }
+
+ // Refresh valid lifetime for addr of prefix1.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, newMinVL, 0))
+ select {
+ case <-ndpDisp.autoGenAddrC:
+ t.Fatal("unexpectedly auto-generated an address when we already have an address for a prefix")
+ default:
+ }
+
+ // Wait for addr of prefix1 to be invalidated.
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr1, invalidatedAddr); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ case <-time.After(newMinVLDuration + defaultAsyncPositiveEventTimeout):
+ t.Fatal("timed out waiting for addr auto gen event")
+ }
+ if containsV6Addr(s.NICInfo()[1].ProtocolAddresses, addr1) {
+ t.Fatalf("Should not have %s in the list of addresses", addr1)
+ }
+ if !containsV6Addr(s.NICInfo()[1].ProtocolAddresses, addr2) {
+ t.Fatalf("Should have %s in the list of addresses", addr2)
+ }
+}
+
+func addressCheck(addrs []tcpip.ProtocolAddress, containList, notContainList []tcpip.AddressWithPrefix) string {
+ ret := ""
+ for _, c := range containList {
+ if !containsV6Addr(addrs, c) {
+ ret += fmt.Sprintf("should have %s in the list of addresses\n", c)
+ }
+ }
+ for _, c := range notContainList {
+ if containsV6Addr(addrs, c) {
+ ret += fmt.Sprintf("should not have %s in the list of addresses\n", c)
+ }
+ }
+ return ret
+}
+
+// TestAutoGenTempAddr tests that temporary SLAAC addresses are generated when
+// configured to do so as part of IPv6 Privacy Extensions.
+func TestAutoGenTempAddr(t *testing.T) {
+ const (
+ nicID = 1
+ newMinVL = 5
+ newMinVLDuration = newMinVL * time.Second
+ )
+
+ savedMinPrefixInformationValidLifetimeForUpdate := stack.MinPrefixInformationValidLifetimeForUpdate
+ savedMaxDesync := stack.MaxDesyncFactor
+ defer func() {
+ stack.MinPrefixInformationValidLifetimeForUpdate = savedMinPrefixInformationValidLifetimeForUpdate
+ stack.MaxDesyncFactor = savedMaxDesync
+ }()
+ stack.MinPrefixInformationValidLifetimeForUpdate = newMinVLDuration
+ stack.MaxDesyncFactor = time.Nanosecond
+
+ prefix1, _, addr1 := prefixSubnetAddr(0, linkAddr1)
+ prefix2, _, addr2 := prefixSubnetAddr(1, linkAddr1)
+
+ tests := []struct {
+ name string
+ dupAddrTransmits uint8
+ retransmitTimer time.Duration
+ }{
+ {
+ name: "DAD disabled",
+ },
+ {
+ name: "DAD enabled",
+ dupAddrTransmits: 1,
+ retransmitTimer: time.Second,
+ },
+ }
+
+ // This Run will not return until the parallel tests finish.
+ //
+ // We need this because we need to do some teardown work after the
+ // parallel tests complete.
+ //
+ // See https://godoc.org/testing#hdr-Subtests_and_Sub_benchmarks for
+ // more details.
+ t.Run("group", func(t *testing.T) {
+ for i, test := range tests {
+ i := i
+ test := test
+
+ t.Run(test.name, func(t *testing.T) {
+ t.Parallel()
+
+ seed := []byte{uint8(i)}
+ var tempIIDHistory [header.IIDSize]byte
+ header.InitialTempIID(tempIIDHistory[:], seed, nicID)
+ newTempAddr := func(stableAddr tcpip.Address) tcpip.AddressWithPrefix {
+ return header.GenerateTempIPv6SLAACAddr(tempIIDHistory[:], stableAddr)
+ }
+
+ ndpDisp := ndpDispatcher{
+ dadC: make(chan ndpDADEvent, 2),
+ autoGenAddrC: make(chan ndpAutoGenAddrEvent, 2),
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ DupAddrDetectTransmits: test.dupAddrTransmits,
+ RetransmitTimer: test.retransmitTimer,
+ HandleRAs: true,
+ AutoGenGlobalAddresses: true,
+ AutoGenTempGlobalAddresses: true,
+ },
+ NDPDisp: &ndpDisp,
+ TempIIDSeed: seed,
+ })
+
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+
+ expectAutoGenAddrEvent := func(addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, eventType); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected addr auto gen event")
+ }
+ }
+
+ expectAutoGenAddrEventAsync := func(addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, eventType); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ case <-time.After(defaultAsyncPositiveEventTimeout):
+ t.Fatal("timed out waiting for addr auto gen event")
+ }
+ }
+
+ expectDADEventAsync := func(addr tcpip.Address) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.dadC:
+ if diff := checkDADEvent(e, nicID, addr, true, nil); diff != "" {
+ t.Errorf("dad event mismatch (-want +got):\n%s", diff)
+ }
+ case <-time.After(time.Duration(test.dupAddrTransmits)*test.retransmitTimer + defaultAsyncPositiveEventTimeout):
+ t.Fatal("timed out waiting for DAD event")
+ }
+ }
+
+ // Receive an RA with prefix1 in an NDP Prefix Information option (PI)
+ // with zero valid lifetime.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, 0, 0))
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ t.Fatalf("unexpectedly auto-generated an address with 0 lifetime; event = %+v", e)
+ default:
+ }
+
+ // Receive an RA with prefix1 in an NDP Prefix Information option (PI)
+ // with non-zero valid lifetime.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, 100, 0))
+ expectAutoGenAddrEvent(addr1, newAddr)
+ expectDADEventAsync(addr1.Address)
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ t.Fatalf("unexpectedly got an auto gen addr event = %+v", e)
+ default:
+ }
+ if mismatch := addressCheck(s.NICInfo()[nicID].ProtocolAddresses, []tcpip.AddressWithPrefix{addr1}, nil); mismatch != "" {
+ t.Fatal(mismatch)
+ }
+
+ // Receive an RA with prefix1 in an NDP Prefix Information option (PI)
+ // with non-zero valid & preferred lifetimes.
+ tempAddr1 := newTempAddr(addr1.Address)
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, 100, 100))
+ expectAutoGenAddrEvent(tempAddr1, newAddr)
+ expectDADEventAsync(tempAddr1.Address)
+ if mismatch := addressCheck(s.NICInfo()[1].ProtocolAddresses, []tcpip.AddressWithPrefix{addr1, tempAddr1}, nil); mismatch != "" {
+ t.Fatal(mismatch)
+ }
+
+ // Receive an RA with prefix2 in an NDP Prefix Information option (PI)
+ // with preferred lifetime > valid lifetime
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, 5, 6))
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ t.Fatalf("unexpectedly auto-generated an address with preferred lifetime > valid lifetime; event = %+v", e)
+ default:
+ }
+ if mismatch := addressCheck(s.NICInfo()[nicID].ProtocolAddresses, []tcpip.AddressWithPrefix{addr1, tempAddr1}, nil); mismatch != "" {
+ t.Fatal(mismatch)
+ }
+
+ // Receive an RA with prefix2 in a PI w/ non-zero valid and preferred
+ // lifetimes.
+ tempAddr2 := newTempAddr(addr2.Address)
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix2, true, true, 100, 100))
+ expectAutoGenAddrEvent(addr2, newAddr)
+ expectDADEventAsync(addr2.Address)
+ expectAutoGenAddrEventAsync(tempAddr2, newAddr)
+ expectDADEventAsync(tempAddr2.Address)
+ if mismatch := addressCheck(s.NICInfo()[nicID].ProtocolAddresses, []tcpip.AddressWithPrefix{addr1, tempAddr1, addr2, tempAddr2}, nil); mismatch != "" {
+ t.Fatal(mismatch)
+ }
+
+ // Deprecate prefix1.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, 100, 0))
+ expectAutoGenAddrEvent(addr1, deprecatedAddr)
+ expectAutoGenAddrEvent(tempAddr1, deprecatedAddr)
+ if mismatch := addressCheck(s.NICInfo()[nicID].ProtocolAddresses, []tcpip.AddressWithPrefix{addr1, tempAddr1, addr2, tempAddr2}, nil); mismatch != "" {
+ t.Fatal(mismatch)
}
- if !e.resolved {
- t.Fatal("got DAD event w/ resolved = false, want = true")
+
+ // Refresh lifetimes for prefix1.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, 100, 100))
+ if mismatch := addressCheck(s.NICInfo()[nicID].ProtocolAddresses, []tcpip.AddressWithPrefix{addr1, tempAddr1, addr2, tempAddr2}, nil); mismatch != "" {
+ t.Fatal(mismatch)
+ }
+
+ // Reduce valid lifetime and deprecate addresses of prefix1.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, newMinVL, 0))
+ expectAutoGenAddrEvent(addr1, deprecatedAddr)
+ expectAutoGenAddrEvent(tempAddr1, deprecatedAddr)
+ if mismatch := addressCheck(s.NICInfo()[nicID].ProtocolAddresses, []tcpip.AddressWithPrefix{addr1, tempAddr1, addr2, tempAddr2}, nil); mismatch != "" {
+ t.Fatal(mismatch)
+ }
+
+ // Wait for addrs of prefix1 to be invalidated. They should be
+ // invalidated at the same time.
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ var nextAddr tcpip.AddressWithPrefix
+ if e.addr == addr1 {
+ if diff := checkAutoGenAddrEvent(e, addr1, invalidatedAddr); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ nextAddr = tempAddr1
+ } else {
+ if diff := checkAutoGenAddrEvent(e, tempAddr1, invalidatedAddr); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ nextAddr = addr1
+ }
+
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, nextAddr, invalidatedAddr); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ case <-time.After(defaultAsyncPositiveEventTimeout):
+ t.Fatal("timed out waiting for addr auto gen event")
+ }
+ case <-time.After(newMinVLDuration + defaultAsyncPositiveEventTimeout):
+ t.Fatal("timed out waiting for addr auto gen event")
+ }
+ if mismatch := addressCheck(s.NICInfo()[nicID].ProtocolAddresses, []tcpip.AddressWithPrefix{addr2, tempAddr2}, []tcpip.AddressWithPrefix{addr1, tempAddr1}); mismatch != "" {
+ t.Fatal(mismatch)
}
+
+ // Receive an RA with prefix2 in a PI w/ 0 lifetimes.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix2, true, true, 0, 0))
+ expectAutoGenAddrEvent(addr2, deprecatedAddr)
+ expectAutoGenAddrEvent(tempAddr2, deprecatedAddr)
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ t.Errorf("got unexpected auto gen addr event = %+v", e)
+ default:
+ }
+ if mismatch := addressCheck(s.NICInfo()[nicID].ProtocolAddresses, []tcpip.AddressWithPrefix{addr2, tempAddr2}, []tcpip.AddressWithPrefix{addr1, tempAddr1}); mismatch != "" {
+ t.Fatal(mismatch)
+ }
+ })
+ }
+ })
+}
+
+// TestNoAutoGenTempAddrForLinkLocal test that temporary SLAAC addresses are not
+// generated for auto generated link-local addresses.
+func TestNoAutoGenTempAddrForLinkLocal(t *testing.T) {
+ const nicID = 1
+
+ savedMaxDesyncFactor := stack.MaxDesyncFactor
+ defer func() {
+ stack.MaxDesyncFactor = savedMaxDesyncFactor
+ }()
+ stack.MaxDesyncFactor = time.Nanosecond
+
+ tests := []struct {
+ name string
+ dupAddrTransmits uint8
+ retransmitTimer time.Duration
+ }{
+ {
+ name: "DAD disabled",
+ },
+ {
+ name: "DAD enabled",
+ dupAddrTransmits: 1,
+ retransmitTimer: time.Second,
+ },
+ }
+
+ // This Run will not return until the parallel tests finish.
+ //
+ // We need this because we need to do some teardown work after the
+ // parallel tests complete.
+ //
+ // See https://godoc.org/testing#hdr-Subtests_and_Sub_benchmarks for
+ // more details.
+ t.Run("group", func(t *testing.T) {
+ for _, test := range tests {
+ test := test
+
+ t.Run(test.name, func(t *testing.T) {
+ t.Parallel()
+
+ ndpDisp := ndpDispatcher{
+ dadC: make(chan ndpDADEvent, 1),
+ autoGenAddrC: make(chan ndpAutoGenAddrEvent, 1),
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ AutoGenTempGlobalAddresses: true,
+ },
+ NDPDisp: &ndpDisp,
+ AutoGenIPv6LinkLocal: true,
+ })
+
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+
+ // The stable link-local address should auto-generate and resolve DAD.
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, tcpip.AddressWithPrefix{Address: llAddr1, PrefixLen: header.IIDOffsetInIPv6Address * 8}, newAddr); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected addr auto gen event")
+ }
+ select {
+ case e := <-ndpDisp.dadC:
+ if diff := checkDADEvent(e, nicID, llAddr1, true, nil); diff != "" {
+ t.Errorf("dad event mismatch (-want +got):\n%s", diff)
+ }
+ case <-time.After(time.Duration(test.dupAddrTransmits)*test.retransmitTimer + defaultAsyncPositiveEventTimeout):
+ t.Fatal("timed out waiting for DAD event")
+ }
+
+ // No new addresses should be generated.
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ t.Errorf("got unxpected auto gen addr event = %+v", e)
+ case <-time.After(defaultAsyncNegativeEventTimeout):
+ }
+ })
+ }
+ })
+}
+
+// TestNoAutoGenTempAddrWithoutStableAddr tests that a temporary SLAAC address
+// will not be generated until after DAD completes, even if a new Router
+// Advertisement is received to refresh lifetimes.
+func TestNoAutoGenTempAddrWithoutStableAddr(t *testing.T) {
+ const (
+ nicID = 1
+ dadTransmits = 1
+ retransmitTimer = 2 * time.Second
+ )
+
+ savedMaxDesyncFactor := stack.MaxDesyncFactor
+ defer func() {
+ stack.MaxDesyncFactor = savedMaxDesyncFactor
+ }()
+ stack.MaxDesyncFactor = 0
+
+ prefix, _, addr := prefixSubnetAddr(0, linkAddr1)
+ var tempIIDHistory [header.IIDSize]byte
+ header.InitialTempIID(tempIIDHistory[:], nil, nicID)
+ tempAddr := header.GenerateTempIPv6SLAACAddr(tempIIDHistory[:], addr.Address)
+
+ ndpDisp := ndpDispatcher{
+ dadC: make(chan ndpDADEvent, 1),
+ autoGenAddrC: make(chan ndpAutoGenAddrEvent, 1),
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ DupAddrDetectTransmits: dadTransmits,
+ RetransmitTimer: retransmitTimer,
+ HandleRAs: true,
+ AutoGenGlobalAddresses: true,
+ AutoGenTempGlobalAddresses: true,
+ },
+ NDPDisp: &ndpDisp,
+ })
+
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+
+ // Receive an RA to trigger SLAAC for prefix.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, 100, 100))
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, newAddr); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected addr auto gen event")
+ }
+
+ // DAD on the stable address for prefix has not yet completed. Receiving a new
+ // RA that would refresh lifetimes should not generate a temporary SLAAC
+ // address for the prefix.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, 100, 100))
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ t.Fatalf("unexpected auto gen addr event = %+v", e)
+ default:
+ }
+
+ // Wait for DAD to complete for the stable address then expect the temporary
+ // address to be generated.
+ select {
+ case e := <-ndpDisp.dadC:
+ if diff := checkDADEvent(e, nicID, addr.Address, true, nil); diff != "" {
+ t.Errorf("dad event mismatch (-want +got):\n%s", diff)
+ }
+ case <-time.After(dadTransmits*retransmitTimer + defaultAsyncPositiveEventTimeout):
+ t.Fatal("timed out waiting for DAD event")
+ }
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, tempAddr, newAddr); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ case <-time.After(defaultAsyncPositiveEventTimeout):
+ t.Fatal("timed out waiting for addr auto gen event")
+ }
+}
+
+// TestAutoGenTempAddrRegen tests that temporary SLAAC addresses are
+// regenerated.
+func TestAutoGenTempAddrRegen(t *testing.T) {
+ const (
+ nicID = 1
+ regenAfter = 2 * time.Second
+ newMinVL = 10
+ newMinVLDuration = newMinVL * time.Second
+ )
+
+ savedMaxDesyncFactor := stack.MaxDesyncFactor
+ savedMinMaxTempAddrPreferredLifetime := stack.MinMaxTempAddrPreferredLifetime
+ savedMinMaxTempAddrValidLifetime := stack.MinMaxTempAddrValidLifetime
+ defer func() {
+ stack.MaxDesyncFactor = savedMaxDesyncFactor
+ stack.MinMaxTempAddrPreferredLifetime = savedMinMaxTempAddrPreferredLifetime
+ stack.MinMaxTempAddrValidLifetime = savedMinMaxTempAddrValidLifetime
+ }()
+ stack.MaxDesyncFactor = 0
+ stack.MinMaxTempAddrPreferredLifetime = newMinVLDuration
+ stack.MinMaxTempAddrValidLifetime = newMinVLDuration
+
+ prefix, _, addr := prefixSubnetAddr(0, linkAddr1)
+ var tempIIDHistory [header.IIDSize]byte
+ header.InitialTempIID(tempIIDHistory[:], nil, nicID)
+ tempAddr1 := header.GenerateTempIPv6SLAACAddr(tempIIDHistory[:], addr.Address)
+ tempAddr2 := header.GenerateTempIPv6SLAACAddr(tempIIDHistory[:], addr.Address)
+ tempAddr3 := header.GenerateTempIPv6SLAACAddr(tempIIDHistory[:], addr.Address)
+
+ ndpDisp := ndpDispatcher{
+ autoGenAddrC: make(chan ndpAutoGenAddrEvent, 2),
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ ndpConfigs := stack.NDPConfigurations{
+ HandleRAs: true,
+ AutoGenGlobalAddresses: true,
+ AutoGenTempGlobalAddresses: true,
+ RegenAdvanceDuration: newMinVLDuration - regenAfter,
+ }
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: ndpConfigs,
+ NDPDisp: &ndpDisp,
+ })
+
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+
+ expectAutoGenAddrEvent := func(addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, eventType); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
}
- addr, err = s.GetMainNICAddress(1, header.IPv6ProtocolNumber)
- if err != nil {
- t.Fatalf("stack.GetMainNICAddress(1, _) err = %s", err)
+ default:
+ t.Fatal("expected addr auto gen event")
+ }
+ }
+
+ expectAutoGenAddrEventAsync := func(addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType, timeout time.Duration) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, eventType); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
}
- if addr.Address != addr1 {
- t.Fatalf("got stack.GetMainNICAddress(1, _) = %s, want = %s", addr, addr1)
+ case <-time.After(timeout):
+ t.Fatal("timed out waiting for addr auto gen event")
+ }
+ }
+
+ // Receive an RA with prefix1 in an NDP Prefix Information option (PI)
+ // with non-zero valid & preferred lifetimes.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, 100, 100))
+ expectAutoGenAddrEvent(addr, newAddr)
+ expectAutoGenAddrEvent(tempAddr1, newAddr)
+ if mismatch := addressCheck(s.NICInfo()[nicID].ProtocolAddresses, []tcpip.AddressWithPrefix{addr, tempAddr1}, nil); mismatch != "" {
+ t.Fatal(mismatch)
+ }
+
+ // Wait for regeneration
+ expectAutoGenAddrEventAsync(tempAddr2, newAddr, regenAfter+defaultAsyncPositiveEventTimeout)
+ if mismatch := addressCheck(s.NICInfo()[nicID].ProtocolAddresses, []tcpip.AddressWithPrefix{addr, tempAddr1, tempAddr2}, nil); mismatch != "" {
+ t.Fatal(mismatch)
+ }
+
+ // Wait for regeneration
+ expectAutoGenAddrEventAsync(tempAddr3, newAddr, regenAfter+defaultAsyncPositiveEventTimeout)
+ if mismatch := addressCheck(s.NICInfo()[nicID].ProtocolAddresses, []tcpip.AddressWithPrefix{addr, tempAddr1, tempAddr2, tempAddr3}, nil); mismatch != "" {
+ t.Fatal(mismatch)
+ }
+
+ // Stop generating temporary addresses
+ ndpConfigs.AutoGenTempGlobalAddresses = false
+ if err := s.SetNDPConfigurations(nicID, ndpConfigs); err != nil {
+ t.Fatalf("s.SetNDPConfigurations(%d, _): %s", nicID, err)
+ }
+
+ // Wait for all the temporary addresses to get invalidated.
+ tempAddrs := []tcpip.AddressWithPrefix{tempAddr1, tempAddr2, tempAddr3}
+ invalidateAfter := newMinVLDuration - 2*regenAfter
+ for _, addr := range tempAddrs {
+ // Wait for a deprecation then invalidation event, or just an invalidation
+ // event. We need to cover both cases but cannot deterministically hit both
+ // cases because the deprecation and invalidation jobs could execute in any
+ // order.
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, deprecatedAddr); diff == "" {
+ // If we get a deprecation event first, we should get an invalidation
+ // event almost immediately after.
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, invalidatedAddr); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ case <-time.After(defaultAsyncPositiveEventTimeout):
+ t.Fatal("timed out waiting for addr auto gen event")
+ }
+ } else if diff := checkAutoGenAddrEvent(e, addr, invalidatedAddr); diff == "" {
+ // If we get an invalidation event first, we shouldn't get a deprecation
+ // event after.
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ t.Fatalf("unexpectedly got an auto-generated event = %+v", e)
+ case <-time.After(defaultAsyncNegativeEventTimeout):
+ }
+ } else {
+ t.Fatalf("got unexpected auto-generated event = %+v", e)
+ }
+ case <-time.After(invalidateAfter + defaultAsyncPositiveEventTimeout):
+ t.Fatal("timed out waiting for addr auto gen event")
+ }
+
+ invalidateAfter = regenAfter
+ }
+ if mismatch := addressCheck(s.NICInfo()[1].ProtocolAddresses, []tcpip.AddressWithPrefix{addr}, tempAddrs); mismatch != "" {
+ t.Fatal(mismatch)
+ }
+}
+
+// TestAutoGenTempAddrRegenJobUpdates tests that a temporary address's
+// regeneration job gets updated when refreshing the address's lifetimes.
+func TestAutoGenTempAddrRegenJobUpdates(t *testing.T) {
+ const (
+ nicID = 1
+ regenAfter = 2 * time.Second
+ newMinVL = 10
+ newMinVLDuration = newMinVL * time.Second
+ )
+
+ savedMaxDesyncFactor := stack.MaxDesyncFactor
+ savedMinMaxTempAddrPreferredLifetime := stack.MinMaxTempAddrPreferredLifetime
+ savedMinMaxTempAddrValidLifetime := stack.MinMaxTempAddrValidLifetime
+ defer func() {
+ stack.MaxDesyncFactor = savedMaxDesyncFactor
+ stack.MinMaxTempAddrPreferredLifetime = savedMinMaxTempAddrPreferredLifetime
+ stack.MinMaxTempAddrValidLifetime = savedMinMaxTempAddrValidLifetime
+ }()
+ stack.MaxDesyncFactor = 0
+ stack.MinMaxTempAddrPreferredLifetime = newMinVLDuration
+ stack.MinMaxTempAddrValidLifetime = newMinVLDuration
+
+ prefix, _, addr := prefixSubnetAddr(0, linkAddr1)
+ var tempIIDHistory [header.IIDSize]byte
+ header.InitialTempIID(tempIIDHistory[:], nil, nicID)
+ tempAddr1 := header.GenerateTempIPv6SLAACAddr(tempIIDHistory[:], addr.Address)
+ tempAddr2 := header.GenerateTempIPv6SLAACAddr(tempIIDHistory[:], addr.Address)
+ tempAddr3 := header.GenerateTempIPv6SLAACAddr(tempIIDHistory[:], addr.Address)
+
+ ndpDisp := ndpDispatcher{
+ autoGenAddrC: make(chan ndpAutoGenAddrEvent, 2),
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ ndpConfigs := stack.NDPConfigurations{
+ HandleRAs: true,
+ AutoGenGlobalAddresses: true,
+ AutoGenTempGlobalAddresses: true,
+ RegenAdvanceDuration: newMinVLDuration - regenAfter,
+ }
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: ndpConfigs,
+ NDPDisp: &ndpDisp,
+ })
+
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+
+ expectAutoGenAddrEvent := func(addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, eventType); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected addr auto gen event")
+ }
+ }
+
+ expectAutoGenAddrEventAsync := func(addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType, timeout time.Duration) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, eventType); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ case <-time.After(timeout):
+ t.Fatal("timed out waiting for addr auto gen event")
+ }
+ }
+
+ // Receive an RA with prefix1 in an NDP Prefix Information option (PI)
+ // with non-zero valid & preferred lifetimes.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, 100, 100))
+ expectAutoGenAddrEvent(addr, newAddr)
+ expectAutoGenAddrEvent(tempAddr1, newAddr)
+ if mismatch := addressCheck(s.NICInfo()[nicID].ProtocolAddresses, []tcpip.AddressWithPrefix{addr, tempAddr1}, nil); mismatch != "" {
+ t.Fatal(mismatch)
+ }
+
+ // Deprecate the prefix.
+ //
+ // A new temporary address should be generated after the regeneration
+ // time has passed since the prefix is deprecated.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, 100, 0))
+ expectAutoGenAddrEvent(addr, deprecatedAddr)
+ expectAutoGenAddrEvent(tempAddr1, deprecatedAddr)
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ t.Fatalf("unexpected auto gen addr event = %+v", e)
+ case <-time.After(regenAfter + defaultAsyncNegativeEventTimeout):
+ }
+
+ // Prefer the prefix again.
+ //
+ // A new temporary address should immediately be generated since the
+ // regeneration time has already passed since the last address was generated
+ // - this regeneration does not depend on a job.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, 100, 100))
+ expectAutoGenAddrEvent(tempAddr2, newAddr)
+
+ // Increase the maximum lifetimes for temporary addresses to large values
+ // then refresh the lifetimes of the prefix.
+ //
+ // A new address should not be generated after the regeneration time that was
+ // expected for the previous check. This is because the preferred lifetime for
+ // the temporary addresses has increased, so it will take more time to
+ // regenerate a new temporary address. Note, new addresses are only
+ // regenerated after the preferred lifetime - the regenerate advance duration
+ // as paased.
+ ndpConfigs.MaxTempAddrValidLifetime = 100 * time.Second
+ ndpConfigs.MaxTempAddrPreferredLifetime = 100 * time.Second
+ if err := s.SetNDPConfigurations(nicID, ndpConfigs); err != nil {
+ t.Fatalf("s.SetNDPConfigurations(%d, _): %s", nicID, err)
+ }
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, 100, 100))
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ t.Fatalf("unexpected auto gen addr event = %+v", e)
+ case <-time.After(regenAfter + defaultAsyncNegativeEventTimeout):
+ }
+
+ // Set the maximum lifetimes for temporary addresses such that on the next
+ // RA, the regeneration job gets scheduled again.
+ //
+ // The maximum lifetime is the sum of the minimum lifetimes for temporary
+ // addresses + the time that has already passed since the last address was
+ // generated so that the regeneration job is needed to generate the next
+ // address.
+ newLifetimes := newMinVLDuration + regenAfter + defaultAsyncNegativeEventTimeout
+ ndpConfigs.MaxTempAddrValidLifetime = newLifetimes
+ ndpConfigs.MaxTempAddrPreferredLifetime = newLifetimes
+ if err := s.SetNDPConfigurations(nicID, ndpConfigs); err != nil {
+ t.Fatalf("s.SetNDPConfigurations(%d, _): %s", nicID, err)
+ }
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, 100, 100))
+ expectAutoGenAddrEventAsync(tempAddr3, newAddr, regenAfter+defaultAsyncPositiveEventTimeout)
+}
+
+// TestMixedSLAACAddrConflictRegen tests SLAAC address regeneration in response
+// to a mix of DAD conflicts and NIC-local conflicts.
+func TestMixedSLAACAddrConflictRegen(t *testing.T) {
+ const (
+ nicID = 1
+ nicName = "nic"
+ lifetimeSeconds = 9999
+ // From stack.maxSLAACAddrLocalRegenAttempts
+ maxSLAACAddrLocalRegenAttempts = 10
+ // We use 2 more addreses than the maximum local regeneration attempts
+ // because we want to also trigger regeneration in response to a DAD
+ // conflicts for this test.
+ maxAddrs = maxSLAACAddrLocalRegenAttempts + 2
+ dupAddrTransmits = 1
+ retransmitTimer = time.Second
+ )
+
+ var tempIIDHistoryWithModifiedEUI64 [header.IIDSize]byte
+ header.InitialTempIID(tempIIDHistoryWithModifiedEUI64[:], nil, nicID)
+
+ var tempIIDHistoryWithOpaqueIID [header.IIDSize]byte
+ header.InitialTempIID(tempIIDHistoryWithOpaqueIID[:], nil, nicID)
+
+ prefix, subnet, stableAddrWithModifiedEUI64 := prefixSubnetAddr(0, linkAddr1)
+ var stableAddrsWithOpaqueIID [maxAddrs]tcpip.AddressWithPrefix
+ var tempAddrsWithOpaqueIID [maxAddrs]tcpip.AddressWithPrefix
+ var tempAddrsWithModifiedEUI64 [maxAddrs]tcpip.AddressWithPrefix
+ addrBytes := []byte(subnet.ID())
+ for i := 0; i < maxAddrs; i++ {
+ stableAddrsWithOpaqueIID[i] = tcpip.AddressWithPrefix{
+ Address: tcpip.Address(header.AppendOpaqueInterfaceIdentifier(addrBytes[:header.IIDOffsetInIPv6Address], subnet, nicName, uint8(i), nil)),
+ PrefixLen: header.IIDOffsetInIPv6Address * 8,
+ }
+ // When generating temporary addresses, the resolved stable address for the
+ // SLAAC prefix will be the first address stable address generated for the
+ // prefix as we will not simulate address conflicts for the stable addresses
+ // in tests involving temporary addresses. Address conflicts for stable
+ // addresses will be done in their own tests.
+ tempAddrsWithOpaqueIID[i] = header.GenerateTempIPv6SLAACAddr(tempIIDHistoryWithOpaqueIID[:], stableAddrsWithOpaqueIID[0].Address)
+ tempAddrsWithModifiedEUI64[i] = header.GenerateTempIPv6SLAACAddr(tempIIDHistoryWithModifiedEUI64[:], stableAddrWithModifiedEUI64.Address)
+ }
+
+ tests := []struct {
+ name string
+ addrs []tcpip.AddressWithPrefix
+ tempAddrs bool
+ initialExpect tcpip.AddressWithPrefix
+ nicNameFromID func(tcpip.NICID, string) string
+ }{
+ {
+ name: "Stable addresses with opaque IIDs",
+ addrs: stableAddrsWithOpaqueIID[:],
+ nicNameFromID: func(tcpip.NICID, string) string {
+ return nicName
+ },
+ },
+ {
+ name: "Temporary addresses with opaque IIDs",
+ addrs: tempAddrsWithOpaqueIID[:],
+ tempAddrs: true,
+ initialExpect: stableAddrsWithOpaqueIID[0],
+ nicNameFromID: func(tcpip.NICID, string) string {
+ return nicName
+ },
+ },
+ {
+ name: "Temporary addresses with modified EUI64",
+ addrs: tempAddrsWithModifiedEUI64[:],
+ tempAddrs: true,
+ initialExpect: stableAddrWithModifiedEUI64,
+ },
+ }
+
+ for _, test := range tests {
+ test := test
+
+ t.Run(test.name, func(t *testing.T) {
+ t.Parallel()
+
+ ndpDisp := ndpDispatcher{
+ autoGenAddrC: make(chan ndpAutoGenAddrEvent, 2),
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ ndpConfigs := stack.NDPConfigurations{
+ HandleRAs: true,
+ AutoGenGlobalAddresses: true,
+ AutoGenTempGlobalAddresses: test.tempAddrs,
+ AutoGenAddressConflictRetries: 1,
+ }
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()},
+ NDPConfigs: ndpConfigs,
+ NDPDisp: &ndpDisp,
+ OpaqueIIDOpts: stack.OpaqueInterfaceIdentifierOptions{
+ NICNameFromID: test.nicNameFromID,
+ },
+ })
+
+ s.SetRouteTable([]tcpip.Route{{
+ Destination: header.IPv6EmptySubnet,
+ Gateway: llAddr2,
+ NIC: nicID,
+ }})
+
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+
+ for j := 0; j < len(test.addrs)-1; j++ {
+ // The NIC will not attempt to generate an address in response to a
+ // NIC-local conflict after some maximum number of attempts. We skip
+ // creating a conflict for the address that would be generated as part
+ // of the last attempt so we can simulate a DAD conflict for this
+ // address and restart the NIC-local generation process.
+ if j == maxSLAACAddrLocalRegenAttempts-1 {
+ continue
+ }
+
+ if err := s.AddAddress(nicID, ipv6.ProtocolNumber, test.addrs[j].Address); err != nil {
+ t.Fatalf("s.AddAddress(%d, %d, %s): %s", nicID, ipv6.ProtocolNumber, test.addrs[j].Address, err)
+ }
+ }
+
+ expectAutoGenAddrEvent := func(addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, eventType); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected addr auto gen event")
+ }
+ }
+
+ expectAutoGenAddrAsyncEvent := func(addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, eventType); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ case <-time.After(defaultAsyncPositiveEventTimeout):
+ t.Fatal("timed out waiting for addr auto gen event")
+ }
+ }
+
+ expectDADEventAsync := func(addr tcpip.Address) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.dadC:
+ if diff := checkDADEvent(e, nicID, addr, true, nil); diff != "" {
+ t.Errorf("dad event mismatch (-want +got):\n%s", diff)
+ }
+ case <-time.After(dupAddrTransmits*retransmitTimer + defaultAsyncPositiveEventTimeout):
+ t.Fatal("timed out waiting for DAD event")
+ }
+ }
+
+ // Enable DAD.
+ ndpDisp.dadC = make(chan ndpDADEvent, 2)
+ ndpConfigs.DupAddrDetectTransmits = dupAddrTransmits
+ ndpConfigs.RetransmitTimer = retransmitTimer
+ if err := s.SetNDPConfigurations(nicID, ndpConfigs); err != nil {
+ t.Fatalf("s.SetNDPConfigurations(%d, _): %s", nicID, err)
+ }
+
+ // Do SLAAC for prefix.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, lifetimeSeconds, lifetimeSeconds))
+ if test.initialExpect != (tcpip.AddressWithPrefix{}) {
+ expectAutoGenAddrEvent(test.initialExpect, newAddr)
+ expectDADEventAsync(test.initialExpect.Address)
+ }
+
+ // The last local generation attempt should succeed, but we introduce a
+ // DAD failure to restart the local generation process.
+ addr := test.addrs[maxSLAACAddrLocalRegenAttempts-1]
+ expectAutoGenAddrAsyncEvent(addr, newAddr)
+ if err := s.DupTentativeAddrDetected(nicID, addr.Address); err != nil {
+ t.Fatalf("s.DupTentativeAddrDetected(%d, %s): %s", nicID, addr.Address, err)
+ }
+ select {
+ case e := <-ndpDisp.dadC:
+ if diff := checkDADEvent(e, nicID, addr.Address, false, nil); diff != "" {
+ t.Errorf("dad event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected DAD event")
+ }
+ expectAutoGenAddrEvent(addr, invalidatedAddr)
+
+ // The last address generated should resolve DAD.
+ addr = test.addrs[len(test.addrs)-1]
+ expectAutoGenAddrAsyncEvent(addr, newAddr)
+ expectDADEventAsync(addr.Address)
+
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ t.Fatalf("unexpected auto gen addr event = %+v", e)
+ default:
+ }
+ })
+ }
+}
+
+// stackAndNdpDispatcherWithDefaultRoute returns an ndpDispatcher,
+// channel.Endpoint and stack.Stack.
+//
+// stack.Stack will have a default route through the router (llAddr3) installed
+// and a static link-address (linkAddr3) added to the link address cache for the
+// router.
+func stackAndNdpDispatcherWithDefaultRoute(t *testing.T, nicID tcpip.NICID) (*ndpDispatcher, *channel.Endpoint, *stack.Stack) {
+ t.Helper()
+ ndpDisp := &ndpDispatcher{
+ autoGenAddrC: make(chan ndpAutoGenAddrEvent, 1),
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ HandleRAs: true,
+ AutoGenGlobalAddresses: true,
+ },
+ NDPDisp: ndpDisp,
+ })
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+ s.SetRouteTable([]tcpip.Route{{
+ Destination: header.IPv6EmptySubnet,
+ Gateway: llAddr3,
+ NIC: nicID,
+ }})
+ s.AddLinkAddress(nicID, llAddr3, linkAddr3)
+ return ndpDisp, e, s
+}
+
+// addrForNewConnectionTo returns the local address used when creating a new
+// connection to addr.
+func addrForNewConnectionTo(t *testing.T, s *stack.Stack, addr tcpip.FullAddress) tcpip.Address {
+ t.Helper()
+
+ wq := waiter.Queue{}
+ we, ch := waiter.NewChannelEntry(nil)
+ wq.EventRegister(&we, waiter.EventIn)
+ defer wq.EventUnregister(&we)
+ defer close(ch)
+ ep, err := s.NewEndpoint(header.UDPProtocolNumber, header.IPv6ProtocolNumber, &wq)
+ if err != nil {
+ t.Fatalf("s.NewEndpoint(%d, %d, _): %s", header.UDPProtocolNumber, header.IPv6ProtocolNumber, err)
+ }
+ defer ep.Close()
+ if err := ep.SetSockOptBool(tcpip.V6OnlyOption, true); err != nil {
+ t.Fatalf("SetSockOpt(tcpip.V6OnlyOption, true): %s", err)
+ }
+ if err := ep.Connect(addr); err != nil {
+ t.Fatalf("ep.Connect(%+v): %s", addr, err)
+ }
+ got, err := ep.GetLocalAddress()
+ if err != nil {
+ t.Fatalf("ep.GetLocalAddress(): %s", err)
+ }
+ return got.Addr
+}
+
+// addrForNewConnection returns the local address used when creating a new
+// connection.
+func addrForNewConnection(t *testing.T, s *stack.Stack) tcpip.Address {
+ t.Helper()
+
+ return addrForNewConnectionTo(t, s, dstAddr)
+}
+
+// addrForNewConnectionWithAddr returns the local address used when creating a
+// new connection with a specific local address.
+func addrForNewConnectionWithAddr(t *testing.T, s *stack.Stack, addr tcpip.FullAddress) tcpip.Address {
+ t.Helper()
+
+ wq := waiter.Queue{}
+ we, ch := waiter.NewChannelEntry(nil)
+ wq.EventRegister(&we, waiter.EventIn)
+ defer wq.EventUnregister(&we)
+ defer close(ch)
+ ep, err := s.NewEndpoint(header.UDPProtocolNumber, header.IPv6ProtocolNumber, &wq)
+ if err != nil {
+ t.Fatalf("s.NewEndpoint(%d, %d, _): %s", header.UDPProtocolNumber, header.IPv6ProtocolNumber, err)
+ }
+ defer ep.Close()
+ if err := ep.SetSockOptBool(tcpip.V6OnlyOption, true); err != nil {
+ t.Fatalf("SetSockOpt(tcpip.V6OnlyOption, true): %s", err)
+ }
+ if err := ep.Bind(addr); err != nil {
+ t.Fatalf("ep.Bind(%+v): %s", addr, err)
+ }
+ if err := ep.Connect(dstAddr); err != nil {
+ t.Fatalf("ep.Connect(%+v): %s", dstAddr, err)
+ }
+ got, err := ep.GetLocalAddress()
+ if err != nil {
+ t.Fatalf("ep.GetLocalAddress(): %s", err)
+ }
+ return got.Addr
+}
+
+// TestAutoGenAddrDeprecateFromPI tests deprecating a SLAAC address when
+// receiving a PI with 0 preferred lifetime.
+func TestAutoGenAddrDeprecateFromPI(t *testing.T) {
+ const nicID = 1
+
+ prefix1, _, addr1 := prefixSubnetAddr(0, linkAddr1)
+ prefix2, _, addr2 := prefixSubnetAddr(1, linkAddr1)
+
+ ndpDisp, e, s := stackAndNdpDispatcherWithDefaultRoute(t, nicID)
+
+ expectAutoGenAddrEvent := func(addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, eventType); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected addr auto gen event")
+ }
+ }
+
+ expectPrimaryAddr := func(addr tcpip.AddressWithPrefix) {
+ t.Helper()
+
+ if got, err := s.GetMainNICAddress(nicID, header.IPv6ProtocolNumber); err != nil {
+ t.Fatalf("s.GetMainNICAddress(%d, %d): %s", nicID, header.IPv6ProtocolNumber, err)
+ } else if got != addr {
+ t.Errorf("got s.GetMainNICAddress(%d, %d) = %s, want = %s", nicID, header.IPv6ProtocolNumber, got, addr)
+ }
+
+ if got := addrForNewConnection(t, s); got != addr.Address {
+ t.Errorf("got addrForNewConnection = %s, want = %s", got, addr.Address)
+ }
+ }
+
+ // Receive PI for prefix1.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, 100, 100))
+ expectAutoGenAddrEvent(addr1, newAddr)
+ if !containsV6Addr(s.NICInfo()[nicID].ProtocolAddresses, addr1) {
+ t.Fatalf("should have %s in the list of addresses", addr1)
+ }
+ expectPrimaryAddr(addr1)
+
+ // Deprecate addr for prefix1 immedaitely.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, 100, 0))
+ expectAutoGenAddrEvent(addr1, deprecatedAddr)
+ if !containsV6Addr(s.NICInfo()[nicID].ProtocolAddresses, addr1) {
+ t.Fatalf("should have %s in the list of addresses", addr1)
+ }
+ // addr should still be the primary endpoint as there are no other addresses.
+ expectPrimaryAddr(addr1)
+
+ // Refresh lifetimes of addr generated from prefix1.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, 100, 100))
+ select {
+ case <-ndpDisp.autoGenAddrC:
+ t.Fatal("unexpectedly got an auto-generated event")
+ default:
+ }
+ expectPrimaryAddr(addr1)
+
+ // Receive PI for prefix2.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix2, true, true, 100, 100))
+ expectAutoGenAddrEvent(addr2, newAddr)
+ if !containsV6Addr(s.NICInfo()[nicID].ProtocolAddresses, addr2) {
+ t.Fatalf("should have %s in the list of addresses", addr2)
+ }
+ expectPrimaryAddr(addr2)
+
+ // Deprecate addr for prefix2 immedaitely.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix2, true, true, 100, 0))
+ expectAutoGenAddrEvent(addr2, deprecatedAddr)
+ if !containsV6Addr(s.NICInfo()[nicID].ProtocolAddresses, addr2) {
+ t.Fatalf("should have %s in the list of addresses", addr2)
+ }
+ // addr1 should be the primary endpoint now since addr2 is deprecated but
+ // addr1 is not.
+ expectPrimaryAddr(addr1)
+ // addr2 is deprecated but if explicitly requested, it should be used.
+ fullAddr2 := tcpip.FullAddress{Addr: addr2.Address, NIC: nicID}
+ if got := addrForNewConnectionWithAddr(t, s, fullAddr2); got != addr2.Address {
+ t.Errorf("got addrForNewConnectionWithAddr(_, _, %+v) = %s, want = %s", fullAddr2, got, addr2.Address)
+ }
+
+ // Another PI w/ 0 preferred lifetime should not result in a deprecation
+ // event.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix2, true, true, 100, 0))
+ select {
+ case <-ndpDisp.autoGenAddrC:
+ t.Fatal("unexpectedly got an auto-generated event")
+ default:
+ }
+ expectPrimaryAddr(addr1)
+ if got := addrForNewConnectionWithAddr(t, s, fullAddr2); got != addr2.Address {
+ t.Errorf("got addrForNewConnectionWithAddr(_, _, %+v) = %s, want = %s", fullAddr2, got, addr2.Address)
+ }
+
+ // Refresh lifetimes of addr generated from prefix2.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix2, true, true, 100, 100))
+ select {
+ case <-ndpDisp.autoGenAddrC:
+ t.Fatal("unexpectedly got an auto-generated event")
+ default:
+ }
+ expectPrimaryAddr(addr2)
+}
+
+// TestAutoGenAddrJobDeprecation tests that an address is properly deprecated
+// when its preferred lifetime expires.
+func TestAutoGenAddrJobDeprecation(t *testing.T) {
+ const nicID = 1
+ const newMinVL = 2
+ newMinVLDuration := newMinVL * time.Second
+ saved := stack.MinPrefixInformationValidLifetimeForUpdate
+ defer func() {
+ stack.MinPrefixInformationValidLifetimeForUpdate = saved
+ }()
+ stack.MinPrefixInformationValidLifetimeForUpdate = newMinVLDuration
+
+ prefix1, _, addr1 := prefixSubnetAddr(0, linkAddr1)
+ prefix2, _, addr2 := prefixSubnetAddr(1, linkAddr1)
+
+ ndpDisp, e, s := stackAndNdpDispatcherWithDefaultRoute(t, nicID)
+
+ expectAutoGenAddrEvent := func(addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, eventType); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected addr auto gen event")
+ }
+ }
+
+ expectAutoGenAddrEventAfter := func(addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType, timeout time.Duration) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, eventType); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ case <-time.After(timeout):
+ t.Fatal("timed out waiting for addr auto gen event")
+ }
+ }
+
+ expectPrimaryAddr := func(addr tcpip.AddressWithPrefix) {
+ t.Helper()
+
+ if got, err := s.GetMainNICAddress(nicID, header.IPv6ProtocolNumber); err != nil {
+ t.Fatalf("s.GetMainNICAddress(%d, %d): %s", nicID, header.IPv6ProtocolNumber, err)
+ } else if got != addr {
+ t.Errorf("got s.GetMainNICAddress(%d, %d) = %s, want = %s", nicID, header.IPv6ProtocolNumber, got, addr)
+ }
+
+ if got := addrForNewConnection(t, s); got != addr.Address {
+ t.Errorf("got addrForNewConnection = %s, want = %s", got, addr.Address)
+ }
+ }
+
+ // Receive PI for prefix2.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix2, true, true, 100, 100))
+ expectAutoGenAddrEvent(addr2, newAddr)
+ if !containsV6Addr(s.NICInfo()[nicID].ProtocolAddresses, addr2) {
+ t.Fatalf("should have %s in the list of addresses", addr2)
+ }
+ expectPrimaryAddr(addr2)
+
+ // Receive a PI for prefix1.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, 100, 90))
+ expectAutoGenAddrEvent(addr1, newAddr)
+ if !containsV6Addr(s.NICInfo()[nicID].ProtocolAddresses, addr1) {
+ t.Fatalf("should have %s in the list of addresses", addr1)
+ }
+ if !containsV6Addr(s.NICInfo()[nicID].ProtocolAddresses, addr2) {
+ t.Fatalf("should have %s in the list of addresses", addr2)
+ }
+ expectPrimaryAddr(addr1)
+
+ // Refresh lifetime for addr of prefix1.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, newMinVL, newMinVL-1))
+ select {
+ case <-ndpDisp.autoGenAddrC:
+ t.Fatal("unexpectedly got an auto-generated event")
+ default:
+ }
+ expectPrimaryAddr(addr1)
+
+ // Wait for addr of prefix1 to be deprecated.
+ expectAutoGenAddrEventAfter(addr1, deprecatedAddr, newMinVLDuration-time.Second+defaultAsyncPositiveEventTimeout)
+ if !containsV6Addr(s.NICInfo()[nicID].ProtocolAddresses, addr1) {
+ t.Fatalf("should not have %s in the list of addresses", addr1)
+ }
+ if !containsV6Addr(s.NICInfo()[nicID].ProtocolAddresses, addr2) {
+ t.Fatalf("should have %s in the list of addresses", addr2)
+ }
+ // addr2 should be the primary endpoint now since addr1 is deprecated but
+ // addr2 is not.
+ expectPrimaryAddr(addr2)
+ // addr1 is deprecated but if explicitly requested, it should be used.
+ fullAddr1 := tcpip.FullAddress{Addr: addr1.Address, NIC: nicID}
+ if got := addrForNewConnectionWithAddr(t, s, fullAddr1); got != addr1.Address {
+ t.Errorf("got addrForNewConnectionWithAddr(_, _, %+v) = %s, want = %s", fullAddr1, got, addr1.Address)
+ }
+
+ // Refresh valid lifetime for addr of prefix1, w/ 0 preferred lifetime to make
+ // sure we do not get a deprecation event again.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, newMinVL, 0))
+ select {
+ case <-ndpDisp.autoGenAddrC:
+ t.Fatal("unexpectedly got an auto-generated event")
+ default:
+ }
+ expectPrimaryAddr(addr2)
+ if got := addrForNewConnectionWithAddr(t, s, fullAddr1); got != addr1.Address {
+ t.Errorf("got addrForNewConnectionWithAddr(_, _, %+v) = %s, want = %s", fullAddr1, got, addr1.Address)
+ }
+
+ // Refresh lifetimes for addr of prefix1.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, newMinVL, newMinVL-1))
+ select {
+ case <-ndpDisp.autoGenAddrC:
+ t.Fatal("unexpectedly got an auto-generated event")
+ default:
+ }
+ // addr1 is the primary endpoint again since it is non-deprecated now.
+ expectPrimaryAddr(addr1)
+
+ // Wait for addr of prefix1 to be deprecated.
+ expectAutoGenAddrEventAfter(addr1, deprecatedAddr, newMinVLDuration-time.Second+defaultAsyncPositiveEventTimeout)
+ if !containsV6Addr(s.NICInfo()[nicID].ProtocolAddresses, addr1) {
+ t.Fatalf("should not have %s in the list of addresses", addr1)
+ }
+ if !containsV6Addr(s.NICInfo()[nicID].ProtocolAddresses, addr2) {
+ t.Fatalf("should have %s in the list of addresses", addr2)
+ }
+ // addr2 should be the primary endpoint now since it is not deprecated.
+ expectPrimaryAddr(addr2)
+ if got := addrForNewConnectionWithAddr(t, s, fullAddr1); got != addr1.Address {
+ t.Errorf("got addrForNewConnectionWithAddr(_, _, %+v) = %s, want = %s", fullAddr1, got, addr1.Address)
+ }
+
+ // Wait for addr of prefix1 to be invalidated.
+ expectAutoGenAddrEventAfter(addr1, invalidatedAddr, time.Second+defaultAsyncPositiveEventTimeout)
+ if containsV6Addr(s.NICInfo()[nicID].ProtocolAddresses, addr1) {
+ t.Fatalf("should not have %s in the list of addresses", addr1)
+ }
+ if !containsV6Addr(s.NICInfo()[nicID].ProtocolAddresses, addr2) {
+ t.Fatalf("should have %s in the list of addresses", addr2)
+ }
+ expectPrimaryAddr(addr2)
+
+ // Refresh both lifetimes for addr of prefix2 to the same value.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix2, true, true, newMinVL, newMinVL))
+ select {
+ case <-ndpDisp.autoGenAddrC:
+ t.Fatal("unexpectedly got an auto-generated event")
+ default:
+ }
+
+ // Wait for a deprecation then invalidation events, or just an invalidation
+ // event. We need to cover both cases but cannot deterministically hit both
+ // cases because the deprecation and invalidation handlers could be handled in
+ // either deprecation then invalidation, or invalidation then deprecation
+ // (which should be cancelled by the invalidation handler).
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr2, deprecatedAddr); diff == "" {
+ // If we get a deprecation event first, we should get an invalidation
+ // event almost immediately after.
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr2, invalidatedAddr); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ case <-time.After(defaultAsyncPositiveEventTimeout):
+ t.Fatal("timed out waiting for addr auto gen event")
+ }
+ } else if diff := checkAutoGenAddrEvent(e, addr2, invalidatedAddr); diff == "" {
+ // If we get an invalidation event first, we should not get a deprecation
+ // event after.
+ select {
+ case <-ndpDisp.autoGenAddrC:
+ t.Fatal("unexpectedly got an auto-generated event")
+ case <-time.After(defaultAsyncNegativeEventTimeout):
+ }
+ } else {
+ t.Fatalf("got unexpected auto-generated event")
+ }
+ case <-time.After(newMinVLDuration + defaultAsyncPositiveEventTimeout):
+ t.Fatal("timed out waiting for addr auto gen event")
+ }
+ if containsV6Addr(s.NICInfo()[nicID].ProtocolAddresses, addr1) {
+ t.Fatalf("should not have %s in the list of addresses", addr1)
+ }
+ if containsV6Addr(s.NICInfo()[nicID].ProtocolAddresses, addr2) {
+ t.Fatalf("should not have %s in the list of addresses", addr2)
+ }
+ // Should not have any primary endpoints.
+ if got, err := s.GetMainNICAddress(nicID, header.IPv6ProtocolNumber); err != nil {
+ t.Fatalf("s.GetMainNICAddress(%d, %d): %s", nicID, header.IPv6ProtocolNumber, err)
+ } else if want := (tcpip.AddressWithPrefix{}); got != want {
+ t.Errorf("got s.GetMainNICAddress(%d, %d) = %s, want = %s", nicID, header.IPv6ProtocolNumber, got, want)
+ }
+ wq := waiter.Queue{}
+ we, ch := waiter.NewChannelEntry(nil)
+ wq.EventRegister(&we, waiter.EventIn)
+ defer wq.EventUnregister(&we)
+ defer close(ch)
+ ep, err := s.NewEndpoint(header.UDPProtocolNumber, header.IPv6ProtocolNumber, &wq)
+ if err != nil {
+ t.Fatalf("s.NewEndpoint(%d, %d, _): %s", header.UDPProtocolNumber, header.IPv6ProtocolNumber, err)
+ }
+ defer ep.Close()
+ if err := ep.SetSockOptBool(tcpip.V6OnlyOption, true); err != nil {
+ t.Fatalf("SetSockOpt(tcpip.V6OnlyOption, true): %s", err)
+ }
+
+ if err := ep.Connect(dstAddr); err != tcpip.ErrNoRoute {
+ t.Errorf("got ep.Connect(%+v) = %v, want = %s", dstAddr, err, tcpip.ErrNoRoute)
+ }
+}
+
+// Tests transitioning a SLAAC address's valid lifetime between finite and
+// infinite values.
+func TestAutoGenAddrFiniteToInfiniteToFiniteVL(t *testing.T) {
+ const infiniteVLSeconds = 2
+ const minVLSeconds = 1
+ savedIL := header.NDPInfiniteLifetime
+ savedMinVL := stack.MinPrefixInformationValidLifetimeForUpdate
+ defer func() {
+ stack.MinPrefixInformationValidLifetimeForUpdate = savedMinVL
+ header.NDPInfiniteLifetime = savedIL
+ }()
+ stack.MinPrefixInformationValidLifetimeForUpdate = minVLSeconds * time.Second
+ header.NDPInfiniteLifetime = infiniteVLSeconds * time.Second
+
+ prefix, _, addr := prefixSubnetAddr(0, linkAddr1)
+
+ tests := []struct {
+ name string
+ infiniteVL uint32
+ }{
+ {
+ name: "EqualToInfiniteVL",
+ infiniteVL: infiniteVLSeconds,
+ },
+ // Our implementation supports changing header.NDPInfiniteLifetime for tests
+ // such that a packet can be received where the lifetime field has a value
+ // greater than header.NDPInfiniteLifetime. Because of this, we test to make
+ // sure that receiving a value greater than header.NDPInfiniteLifetime is
+ // handled the same as when receiving a value equal to
+ // header.NDPInfiniteLifetime.
+ {
+ name: "MoreThanInfiniteVL",
+ infiniteVL: infiniteVLSeconds + 1,
+ },
+ }
+
+ // This Run will not return until the parallel tests finish.
+ //
+ // We need this because we need to do some teardown work after the
+ // parallel tests complete.
+ //
+ // See https://godoc.org/testing#hdr-Subtests_and_Sub_benchmarks for
+ // more details.
+ t.Run("group", func(t *testing.T) {
+ for _, test := range tests {
+ test := test
+
+ t.Run(test.name, func(t *testing.T) {
+ t.Parallel()
+
+ ndpDisp := ndpDispatcher{
+ autoGenAddrC: make(chan ndpAutoGenAddrEvent, 1),
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ HandleRAs: true,
+ AutoGenGlobalAddresses: true,
+ },
+ NDPDisp: &ndpDisp,
+ })
+
+ if err := s.CreateNIC(1, e); err != nil {
+ t.Fatalf("CreateNIC(1) = %s", err)
+ }
+
+ // Receive an RA with finite prefix.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, minVLSeconds, 0))
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, newAddr); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+
+ default:
+ t.Fatal("expected addr auto gen event")
+ }
+
+ // Receive an new RA with prefix with infinite VL.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, test.infiniteVL, 0))
+
+ // Receive a new RA with prefix with finite VL.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, minVLSeconds, 0))
+
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, invalidatedAddr); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+
+ case <-time.After(minVLSeconds*time.Second + defaultAsyncPositiveEventTimeout):
+ t.Fatal("timeout waiting for addr auto gen event")
+ }
+ })
+ }
+ })
+}
+
+// TestAutoGenAddrValidLifetimeUpdates tests that the valid lifetime of an
+// auto-generated address only gets updated when required to, as specified in
+// RFC 4862 section 5.5.3.e.
+func TestAutoGenAddrValidLifetimeUpdates(t *testing.T) {
+ const infiniteVL = 4294967295
+ const newMinVL = 4
+ saved := stack.MinPrefixInformationValidLifetimeForUpdate
+ defer func() {
+ stack.MinPrefixInformationValidLifetimeForUpdate = saved
+ }()
+ stack.MinPrefixInformationValidLifetimeForUpdate = newMinVL * time.Second
+
+ prefix, _, addr := prefixSubnetAddr(0, linkAddr1)
+
+ tests := []struct {
+ name string
+ ovl uint32
+ nvl uint32
+ evl uint32
+ }{
+ // Should update the VL to the minimum VL for updating if the
+ // new VL is less than newMinVL but was originally greater than
+ // it.
+ {
+ "LargeVLToVLLessThanMinVLForUpdate",
+ 9999,
+ 1,
+ newMinVL,
+ },
+ {
+ "LargeVLTo0",
+ 9999,
+ 0,
+ newMinVL,
+ },
+ {
+ "InfiniteVLToVLLessThanMinVLForUpdate",
+ infiniteVL,
+ 1,
+ newMinVL,
+ },
+ {
+ "InfiniteVLTo0",
+ infiniteVL,
+ 0,
+ newMinVL,
+ },
+
+ // Should not update VL if original VL was less than newMinVL
+ // and the new VL is also less than newMinVL.
+ {
+ "ShouldNotUpdateWhenBothOldAndNewAreLessThanMinVLForUpdate",
+ newMinVL - 1,
+ newMinVL - 3,
+ newMinVL - 1,
+ },
+
+ // Should take the new VL if the new VL is greater than the
+ // remaining time or is greater than newMinVL.
+ {
+ "MorethanMinVLToLesserButStillMoreThanMinVLForUpdate",
+ newMinVL + 5,
+ newMinVL + 3,
+ newMinVL + 3,
+ },
+ {
+ "SmallVLToGreaterVLButStillLessThanMinVLForUpdate",
+ newMinVL - 3,
+ newMinVL - 1,
+ newMinVL - 1,
+ },
+ {
+ "SmallVLToGreaterVLThatIsMoreThaMinVLForUpdate",
+ newMinVL - 3,
+ newMinVL + 1,
+ newMinVL + 1,
+ },
+ }
+
+ // This Run will not return until the parallel tests finish.
+ //
+ // We need this because we need to do some teardown work after the
+ // parallel tests complete.
+ //
+ // See https://godoc.org/testing#hdr-Subtests_and_Sub_benchmarks for
+ // more details.
+ t.Run("group", func(t *testing.T) {
+ for _, test := range tests {
+ test := test
+
+ t.Run(test.name, func(t *testing.T) {
+ t.Parallel()
+
+ ndpDisp := ndpDispatcher{
+ autoGenAddrC: make(chan ndpAutoGenAddrEvent, 10),
+ }
+ e := channel.New(10, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ HandleRAs: true,
+ AutoGenGlobalAddresses: true,
+ },
+ NDPDisp: &ndpDisp,
+ })
+
+ if err := s.CreateNIC(1, e); err != nil {
+ t.Fatalf("CreateNIC(1) = %s", err)
+ }
+
+ // Receive an RA with prefix with initial VL,
+ // test.ovl.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, test.ovl, 0))
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, newAddr); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected addr auto gen event")
+ }
+
+ // Receive an new RA with prefix with new VL,
+ // test.nvl.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, test.nvl, 0))
+
+ //
+ // Validate that the VL for the address got set
+ // to test.evl.
+ //
+
+ // The address should not be invalidated until the effective valid
+ // lifetime has passed.
+ select {
+ case <-ndpDisp.autoGenAddrC:
+ t.Fatal("unexpectedly received an auto gen addr event")
+ case <-time.After(time.Duration(test.evl)*time.Second - defaultAsyncNegativeEventTimeout):
+ }
+
+ // Wait for the invalidation event.
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, invalidatedAddr); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ case <-time.After(defaultAsyncPositiveEventTimeout):
+ t.Fatal("timeout waiting for addr auto gen event")
+ }
+ })
+ }
+ })
+}
+
+// TestAutoGenAddrRemoval tests that when auto-generated addresses are removed
+// by the user, its resources will be cleaned up and an invalidation event will
+// be sent to the integrator.
+func TestAutoGenAddrRemoval(t *testing.T) {
+ prefix, _, addr := prefixSubnetAddr(0, linkAddr1)
+
+ ndpDisp := ndpDispatcher{
+ autoGenAddrC: make(chan ndpAutoGenAddrEvent, 1),
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ HandleRAs: true,
+ AutoGenGlobalAddresses: true,
+ },
+ NDPDisp: &ndpDisp,
+ })
+
+ if err := s.CreateNIC(1, e); err != nil {
+ t.Fatalf("CreateNIC(1) = %s", err)
+ }
+
+ expectAutoGenAddrEvent := func(addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, eventType); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected addr auto gen event")
+ }
+ }
+
+ // Receive a PI to auto-generate an address.
+ const lifetimeSeconds = 1
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, lifetimeSeconds, 0))
+ expectAutoGenAddrEvent(addr, newAddr)
+
+ // Removing the address should result in an invalidation event
+ // immediately.
+ if err := s.RemoveAddress(1, addr.Address); err != nil {
+ t.Fatalf("RemoveAddress(_, %s) = %s", addr.Address, err)
+ }
+ expectAutoGenAddrEvent(addr, invalidatedAddr)
+
+ // Wait for the original valid lifetime to make sure the original job got
+ // cancelled/cleaned up.
+ select {
+ case <-ndpDisp.autoGenAddrC:
+ t.Fatal("unexpectedly received an auto gen addr event")
+ case <-time.After(lifetimeSeconds*time.Second + defaultAsyncNegativeEventTimeout):
+ }
+}
+
+// TestAutoGenAddrAfterRemoval tests adding a SLAAC address that was previously
+// assigned to the NIC but is in the permanentExpired state.
+func TestAutoGenAddrAfterRemoval(t *testing.T) {
+ const nicID = 1
+
+ prefix1, _, addr1 := prefixSubnetAddr(0, linkAddr1)
+ prefix2, _, addr2 := prefixSubnetAddr(1, linkAddr1)
+ ndpDisp, e, s := stackAndNdpDispatcherWithDefaultRoute(t, nicID)
+
+ expectAutoGenAddrEvent := func(addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, eventType); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected addr auto gen event")
+ }
+ }
+
+ expectPrimaryAddr := func(addr tcpip.AddressWithPrefix) {
+ t.Helper()
+
+ if got, err := s.GetMainNICAddress(nicID, header.IPv6ProtocolNumber); err != nil {
+ t.Fatalf("s.GetMainNICAddress(%d, %d): %s", nicID, header.IPv6ProtocolNumber, err)
+ } else if got != addr {
+ t.Errorf("got s.GetMainNICAddress(%d, %d) = %s, want = %s", nicID, header.IPv6ProtocolNumber, got, addr)
+ }
+
+ if got := addrForNewConnection(t, s); got != addr.Address {
+ t.Errorf("got addrForNewConnection = %s, want = %s", got, addr.Address)
+ }
+ }
+
+ // Receive a PI to auto-generate addr1 with a large valid and preferred
+ // lifetime.
+ const largeLifetimeSeconds = 999
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr3, 0, prefix1, true, true, largeLifetimeSeconds, largeLifetimeSeconds))
+ expectAutoGenAddrEvent(addr1, newAddr)
+ expectPrimaryAddr(addr1)
+
+ // Add addr2 as a static address.
+ protoAddr2 := tcpip.ProtocolAddress{
+ Protocol: header.IPv6ProtocolNumber,
+ AddressWithPrefix: addr2,
+ }
+ if err := s.AddProtocolAddressWithOptions(nicID, protoAddr2, stack.FirstPrimaryEndpoint); err != nil {
+ t.Fatalf("AddProtocolAddressWithOptions(%d, %+v, %d) = %s", nicID, protoAddr2, stack.FirstPrimaryEndpoint, err)
+ }
+ // addr2 should be more preferred now since it is at the front of the primary
+ // list.
+ expectPrimaryAddr(addr2)
+
+ // Get a route using addr2 to increment its reference count then remove it
+ // to leave it in the permanentExpired state.
+ r, err := s.FindRoute(nicID, addr2.Address, addr3, header.IPv6ProtocolNumber, false)
+ if err != nil {
+ t.Fatalf("FindRoute(%d, %s, %s, %d, false): %s", nicID, addr2.Address, addr3, header.IPv6ProtocolNumber, err)
+ }
+ defer r.Release()
+ if err := s.RemoveAddress(nicID, addr2.Address); err != nil {
+ t.Fatalf("s.RemoveAddress(%d, %s): %s", nicID, addr2.Address, err)
+ }
+ // addr1 should be preferred again since addr2 is in the expired state.
+ expectPrimaryAddr(addr1)
+
+ // Receive a PI to auto-generate addr2 as valid and preferred.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr3, 0, prefix2, true, true, largeLifetimeSeconds, largeLifetimeSeconds))
+ expectAutoGenAddrEvent(addr2, newAddr)
+ // addr2 should be more preferred now that it is closer to the front of the
+ // primary list and not deprecated.
+ expectPrimaryAddr(addr2)
+
+ // Removing the address should result in an invalidation event immediately.
+ // It should still be in the permanentExpired state because r is still held.
+ //
+ // We remove addr2 here to make sure addr2 was marked as a SLAAC address
+ // (it was previously marked as a static address).
+ if err := s.RemoveAddress(1, addr2.Address); err != nil {
+ t.Fatalf("RemoveAddress(_, %s) = %s", addr2.Address, err)
+ }
+ expectAutoGenAddrEvent(addr2, invalidatedAddr)
+ // addr1 should be more preferred since addr2 is in the expired state.
+ expectPrimaryAddr(addr1)
+
+ // Receive a PI to auto-generate addr2 as valid and deprecated.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr3, 0, prefix2, true, true, largeLifetimeSeconds, 0))
+ expectAutoGenAddrEvent(addr2, newAddr)
+ // addr1 should still be more preferred since addr2 is deprecated, even though
+ // it is closer to the front of the primary list.
+ expectPrimaryAddr(addr1)
+
+ // Receive a PI to refresh addr2's preferred lifetime.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr3, 0, prefix2, true, true, largeLifetimeSeconds, largeLifetimeSeconds))
+ select {
+ case <-ndpDisp.autoGenAddrC:
+ t.Fatal("unexpectedly got an auto gen addr event")
+ default:
+ }
+ // addr2 should be more preferred now that it is not deprecated.
+ expectPrimaryAddr(addr2)
+
+ if err := s.RemoveAddress(1, addr2.Address); err != nil {
+ t.Fatalf("RemoveAddress(_, %s) = %s", addr2.Address, err)
+ }
+ expectAutoGenAddrEvent(addr2, invalidatedAddr)
+ expectPrimaryAddr(addr1)
+}
+
+// TestAutoGenAddrStaticConflict tests that if SLAAC generates an address that
+// is already assigned to the NIC, the static address remains.
+func TestAutoGenAddrStaticConflict(t *testing.T) {
+ prefix, _, addr := prefixSubnetAddr(0, linkAddr1)
+
+ ndpDisp := ndpDispatcher{
+ autoGenAddrC: make(chan ndpAutoGenAddrEvent, 1),
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ HandleRAs: true,
+ AutoGenGlobalAddresses: true,
+ },
+ NDPDisp: &ndpDisp,
+ })
+
+ if err := s.CreateNIC(1, e); err != nil {
+ t.Fatalf("CreateNIC(1) = %s", err)
+ }
+
+ // Add the address as a static address before SLAAC tries to add it.
+ if err := s.AddProtocolAddress(1, tcpip.ProtocolAddress{Protocol: header.IPv6ProtocolNumber, AddressWithPrefix: addr}); err != nil {
+ t.Fatalf("AddAddress(_, %d, %s) = %s", header.IPv6ProtocolNumber, addr.Address, err)
+ }
+ if !containsV6Addr(s.NICInfo()[1].ProtocolAddresses, addr) {
+ t.Fatalf("Should have %s in the list of addresses", addr1)
+ }
+
+ // Receive a PI where the generated address will be the same as the one
+ // that we already have assigned statically.
+ const lifetimeSeconds = 1
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, lifetimeSeconds, 0))
+ select {
+ case <-ndpDisp.autoGenAddrC:
+ t.Fatal("unexpectedly received an auto gen addr event for an address we already have statically")
+ default:
+ }
+ if !containsV6Addr(s.NICInfo()[1].ProtocolAddresses, addr) {
+ t.Fatalf("Should have %s in the list of addresses", addr1)
+ }
+
+ // Should not get an invalidation event after the PI's invalidation
+ // time.
+ select {
+ case <-ndpDisp.autoGenAddrC:
+ t.Fatal("unexpectedly received an auto gen addr event")
+ case <-time.After(lifetimeSeconds*time.Second + defaultAsyncNegativeEventTimeout):
+ }
+ if !containsV6Addr(s.NICInfo()[1].ProtocolAddresses, addr) {
+ t.Fatalf("Should have %s in the list of addresses", addr1)
+ }
+}
+
+// TestAutoGenAddrWithOpaqueIID tests that SLAAC generated addresses will use
+// opaque interface identifiers when configured to do so.
+func TestAutoGenAddrWithOpaqueIID(t *testing.T) {
+ const nicID = 1
+ const nicName = "nic1"
+ var secretKeyBuf [header.OpaqueIIDSecretKeyMinBytes]byte
+ secretKey := secretKeyBuf[:]
+ n, err := rand.Read(secretKey)
+ if err != nil {
+ t.Fatalf("rand.Read(_): %s", err)
+ }
+ if n != header.OpaqueIIDSecretKeyMinBytes {
+ t.Fatalf("got rand.Read(_) = (%d, _), want = (%d, _)", n, header.OpaqueIIDSecretKeyMinBytes)
+ }
+
+ prefix1, subnet1, _ := prefixSubnetAddr(0, linkAddr1)
+ prefix2, subnet2, _ := prefixSubnetAddr(1, linkAddr1)
+ // addr1 and addr2 are the addresses that are expected to be generated when
+ // stack.Stack is configured to generate opaque interface identifiers as
+ // defined by RFC 7217.
+ addrBytes := []byte(subnet1.ID())
+ addr1 := tcpip.AddressWithPrefix{
+ Address: tcpip.Address(header.AppendOpaqueInterfaceIdentifier(addrBytes[:header.IIDOffsetInIPv6Address], subnet1, nicName, 0, secretKey)),
+ PrefixLen: 64,
+ }
+ addrBytes = []byte(subnet2.ID())
+ addr2 := tcpip.AddressWithPrefix{
+ Address: tcpip.Address(header.AppendOpaqueInterfaceIdentifier(addrBytes[:header.IIDOffsetInIPv6Address], subnet2, nicName, 0, secretKey)),
+ PrefixLen: 64,
+ }
+
+ ndpDisp := ndpDispatcher{
+ autoGenAddrC: make(chan ndpAutoGenAddrEvent, 1),
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ HandleRAs: true,
+ AutoGenGlobalAddresses: true,
+ },
+ NDPDisp: &ndpDisp,
+ OpaqueIIDOpts: stack.OpaqueInterfaceIdentifierOptions{
+ NICNameFromID: func(_ tcpip.NICID, nicName string) string {
+ return nicName
+ },
+ SecretKey: secretKey,
+ },
+ })
+ opts := stack.NICOptions{Name: nicName}
+ if err := s.CreateNICWithOptions(nicID, e, opts); err != nil {
+ t.Fatalf("CreateNICWithOptions(%d, _, %+v, _) = %s", nicID, opts, err)
+ }
+
+ expectAutoGenAddrEvent := func(addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, eventType); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected addr auto gen event")
+ }
+ }
+
+ // Receive an RA with prefix1 in a PI.
+ const validLifetimeSecondPrefix1 = 1
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, validLifetimeSecondPrefix1, 0))
+ expectAutoGenAddrEvent(addr1, newAddr)
+ if !containsV6Addr(s.NICInfo()[nicID].ProtocolAddresses, addr1) {
+ t.Fatalf("should have %s in the list of addresses", addr1)
+ }
+
+ // Receive an RA with prefix2 in a PI with a large valid lifetime.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix2, true, true, 100, 0))
+ expectAutoGenAddrEvent(addr2, newAddr)
+ if !containsV6Addr(s.NICInfo()[nicID].ProtocolAddresses, addr1) {
+ t.Fatalf("should have %s in the list of addresses", addr1)
+ }
+ if !containsV6Addr(s.NICInfo()[nicID].ProtocolAddresses, addr2) {
+ t.Fatalf("should have %s in the list of addresses", addr2)
+ }
+
+ // Wait for addr of prefix1 to be invalidated.
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr1, invalidatedAddr); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ case <-time.After(validLifetimeSecondPrefix1*time.Second + defaultAsyncPositiveEventTimeout):
+ t.Fatal("timed out waiting for addr auto gen event")
+ }
+ if containsV6Addr(s.NICInfo()[nicID].ProtocolAddresses, addr1) {
+ t.Fatalf("should not have %s in the list of addresses", addr1)
+ }
+ if !containsV6Addr(s.NICInfo()[nicID].ProtocolAddresses, addr2) {
+ t.Fatalf("should have %s in the list of addresses", addr2)
+ }
+}
+
+func TestAutoGenAddrInResponseToDADConflicts(t *testing.T) {
+ const nicID = 1
+ const nicName = "nic"
+ const dadTransmits = 1
+ const retransmitTimer = time.Second
+ const maxMaxRetries = 3
+ const lifetimeSeconds = 10
+
+ // Needed for the temporary address sub test.
+ savedMaxDesync := stack.MaxDesyncFactor
+ defer func() {
+ stack.MaxDesyncFactor = savedMaxDesync
+ }()
+ stack.MaxDesyncFactor = time.Nanosecond
+
+ var secretKeyBuf [header.OpaqueIIDSecretKeyMinBytes]byte
+ secretKey := secretKeyBuf[:]
+ n, err := rand.Read(secretKey)
+ if err != nil {
+ t.Fatalf("rand.Read(_): %s", err)
+ }
+ if n != header.OpaqueIIDSecretKeyMinBytes {
+ t.Fatalf("got rand.Read(_) = (%d, _), want = (%d, _)", n, header.OpaqueIIDSecretKeyMinBytes)
+ }
+
+ prefix, subnet, _ := prefixSubnetAddr(0, linkAddr1)
+
+ addrForSubnet := func(subnet tcpip.Subnet, dadCounter uint8) tcpip.AddressWithPrefix {
+ addrBytes := []byte(subnet.ID())
+ return tcpip.AddressWithPrefix{
+ Address: tcpip.Address(header.AppendOpaqueInterfaceIdentifier(addrBytes[:header.IIDOffsetInIPv6Address], subnet, nicName, dadCounter, secretKey)),
+ PrefixLen: 64,
+ }
+ }
+
+ expectAutoGenAddrEvent := func(t *testing.T, ndpDisp *ndpDispatcher, addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, eventType); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected addr auto gen event")
+ }
+ }
+
+ expectAutoGenAddrEventAsync := func(t *testing.T, ndpDisp *ndpDispatcher, addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, eventType); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ case <-time.After(defaultAsyncPositiveEventTimeout):
+ t.Fatal("timed out waiting for addr auto gen event")
+ }
+ }
+
+ expectDADEvent := func(t *testing.T, ndpDisp *ndpDispatcher, addr tcpip.Address, resolved bool) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.dadC:
+ if diff := checkDADEvent(e, nicID, addr, resolved, nil); diff != "" {
+ t.Errorf("dad event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected DAD event")
+ }
+ }
+
+ expectDADEventAsync := func(t *testing.T, ndpDisp *ndpDispatcher, addr tcpip.Address, resolved bool) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.dadC:
+ if diff := checkDADEvent(e, nicID, addr, resolved, nil); diff != "" {
+ t.Errorf("dad event mismatch (-want +got):\n%s", diff)
+ }
+ case <-time.After(dadTransmits*retransmitTimer + defaultAsyncPositiveEventTimeout):
+ t.Fatal("timed out waiting for DAD event")
+ }
+ }
+
+ stableAddrForTempAddrTest := addrForSubnet(subnet, 0)
+
+ addrTypes := []struct {
+ name string
+ ndpConfigs stack.NDPConfigurations
+ autoGenLinkLocal bool
+ prepareFn func(t *testing.T, ndpDisp *ndpDispatcher, e *channel.Endpoint, tempIIDHistory []byte) []tcpip.AddressWithPrefix
+ addrGenFn func(dadCounter uint8, tempIIDHistory []byte) tcpip.AddressWithPrefix
+ }{
+ {
+ name: "Global address",
+ ndpConfigs: stack.NDPConfigurations{
+ DupAddrDetectTransmits: dadTransmits,
+ RetransmitTimer: retransmitTimer,
+ HandleRAs: true,
+ AutoGenGlobalAddresses: true,
+ },
+ prepareFn: func(_ *testing.T, _ *ndpDispatcher, e *channel.Endpoint, _ []byte) []tcpip.AddressWithPrefix {
+ // Receive an RA with prefix1 in a PI.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, lifetimeSeconds, lifetimeSeconds))
+ return nil
+
+ },
+ addrGenFn: func(dadCounter uint8, _ []byte) tcpip.AddressWithPrefix {
+ return addrForSubnet(subnet, dadCounter)
+ },
+ },
+ {
+ name: "LinkLocal address",
+ ndpConfigs: stack.NDPConfigurations{
+ DupAddrDetectTransmits: dadTransmits,
+ RetransmitTimer: retransmitTimer,
+ },
+ autoGenLinkLocal: true,
+ prepareFn: func(*testing.T, *ndpDispatcher, *channel.Endpoint, []byte) []tcpip.AddressWithPrefix {
+ return nil
+ },
+ addrGenFn: func(dadCounter uint8, _ []byte) tcpip.AddressWithPrefix {
+ return addrForSubnet(header.IPv6LinkLocalPrefix.Subnet(), dadCounter)
+ },
+ },
+ {
+ name: "Temporary address",
+ ndpConfigs: stack.NDPConfigurations{
+ DupAddrDetectTransmits: dadTransmits,
+ RetransmitTimer: retransmitTimer,
+ HandleRAs: true,
+ AutoGenGlobalAddresses: true,
+ AutoGenTempGlobalAddresses: true,
+ },
+ prepareFn: func(t *testing.T, ndpDisp *ndpDispatcher, e *channel.Endpoint, tempIIDHistory []byte) []tcpip.AddressWithPrefix {
+ header.InitialTempIID(tempIIDHistory, nil, nicID)
+
+ // Generate a stable SLAAC address so temporary addresses will be
+ // generated.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, 100, 100))
+ expectAutoGenAddrEvent(t, ndpDisp, stableAddrForTempAddrTest, newAddr)
+ expectDADEventAsync(t, ndpDisp, stableAddrForTempAddrTest.Address, true)
+
+ // The stable address will be assigned throughout the test.
+ return []tcpip.AddressWithPrefix{stableAddrForTempAddrTest}
+ },
+ addrGenFn: func(_ uint8, tempIIDHistory []byte) tcpip.AddressWithPrefix {
+ return header.GenerateTempIPv6SLAACAddr(tempIIDHistory, stableAddrForTempAddrTest.Address)
+ },
+ },
+ }
+
+ for _, addrType := range addrTypes {
+ // This Run will not return until the parallel tests finish.
+ //
+ // We need this because we need to do some teardown work after the parallel
+ // tests complete and limit the number of parallel tests running at the same
+ // time to reduce flakes.
+ //
+ // See https://godoc.org/testing#hdr-Subtests_and_Sub_benchmarks for
+ // more details.
+ t.Run(addrType.name, func(t *testing.T) {
+ for maxRetries := uint8(0); maxRetries <= maxMaxRetries; maxRetries++ {
+ for numFailures := uint8(0); numFailures <= maxRetries+1; numFailures++ {
+ maxRetries := maxRetries
+ numFailures := numFailures
+ addrType := addrType
+
+ t.Run(fmt.Sprintf("%d max retries and %d failures", maxRetries, numFailures), func(t *testing.T) {
+ t.Parallel()
+
+ ndpDisp := ndpDispatcher{
+ dadC: make(chan ndpDADEvent, 1),
+ autoGenAddrC: make(chan ndpAutoGenAddrEvent, 2),
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ ndpConfigs := addrType.ndpConfigs
+ ndpConfigs.AutoGenAddressConflictRetries = maxRetries
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ AutoGenIPv6LinkLocal: addrType.autoGenLinkLocal,
+ NDPConfigs: ndpConfigs,
+ NDPDisp: &ndpDisp,
+ OpaqueIIDOpts: stack.OpaqueInterfaceIdentifierOptions{
+ NICNameFromID: func(_ tcpip.NICID, nicName string) string {
+ return nicName
+ },
+ SecretKey: secretKey,
+ },
+ })
+ opts := stack.NICOptions{Name: nicName}
+ if err := s.CreateNICWithOptions(nicID, e, opts); err != nil {
+ t.Fatalf("CreateNICWithOptions(%d, _, %+v) = %s", nicID, opts, err)
+ }
+
+ var tempIIDHistory [header.IIDSize]byte
+ stableAddrs := addrType.prepareFn(t, &ndpDisp, e, tempIIDHistory[:])
+
+ // Simulate DAD conflicts so the address is regenerated.
+ for i := uint8(0); i < numFailures; i++ {
+ addr := addrType.addrGenFn(i, tempIIDHistory[:])
+ expectAutoGenAddrEventAsync(t, &ndpDisp, addr, newAddr)
+
+ // Should not have any new addresses assigned to the NIC.
+ if mismatch := addressCheck(s.NICInfo()[nicID].ProtocolAddresses, stableAddrs, nil); mismatch != "" {
+ t.Fatal(mismatch)
+ }
+
+ // Simulate a DAD conflict.
+ if err := s.DupTentativeAddrDetected(nicID, addr.Address); err != nil {
+ t.Fatalf("s.DupTentativeAddrDetected(%d, %s): %s", nicID, addr.Address, err)
+ }
+ expectAutoGenAddrEvent(t, &ndpDisp, addr, invalidatedAddr)
+ expectDADEvent(t, &ndpDisp, addr.Address, false)
+
+ // Attempting to add the address manually should not fail if the
+ // address's state was cleaned up when DAD failed.
+ if err := s.AddAddress(nicID, header.IPv6ProtocolNumber, addr.Address); err != nil {
+ t.Fatalf("AddAddress(%d, %d, %s) = %s", nicID, header.IPv6ProtocolNumber, addr.Address, err)
+ }
+ if err := s.RemoveAddress(nicID, addr.Address); err != nil {
+ t.Fatalf("RemoveAddress(%d, %s) = %s", nicID, addr.Address, err)
+ }
+ expectDADEvent(t, &ndpDisp, addr.Address, false)
+ }
+
+ // Should not have any new addresses assigned to the NIC.
+ if mismatch := addressCheck(s.NICInfo()[nicID].ProtocolAddresses, stableAddrs, nil); mismatch != "" {
+ t.Fatal(mismatch)
+ }
+
+ // If we had less failures than generation attempts, we should have
+ // an address after DAD resolves.
+ if maxRetries+1 > numFailures {
+ addr := addrType.addrGenFn(numFailures, tempIIDHistory[:])
+ expectAutoGenAddrEventAsync(t, &ndpDisp, addr, newAddr)
+ expectDADEventAsync(t, &ndpDisp, addr.Address, true)
+ if mismatch := addressCheck(s.NICInfo()[nicID].ProtocolAddresses, append(stableAddrs, addr), nil); mismatch != "" {
+ t.Fatal(mismatch)
+ }
+ }
+
+ // Should not attempt address generation again.
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ t.Fatalf("unexpectedly got an auto-generated address event = %+v", e)
+ case <-time.After(defaultAsyncNegativeEventTimeout):
+ }
+ })
+ }
+ }
+ })
+ }
+}
+
+// TestAutoGenAddrWithEUI64IIDNoDADRetries tests that a regeneration attempt is
+// not made for SLAAC addresses generated with an IID based on the NIC's link
+// address.
+func TestAutoGenAddrWithEUI64IIDNoDADRetries(t *testing.T) {
+ const nicID = 1
+ const dadTransmits = 1
+ const retransmitTimer = time.Second
+ const maxRetries = 3
+ const lifetimeSeconds = 10
+
+ prefix, subnet, _ := prefixSubnetAddr(0, linkAddr1)
+
+ addrTypes := []struct {
+ name string
+ ndpConfigs stack.NDPConfigurations
+ autoGenLinkLocal bool
+ subnet tcpip.Subnet
+ triggerSLAACFn func(e *channel.Endpoint)
+ }{
+ {
+ name: "Global address",
+ ndpConfigs: stack.NDPConfigurations{
+ DupAddrDetectTransmits: dadTransmits,
+ RetransmitTimer: retransmitTimer,
+ HandleRAs: true,
+ AutoGenGlobalAddresses: true,
+ AutoGenAddressConflictRetries: maxRetries,
+ },
+ subnet: subnet,
+ triggerSLAACFn: func(e *channel.Endpoint) {
+ // Receive an RA with prefix1 in a PI.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, lifetimeSeconds, lifetimeSeconds))
+
+ },
+ },
+ {
+ name: "LinkLocal address",
+ ndpConfigs: stack.NDPConfigurations{
+ DupAddrDetectTransmits: dadTransmits,
+ RetransmitTimer: retransmitTimer,
+ AutoGenAddressConflictRetries: maxRetries,
+ },
+ autoGenLinkLocal: true,
+ subnet: header.IPv6LinkLocalPrefix.Subnet(),
+ triggerSLAACFn: func(e *channel.Endpoint) {},
+ },
+ }
+
+ for _, addrType := range addrTypes {
+ addrType := addrType
+
+ t.Run(addrType.name, func(t *testing.T) {
+ t.Parallel()
+
+ ndpDisp := ndpDispatcher{
+ dadC: make(chan ndpDADEvent, 1),
+ autoGenAddrC: make(chan ndpAutoGenAddrEvent, 2),
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ AutoGenIPv6LinkLocal: addrType.autoGenLinkLocal,
+ NDPConfigs: addrType.ndpConfigs,
+ NDPDisp: &ndpDisp,
+ })
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+
+ expectAutoGenAddrEvent := func(addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, eventType); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected addr auto gen event")
+ }
+ }
+
+ addrType.triggerSLAACFn(e)
+
+ addrBytes := []byte(addrType.subnet.ID())
+ header.EthernetAdddressToModifiedEUI64IntoBuf(linkAddr1, addrBytes[header.IIDOffsetInIPv6Address:])
+ addr := tcpip.AddressWithPrefix{
+ Address: tcpip.Address(addrBytes),
+ PrefixLen: 64,
+ }
+ expectAutoGenAddrEvent(addr, newAddr)
+
+ // Simulate a DAD conflict.
+ if err := s.DupTentativeAddrDetected(nicID, addr.Address); err != nil {
+ t.Fatalf("s.DupTentativeAddrDetected(%d, %s): %s", nicID, addr.Address, err)
+ }
+ expectAutoGenAddrEvent(addr, invalidatedAddr)
+ select {
+ case e := <-ndpDisp.dadC:
+ if diff := checkDADEvent(e, nicID, addr.Address, false, nil); diff != "" {
+ t.Errorf("dad event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected DAD event")
+ }
+
+ // Should not attempt address regeneration.
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ t.Fatalf("unexpectedly got an auto-generated address event = %+v", e)
+ case <-time.After(defaultAsyncNegativeEventTimeout):
+ }
+ })
+ }
+}
+
+// TestAutoGenAddrContinuesLifetimesAfterRetry tests that retrying address
+// generation in response to DAD conflicts does not refresh the lifetimes.
+func TestAutoGenAddrContinuesLifetimesAfterRetry(t *testing.T) {
+ const nicID = 1
+ const nicName = "nic"
+ const dadTransmits = 1
+ const retransmitTimer = 2 * time.Second
+ const failureTimer = time.Second
+ const maxRetries = 1
+ const lifetimeSeconds = 5
+
+ var secretKeyBuf [header.OpaqueIIDSecretKeyMinBytes]byte
+ secretKey := secretKeyBuf[:]
+ n, err := rand.Read(secretKey)
+ if err != nil {
+ t.Fatalf("rand.Read(_): %s", err)
+ }
+ if n != header.OpaqueIIDSecretKeyMinBytes {
+ t.Fatalf("got rand.Read(_) = (%d, _), want = (%d, _)", n, header.OpaqueIIDSecretKeyMinBytes)
+ }
+
+ prefix, subnet, _ := prefixSubnetAddr(0, linkAddr1)
+
+ ndpDisp := ndpDispatcher{
+ dadC: make(chan ndpDADEvent, 1),
+ autoGenAddrC: make(chan ndpAutoGenAddrEvent, 2),
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ DupAddrDetectTransmits: dadTransmits,
+ RetransmitTimer: retransmitTimer,
+ HandleRAs: true,
+ AutoGenGlobalAddresses: true,
+ AutoGenAddressConflictRetries: maxRetries,
+ },
+ NDPDisp: &ndpDisp,
+ OpaqueIIDOpts: stack.OpaqueInterfaceIdentifierOptions{
+ NICNameFromID: func(_ tcpip.NICID, nicName string) string {
+ return nicName
+ },
+ SecretKey: secretKey,
+ },
+ })
+ opts := stack.NICOptions{Name: nicName}
+ if err := s.CreateNICWithOptions(nicID, e, opts); err != nil {
+ t.Fatalf("CreateNICWithOptions(%d, _, %+v) = %s", nicID, opts, err)
+ }
+
+ expectAutoGenAddrEvent := func(addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType) {
+ t.Helper()
+
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, eventType); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected addr auto gen event")
+ }
+ }
+
+ // Receive an RA with prefix in a PI.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, lifetimeSeconds, lifetimeSeconds))
+
+ addrBytes := []byte(subnet.ID())
+ addr := tcpip.AddressWithPrefix{
+ Address: tcpip.Address(header.AppendOpaqueInterfaceIdentifier(addrBytes[:header.IIDOffsetInIPv6Address], subnet, nicName, 0, secretKey)),
+ PrefixLen: 64,
+ }
+ expectAutoGenAddrEvent(addr, newAddr)
+
+ // Simulate a DAD conflict after some time has passed.
+ time.Sleep(failureTimer)
+ if err := s.DupTentativeAddrDetected(nicID, addr.Address); err != nil {
+ t.Fatalf("s.DupTentativeAddrDetected(%d, %s): %s", nicID, addr.Address, err)
+ }
+ expectAutoGenAddrEvent(addr, invalidatedAddr)
+ select {
+ case e := <-ndpDisp.dadC:
+ if diff := checkDADEvent(e, nicID, addr.Address, false, nil); diff != "" {
+ t.Errorf("dad event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected DAD event")
+ }
+
+ // Let the next address resolve.
+ addr.Address = tcpip.Address(header.AppendOpaqueInterfaceIdentifier(addrBytes[:header.IIDOffsetInIPv6Address], subnet, nicName, 1, secretKey))
+ expectAutoGenAddrEvent(addr, newAddr)
+ select {
+ case e := <-ndpDisp.dadC:
+ if diff := checkDADEvent(e, nicID, addr.Address, true, nil); diff != "" {
+ t.Errorf("dad event mismatch (-want +got):\n%s", diff)
+ }
+ case <-time.After(dadTransmits*retransmitTimer + defaultAsyncPositiveEventTimeout):
+ t.Fatal("timed out waiting for DAD event")
+ }
+
+ // Address should be deprecated/invalidated after the lifetime expires.
+ //
+ // Note, the remaining lifetime is calculated from when the PI was first
+ // processed. Since we wait for some time before simulating a DAD conflict
+ // and more time for the new address to resolve, the new address is only
+ // expected to be valid for the remaining time. The DAD conflict should
+ // not have reset the lifetimes.
+ //
+ // We expect either just the invalidation event or the deprecation event
+ // followed by the invalidation event.
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if e.eventType == deprecatedAddr {
+ if diff := checkAutoGenAddrEvent(e, addr, deprecatedAddr); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, addr, invalidatedAddr); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ case <-time.After(defaultAsyncPositiveEventTimeout):
+ t.Fatal("timed out waiting for invalidated auto gen addr event after deprecation")
+ }
+ } else {
+ if diff := checkAutoGenAddrEvent(e, addr, invalidatedAddr); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ }
+ case <-time.After(lifetimeSeconds*time.Second - failureTimer - dadTransmits*retransmitTimer + defaultAsyncPositiveEventTimeout):
+ t.Fatal("timed out waiting for auto gen addr event")
+ }
+}
+
+// TestNDPRecursiveDNSServerDispatch tests that we properly dispatch an event
+// to the integrator when an RA is received with the NDP Recursive DNS Server
+// option with at least one valid address.
+func TestNDPRecursiveDNSServerDispatch(t *testing.T) {
+ tests := []struct {
+ name string
+ opt header.NDPRecursiveDNSServer
+ expected *ndpRDNSS
+ }{
+ {
+ "Unspecified",
+ header.NDPRecursiveDNSServer([]byte{
+ 0, 0,
+ 0, 0, 0, 2,
+ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+ }),
+ nil,
+ },
+ {
+ "Multicast",
+ header.NDPRecursiveDNSServer([]byte{
+ 0, 0,
+ 0, 0, 0, 2,
+ 255, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1,
+ }),
+ nil,
+ },
+ {
+ "OptionTooSmall",
+ header.NDPRecursiveDNSServer([]byte{
+ 0, 0,
+ 0, 0, 0, 2,
+ 1, 2, 3, 4, 5, 6, 7, 8,
+ }),
+ nil,
+ },
+ {
+ "0Addresses",
+ header.NDPRecursiveDNSServer([]byte{
+ 0, 0,
+ 0, 0, 0, 2,
+ }),
+ nil,
+ },
+ {
+ "Valid1Address",
+ header.NDPRecursiveDNSServer([]byte{
+ 0, 0,
+ 0, 0, 0, 2,
+ 1, 2, 3, 4, 5, 6, 7, 8, 0, 0, 0, 0, 0, 0, 0, 1,
+ }),
+ &ndpRDNSS{
+ []tcpip.Address{
+ "\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x01",
+ },
+ 2 * time.Second,
+ },
+ },
+ {
+ "Valid2Addresses",
+ header.NDPRecursiveDNSServer([]byte{
+ 0, 0,
+ 0, 0, 0, 1,
+ 1, 2, 3, 4, 5, 6, 7, 8, 0, 0, 0, 0, 0, 0, 0, 1,
+ 1, 2, 3, 4, 5, 6, 7, 8, 0, 0, 0, 0, 0, 0, 0, 2,
+ }),
+ &ndpRDNSS{
+ []tcpip.Address{
+ "\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x01",
+ "\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x02",
+ },
+ time.Second,
+ },
+ },
+ {
+ "Valid3Addresses",
+ header.NDPRecursiveDNSServer([]byte{
+ 0, 0,
+ 0, 0, 0, 0,
+ 1, 2, 3, 4, 5, 6, 7, 8, 0, 0, 0, 0, 0, 0, 0, 1,
+ 1, 2, 3, 4, 5, 6, 7, 8, 0, 0, 0, 0, 0, 0, 0, 2,
+ 1, 2, 3, 4, 5, 6, 7, 8, 0, 0, 0, 0, 0, 0, 0, 3,
+ }),
+ &ndpRDNSS{
+ []tcpip.Address{
+ "\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x01",
+ "\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x02",
+ "\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x03",
+ },
+ 0,
+ },
+ },
+ }
+
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ ndpDisp := ndpDispatcher{
+ // We do not expect more than a single RDNSS
+ // event at any time for this test.
+ rdnssC: make(chan ndpRDNSSEvent, 1),
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ HandleRAs: true,
+ },
+ NDPDisp: &ndpDisp,
+ })
+ if err := s.CreateNIC(1, e); err != nil {
+ t.Fatalf("CreateNIC(1) = %s", err)
+ }
+
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithOpts(llAddr1, 0, header.NDPOptionsSerializer{test.opt}))
+
+ if test.expected != nil {
+ select {
+ case e := <-ndpDisp.rdnssC:
+ if e.nicID != 1 {
+ t.Errorf("got rdnss nicID = %d, want = 1", e.nicID)
+ }
+ if diff := cmp.Diff(e.rdnss.addrs, test.expected.addrs); diff != "" {
+ t.Errorf("rdnss addrs mismatch (-want +got):\n%s", diff)
+ }
+ if e.rdnss.lifetime != test.expected.lifetime {
+ t.Errorf("got rdnss lifetime = %s, want = %s", e.rdnss.lifetime, test.expected.lifetime)
+ }
+ default:
+ t.Fatal("expected an RDNSS option event")
+ }
+ }
+
+ // Should have no more RDNSS options.
+ select {
+ case e := <-ndpDisp.rdnssC:
+ t.Fatalf("unexpectedly got a new RDNSS option event: %+v", e)
+ default:
+ }
+ })
+ }
+}
+
+// TestNDPDNSSearchListDispatch tests that the integrator is informed when an
+// NDP DNS Search List option is received with at least one domain name in the
+// search list.
+func TestNDPDNSSearchListDispatch(t *testing.T) {
+ const nicID = 1
+
+ ndpDisp := ndpDispatcher{
+ dnsslC: make(chan ndpDNSSLEvent, 3),
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ HandleRAs: true,
+ },
+ NDPDisp: &ndpDisp,
+ })
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+
+ optSer := header.NDPOptionsSerializer{
+ header.NDPDNSSearchList([]byte{
+ 0, 0,
+ 0, 0, 0, 0,
+ 2, 'h', 'i',
+ 0,
+ }),
+ header.NDPDNSSearchList([]byte{
+ 0, 0,
+ 0, 0, 0, 1,
+ 1, 'i',
+ 0,
+ 2, 'a', 'm',
+ 2, 'm', 'e',
+ 0,
+ }),
+ header.NDPDNSSearchList([]byte{
+ 0, 0,
+ 0, 0, 1, 0,
+ 3, 'x', 'y', 'z',
+ 0,
+ 5, 'h', 'e', 'l', 'l', 'o',
+ 5, 'w', 'o', 'r', 'l', 'd',
+ 0,
+ 4, 't', 'h', 'i', 's',
+ 2, 'i', 's',
+ 1, 'a',
+ 4, 't', 'e', 's', 't',
+ 0,
+ }),
+ }
+ expected := []struct {
+ domainNames []string
+ lifetime time.Duration
+ }{
+ {
+ domainNames: []string{
+ "hi",
+ },
+ lifetime: 0,
+ },
+ {
+ domainNames: []string{
+ "i",
+ "am.me",
+ },
+ lifetime: time.Second,
+ },
+ {
+ domainNames: []string{
+ "xyz",
+ "hello.world",
+ "this.is.a.test",
+ },
+ lifetime: 256 * time.Second,
+ },
+ }
+
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithOpts(llAddr1, 0, optSer))
+
+ for i, expected := range expected {
+ select {
+ case dnssl := <-ndpDisp.dnsslC:
+ if dnssl.nicID != nicID {
+ t.Errorf("got %d-th dnssl nicID = %d, want = %d", i, dnssl.nicID, nicID)
+ }
+ if diff := cmp.Diff(dnssl.domainNames, expected.domainNames); diff != "" {
+ t.Errorf("%d-th dnssl domain names mismatch (-want +got):\n%s", i, diff)
+ }
+ if dnssl.lifetime != expected.lifetime {
+ t.Errorf("got %d-th dnssl lifetime = %s, want = %s", i, dnssl.lifetime, expected.lifetime)
+ }
+ default:
+ t.Fatal("expected a DNSSL event")
+ }
+ }
+
+ // Should have no more DNSSL options.
+ select {
+ case <-ndpDisp.dnsslC:
+ t.Fatal("unexpectedly got a DNSSL event")
+ default:
+ }
+}
+
+// TestCleanupNDPState tests that all discovered routers and prefixes, and
+// auto-generated addresses are invalidated when a NIC becomes a router.
+func TestCleanupNDPState(t *testing.T) {
+ const (
+ lifetimeSeconds = 5
+ maxRouterAndPrefixEvents = 4
+ nicID1 = 1
+ nicID2 = 2
+ )
+
+ prefix1, subnet1, e1Addr1 := prefixSubnetAddr(0, linkAddr1)
+ prefix2, subnet2, e1Addr2 := prefixSubnetAddr(1, linkAddr1)
+ e2Addr1 := addrForSubnet(subnet1, linkAddr2)
+ e2Addr2 := addrForSubnet(subnet2, linkAddr2)
+ llAddrWithPrefix1 := tcpip.AddressWithPrefix{
+ Address: llAddr1,
+ PrefixLen: 64,
+ }
+ llAddrWithPrefix2 := tcpip.AddressWithPrefix{
+ Address: llAddr2,
+ PrefixLen: 64,
+ }
+
+ tests := []struct {
+ name string
+ cleanupFn func(t *testing.T, s *stack.Stack)
+ keepAutoGenLinkLocal bool
+ maxAutoGenAddrEvents int
+ skipFinalAddrCheck bool
+ }{
+ // A NIC should still keep its auto-generated link-local address when
+ // becoming a router.
+ {
+ name: "Enable forwarding",
+ cleanupFn: func(t *testing.T, s *stack.Stack) {
+ t.Helper()
+ s.SetForwarding(ipv6.ProtocolNumber, true)
+ },
+ keepAutoGenLinkLocal: true,
+ maxAutoGenAddrEvents: 4,
+ },
+
+ // A NIC should cleanup all NDP state when it is disabled.
+ {
+ name: "Disable NIC",
+ cleanupFn: func(t *testing.T, s *stack.Stack) {
+ t.Helper()
+
+ if err := s.DisableNIC(nicID1); err != nil {
+ t.Fatalf("s.DisableNIC(%d): %s", nicID1, err)
+ }
+ if err := s.DisableNIC(nicID2); err != nil {
+ t.Fatalf("s.DisableNIC(%d): %s", nicID2, err)
+ }
+ },
+ keepAutoGenLinkLocal: false,
+ maxAutoGenAddrEvents: 6,
+ },
+
+ // A NIC should cleanup all NDP state when it is removed.
+ {
+ name: "Remove NIC",
+ cleanupFn: func(t *testing.T, s *stack.Stack) {
+ t.Helper()
+
+ if err := s.RemoveNIC(nicID1); err != nil {
+ t.Fatalf("s.RemoveNIC(%d): %s", nicID1, err)
+ }
+ if err := s.RemoveNIC(nicID2); err != nil {
+ t.Fatalf("s.RemoveNIC(%d): %s", nicID2, err)
+ }
+ },
+ keepAutoGenLinkLocal: false,
+ maxAutoGenAddrEvents: 6,
+ // The NICs are removed so we can't check their addresses after calling
+ // stopFn.
+ skipFinalAddrCheck: true,
+ },
+ }
+
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ ndpDisp := ndpDispatcher{
+ routerC: make(chan ndpRouterEvent, maxRouterAndPrefixEvents),
+ rememberRouter: true,
+ prefixC: make(chan ndpPrefixEvent, maxRouterAndPrefixEvents),
+ rememberPrefix: true,
+ autoGenAddrC: make(chan ndpAutoGenAddrEvent, test.maxAutoGenAddrEvents),
+ }
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ AutoGenIPv6LinkLocal: true,
+ NDPConfigs: stack.NDPConfigurations{
+ HandleRAs: true,
+ DiscoverDefaultRouters: true,
+ DiscoverOnLinkPrefixes: true,
+ AutoGenGlobalAddresses: true,
+ },
+ NDPDisp: &ndpDisp,
+ })
+
+ expectRouterEvent := func() (bool, ndpRouterEvent) {
+ select {
+ case e := <-ndpDisp.routerC:
+ return true, e
+ default:
+ }
+
+ return false, ndpRouterEvent{}
+ }
+
+ expectPrefixEvent := func() (bool, ndpPrefixEvent) {
+ select {
+ case e := <-ndpDisp.prefixC:
+ return true, e
+ default:
+ }
+
+ return false, ndpPrefixEvent{}
+ }
+
+ expectAutoGenAddrEvent := func() (bool, ndpAutoGenAddrEvent) {
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ return true, e
+ default:
+ }
+
+ return false, ndpAutoGenAddrEvent{}
+ }
+
+ e1 := channel.New(0, 1280, linkAddr1)
+ if err := s.CreateNIC(nicID1, e1); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID1, err)
+ }
+ // We have other tests that make sure we receive the *correct* events
+ // on normal discovery of routers/prefixes, and auto-generated
+ // addresses. Here we just make sure we get an event and let other tests
+ // handle the correctness check.
+ expectAutoGenAddrEvent()
+
+ e2 := channel.New(0, 1280, linkAddr2)
+ if err := s.CreateNIC(nicID2, e2); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID2, err)
+ }
+ expectAutoGenAddrEvent()
+
+ // Receive RAs on NIC(1) and NIC(2) from default routers (llAddr3 and
+ // llAddr4) w/ PI (for prefix1 in RA from llAddr3 and prefix2 in RA from
+ // llAddr4) to discover multiple routers and prefixes, and auto-gen
+ // multiple addresses.
+
+ e1.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr3, lifetimeSeconds, prefix1, true, true, lifetimeSeconds, lifetimeSeconds))
+ if ok, _ := expectRouterEvent(); !ok {
+ t.Errorf("expected router event for %s on NIC(%d)", llAddr3, nicID1)
+ }
+ if ok, _ := expectPrefixEvent(); !ok {
+ t.Errorf("expected prefix event for %s on NIC(%d)", prefix1, nicID1)
+ }
+ if ok, _ := expectAutoGenAddrEvent(); !ok {
+ t.Errorf("expected auto-gen addr event for %s on NIC(%d)", e1Addr1, nicID1)
+ }
+
+ e1.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr4, lifetimeSeconds, prefix2, true, true, lifetimeSeconds, lifetimeSeconds))
+ if ok, _ := expectRouterEvent(); !ok {
+ t.Errorf("expected router event for %s on NIC(%d)", llAddr4, nicID1)
+ }
+ if ok, _ := expectPrefixEvent(); !ok {
+ t.Errorf("expected prefix event for %s on NIC(%d)", prefix2, nicID1)
+ }
+ if ok, _ := expectAutoGenAddrEvent(); !ok {
+ t.Errorf("expected auto-gen addr event for %s on NIC(%d)", e1Addr2, nicID1)
+ }
+
+ e2.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr3, lifetimeSeconds, prefix1, true, true, lifetimeSeconds, lifetimeSeconds))
+ if ok, _ := expectRouterEvent(); !ok {
+ t.Errorf("expected router event for %s on NIC(%d)", llAddr3, nicID2)
+ }
+ if ok, _ := expectPrefixEvent(); !ok {
+ t.Errorf("expected prefix event for %s on NIC(%d)", prefix1, nicID2)
+ }
+ if ok, _ := expectAutoGenAddrEvent(); !ok {
+ t.Errorf("expected auto-gen addr event for %s on NIC(%d)", e1Addr2, nicID2)
+ }
+
+ e2.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr4, lifetimeSeconds, prefix2, true, true, lifetimeSeconds, lifetimeSeconds))
+ if ok, _ := expectRouterEvent(); !ok {
+ t.Errorf("expected router event for %s on NIC(%d)", llAddr4, nicID2)
+ }
+ if ok, _ := expectPrefixEvent(); !ok {
+ t.Errorf("expected prefix event for %s on NIC(%d)", prefix2, nicID2)
+ }
+ if ok, _ := expectAutoGenAddrEvent(); !ok {
+ t.Errorf("expected auto-gen addr event for %s on NIC(%d)", e2Addr2, nicID2)
+ }
+
+ // We should have the auto-generated addresses added.
+ nicinfo := s.NICInfo()
+ nic1Addrs := nicinfo[nicID1].ProtocolAddresses
+ nic2Addrs := nicinfo[nicID2].ProtocolAddresses
+ if !containsV6Addr(nic1Addrs, llAddrWithPrefix1) {
+ t.Errorf("missing %s from the list of addresses for NIC(%d): %+v", llAddrWithPrefix1, nicID1, nic1Addrs)
+ }
+ if !containsV6Addr(nic1Addrs, e1Addr1) {
+ t.Errorf("missing %s from the list of addresses for NIC(%d): %+v", e1Addr1, nicID1, nic1Addrs)
+ }
+ if !containsV6Addr(nic1Addrs, e1Addr2) {
+ t.Errorf("missing %s from the list of addresses for NIC(%d): %+v", e1Addr2, nicID1, nic1Addrs)
+ }
+ if !containsV6Addr(nic2Addrs, llAddrWithPrefix2) {
+ t.Errorf("missing %s from the list of addresses for NIC(%d): %+v", llAddrWithPrefix2, nicID2, nic2Addrs)
+ }
+ if !containsV6Addr(nic2Addrs, e2Addr1) {
+ t.Errorf("missing %s from the list of addresses for NIC(%d): %+v", e2Addr1, nicID2, nic2Addrs)
+ }
+ if !containsV6Addr(nic2Addrs, e2Addr2) {
+ t.Errorf("missing %s from the list of addresses for NIC(%d): %+v", e2Addr2, nicID2, nic2Addrs)
+ }
+
+ // We can't proceed any further if we already failed the test (missing
+ // some discovery/auto-generated address events or addresses).
+ if t.Failed() {
+ t.FailNow()
+ }
+
+ test.cleanupFn(t, s)
+
+ // Collect invalidation events after having NDP state cleaned up.
+ gotRouterEvents := make(map[ndpRouterEvent]int)
+ for i := 0; i < maxRouterAndPrefixEvents; i++ {
+ ok, e := expectRouterEvent()
+ if !ok {
+ t.Errorf("expected %d router events after becoming a router; got = %d", maxRouterAndPrefixEvents, i)
+ break
+ }
+ gotRouterEvents[e]++
+ }
+ gotPrefixEvents := make(map[ndpPrefixEvent]int)
+ for i := 0; i < maxRouterAndPrefixEvents; i++ {
+ ok, e := expectPrefixEvent()
+ if !ok {
+ t.Errorf("expected %d prefix events after becoming a router; got = %d", maxRouterAndPrefixEvents, i)
+ break
+ }
+ gotPrefixEvents[e]++
+ }
+ gotAutoGenAddrEvents := make(map[ndpAutoGenAddrEvent]int)
+ for i := 0; i < test.maxAutoGenAddrEvents; i++ {
+ ok, e := expectAutoGenAddrEvent()
+ if !ok {
+ t.Errorf("expected %d auto-generated address events after becoming a router; got = %d", test.maxAutoGenAddrEvents, i)
+ break
+ }
+ gotAutoGenAddrEvents[e]++
+ }
+
+ // No need to proceed any further if we already failed the test (missing
+ // some invalidation events).
+ if t.Failed() {
+ t.FailNow()
+ }
+
+ expectedRouterEvents := map[ndpRouterEvent]int{
+ {nicID: nicID1, addr: llAddr3, discovered: false}: 1,
+ {nicID: nicID1, addr: llAddr4, discovered: false}: 1,
+ {nicID: nicID2, addr: llAddr3, discovered: false}: 1,
+ {nicID: nicID2, addr: llAddr4, discovered: false}: 1,
+ }
+ if diff := cmp.Diff(expectedRouterEvents, gotRouterEvents); diff != "" {
+ t.Errorf("router events mismatch (-want +got):\n%s", diff)
+ }
+ expectedPrefixEvents := map[ndpPrefixEvent]int{
+ {nicID: nicID1, prefix: subnet1, discovered: false}: 1,
+ {nicID: nicID1, prefix: subnet2, discovered: false}: 1,
+ {nicID: nicID2, prefix: subnet1, discovered: false}: 1,
+ {nicID: nicID2, prefix: subnet2, discovered: false}: 1,
+ }
+ if diff := cmp.Diff(expectedPrefixEvents, gotPrefixEvents); diff != "" {
+ t.Errorf("prefix events mismatch (-want +got):\n%s", diff)
+ }
+ expectedAutoGenAddrEvents := map[ndpAutoGenAddrEvent]int{
+ {nicID: nicID1, addr: e1Addr1, eventType: invalidatedAddr}: 1,
+ {nicID: nicID1, addr: e1Addr2, eventType: invalidatedAddr}: 1,
+ {nicID: nicID2, addr: e2Addr1, eventType: invalidatedAddr}: 1,
+ {nicID: nicID2, addr: e2Addr2, eventType: invalidatedAddr}: 1,
+ }
+
+ if !test.keepAutoGenLinkLocal {
+ expectedAutoGenAddrEvents[ndpAutoGenAddrEvent{nicID: nicID1, addr: llAddrWithPrefix1, eventType: invalidatedAddr}] = 1
+ expectedAutoGenAddrEvents[ndpAutoGenAddrEvent{nicID: nicID2, addr: llAddrWithPrefix2, eventType: invalidatedAddr}] = 1
+ }
+
+ if diff := cmp.Diff(expectedAutoGenAddrEvents, gotAutoGenAddrEvents); diff != "" {
+ t.Errorf("auto-generated address events mismatch (-want +got):\n%s", diff)
+ }
+
+ if !test.skipFinalAddrCheck {
+ // Make sure the auto-generated addresses got removed.
+ nicinfo = s.NICInfo()
+ nic1Addrs = nicinfo[nicID1].ProtocolAddresses
+ nic2Addrs = nicinfo[nicID2].ProtocolAddresses
+ if containsV6Addr(nic1Addrs, llAddrWithPrefix1) != test.keepAutoGenLinkLocal {
+ if test.keepAutoGenLinkLocal {
+ t.Errorf("missing %s from the list of addresses for NIC(%d): %+v", llAddrWithPrefix1, nicID1, nic1Addrs)
+ } else {
+ t.Errorf("still have %s in the list of addresses for NIC(%d): %+v", llAddrWithPrefix1, nicID1, nic1Addrs)
+ }
+ }
+ if containsV6Addr(nic1Addrs, e1Addr1) {
+ t.Errorf("still have %s in the list of addresses for NIC(%d): %+v", e1Addr1, nicID1, nic1Addrs)
+ }
+ if containsV6Addr(nic1Addrs, e1Addr2) {
+ t.Errorf("still have %s in the list of addresses for NIC(%d): %+v", e1Addr2, nicID1, nic1Addrs)
+ }
+ if containsV6Addr(nic2Addrs, llAddrWithPrefix2) != test.keepAutoGenLinkLocal {
+ if test.keepAutoGenLinkLocal {
+ t.Errorf("missing %s from the list of addresses for NIC(%d): %+v", llAddrWithPrefix2, nicID2, nic2Addrs)
+ } else {
+ t.Errorf("still have %s in the list of addresses for NIC(%d): %+v", llAddrWithPrefix2, nicID2, nic2Addrs)
+ }
+ }
+ if containsV6Addr(nic2Addrs, e2Addr1) {
+ t.Errorf("still have %s in the list of addresses for NIC(%d): %+v", e2Addr1, nicID2, nic2Addrs)
+ }
+ if containsV6Addr(nic2Addrs, e2Addr2) {
+ t.Errorf("still have %s in the list of addresses for NIC(%d): %+v", e2Addr2, nicID2, nic2Addrs)
+ }
+ }
+
+ // Should not get any more events (invalidation timers should have been
+ // cancelled when the NDP state was cleaned up).
+ time.Sleep(lifetimeSeconds*time.Second + defaultAsyncNegativeEventTimeout)
+ select {
+ case <-ndpDisp.routerC:
+ t.Error("unexpected router event")
+ default:
+ }
+ select {
+ case <-ndpDisp.prefixC:
+ t.Error("unexpected prefix event")
+ default:
+ }
+ select {
+ case <-ndpDisp.autoGenAddrC:
+ t.Error("unexpected auto-generated address event")
+ default:
+ }
+ })
+ }
+}
+
+// TestDHCPv6ConfigurationFromNDPDA tests that the NDPDispatcher is properly
+// informed when new information about what configurations are available via
+// DHCPv6 is learned.
+func TestDHCPv6ConfigurationFromNDPDA(t *testing.T) {
+ const nicID = 1
+
+ ndpDisp := ndpDispatcher{
+ dhcpv6ConfigurationC: make(chan ndpDHCPv6Event, 1),
+ rememberRouter: true,
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ HandleRAs: true,
+ },
+ NDPDisp: &ndpDisp,
+ })
+
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+
+ expectDHCPv6Event := func(configuration stack.DHCPv6ConfigurationFromNDPRA) {
+ t.Helper()
+ select {
+ case e := <-ndpDisp.dhcpv6ConfigurationC:
+ if diff := cmp.Diff(ndpDHCPv6Event{nicID: nicID, configuration: configuration}, e, cmp.AllowUnexported(e)); diff != "" {
+ t.Errorf("dhcpv6 event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected DHCPv6 configuration event")
+ }
+ }
+
+ expectNoDHCPv6Event := func() {
+ t.Helper()
+ select {
+ case <-ndpDisp.dhcpv6ConfigurationC:
+ t.Fatal("unexpected DHCPv6 configuration event")
+ default:
+ }
+ }
+
+ // Even if the first RA reports no DHCPv6 configurations are available, the
+ // dispatcher should get an event.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, false))
+ expectDHCPv6Event(stack.DHCPv6NoConfiguration)
+ // Receiving the same update again should not result in an event to the
+ // dispatcher.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, false))
+ expectNoDHCPv6Event()
+
+ // Receive an RA that updates the DHCPv6 configuration to Other
+ // Configurations.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true))
+ expectDHCPv6Event(stack.DHCPv6OtherConfigurations)
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true))
+ expectNoDHCPv6Event()
+
+ // Receive an RA that updates the DHCPv6 configuration to Managed Address.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, true, false))
+ expectDHCPv6Event(stack.DHCPv6ManagedAddress)
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, true, false))
+ expectNoDHCPv6Event()
+
+ // Receive an RA that updates the DHCPv6 configuration to none.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, false))
+ expectDHCPv6Event(stack.DHCPv6NoConfiguration)
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, false))
+ expectNoDHCPv6Event()
+
+ // Receive an RA that updates the DHCPv6 configuration to Managed Address.
+ //
+ // Note, when the M flag is set, the O flag is redundant.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, true, true))
+ expectDHCPv6Event(stack.DHCPv6ManagedAddress)
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, true, true))
+ expectNoDHCPv6Event()
+ // Even though the DHCPv6 flags are different, the effective configuration is
+ // the same so we should not receive a new event.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, true, false))
+ expectNoDHCPv6Event()
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, true, true))
+ expectNoDHCPv6Event()
+
+ // Receive an RA that updates the DHCPv6 configuration to Other
+ // Configurations.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true))
+ expectDHCPv6Event(stack.DHCPv6OtherConfigurations)
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true))
+ expectNoDHCPv6Event()
+
+ // Cycling the NIC should cause the last DHCPv6 configuration to be cleared.
+ if err := s.DisableNIC(nicID); err != nil {
+ t.Fatalf("s.DisableNIC(%d): %s", nicID, err)
+ }
+ if err := s.EnableNIC(nicID); err != nil {
+ t.Fatalf("s.EnableNIC(%d): %s", nicID, err)
+ }
+
+ // Receive an RA that updates the DHCPv6 configuration to Other
+ // Configurations.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true))
+ expectDHCPv6Event(stack.DHCPv6OtherConfigurations)
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true))
+ expectNoDHCPv6Event()
+}
+
+// TestRouterSolicitation tests the initial Router Solicitations that are sent
+// when a NIC newly becomes enabled.
+func TestRouterSolicitation(t *testing.T) {
+ const nicID = 1
+
+ tests := []struct {
+ name string
+ linkHeaderLen uint16
+ linkAddr tcpip.LinkAddress
+ nicAddr tcpip.Address
+ expectedSrcAddr tcpip.Address
+ expectedNDPOpts []header.NDPOption
+ maxRtrSolicit uint8
+ rtrSolicitInt time.Duration
+ effectiveRtrSolicitInt time.Duration
+ maxRtrSolicitDelay time.Duration
+ effectiveMaxRtrSolicitDelay time.Duration
+ }{
+ {
+ name: "Single RS with 2s delay and interval",
+ expectedSrcAddr: header.IPv6Any,
+ maxRtrSolicit: 1,
+ rtrSolicitInt: 2 * time.Second,
+ effectiveRtrSolicitInt: 2 * time.Second,
+ maxRtrSolicitDelay: 2 * time.Second,
+ effectiveMaxRtrSolicitDelay: 2 * time.Second,
+ },
+ {
+ name: "Single RS with 4s delay and interval",
+ expectedSrcAddr: header.IPv6Any,
+ maxRtrSolicit: 1,
+ rtrSolicitInt: 4 * time.Second,
+ effectiveRtrSolicitInt: 4 * time.Second,
+ maxRtrSolicitDelay: 4 * time.Second,
+ effectiveMaxRtrSolicitDelay: 4 * time.Second,
+ },
+ {
+ name: "Two RS with delay",
+ linkHeaderLen: 1,
+ nicAddr: llAddr1,
+ expectedSrcAddr: llAddr1,
+ maxRtrSolicit: 2,
+ rtrSolicitInt: 2 * time.Second,
+ effectiveRtrSolicitInt: 2 * time.Second,
+ maxRtrSolicitDelay: 500 * time.Millisecond,
+ effectiveMaxRtrSolicitDelay: 500 * time.Millisecond,
+ },
+ {
+ name: "Single RS without delay",
+ linkHeaderLen: 2,
+ linkAddr: linkAddr1,
+ nicAddr: llAddr1,
+ expectedSrcAddr: llAddr1,
+ expectedNDPOpts: []header.NDPOption{
+ header.NDPSourceLinkLayerAddressOption(linkAddr1),
+ },
+ maxRtrSolicit: 1,
+ rtrSolicitInt: 2 * time.Second,
+ effectiveRtrSolicitInt: 2 * time.Second,
+ maxRtrSolicitDelay: 0,
+ effectiveMaxRtrSolicitDelay: 0,
+ },
+ {
+ name: "Two RS without delay and invalid zero interval",
+ linkHeaderLen: 3,
+ linkAddr: linkAddr1,
+ expectedSrcAddr: header.IPv6Any,
+ maxRtrSolicit: 2,
+ rtrSolicitInt: 0,
+ effectiveRtrSolicitInt: 4 * time.Second,
+ maxRtrSolicitDelay: 0,
+ effectiveMaxRtrSolicitDelay: 0,
+ },
+ {
+ name: "Three RS without delay",
+ linkAddr: linkAddr1,
+ expectedSrcAddr: header.IPv6Any,
+ maxRtrSolicit: 3,
+ rtrSolicitInt: 500 * time.Millisecond,
+ effectiveRtrSolicitInt: 500 * time.Millisecond,
+ maxRtrSolicitDelay: 0,
+ effectiveMaxRtrSolicitDelay: 0,
+ },
+ {
+ name: "Two RS with invalid negative delay",
+ linkAddr: linkAddr1,
+ expectedSrcAddr: header.IPv6Any,
+ maxRtrSolicit: 2,
+ rtrSolicitInt: time.Second,
+ effectiveRtrSolicitInt: time.Second,
+ maxRtrSolicitDelay: -3 * time.Second,
+ effectiveMaxRtrSolicitDelay: time.Second,
+ },
+ }
+
+ // This Run will not return until the parallel tests finish.
+ //
+ // We need this because we need to do some teardown work after the
+ // parallel tests complete.
+ //
+ // See https://godoc.org/testing#hdr-Subtests_and_Sub_benchmarks for
+ // more details.
+ t.Run("group", func(t *testing.T) {
+ for _, test := range tests {
+ test := test
+
+ t.Run(test.name, func(t *testing.T) {
+ t.Parallel()
+
+ e := channelLinkWithHeaderLength{
+ Endpoint: channel.New(int(test.maxRtrSolicit), 1280, test.linkAddr),
+ headerLength: test.linkHeaderLen,
+ }
+ e.Endpoint.LinkEPCapabilities |= stack.CapabilityResolutionRequired
+ waitForPkt := func(timeout time.Duration) {
+ t.Helper()
+ ctx, cancel := context.WithTimeout(context.Background(), timeout)
+ defer cancel()
+ p, ok := e.ReadContext(ctx)
+ if !ok {
+ t.Fatal("timed out waiting for packet")
+ return
+ }
+
+ if p.Proto != header.IPv6ProtocolNumber {
+ t.Fatalf("got Proto = %d, want = %d", p.Proto, header.IPv6ProtocolNumber)
+ }
+
+ // Make sure the right remote link address is used.
+ if want := header.EthernetAddressFromMulticastIPv6Address(header.IPv6AllRoutersMulticastAddress); p.Route.RemoteLinkAddress != want {
+ t.Errorf("got remote link address = %s, want = %s", p.Route.RemoteLinkAddress, want)
+ }
+
+ checker.IPv6(t, stack.PayloadSince(p.Pkt.NetworkHeader()),
+ checker.SrcAddr(test.expectedSrcAddr),
+ checker.DstAddr(header.IPv6AllRoutersMulticastAddress),
+ checker.TTL(header.NDPHopLimit),
+ checker.NDPRS(checker.NDPRSOptions(test.expectedNDPOpts)),
+ )
+
+ if l, want := p.Pkt.AvailableHeaderBytes(), int(test.linkHeaderLen); l != want {
+ t.Errorf("got p.Pkt.AvailableHeaderBytes() = %d; want = %d", l, want)
+ }
+ }
+ waitForNothing := func(timeout time.Duration) {
+ t.Helper()
+ ctx, cancel := context.WithTimeout(context.Background(), timeout)
+ defer cancel()
+ if _, ok := e.ReadContext(ctx); ok {
+ t.Fatal("unexpectedly got a packet")
+ }
+ }
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ MaxRtrSolicitations: test.maxRtrSolicit,
+ RtrSolicitationInterval: test.rtrSolicitInt,
+ MaxRtrSolicitationDelay: test.maxRtrSolicitDelay,
+ },
+ })
+ if err := s.CreateNIC(nicID, &e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+
+ if addr := test.nicAddr; addr != "" {
+ if err := s.AddAddress(nicID, header.IPv6ProtocolNumber, addr); err != nil {
+ t.Fatalf("AddAddress(%d, %d, %s) = %s", nicID, header.IPv6ProtocolNumber, addr, err)
+ }
+ }
+
+ // Make sure each RS is sent at the right time.
+ remaining := test.maxRtrSolicit
+ if remaining > 0 {
+ waitForPkt(test.effectiveMaxRtrSolicitDelay + defaultAsyncPositiveEventTimeout)
+ remaining--
+ }
+
+ for ; remaining > 0; remaining-- {
+ if test.effectiveRtrSolicitInt > defaultAsyncPositiveEventTimeout {
+ waitForNothing(test.effectiveRtrSolicitInt - defaultAsyncNegativeEventTimeout)
+ waitForPkt(defaultAsyncPositiveEventTimeout)
+ } else {
+ waitForPkt(test.effectiveRtrSolicitInt + defaultAsyncPositiveEventTimeout)
+ }
+ }
+
+ // Make sure no more RS.
+ if test.effectiveRtrSolicitInt > test.effectiveMaxRtrSolicitDelay {
+ waitForNothing(test.effectiveRtrSolicitInt + defaultAsyncNegativeEventTimeout)
+ } else {
+ waitForNothing(test.effectiveMaxRtrSolicitDelay + defaultAsyncNegativeEventTimeout)
+ }
+
+ // Make sure the counter got properly
+ // incremented.
+ if got, want := s.Stats().ICMP.V6PacketsSent.RouterSolicit.Value(), uint64(test.maxRtrSolicit); got != want {
+ t.Fatalf("got sent RouterSolicit = %d, want = %d", got, want)
+ }
+ })
+ }
+ })
+}
+
+func TestStopStartSolicitingRouters(t *testing.T) {
+ const nicID = 1
+ const delay = 0
+ const interval = 500 * time.Millisecond
+ const maxRtrSolicitations = 3
+
+ tests := []struct {
+ name string
+ startFn func(t *testing.T, s *stack.Stack)
+ // first is used to tell stopFn that it is being called for the first time
+ // after router solicitations were last enabled.
+ stopFn func(t *testing.T, s *stack.Stack, first bool)
+ }{
+ // Tests that when forwarding is enabled or disabled, router solicitations
+ // are stopped or started, respectively.
+ {
+ name: "Enable and disable forwarding",
+ startFn: func(t *testing.T, s *stack.Stack) {
+ t.Helper()
+ s.SetForwarding(ipv6.ProtocolNumber, false)
+ },
+ stopFn: func(t *testing.T, s *stack.Stack, _ bool) {
+ t.Helper()
+ s.SetForwarding(ipv6.ProtocolNumber, true)
+ },
+ },
+
+ // Tests that when a NIC is enabled or disabled, router solicitations
+ // are started or stopped, respectively.
+ {
+ name: "Enable and disable NIC",
+ startFn: func(t *testing.T, s *stack.Stack) {
+ t.Helper()
+
+ if err := s.EnableNIC(nicID); err != nil {
+ t.Fatalf("s.EnableNIC(%d): %s", nicID, err)
+ }
+ },
+ stopFn: func(t *testing.T, s *stack.Stack, _ bool) {
+ t.Helper()
+
+ if err := s.DisableNIC(nicID); err != nil {
+ t.Fatalf("s.DisableNIC(%d): %s", nicID, err)
+ }
+ },
+ },
+
+ // Tests that when a NIC is removed, router solicitations are stopped. We
+ // cannot start router solications on a removed NIC.
+ {
+ name: "Remove NIC",
+ stopFn: func(t *testing.T, s *stack.Stack, first bool) {
+ t.Helper()
+
+ // Only try to remove the NIC the first time stopFn is called since it's
+ // impossible to remove an already removed NIC.
+ if !first {
+ return
+ }
+
+ if err := s.RemoveNIC(nicID); err != nil {
+ t.Fatalf("s.RemoveNIC(%d): %s", nicID, err)
+ }
+ },
+ },
+ }
+
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ e := channel.New(maxRtrSolicitations, 1280, linkAddr1)
+ waitForPkt := func(timeout time.Duration) {
+ t.Helper()
+
+ ctx, cancel := context.WithTimeout(context.Background(), timeout)
+ defer cancel()
+ p, ok := e.ReadContext(ctx)
+ if !ok {
+ t.Fatal("timed out waiting for packet")
+ }
+
+ if p.Proto != header.IPv6ProtocolNumber {
+ t.Fatalf("got Proto = %d, want = %d", p.Proto, header.IPv6ProtocolNumber)
+ }
+ checker.IPv6(t, stack.PayloadSince(p.Pkt.NetworkHeader()),
+ checker.SrcAddr(header.IPv6Any),
+ checker.DstAddr(header.IPv6AllRoutersMulticastAddress),
+ checker.TTL(header.NDPHopLimit),
+ checker.NDPRS())
+ }
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ MaxRtrSolicitations: maxRtrSolicitations,
+ RtrSolicitationInterval: interval,
+ MaxRtrSolicitationDelay: delay,
+ },
+ })
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+
+ // Stop soliciting routers.
+ test.stopFn(t, s, true /* first */)
+ ctx, cancel := context.WithTimeout(context.Background(), delay+defaultAsyncNegativeEventTimeout)
+ defer cancel()
+ if _, ok := e.ReadContext(ctx); ok {
+ // A single RS may have been sent before solicitations were stopped.
+ ctx, cancel := context.WithTimeout(context.Background(), interval+defaultAsyncNegativeEventTimeout)
+ defer cancel()
+ if _, ok = e.ReadContext(ctx); ok {
+ t.Fatal("should not have sent more than one RS message")
+ }
+ }
+
+ // Stopping router solicitations after it has already been stopped should
+ // do nothing.
+ test.stopFn(t, s, false /* first */)
+ ctx, cancel = context.WithTimeout(context.Background(), delay+defaultAsyncNegativeEventTimeout)
+ defer cancel()
+ if _, ok := e.ReadContext(ctx); ok {
+ t.Fatal("unexpectedly got a packet after router solicitation has been stopepd")
+ }
+
+ // If test.startFn is nil, there is no way to restart router solications.
+ if test.startFn == nil {
+ return
+ }
+
+ // Start soliciting routers.
+ test.startFn(t, s)
+ waitForPkt(delay + defaultAsyncPositiveEventTimeout)
+ waitForPkt(interval + defaultAsyncPositiveEventTimeout)
+ waitForPkt(interval + defaultAsyncPositiveEventTimeout)
+ ctx, cancel = context.WithTimeout(context.Background(), interval+defaultAsyncNegativeEventTimeout)
+ defer cancel()
+ if _, ok := e.ReadContext(ctx); ok {
+ t.Fatal("unexpectedly got an extra packet after sending out the expected RSs")
+ }
+
+ // Starting router solicitations after it has already completed should do
+ // nothing.
+ test.startFn(t, s)
+ ctx, cancel = context.WithTimeout(context.Background(), delay+defaultAsyncNegativeEventTimeout)
+ defer cancel()
+ if _, ok := e.ReadContext(ctx); ok {
+ t.Fatal("unexpectedly got a packet after finishing router solicitations")
}
})
}
diff --git a/pkg/tcpip/stack/neighbor_cache.go b/pkg/tcpip/stack/neighbor_cache.go
new file mode 100644
index 000000000..27e1feec0
--- /dev/null
+++ b/pkg/tcpip/stack/neighbor_cache.go
@@ -0,0 +1,333 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package stack
+
+import (
+ "fmt"
+ "time"
+
+ "gvisor.dev/gvisor/pkg/sleep"
+ "gvisor.dev/gvisor/pkg/sync"
+ "gvisor.dev/gvisor/pkg/tcpip"
+)
+
+const neighborCacheSize = 512 // max entries per interface
+
+// neighborCache maps IP addresses to link addresses. It uses the Least
+// Recently Used (LRU) eviction strategy to implement a bounded cache for
+// dynmically acquired entries. It contains the state machine and configuration
+// for running Neighbor Unreachability Detection (NUD).
+//
+// There are two types of entries in the neighbor cache:
+// 1. Dynamic entries are discovered automatically by neighbor discovery
+// protocols (e.g. ARP, NDP). These protocols will attempt to reconfirm
+// reachability with the device once the entry's state becomes Stale.
+// 2. Static entries are explicitly added by a user and have no expiration.
+// Their state is always Static. The amount of static entries stored in the
+// cache is unbounded.
+//
+// neighborCache implements NUDHandler.
+type neighborCache struct {
+ nic *NIC
+ state *NUDState
+
+ // mu protects the fields below.
+ mu sync.RWMutex
+
+ cache map[tcpip.Address]*neighborEntry
+ dynamic struct {
+ lru neighborEntryList
+
+ // count tracks the amount of dynamic entries in the cache. This is
+ // needed since static entries do not count towards the LRU cache
+ // eviction strategy.
+ count uint16
+ }
+}
+
+var _ NUDHandler = (*neighborCache)(nil)
+
+// getOrCreateEntry retrieves a cache entry associated with addr. The
+// returned entry is always refreshed in the cache (it is reachable via the
+// map, and its place is bumped in LRU).
+//
+// If a matching entry exists in the cache, it is returned. If no matching
+// entry exists and the cache is full, an existing entry is evicted via LRU,
+// reset to state incomplete, and returned. If no matching entry exists and the
+// cache is not full, a new entry with state incomplete is allocated and
+// returned.
+func (n *neighborCache) getOrCreateEntry(remoteAddr, localAddr tcpip.Address, linkRes LinkAddressResolver) *neighborEntry {
+ n.mu.Lock()
+ defer n.mu.Unlock()
+
+ if entry, ok := n.cache[remoteAddr]; ok {
+ entry.mu.RLock()
+ if entry.neigh.State != Static {
+ n.dynamic.lru.Remove(entry)
+ n.dynamic.lru.PushFront(entry)
+ }
+ entry.mu.RUnlock()
+ return entry
+ }
+
+ // The entry that needs to be created must be dynamic since all static
+ // entries are directly added to the cache via addStaticEntry.
+ entry := newNeighborEntry(n.nic, remoteAddr, localAddr, n.state, linkRes)
+ if n.dynamic.count == neighborCacheSize {
+ e := n.dynamic.lru.Back()
+ e.mu.Lock()
+
+ delete(n.cache, e.neigh.Addr)
+ n.dynamic.lru.Remove(e)
+ n.dynamic.count--
+
+ e.dispatchRemoveEventLocked()
+ e.setStateLocked(Unknown)
+ e.notifyWakersLocked()
+ e.mu.Unlock()
+ }
+ n.cache[remoteAddr] = entry
+ n.dynamic.lru.PushFront(entry)
+ n.dynamic.count++
+ return entry
+}
+
+// entry looks up the neighbor cache for translating address to link address
+// (e.g. IP -> MAC). If the LinkEndpoint requests address resolution and there
+// is a LinkAddressResolver registered with the network protocol, the cache
+// attempts to resolve the address and returns ErrWouldBlock. If a Waker is
+// provided, it will be notified when address resolution is complete (success
+// or not).
+//
+// If address resolution is required, ErrNoLinkAddress and a notification
+// channel is returned for the top level caller to block. Channel is closed
+// once address resolution is complete (success or not).
+func (n *neighborCache) entry(remoteAddr, localAddr tcpip.Address, linkRes LinkAddressResolver, w *sleep.Waker) (NeighborEntry, <-chan struct{}, *tcpip.Error) {
+ if linkAddr, ok := linkRes.ResolveStaticAddress(remoteAddr); ok {
+ e := NeighborEntry{
+ Addr: remoteAddr,
+ LocalAddr: localAddr,
+ LinkAddr: linkAddr,
+ State: Static,
+ UpdatedAt: time.Now(),
+ }
+ return e, nil, nil
+ }
+
+ entry := n.getOrCreateEntry(remoteAddr, localAddr, linkRes)
+ entry.mu.Lock()
+ defer entry.mu.Unlock()
+
+ switch s := entry.neigh.State; s {
+ case Reachable, Static:
+ return entry.neigh, nil, nil
+
+ case Unknown, Incomplete, Stale, Delay, Probe:
+ entry.addWakerLocked(w)
+
+ if entry.done == nil {
+ // Address resolution needs to be initiated.
+ if linkRes == nil {
+ return entry.neigh, nil, tcpip.ErrNoLinkAddress
+ }
+ entry.done = make(chan struct{})
+ }
+
+ entry.handlePacketQueuedLocked()
+ return entry.neigh, entry.done, tcpip.ErrWouldBlock
+
+ case Failed:
+ return entry.neigh, nil, tcpip.ErrNoLinkAddress
+
+ default:
+ panic(fmt.Sprintf("Invalid cache entry state: %s", s))
+ }
+}
+
+// removeWaker removes a waker that has been added when link resolution for
+// addr was requested.
+func (n *neighborCache) removeWaker(addr tcpip.Address, waker *sleep.Waker) {
+ n.mu.Lock()
+ if entry, ok := n.cache[addr]; ok {
+ delete(entry.wakers, waker)
+ }
+ n.mu.Unlock()
+}
+
+// entries returns all entries in the neighbor cache.
+func (n *neighborCache) entries() []NeighborEntry {
+ entries := make([]NeighborEntry, 0, len(n.cache))
+ n.mu.RLock()
+ for _, entry := range n.cache {
+ entry.mu.RLock()
+ entries = append(entries, entry.neigh)
+ entry.mu.RUnlock()
+ }
+ n.mu.RUnlock()
+ return entries
+}
+
+// addStaticEntry adds a static entry to the neighbor cache, mapping an IP
+// address to a link address. If a dynamic entry exists in the neighbor cache
+// with the same address, it will be replaced with this static entry. If a
+// static entry exists with the same address but different link address, it
+// will be updated with the new link address. If a static entry exists with the
+// same address and link address, nothing will happen.
+func (n *neighborCache) addStaticEntry(addr tcpip.Address, linkAddr tcpip.LinkAddress) {
+ n.mu.Lock()
+ defer n.mu.Unlock()
+
+ if entry, ok := n.cache[addr]; ok {
+ entry.mu.Lock()
+ if entry.neigh.State != Static {
+ // Dynamic entry found with the same address.
+ n.dynamic.lru.Remove(entry)
+ n.dynamic.count--
+ } else if entry.neigh.LinkAddr == linkAddr {
+ // Static entry found with the same address and link address.
+ entry.mu.Unlock()
+ return
+ } else {
+ // Static entry found with the same address but different link address.
+ entry.neigh.LinkAddr = linkAddr
+ entry.dispatchChangeEventLocked(entry.neigh.State)
+ entry.mu.Unlock()
+ return
+ }
+
+ // Notify that resolution has been interrupted, just in case the entry was
+ // in the Incomplete or Probe state.
+ entry.dispatchRemoveEventLocked()
+ entry.setStateLocked(Unknown)
+ entry.notifyWakersLocked()
+ entry.mu.Unlock()
+ }
+
+ entry := newStaticNeighborEntry(n.nic, addr, linkAddr, n.state)
+ n.cache[addr] = entry
+}
+
+// removeEntryLocked removes the specified entry from the neighbor cache.
+func (n *neighborCache) removeEntryLocked(entry *neighborEntry) {
+ if entry.neigh.State != Static {
+ n.dynamic.lru.Remove(entry)
+ n.dynamic.count--
+ }
+ if entry.neigh.State != Failed {
+ entry.dispatchRemoveEventLocked()
+ }
+ entry.setStateLocked(Unknown)
+ entry.notifyWakersLocked()
+
+ delete(n.cache, entry.neigh.Addr)
+}
+
+// removeEntry removes a dynamic or static entry by address from the neighbor
+// cache. Returns true if the entry was found and deleted.
+func (n *neighborCache) removeEntry(addr tcpip.Address) bool {
+ n.mu.Lock()
+ defer n.mu.Unlock()
+
+ entry, ok := n.cache[addr]
+ if !ok {
+ return false
+ }
+
+ entry.mu.Lock()
+ defer entry.mu.Unlock()
+
+ n.removeEntryLocked(entry)
+ return true
+}
+
+// clear removes all dynamic and static entries from the neighbor cache.
+func (n *neighborCache) clear() {
+ n.mu.Lock()
+ defer n.mu.Unlock()
+
+ for _, entry := range n.cache {
+ entry.mu.Lock()
+ entry.dispatchRemoveEventLocked()
+ entry.setStateLocked(Unknown)
+ entry.notifyWakersLocked()
+ entry.mu.Unlock()
+ }
+
+ n.dynamic.lru = neighborEntryList{}
+ n.cache = make(map[tcpip.Address]*neighborEntry)
+ n.dynamic.count = 0
+}
+
+// config returns the NUD configuration.
+func (n *neighborCache) config() NUDConfigurations {
+ return n.state.Config()
+}
+
+// setConfig changes the NUD configuration.
+//
+// If config contains invalid NUD configuration values, it will be fixed to
+// use default values for the erroneous values.
+func (n *neighborCache) setConfig(config NUDConfigurations) {
+ config.resetInvalidFields()
+ n.state.SetConfig(config)
+}
+
+// HandleProbe implements NUDHandler.HandleProbe by following the logic defined
+// in RFC 4861 section 7.2.3. Validation of the probe is expected to be handled
+// by the caller.
+func (n *neighborCache) HandleProbe(remoteAddr, localAddr tcpip.Address, protocol tcpip.NetworkProtocolNumber, remoteLinkAddr tcpip.LinkAddress, linkRes LinkAddressResolver) {
+ entry := n.getOrCreateEntry(remoteAddr, localAddr, linkRes)
+ entry.mu.Lock()
+ entry.handleProbeLocked(remoteLinkAddr)
+ entry.mu.Unlock()
+}
+
+// HandleConfirmation implements NUDHandler.HandleConfirmation by following the
+// logic defined in RFC 4861 section 7.2.5.
+//
+// TODO(gvisor.dev/issue/2277): To protect against ARP poisoning and other
+// attacks against NDP functions, Secure Neighbor Discovery (SEND) Protocol
+// should be deployed where preventing access to the broadcast segment might
+// not be possible. SEND uses RSA key pairs to produce cryptographically
+// generated addresses, as defined in RFC 3972, Cryptographically Generated
+// Addresses (CGA). This ensures that the claimed source of an NDP message is
+// the owner of the claimed address.
+func (n *neighborCache) HandleConfirmation(addr tcpip.Address, linkAddr tcpip.LinkAddress, flags ReachabilityConfirmationFlags) {
+ n.mu.RLock()
+ entry, ok := n.cache[addr]
+ n.mu.RUnlock()
+ if ok {
+ entry.mu.Lock()
+ entry.handleConfirmationLocked(linkAddr, flags)
+ entry.mu.Unlock()
+ }
+ // The confirmation SHOULD be silently discarded if the recipient did not
+ // initiate any communication with the target. This is indicated if there is
+ // no matching entry for the remote address.
+}
+
+// HandleUpperLevelConfirmation implements
+// NUDHandler.HandleUpperLevelConfirmation by following the logic defined in
+// RFC 4861 section 7.3.1.
+func (n *neighborCache) HandleUpperLevelConfirmation(addr tcpip.Address) {
+ n.mu.RLock()
+ entry, ok := n.cache[addr]
+ n.mu.RUnlock()
+ if ok {
+ entry.mu.Lock()
+ entry.handleUpperLevelConfirmationLocked()
+ entry.mu.Unlock()
+ }
+}
diff --git a/pkg/tcpip/stack/neighbor_cache_test.go b/pkg/tcpip/stack/neighbor_cache_test.go
new file mode 100644
index 000000000..b4fa69e3e
--- /dev/null
+++ b/pkg/tcpip/stack/neighbor_cache_test.go
@@ -0,0 +1,1726 @@
+// Copyright 2019 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package stack
+
+import (
+ "bytes"
+ "encoding/binary"
+ "fmt"
+ "math"
+ "math/rand"
+ "strings"
+ "sync"
+ "sync/atomic"
+ "testing"
+ "time"
+
+ "github.com/google/go-cmp/cmp"
+ "github.com/google/go-cmp/cmp/cmpopts"
+ "gvisor.dev/gvisor/pkg/sleep"
+ "gvisor.dev/gvisor/pkg/tcpip"
+)
+
+const (
+ // entryStoreSize is the default number of entries that will be generated and
+ // added to the entry store. This number needs to be larger than the size of
+ // the neighbor cache to give ample opportunity for verifying behavior during
+ // cache overflows. Four times the size of the neighbor cache allows for
+ // three complete cache overflows.
+ entryStoreSize = 4 * neighborCacheSize
+
+ // typicalLatency is the typical latency for an ARP or NDP packet to travel
+ // to a router and back.
+ typicalLatency = time.Millisecond
+
+ // testEntryBroadcastAddr is a special address that indicates a packet should
+ // be sent to all nodes.
+ testEntryBroadcastAddr = tcpip.Address("broadcast")
+
+ // testEntryLocalAddr is the source address of neighbor probes.
+ testEntryLocalAddr = tcpip.Address("local_addr")
+
+ // testEntryBroadcastLinkAddr is a special link address sent back to
+ // multicast neighbor probes.
+ testEntryBroadcastLinkAddr = tcpip.LinkAddress("mac_broadcast")
+
+ // infiniteDuration indicates that a task will not occur in our lifetime.
+ infiniteDuration = time.Duration(math.MaxInt64)
+)
+
+// entryDiffOpts returns the options passed to cmp.Diff to compare neighbor
+// entries. The UpdatedAt field is ignored due to a lack of a deterministic
+// method to predict the time that an event will be dispatched.
+func entryDiffOpts() []cmp.Option {
+ return []cmp.Option{
+ cmpopts.IgnoreFields(NeighborEntry{}, "UpdatedAt"),
+ }
+}
+
+// entryDiffOptsWithSort is like entryDiffOpts but also includes an option to
+// sort slices of entries for cases where ordering must be ignored.
+func entryDiffOptsWithSort() []cmp.Option {
+ return []cmp.Option{
+ cmpopts.IgnoreFields(NeighborEntry{}, "UpdatedAt"),
+ cmpopts.SortSlices(func(a, b NeighborEntry) bool {
+ return strings.Compare(string(a.Addr), string(b.Addr)) < 0
+ }),
+ }
+}
+
+func newTestNeighborCache(nudDisp NUDDispatcher, config NUDConfigurations, clock tcpip.Clock) *neighborCache {
+ config.resetInvalidFields()
+ rng := rand.New(rand.NewSource(time.Now().UnixNano()))
+ return &neighborCache{
+ nic: &NIC{
+ stack: &Stack{
+ clock: clock,
+ nudDisp: nudDisp,
+ },
+ id: 1,
+ },
+ state: NewNUDState(config, rng),
+ cache: make(map[tcpip.Address]*neighborEntry, neighborCacheSize),
+ }
+}
+
+// testEntryStore contains a set of IP to NeighborEntry mappings.
+type testEntryStore struct {
+ mu sync.RWMutex
+ entriesMap map[tcpip.Address]NeighborEntry
+}
+
+func toAddress(i int) tcpip.Address {
+ buf := new(bytes.Buffer)
+ binary.Write(buf, binary.BigEndian, uint8(1))
+ binary.Write(buf, binary.BigEndian, uint8(0))
+ binary.Write(buf, binary.BigEndian, uint16(i))
+ return tcpip.Address(buf.String())
+}
+
+func toLinkAddress(i int) tcpip.LinkAddress {
+ buf := new(bytes.Buffer)
+ binary.Write(buf, binary.BigEndian, uint8(1))
+ binary.Write(buf, binary.BigEndian, uint8(0))
+ binary.Write(buf, binary.BigEndian, uint32(i))
+ return tcpip.LinkAddress(buf.String())
+}
+
+// newTestEntryStore returns a testEntryStore pre-populated with entries.
+func newTestEntryStore() *testEntryStore {
+ store := &testEntryStore{
+ entriesMap: make(map[tcpip.Address]NeighborEntry),
+ }
+ for i := 0; i < entryStoreSize; i++ {
+ addr := toAddress(i)
+ linkAddr := toLinkAddress(i)
+
+ store.entriesMap[addr] = NeighborEntry{
+ Addr: addr,
+ LocalAddr: testEntryLocalAddr,
+ LinkAddr: linkAddr,
+ }
+ }
+ return store
+}
+
+// size returns the number of entries in the store.
+func (s *testEntryStore) size() int {
+ s.mu.RLock()
+ defer s.mu.RUnlock()
+ return len(s.entriesMap)
+}
+
+// entry returns the entry at index i. Returns an empty entry and false if i is
+// out of bounds.
+func (s *testEntryStore) entry(i int) (NeighborEntry, bool) {
+ return s.entryByAddr(toAddress(i))
+}
+
+// entryByAddr returns the entry matching addr for situations when the index is
+// not available. Returns an empty entry and false if no entries match addr.
+func (s *testEntryStore) entryByAddr(addr tcpip.Address) (NeighborEntry, bool) {
+ s.mu.RLock()
+ defer s.mu.RUnlock()
+ entry, ok := s.entriesMap[addr]
+ return entry, ok
+}
+
+// entries returns all entries in the store.
+func (s *testEntryStore) entries() []NeighborEntry {
+ entries := make([]NeighborEntry, 0, len(s.entriesMap))
+ s.mu.RLock()
+ defer s.mu.RUnlock()
+ for i := 0; i < entryStoreSize; i++ {
+ addr := toAddress(i)
+ if entry, ok := s.entriesMap[addr]; ok {
+ entries = append(entries, entry)
+ }
+ }
+ return entries
+}
+
+// set modifies the link addresses of an entry.
+func (s *testEntryStore) set(i int, linkAddr tcpip.LinkAddress) {
+ addr := toAddress(i)
+ s.mu.Lock()
+ defer s.mu.Unlock()
+ if entry, ok := s.entriesMap[addr]; ok {
+ entry.LinkAddr = linkAddr
+ s.entriesMap[addr] = entry
+ }
+}
+
+// testNeighborResolver implements LinkAddressResolver to emulate sending a
+// neighbor probe.
+type testNeighborResolver struct {
+ clock tcpip.Clock
+ neigh *neighborCache
+ entries *testEntryStore
+ delay time.Duration
+ onLinkAddressRequest func()
+}
+
+var _ LinkAddressResolver = (*testNeighborResolver)(nil)
+
+func (r *testNeighborResolver) LinkAddressRequest(addr, localAddr tcpip.Address, linkAddr tcpip.LinkAddress, linkEP LinkEndpoint) *tcpip.Error {
+ // Delay handling the request to emulate network latency.
+ r.clock.AfterFunc(r.delay, func() {
+ r.fakeRequest(addr)
+ })
+
+ // Execute post address resolution action, if available.
+ if f := r.onLinkAddressRequest; f != nil {
+ f()
+ }
+ return nil
+}
+
+// fakeRequest emulates handling a response for a link address request.
+func (r *testNeighborResolver) fakeRequest(addr tcpip.Address) {
+ if entry, ok := r.entries.entryByAddr(addr); ok {
+ r.neigh.HandleConfirmation(addr, entry.LinkAddr, ReachabilityConfirmationFlags{
+ Solicited: true,
+ Override: false,
+ IsRouter: false,
+ })
+ }
+}
+
+func (*testNeighborResolver) ResolveStaticAddress(addr tcpip.Address) (tcpip.LinkAddress, bool) {
+ if addr == testEntryBroadcastAddr {
+ return testEntryBroadcastLinkAddr, true
+ }
+ return "", false
+}
+
+func (*testNeighborResolver) LinkAddressProtocol() tcpip.NetworkProtocolNumber {
+ return 0
+}
+
+type entryEvent struct {
+ nicID tcpip.NICID
+ address tcpip.Address
+ linkAddr tcpip.LinkAddress
+ state NeighborState
+}
+
+func TestNeighborCacheGetConfig(t *testing.T) {
+ nudDisp := testNUDDispatcher{}
+ c := DefaultNUDConfigurations()
+ clock := newFakeClock()
+ neigh := newTestNeighborCache(&nudDisp, c, clock)
+
+ if got, want := neigh.config(), c; got != want {
+ t.Errorf("got neigh.config() = %+v, want = %+v", got, want)
+ }
+
+ // No events should have been dispatched.
+ nudDisp.mu.Lock()
+ defer nudDisp.mu.Unlock()
+ if diff := cmp.Diff(nudDisp.events, []testEntryEventInfo(nil)); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+}
+
+func TestNeighborCacheSetConfig(t *testing.T) {
+ nudDisp := testNUDDispatcher{}
+ c := DefaultNUDConfigurations()
+ clock := newFakeClock()
+ neigh := newTestNeighborCache(&nudDisp, c, clock)
+
+ c.MinRandomFactor = 1
+ c.MaxRandomFactor = 1
+ neigh.setConfig(c)
+
+ if got, want := neigh.config(), c; got != want {
+ t.Errorf("got neigh.config() = %+v, want = %+v", got, want)
+ }
+
+ // No events should have been dispatched.
+ nudDisp.mu.Lock()
+ defer nudDisp.mu.Unlock()
+ if diff := cmp.Diff(nudDisp.events, []testEntryEventInfo(nil)); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+}
+
+func TestNeighborCacheEntry(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ nudDisp := testNUDDispatcher{}
+ clock := newFakeClock()
+ neigh := newTestNeighborCache(&nudDisp, c, clock)
+ store := newTestEntryStore()
+ linkRes := &testNeighborResolver{
+ clock: clock,
+ neigh: neigh,
+ entries: store,
+ delay: typicalLatency,
+ }
+
+ entry, ok := store.entry(0)
+ if !ok {
+ t.Fatalf("store.entry(0) not found")
+ }
+ _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil)
+ if err != tcpip.ErrWouldBlock {
+ t.Errorf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock)
+ }
+
+ clock.advance(typicalLatency)
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: 1,
+ Addr: entry.Addr,
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: 1,
+ Addr: entry.Addr,
+ LinkAddr: entry.LinkAddr,
+ State: Reachable,
+ },
+ }
+ nudDisp.mu.Lock()
+ diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...)
+ nudDisp.events = nil
+ nudDisp.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+
+ if _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil); err != nil {
+ t.Fatalf("unexpected error from neigh.entry(%s, %s, _, nil): %s", entry.Addr, entry.LocalAddr, err)
+ }
+
+ // No more events should have been dispatched.
+ nudDisp.mu.Lock()
+ defer nudDisp.mu.Unlock()
+ if diff := cmp.Diff(nudDisp.events, []testEntryEventInfo(nil)); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+}
+
+func TestNeighborCacheRemoveEntry(t *testing.T) {
+ config := DefaultNUDConfigurations()
+
+ nudDisp := testNUDDispatcher{}
+ clock := newFakeClock()
+ neigh := newTestNeighborCache(&nudDisp, config, clock)
+ store := newTestEntryStore()
+ linkRes := &testNeighborResolver{
+ clock: clock,
+ neigh: neigh,
+ entries: store,
+ delay: typicalLatency,
+ }
+
+ entry, ok := store.entry(0)
+ if !ok {
+ t.Fatalf("store.entry(0) not found")
+ }
+ _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil)
+ if err != tcpip.ErrWouldBlock {
+ t.Errorf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock)
+ }
+
+ clock.advance(typicalLatency)
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: 1,
+ Addr: entry.Addr,
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: 1,
+ Addr: entry.Addr,
+ LinkAddr: entry.LinkAddr,
+ State: Reachable,
+ },
+ }
+ nudDisp.mu.Lock()
+ diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...)
+ nudDisp.events = nil
+ nudDisp.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+
+ neigh.removeEntry(entry.Addr)
+
+ {
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestRemoved,
+ NICID: 1,
+ Addr: entry.Addr,
+ LinkAddr: entry.LinkAddr,
+ State: Reachable,
+ },
+ }
+ nudDisp.mu.Lock()
+ diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...)
+ nudDisp.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ }
+
+ if _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil); err != tcpip.ErrWouldBlock {
+ t.Errorf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock)
+ }
+}
+
+type testContext struct {
+ clock *fakeClock
+ neigh *neighborCache
+ store *testEntryStore
+ linkRes *testNeighborResolver
+ nudDisp *testNUDDispatcher
+}
+
+func newTestContext(c NUDConfigurations) testContext {
+ nudDisp := &testNUDDispatcher{}
+ clock := newFakeClock()
+ neigh := newTestNeighborCache(nudDisp, c, clock)
+ store := newTestEntryStore()
+ linkRes := &testNeighborResolver{
+ clock: clock,
+ neigh: neigh,
+ entries: store,
+ delay: typicalLatency,
+ }
+
+ return testContext{
+ clock: clock,
+ neigh: neigh,
+ store: store,
+ linkRes: linkRes,
+ nudDisp: nudDisp,
+ }
+}
+
+type overflowOptions struct {
+ startAtEntryIndex int
+ wantStaticEntries []NeighborEntry
+}
+
+func (c *testContext) overflowCache(opts overflowOptions) error {
+ // Fill the neighbor cache to capacity to verify the LRU eviction strategy is
+ // working properly after the entry removal.
+ for i := opts.startAtEntryIndex; i < c.store.size(); i++ {
+ // Add a new entry
+ entry, ok := c.store.entry(i)
+ if !ok {
+ return fmt.Errorf("c.store.entry(%d) not found", i)
+ }
+ if _, _, err := c.neigh.entry(entry.Addr, entry.LocalAddr, c.linkRes, nil); err != tcpip.ErrWouldBlock {
+ return fmt.Errorf("got c.neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock)
+ }
+ c.clock.advance(c.neigh.config().RetransmitTimer)
+
+ var wantEvents []testEntryEventInfo
+
+ // When beyond the full capacity, the cache will evict an entry as per the
+ // LRU eviction strategy. Note that the number of static entries should not
+ // affect the total number of dynamic entries that can be added.
+ if i >= neighborCacheSize+opts.startAtEntryIndex {
+ removedEntry, ok := c.store.entry(i - neighborCacheSize)
+ if !ok {
+ return fmt.Errorf("store.entry(%d) not found", i-neighborCacheSize)
+ }
+ wantEvents = append(wantEvents, testEntryEventInfo{
+ EventType: entryTestRemoved,
+ NICID: 1,
+ Addr: removedEntry.Addr,
+ LinkAddr: removedEntry.LinkAddr,
+ State: Reachable,
+ })
+ }
+
+ wantEvents = append(wantEvents, testEntryEventInfo{
+ EventType: entryTestAdded,
+ NICID: 1,
+ Addr: entry.Addr,
+ State: Incomplete,
+ }, testEntryEventInfo{
+ EventType: entryTestChanged,
+ NICID: 1,
+ Addr: entry.Addr,
+ LinkAddr: entry.LinkAddr,
+ State: Reachable,
+ })
+
+ c.nudDisp.mu.Lock()
+ diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...)
+ c.nudDisp.events = nil
+ c.nudDisp.mu.Unlock()
+ if diff != "" {
+ return fmt.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ }
+
+ // Expect to find only the most recent entries. The order of entries reported
+ // by entries() is undeterministic, so entries have to be sorted before
+ // comparison.
+ wantUnsortedEntries := opts.wantStaticEntries
+ for i := c.store.size() - neighborCacheSize; i < c.store.size(); i++ {
+ entry, ok := c.store.entry(i)
+ if !ok {
+ return fmt.Errorf("c.store.entry(%d) not found", i)
+ }
+ wantEntry := NeighborEntry{
+ Addr: entry.Addr,
+ LocalAddr: entry.LocalAddr,
+ LinkAddr: entry.LinkAddr,
+ State: Reachable,
+ }
+ wantUnsortedEntries = append(wantUnsortedEntries, wantEntry)
+ }
+
+ if diff := cmp.Diff(c.neigh.entries(), wantUnsortedEntries, entryDiffOptsWithSort()...); diff != "" {
+ return fmt.Errorf("neighbor entries mismatch (-got, +want):\n%s", diff)
+ }
+
+ // No more events should have been dispatched.
+ c.nudDisp.mu.Lock()
+ defer c.nudDisp.mu.Unlock()
+ if diff := cmp.Diff(c.nudDisp.events, []testEntryEventInfo(nil)); diff != "" {
+ return fmt.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+
+ return nil
+}
+
+// TestNeighborCacheOverflow verifies that the LRU cache eviction strategy
+// respects the dynamic entry count.
+func TestNeighborCacheOverflow(t *testing.T) {
+ config := DefaultNUDConfigurations()
+ // Stay in Reachable so the cache can overflow
+ config.BaseReachableTime = infiniteDuration
+ config.MinRandomFactor = 1
+ config.MaxRandomFactor = 1
+
+ c := newTestContext(config)
+ opts := overflowOptions{
+ startAtEntryIndex: 0,
+ }
+ if err := c.overflowCache(opts); err != nil {
+ t.Errorf("c.overflowCache(%+v): %s", opts, err)
+ }
+}
+
+// TestNeighborCacheRemoveEntryThenOverflow verifies that the LRU cache
+// eviction strategy respects the dynamic entry count when an entry is removed.
+func TestNeighborCacheRemoveEntryThenOverflow(t *testing.T) {
+ config := DefaultNUDConfigurations()
+ // Stay in Reachable so the cache can overflow
+ config.BaseReachableTime = infiniteDuration
+ config.MinRandomFactor = 1
+ config.MaxRandomFactor = 1
+
+ c := newTestContext(config)
+
+ // Add a dynamic entry
+ entry, ok := c.store.entry(0)
+ if !ok {
+ t.Fatalf("c.store.entry(0) not found")
+ }
+ _, _, err := c.neigh.entry(entry.Addr, entry.LocalAddr, c.linkRes, nil)
+ if err != tcpip.ErrWouldBlock {
+ t.Errorf("got c.neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock)
+ }
+ c.clock.advance(c.neigh.config().RetransmitTimer)
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: 1,
+ Addr: entry.Addr,
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: 1,
+ Addr: entry.Addr,
+ LinkAddr: entry.LinkAddr,
+ State: Reachable,
+ },
+ }
+ c.nudDisp.mu.Lock()
+ diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...)
+ c.nudDisp.events = nil
+ c.nudDisp.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+
+ // Remove the entry
+ c.neigh.removeEntry(entry.Addr)
+
+ {
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestRemoved,
+ NICID: 1,
+ Addr: entry.Addr,
+ LinkAddr: entry.LinkAddr,
+ State: Reachable,
+ },
+ }
+ c.nudDisp.mu.Lock()
+ diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...)
+ c.nudDisp.events = nil
+ c.nudDisp.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ }
+
+ opts := overflowOptions{
+ startAtEntryIndex: 0,
+ }
+ if err := c.overflowCache(opts); err != nil {
+ t.Errorf("c.overflowCache(%+v): %s", opts, err)
+ }
+}
+
+// TestNeighborCacheDuplicateStaticEntryWithSameLinkAddress verifies that
+// adding a duplicate static entry with the same link address does not dispatch
+// any events.
+func TestNeighborCacheDuplicateStaticEntryWithSameLinkAddress(t *testing.T) {
+ config := DefaultNUDConfigurations()
+ c := newTestContext(config)
+
+ // Add a static entry
+ entry, ok := c.store.entry(0)
+ if !ok {
+ t.Fatalf("c.store.entry(0) not found")
+ }
+ staticLinkAddr := entry.LinkAddr + "static"
+ c.neigh.addStaticEntry(entry.Addr, staticLinkAddr)
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: 1,
+ Addr: entry.Addr,
+ LinkAddr: staticLinkAddr,
+ State: Static,
+ },
+ }
+ c.nudDisp.mu.Lock()
+ diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...)
+ c.nudDisp.events = nil
+ c.nudDisp.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+
+ // Remove the static entry that was just added
+ c.neigh.addStaticEntry(entry.Addr, staticLinkAddr)
+
+ // No more events should have been dispatched.
+ c.nudDisp.mu.Lock()
+ defer c.nudDisp.mu.Unlock()
+ if diff := cmp.Diff(c.nudDisp.events, []testEntryEventInfo(nil)); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+}
+
+// TestNeighborCacheDuplicateStaticEntryWithDifferentLinkAddress verifies that
+// adding a duplicate static entry with a different link address dispatches a
+// change event.
+func TestNeighborCacheDuplicateStaticEntryWithDifferentLinkAddress(t *testing.T) {
+ config := DefaultNUDConfigurations()
+ c := newTestContext(config)
+
+ // Add a static entry
+ entry, ok := c.store.entry(0)
+ if !ok {
+ t.Fatalf("c.store.entry(0) not found")
+ }
+ staticLinkAddr := entry.LinkAddr + "static"
+ c.neigh.addStaticEntry(entry.Addr, staticLinkAddr)
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: 1,
+ Addr: entry.Addr,
+ LinkAddr: staticLinkAddr,
+ State: Static,
+ },
+ }
+ c.nudDisp.mu.Lock()
+ diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...)
+ c.nudDisp.events = nil
+ c.nudDisp.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+
+ // Add a duplicate entry with a different link address
+ staticLinkAddr += "duplicate"
+ c.neigh.addStaticEntry(entry.Addr, staticLinkAddr)
+ {
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestChanged,
+ NICID: 1,
+ Addr: entry.Addr,
+ LinkAddr: staticLinkAddr,
+ State: Static,
+ },
+ }
+ c.nudDisp.mu.Lock()
+ defer c.nudDisp.mu.Unlock()
+ if diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ }
+}
+
+// TestNeighborCacheRemoveStaticEntryThenOverflow verifies that the LRU cache
+// eviction strategy respects the dynamic entry count when a static entry is
+// added then removed. In this case, the dynamic entry count shouldn't have
+// been touched.
+func TestNeighborCacheRemoveStaticEntryThenOverflow(t *testing.T) {
+ config := DefaultNUDConfigurations()
+ // Stay in Reachable so the cache can overflow
+ config.BaseReachableTime = infiniteDuration
+ config.MinRandomFactor = 1
+ config.MaxRandomFactor = 1
+
+ c := newTestContext(config)
+
+ // Add a static entry
+ entry, ok := c.store.entry(0)
+ if !ok {
+ t.Fatalf("c.store.entry(0) not found")
+ }
+ staticLinkAddr := entry.LinkAddr + "static"
+ c.neigh.addStaticEntry(entry.Addr, staticLinkAddr)
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: 1,
+ Addr: entry.Addr,
+ LinkAddr: staticLinkAddr,
+ State: Static,
+ },
+ }
+ c.nudDisp.mu.Lock()
+ diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...)
+ c.nudDisp.events = nil
+ c.nudDisp.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+
+ // Remove the static entry that was just added
+ c.neigh.removeEntry(entry.Addr)
+ {
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestRemoved,
+ NICID: 1,
+ Addr: entry.Addr,
+ LinkAddr: staticLinkAddr,
+ State: Static,
+ },
+ }
+ c.nudDisp.mu.Lock()
+ diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...)
+ c.nudDisp.events = nil
+ c.nudDisp.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ }
+
+ opts := overflowOptions{
+ startAtEntryIndex: 0,
+ }
+ if err := c.overflowCache(opts); err != nil {
+ t.Errorf("c.overflowCache(%+v): %s", opts, err)
+ }
+}
+
+// TestNeighborCacheOverwriteWithStaticEntryThenOverflow verifies that the LRU
+// cache eviction strategy keeps count of the dynamic entry count when an entry
+// is overwritten by a static entry. Static entries should not count towards
+// the size of the LRU cache.
+func TestNeighborCacheOverwriteWithStaticEntryThenOverflow(t *testing.T) {
+ config := DefaultNUDConfigurations()
+ // Stay in Reachable so the cache can overflow
+ config.BaseReachableTime = infiniteDuration
+ config.MinRandomFactor = 1
+ config.MaxRandomFactor = 1
+
+ c := newTestContext(config)
+
+ // Add a dynamic entry
+ entry, ok := c.store.entry(0)
+ if !ok {
+ t.Fatalf("c.store.entry(0) not found")
+ }
+ _, _, err := c.neigh.entry(entry.Addr, entry.LocalAddr, c.linkRes, nil)
+ if err != tcpip.ErrWouldBlock {
+ t.Errorf("got c.neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock)
+ }
+ c.clock.advance(typicalLatency)
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: 1,
+ Addr: entry.Addr,
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: 1,
+ Addr: entry.Addr,
+ LinkAddr: entry.LinkAddr,
+ State: Reachable,
+ },
+ }
+ c.nudDisp.mu.Lock()
+ diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...)
+ c.nudDisp.events = nil
+ c.nudDisp.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+
+ // Override the entry with a static one using the same address
+ staticLinkAddr := entry.LinkAddr + "static"
+ c.neigh.addStaticEntry(entry.Addr, staticLinkAddr)
+ {
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestRemoved,
+ NICID: 1,
+ Addr: entry.Addr,
+ LinkAddr: entry.LinkAddr,
+ State: Reachable,
+ },
+ {
+ EventType: entryTestAdded,
+ NICID: 1,
+ Addr: entry.Addr,
+ LinkAddr: staticLinkAddr,
+ State: Static,
+ },
+ }
+ c.nudDisp.mu.Lock()
+ diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...)
+ c.nudDisp.events = nil
+ c.nudDisp.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ }
+
+ opts := overflowOptions{
+ startAtEntryIndex: 1,
+ wantStaticEntries: []NeighborEntry{
+ {
+ Addr: entry.Addr,
+ LocalAddr: "", // static entries don't need a local address
+ LinkAddr: staticLinkAddr,
+ State: Static,
+ },
+ },
+ }
+ if err := c.overflowCache(opts); err != nil {
+ t.Errorf("c.overflowCache(%+v): %s", opts, err)
+ }
+}
+
+func TestNeighborCacheNotifiesWaker(t *testing.T) {
+ config := DefaultNUDConfigurations()
+
+ nudDisp := testNUDDispatcher{}
+ clock := newFakeClock()
+ neigh := newTestNeighborCache(&nudDisp, config, clock)
+ store := newTestEntryStore()
+ linkRes := &testNeighborResolver{
+ clock: clock,
+ neigh: neigh,
+ entries: store,
+ delay: typicalLatency,
+ }
+
+ w := sleep.Waker{}
+ s := sleep.Sleeper{}
+ const wakerID = 1
+ s.AddWaker(&w, wakerID)
+
+ entry, ok := store.entry(0)
+ if !ok {
+ t.Fatalf("store.entry(0) not found")
+ }
+ _, doneCh, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, &w)
+ if err != tcpip.ErrWouldBlock {
+ t.Fatalf("got neigh.entry(%s, %s, _, _ = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock)
+ }
+ if doneCh == nil {
+ t.Fatalf("expected done channel from neigh.entry(%s, %s, _, _)", entry.Addr, entry.LocalAddr)
+ }
+ clock.advance(typicalLatency)
+
+ select {
+ case <-doneCh:
+ default:
+ t.Fatal("expected notification from done channel")
+ }
+
+ id, ok := s.Fetch(false /* block */)
+ if !ok {
+ t.Errorf("expected waker to be notified after neigh.entry(%s, %s, _, _)", entry.Addr, entry.LocalAddr)
+ }
+ if id != wakerID {
+ t.Errorf("got s.Fetch(false) = %d, want = %d", id, wakerID)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: 1,
+ Addr: entry.Addr,
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: 1,
+ Addr: entry.Addr,
+ LinkAddr: entry.LinkAddr,
+ State: Reachable,
+ },
+ }
+ nudDisp.mu.Lock()
+ defer nudDisp.mu.Unlock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+}
+
+func TestNeighborCacheRemoveWaker(t *testing.T) {
+ config := DefaultNUDConfigurations()
+
+ nudDisp := testNUDDispatcher{}
+ clock := newFakeClock()
+ neigh := newTestNeighborCache(&nudDisp, config, clock)
+ store := newTestEntryStore()
+ linkRes := &testNeighborResolver{
+ clock: clock,
+ neigh: neigh,
+ entries: store,
+ delay: typicalLatency,
+ }
+
+ w := sleep.Waker{}
+ s := sleep.Sleeper{}
+ const wakerID = 1
+ s.AddWaker(&w, wakerID)
+
+ entry, ok := store.entry(0)
+ if !ok {
+ t.Fatalf("store.entry(0) not found")
+ }
+ _, doneCh, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, &w)
+ if err != tcpip.ErrWouldBlock {
+ t.Fatalf("got neigh.entry(%s, %s, _, _) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock)
+ }
+ if doneCh == nil {
+ t.Fatalf("expected done channel from neigh.entry(%s, %s, _, _)", entry.Addr, entry.LocalAddr)
+ }
+
+ // Remove the waker before the neighbor cache has the opportunity to send a
+ // notification.
+ neigh.removeWaker(entry.Addr, &w)
+ clock.advance(typicalLatency)
+
+ select {
+ case <-doneCh:
+ default:
+ t.Fatal("expected notification from done channel")
+ }
+
+ if id, ok := s.Fetch(false /* block */); ok {
+ t.Errorf("unexpected notification from waker with id %d", id)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: 1,
+ Addr: entry.Addr,
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: 1,
+ Addr: entry.Addr,
+ LinkAddr: entry.LinkAddr,
+ State: Reachable,
+ },
+ }
+ nudDisp.mu.Lock()
+ defer nudDisp.mu.Unlock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+}
+
+func TestNeighborCacheAddStaticEntryThenOverflow(t *testing.T) {
+ config := DefaultNUDConfigurations()
+ // Stay in Reachable so the cache can overflow
+ config.BaseReachableTime = infiniteDuration
+ config.MinRandomFactor = 1
+ config.MaxRandomFactor = 1
+
+ c := newTestContext(config)
+
+ entry, ok := c.store.entry(0)
+ if !ok {
+ t.Fatalf("c.store.entry(0) not found")
+ }
+ c.neigh.addStaticEntry(entry.Addr, entry.LinkAddr)
+ e, _, err := c.neigh.entry(entry.Addr, "", c.linkRes, nil)
+ if err != nil {
+ t.Errorf("unexpected error from c.neigh.entry(%s, \"\", _, nil): %s", entry.Addr, err)
+ }
+ want := NeighborEntry{
+ Addr: entry.Addr,
+ LocalAddr: "", // static entries don't need a local address
+ LinkAddr: entry.LinkAddr,
+ State: Static,
+ }
+ if diff := cmp.Diff(e, want, entryDiffOpts()...); diff != "" {
+ t.Errorf("c.neigh.entry(%s, \"\", _, nil) mismatch (-got, +want):\n%s", entry.Addr, diff)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: 1,
+ Addr: entry.Addr,
+ LinkAddr: entry.LinkAddr,
+ State: Static,
+ },
+ }
+ c.nudDisp.mu.Lock()
+ diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...)
+ c.nudDisp.events = nil
+ c.nudDisp.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+
+ opts := overflowOptions{
+ startAtEntryIndex: 1,
+ wantStaticEntries: []NeighborEntry{
+ {
+ Addr: entry.Addr,
+ LocalAddr: "", // static entries don't need a local address
+ LinkAddr: entry.LinkAddr,
+ State: Static,
+ },
+ },
+ }
+ if err := c.overflowCache(opts); err != nil {
+ t.Errorf("c.overflowCache(%+v): %s", opts, err)
+ }
+}
+
+func TestNeighborCacheClear(t *testing.T) {
+ config := DefaultNUDConfigurations()
+
+ nudDisp := testNUDDispatcher{}
+ clock := newFakeClock()
+ neigh := newTestNeighborCache(&nudDisp, config, clock)
+ store := newTestEntryStore()
+ linkRes := &testNeighborResolver{
+ clock: clock,
+ neigh: neigh,
+ entries: store,
+ delay: typicalLatency,
+ }
+
+ // Add a dynamic entry.
+ entry, ok := store.entry(0)
+ if !ok {
+ t.Fatalf("store.entry(0) not found")
+ }
+ _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil)
+ if err != tcpip.ErrWouldBlock {
+ t.Errorf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock)
+ }
+ clock.advance(typicalLatency)
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: 1,
+ Addr: entry.Addr,
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: 1,
+ Addr: entry.Addr,
+ LinkAddr: entry.LinkAddr,
+ State: Reachable,
+ },
+ }
+ nudDisp.mu.Lock()
+ diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...)
+ nudDisp.events = nil
+ nudDisp.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+
+ // Add a static entry.
+ neigh.addStaticEntry(entryTestAddr1, entryTestLinkAddr1)
+
+ {
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: 1,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Static,
+ },
+ }
+ nudDisp.mu.Lock()
+ diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...)
+ nudDisp.events = nil
+ nudDisp.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ }
+
+ // Clear shoud remove both dynamic and static entries.
+ neigh.clear()
+
+ // Remove events dispatched from clear() have no deterministic order so they
+ // need to be sorted beforehand.
+ wantUnsortedEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestRemoved,
+ NICID: 1,
+ Addr: entry.Addr,
+ LinkAddr: entry.LinkAddr,
+ State: Reachable,
+ },
+ {
+ EventType: entryTestRemoved,
+ NICID: 1,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Static,
+ },
+ }
+ nudDisp.mu.Lock()
+ defer nudDisp.mu.Unlock()
+ if diff := cmp.Diff(nudDisp.events, wantUnsortedEvents, eventDiffOptsWithSort()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+}
+
+// TestNeighborCacheClearThenOverflow verifies that the LRU cache eviction
+// strategy keeps count of the dynamic entry count when all entries are
+// cleared.
+func TestNeighborCacheClearThenOverflow(t *testing.T) {
+ config := DefaultNUDConfigurations()
+ // Stay in Reachable so the cache can overflow
+ config.BaseReachableTime = infiniteDuration
+ config.MinRandomFactor = 1
+ config.MaxRandomFactor = 1
+
+ c := newTestContext(config)
+
+ // Add a dynamic entry
+ entry, ok := c.store.entry(0)
+ if !ok {
+ t.Fatalf("c.store.entry(0) not found")
+ }
+ _, _, err := c.neigh.entry(entry.Addr, entry.LocalAddr, c.linkRes, nil)
+ if err != tcpip.ErrWouldBlock {
+ t.Errorf("got c.neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock)
+ }
+ c.clock.advance(typicalLatency)
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: 1,
+ Addr: entry.Addr,
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: 1,
+ Addr: entry.Addr,
+ LinkAddr: entry.LinkAddr,
+ State: Reachable,
+ },
+ }
+ c.nudDisp.mu.Lock()
+ diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...)
+ c.nudDisp.events = nil
+ c.nudDisp.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+
+ // Clear the cache.
+ c.neigh.clear()
+ {
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestRemoved,
+ NICID: 1,
+ Addr: entry.Addr,
+ LinkAddr: entry.LinkAddr,
+ State: Reachable,
+ },
+ }
+ c.nudDisp.mu.Lock()
+ diff := cmp.Diff(c.nudDisp.events, wantEvents, eventDiffOpts()...)
+ c.nudDisp.events = nil
+ c.nudDisp.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ }
+
+ opts := overflowOptions{
+ startAtEntryIndex: 0,
+ }
+ if err := c.overflowCache(opts); err != nil {
+ t.Errorf("c.overflowCache(%+v): %s", opts, err)
+ }
+}
+
+func TestNeighborCacheKeepFrequentlyUsed(t *testing.T) {
+ config := DefaultNUDConfigurations()
+ // Stay in Reachable so the cache can overflow
+ config.BaseReachableTime = infiniteDuration
+ config.MinRandomFactor = 1
+ config.MaxRandomFactor = 1
+
+ nudDisp := testNUDDispatcher{}
+ clock := newFakeClock()
+ neigh := newTestNeighborCache(&nudDisp, config, clock)
+ store := newTestEntryStore()
+ linkRes := &testNeighborResolver{
+ clock: clock,
+ neigh: neigh,
+ entries: store,
+ delay: typicalLatency,
+ }
+
+ frequentlyUsedEntry, ok := store.entry(0)
+ if !ok {
+ t.Fatalf("store.entry(0) not found")
+ }
+
+ // The following logic is very similar to overflowCache, but
+ // periodically refreshes the frequently used entry.
+
+ // Fill the neighbor cache to capacity
+ for i := 0; i < neighborCacheSize; i++ {
+ entry, ok := store.entry(i)
+ if !ok {
+ t.Fatalf("store.entry(%d) not found", i)
+ }
+ _, doneCh, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil)
+ if err != tcpip.ErrWouldBlock {
+ t.Errorf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock)
+ }
+ clock.advance(typicalLatency)
+ select {
+ case <-doneCh:
+ default:
+ t.Fatalf("expected notification from done channel returned by neigh.entry(%s, %s, _, nil)", entry.Addr, entry.LocalAddr)
+ }
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: 1,
+ Addr: entry.Addr,
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: 1,
+ Addr: entry.Addr,
+ LinkAddr: entry.LinkAddr,
+ State: Reachable,
+ },
+ }
+ nudDisp.mu.Lock()
+ diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...)
+ nudDisp.events = nil
+ nudDisp.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ }
+
+ // Keep adding more entries
+ for i := neighborCacheSize; i < store.size(); i++ {
+ // Periodically refresh the frequently used entry
+ if i%(neighborCacheSize/2) == 0 {
+ _, _, err := neigh.entry(frequentlyUsedEntry.Addr, frequentlyUsedEntry.LocalAddr, linkRes, nil)
+ if err != nil {
+ t.Errorf("unexpected error from neigh.entry(%s, %s, _, nil): %s", frequentlyUsedEntry.Addr, frequentlyUsedEntry.LocalAddr, err)
+ }
+ }
+
+ entry, ok := store.entry(i)
+ if !ok {
+ t.Fatalf("store.entry(%d) not found", i)
+ }
+ _, doneCh, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil)
+ if err != tcpip.ErrWouldBlock {
+ t.Errorf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock)
+ }
+ clock.advance(typicalLatency)
+ select {
+ case <-doneCh:
+ default:
+ t.Fatalf("expected notification from done channel returned by neigh.entry(%s, %s, _, nil)", entry.Addr, entry.LocalAddr)
+ }
+
+ // An entry should have been removed, as per the LRU eviction strategy
+ removedEntry, ok := store.entry(i - neighborCacheSize + 1)
+ if !ok {
+ t.Fatalf("store.entry(%d) not found", i-neighborCacheSize+1)
+ }
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestRemoved,
+ NICID: 1,
+ Addr: removedEntry.Addr,
+ LinkAddr: removedEntry.LinkAddr,
+ State: Reachable,
+ },
+ {
+ EventType: entryTestAdded,
+ NICID: 1,
+ Addr: entry.Addr,
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: 1,
+ Addr: entry.Addr,
+ LinkAddr: entry.LinkAddr,
+ State: Reachable,
+ },
+ }
+ nudDisp.mu.Lock()
+ diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...)
+ nudDisp.events = nil
+ nudDisp.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ }
+
+ // Expect to find only the frequently used entry and the most recent entries.
+ // The order of entries reported by entries() is undeterministic, so entries
+ // have to be sorted before comparison.
+ wantUnsortedEntries := []NeighborEntry{
+ {
+ Addr: frequentlyUsedEntry.Addr,
+ LocalAddr: frequentlyUsedEntry.LocalAddr,
+ LinkAddr: frequentlyUsedEntry.LinkAddr,
+ State: Reachable,
+ },
+ }
+
+ for i := store.size() - neighborCacheSize + 1; i < store.size(); i++ {
+ entry, ok := store.entry(i)
+ if !ok {
+ t.Fatalf("store.entry(%d) not found", i)
+ }
+ wantEntry := NeighborEntry{
+ Addr: entry.Addr,
+ LocalAddr: entry.LocalAddr,
+ LinkAddr: entry.LinkAddr,
+ State: Reachable,
+ }
+ wantUnsortedEntries = append(wantUnsortedEntries, wantEntry)
+ }
+
+ if diff := cmp.Diff(neigh.entries(), wantUnsortedEntries, entryDiffOptsWithSort()...); diff != "" {
+ t.Errorf("neighbor entries mismatch (-got, +want):\n%s", diff)
+ }
+
+ // No more events should have been dispatched.
+ nudDisp.mu.Lock()
+ defer nudDisp.mu.Unlock()
+ if diff := cmp.Diff(nudDisp.events, []testEntryEventInfo(nil)); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+}
+
+func TestNeighborCacheConcurrent(t *testing.T) {
+ const concurrentProcesses = 16
+
+ config := DefaultNUDConfigurations()
+
+ nudDisp := testNUDDispatcher{}
+ clock := newFakeClock()
+ neigh := newTestNeighborCache(&nudDisp, config, clock)
+ store := newTestEntryStore()
+ linkRes := &testNeighborResolver{
+ clock: clock,
+ neigh: neigh,
+ entries: store,
+ delay: typicalLatency,
+ }
+
+ storeEntries := store.entries()
+ for _, entry := range storeEntries {
+ var wg sync.WaitGroup
+ for r := 0; r < concurrentProcesses; r++ {
+ wg.Add(1)
+ go func(entry NeighborEntry) {
+ defer wg.Done()
+ e, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil)
+ if err != nil && err != tcpip.ErrWouldBlock {
+ t.Errorf("got neigh.entry(%s, %s, _, nil) = (%+v, _, %s), want (_, _, nil) or (_, _, %s)", entry.Addr, entry.LocalAddr, e, err, tcpip.ErrWouldBlock)
+ }
+ }(entry)
+ }
+
+ // Wait for all gorountines to send a request
+ wg.Wait()
+
+ // Process all the requests for a single entry concurrently
+ clock.advance(typicalLatency)
+ }
+
+ // All goroutines add in the same order and add more values than can fit in
+ // the cache. Our eviction strategy requires that the last entries are
+ // present, up to the size of the neighbor cache, and the rest are missing.
+ // The order of entries reported by entries() is undeterministic, so entries
+ // have to be sorted before comparison.
+ var wantUnsortedEntries []NeighborEntry
+ for i := store.size() - neighborCacheSize; i < store.size(); i++ {
+ entry, ok := store.entry(i)
+ if !ok {
+ t.Errorf("store.entry(%d) not found", i)
+ }
+ wantEntry := NeighborEntry{
+ Addr: entry.Addr,
+ LocalAddr: entry.LocalAddr,
+ LinkAddr: entry.LinkAddr,
+ State: Reachable,
+ }
+ wantUnsortedEntries = append(wantUnsortedEntries, wantEntry)
+ }
+
+ if diff := cmp.Diff(neigh.entries(), wantUnsortedEntries, entryDiffOptsWithSort()...); diff != "" {
+ t.Errorf("neighbor entries mismatch (-got, +want):\n%s", diff)
+ }
+}
+
+func TestNeighborCacheReplace(t *testing.T) {
+ config := DefaultNUDConfigurations()
+
+ nudDisp := testNUDDispatcher{}
+ clock := newFakeClock()
+ neigh := newTestNeighborCache(&nudDisp, config, clock)
+ store := newTestEntryStore()
+ linkRes := &testNeighborResolver{
+ clock: clock,
+ neigh: neigh,
+ entries: store,
+ delay: typicalLatency,
+ }
+
+ // Add an entry
+ entry, ok := store.entry(0)
+ if !ok {
+ t.Fatalf("store.entry(0) not found")
+ }
+ _, doneCh, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil)
+ if err != tcpip.ErrWouldBlock {
+ t.Fatalf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock)
+ }
+ clock.advance(typicalLatency)
+ select {
+ case <-doneCh:
+ default:
+ t.Fatalf("expected notification from done channel returned by neigh.entry(%s, %s, _, nil)", entry.Addr, entry.LocalAddr)
+ }
+
+ // Verify the entry exists
+ e, doneCh, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil)
+ if err != nil {
+ t.Errorf("unexpected error from neigh.entry(%s, %s, _, nil): %s", entry.Addr, entry.LocalAddr, err)
+ }
+ if doneCh != nil {
+ t.Errorf("unexpected done channel from neigh.entry(%s, %s, _, nil): %v", entry.Addr, entry.LocalAddr, doneCh)
+ }
+ if t.Failed() {
+ t.FailNow()
+ }
+ want := NeighborEntry{
+ Addr: entry.Addr,
+ LocalAddr: entry.LocalAddr,
+ LinkAddr: entry.LinkAddr,
+ State: Reachable,
+ }
+ if diff := cmp.Diff(e, want, entryDiffOpts()...); diff != "" {
+ t.Errorf("neigh.entry(%s, %s, _, nil) mismatch (-got, +want):\n%s", entry.Addr, entry.LinkAddr, diff)
+ }
+
+ // Notify of a link address change
+ var updatedLinkAddr tcpip.LinkAddress
+ {
+ entry, ok := store.entry(1)
+ if !ok {
+ t.Fatalf("store.entry(1) not found")
+ }
+ updatedLinkAddr = entry.LinkAddr
+ }
+ store.set(0, updatedLinkAddr)
+ neigh.HandleConfirmation(entry.Addr, updatedLinkAddr, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: true,
+ IsRouter: false,
+ })
+
+ // Requesting the entry again should start address resolution
+ {
+ _, doneCh, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil)
+ if err != tcpip.ErrWouldBlock {
+ t.Fatalf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock)
+ }
+ clock.advance(config.DelayFirstProbeTime + typicalLatency)
+ select {
+ case <-doneCh:
+ default:
+ t.Fatalf("expected notification from done channel returned by neigh.entry(%s, %s, _, nil)", entry.Addr, entry.LocalAddr)
+ }
+ }
+
+ // Verify the entry's new link address
+ {
+ e, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil)
+ clock.advance(typicalLatency)
+ if err != nil {
+ t.Errorf("unexpected error from neigh.entry(%s, %s, _, nil): %s", entry.Addr, entry.LocalAddr, err)
+ }
+ want = NeighborEntry{
+ Addr: entry.Addr,
+ LocalAddr: entry.LocalAddr,
+ LinkAddr: updatedLinkAddr,
+ State: Reachable,
+ }
+ if diff := cmp.Diff(e, want, entryDiffOpts()...); diff != "" {
+ t.Errorf("neigh.entry(%s, %s, _, nil) mismatch (-got, +want):\n%s", entry.Addr, entry.LocalAddr, diff)
+ }
+ }
+}
+
+func TestNeighborCacheResolutionFailed(t *testing.T) {
+ config := DefaultNUDConfigurations()
+
+ nudDisp := testNUDDispatcher{}
+ clock := newFakeClock()
+ neigh := newTestNeighborCache(&nudDisp, config, clock)
+ store := newTestEntryStore()
+
+ var requestCount uint32
+ linkRes := &testNeighborResolver{
+ clock: clock,
+ neigh: neigh,
+ entries: store,
+ delay: typicalLatency,
+ onLinkAddressRequest: func() {
+ atomic.AddUint32(&requestCount, 1)
+ },
+ }
+
+ // First, sanity check that resolution is working
+ entry, ok := store.entry(0)
+ if !ok {
+ t.Fatalf("store.entry(0) not found")
+ }
+ if _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil); err != tcpip.ErrWouldBlock {
+ t.Fatalf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock)
+ }
+ clock.advance(typicalLatency)
+ got, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil)
+ if err != nil {
+ t.Fatalf("unexpected error from neigh.entry(%s, %s, _, nil): %s", entry.Addr, entry.LocalAddr, err)
+ }
+ want := NeighborEntry{
+ Addr: entry.Addr,
+ LocalAddr: entry.LocalAddr,
+ LinkAddr: entry.LinkAddr,
+ State: Reachable,
+ }
+ if diff := cmp.Diff(got, want, entryDiffOpts()...); diff != "" {
+ t.Errorf("neigh.entry(%s, %s, _, nil) mismatch (-got, +want):\n%s", entry.Addr, entry.LocalAddr, diff)
+ }
+
+ // Verify that address resolution for an unknown address returns ErrNoLinkAddress
+ before := atomic.LoadUint32(&requestCount)
+
+ entry.Addr += "2"
+ if _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil); err != tcpip.ErrWouldBlock {
+ t.Fatalf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock)
+ }
+ waitFor := config.DelayFirstProbeTime + typicalLatency*time.Duration(config.MaxMulticastProbes)
+ clock.advance(waitFor)
+ if _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil); err != tcpip.ErrNoLinkAddress {
+ t.Fatalf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrNoLinkAddress)
+ }
+
+ maxAttempts := neigh.config().MaxUnicastProbes
+ if got, want := atomic.LoadUint32(&requestCount)-before, maxAttempts; got != want {
+ t.Errorf("got link address request count = %d, want = %d", got, want)
+ }
+}
+
+// TestNeighborCacheResolutionTimeout simulates sending MaxMulticastProbes
+// probes and not retrieving a confirmation before the duration defined by
+// MaxMulticastProbes * RetransmitTimer.
+func TestNeighborCacheResolutionTimeout(t *testing.T) {
+ config := DefaultNUDConfigurations()
+ config.RetransmitTimer = time.Millisecond // small enough to cause timeout
+
+ clock := newFakeClock()
+ neigh := newTestNeighborCache(nil, config, clock)
+ store := newTestEntryStore()
+ linkRes := &testNeighborResolver{
+ clock: clock,
+ neigh: neigh,
+ entries: store,
+ delay: time.Minute, // large enough to cause timeout
+ }
+
+ entry, ok := store.entry(0)
+ if !ok {
+ t.Fatalf("store.entry(0) not found")
+ }
+ if _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil); err != tcpip.ErrWouldBlock {
+ t.Fatalf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock)
+ }
+ waitFor := config.RetransmitTimer * time.Duration(config.MaxMulticastProbes)
+ clock.advance(waitFor)
+ if _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil); err != tcpip.ErrNoLinkAddress {
+ t.Fatalf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrNoLinkAddress)
+ }
+}
+
+// TestNeighborCacheStaticResolution checks that static link addresses are
+// resolved immediately and don't send resolution requests.
+func TestNeighborCacheStaticResolution(t *testing.T) {
+ config := DefaultNUDConfigurations()
+ clock := newFakeClock()
+ neigh := newTestNeighborCache(nil, config, clock)
+ store := newTestEntryStore()
+ linkRes := &testNeighborResolver{
+ clock: clock,
+ neigh: neigh,
+ entries: store,
+ delay: typicalLatency,
+ }
+
+ got, _, err := neigh.entry(testEntryBroadcastAddr, testEntryLocalAddr, linkRes, nil)
+ if err != nil {
+ t.Fatalf("unexpected error from neigh.entry(%s, %s, _, nil): %s", testEntryBroadcastAddr, testEntryLocalAddr, err)
+ }
+ want := NeighborEntry{
+ Addr: testEntryBroadcastAddr,
+ LocalAddr: testEntryLocalAddr,
+ LinkAddr: testEntryBroadcastLinkAddr,
+ State: Static,
+ }
+ if diff := cmp.Diff(got, want, entryDiffOpts()...); diff != "" {
+ t.Errorf("neigh.entry(%s, %s, _, nil) mismatch (-got, +want):\n%s", testEntryBroadcastAddr, testEntryLocalAddr, diff)
+ }
+}
+
+func BenchmarkCacheClear(b *testing.B) {
+ b.StopTimer()
+ config := DefaultNUDConfigurations()
+ clock := &tcpip.StdClock{}
+ neigh := newTestNeighborCache(nil, config, clock)
+ store := newTestEntryStore()
+ linkRes := &testNeighborResolver{
+ clock: clock,
+ neigh: neigh,
+ entries: store,
+ delay: 0,
+ }
+
+ // Clear for every possible size of the cache
+ for cacheSize := 0; cacheSize < neighborCacheSize; cacheSize++ {
+ // Fill the neighbor cache to capacity.
+ for i := 0; i < cacheSize; i++ {
+ entry, ok := store.entry(i)
+ if !ok {
+ b.Fatalf("store.entry(%d) not found", i)
+ }
+ _, doneCh, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil)
+ if err != tcpip.ErrWouldBlock {
+ b.Fatalf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock)
+ }
+ if doneCh != nil {
+ <-doneCh
+ }
+ }
+
+ b.StartTimer()
+ neigh.clear()
+ b.StopTimer()
+ }
+}
diff --git a/pkg/tcpip/stack/neighbor_entry.go b/pkg/tcpip/stack/neighbor_entry.go
new file mode 100644
index 000000000..0068cacb8
--- /dev/null
+++ b/pkg/tcpip/stack/neighbor_entry.go
@@ -0,0 +1,482 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package stack
+
+import (
+ "fmt"
+ "sync"
+ "time"
+
+ "gvisor.dev/gvisor/pkg/sleep"
+ "gvisor.dev/gvisor/pkg/tcpip"
+)
+
+// NeighborEntry describes a neighboring device in the local network.
+type NeighborEntry struct {
+ Addr tcpip.Address
+ LocalAddr tcpip.Address
+ LinkAddr tcpip.LinkAddress
+ State NeighborState
+ UpdatedAt time.Time
+}
+
+// NeighborState defines the state of a NeighborEntry within the Neighbor
+// Unreachability Detection state machine, as per RFC 4861 section 7.3.2.
+type NeighborState uint8
+
+const (
+ // Unknown means reachability has not been verified yet. This is the initial
+ // state of entries that have been created automatically by the Neighbor
+ // Unreachability Detection state machine.
+ Unknown NeighborState = iota
+ // Incomplete means that there is an outstanding request to resolve the
+ // address.
+ Incomplete
+ // Reachable means the path to the neighbor is functioning properly for both
+ // receive and transmit paths.
+ Reachable
+ // Stale means reachability to the neighbor is unknown, but packets are still
+ // able to be transmitted to the possibly stale link address.
+ Stale
+ // Delay means reachability to the neighbor is unknown and pending
+ // confirmation from an upper-level protocol like TCP, but packets are still
+ // able to be transmitted to the possibly stale link address.
+ Delay
+ // Probe means a reachability confirmation is actively being sought by
+ // periodically retransmitting reachability probes until a reachability
+ // confirmation is received, or until the max amount of probes has been sent.
+ Probe
+ // Static describes entries that have been explicitly added by the user. They
+ // do not expire and are not deleted until explicitly removed.
+ Static
+ // Failed means traffic should not be sent to this neighbor since attempts of
+ // reachability have returned inconclusive.
+ Failed
+)
+
+// neighborEntry implements a neighbor entry's individual node behavior, as per
+// RFC 4861 section 7.3.3. Neighbor Unreachability Detection operates in
+// parallel with the sending of packets to a neighbor, necessitating the
+// entry's lock to be acquired for all operations.
+type neighborEntry struct {
+ neighborEntryEntry
+
+ nic *NIC
+ protocol tcpip.NetworkProtocolNumber
+
+ // linkRes provides the functionality to send reachability probes, used in
+ // Neighbor Unreachability Detection.
+ linkRes LinkAddressResolver
+
+ // nudState points to the Neighbor Unreachability Detection configuration.
+ nudState *NUDState
+
+ // mu protects the fields below.
+ mu sync.RWMutex
+
+ neigh NeighborEntry
+
+ // wakers is a set of waiters for address resolution result. Anytime state
+ // transitions out of incomplete these waiters are notified. It is nil iff
+ // address resolution is ongoing and no clients are waiting for the result.
+ wakers map[*sleep.Waker]struct{}
+
+ // done is used to allow callers to wait on address resolution. It is nil
+ // iff nudState is not Reachable and address resolution is not yet in
+ // progress.
+ done chan struct{}
+
+ isRouter bool
+ job *tcpip.Job
+}
+
+// newNeighborEntry creates a neighbor cache entry starting at the default
+// state, Unknown. Transition out of Unknown by calling either
+// `handlePacketQueuedLocked` or `handleProbeLocked` on the newly created
+// neighborEntry.
+func newNeighborEntry(nic *NIC, remoteAddr tcpip.Address, localAddr tcpip.Address, nudState *NUDState, linkRes LinkAddressResolver) *neighborEntry {
+ return &neighborEntry{
+ nic: nic,
+ linkRes: linkRes,
+ nudState: nudState,
+ neigh: NeighborEntry{
+ Addr: remoteAddr,
+ LocalAddr: localAddr,
+ State: Unknown,
+ },
+ }
+}
+
+// newStaticNeighborEntry creates a neighbor cache entry starting at the Static
+// state. The entry can only transition out of Static by directly calling
+// `setStateLocked`.
+func newStaticNeighborEntry(nic *NIC, addr tcpip.Address, linkAddr tcpip.LinkAddress, state *NUDState) *neighborEntry {
+ if nic.stack.nudDisp != nil {
+ nic.stack.nudDisp.OnNeighborAdded(nic.id, addr, linkAddr, Static, time.Now())
+ }
+ return &neighborEntry{
+ nic: nic,
+ nudState: state,
+ neigh: NeighborEntry{
+ Addr: addr,
+ LinkAddr: linkAddr,
+ State: Static,
+ UpdatedAt: time.Now(),
+ },
+ }
+}
+
+// addWaker adds w to the list of wakers waiting for address resolution.
+// Assumes the entry has already been appropriately locked.
+func (e *neighborEntry) addWakerLocked(w *sleep.Waker) {
+ if w == nil {
+ return
+ }
+ if e.wakers == nil {
+ e.wakers = make(map[*sleep.Waker]struct{})
+ }
+ e.wakers[w] = struct{}{}
+}
+
+// notifyWakersLocked notifies those waiting for address resolution, whether it
+// succeeded or failed. Assumes the entry has already been appropriately locked.
+func (e *neighborEntry) notifyWakersLocked() {
+ for w := range e.wakers {
+ w.Assert()
+ }
+ e.wakers = nil
+ if ch := e.done; ch != nil {
+ close(ch)
+ e.done = nil
+ }
+}
+
+// dispatchAddEventLocked signals to stack's NUD Dispatcher that the entry has
+// been added.
+func (e *neighborEntry) dispatchAddEventLocked(nextState NeighborState) {
+ if nudDisp := e.nic.stack.nudDisp; nudDisp != nil {
+ nudDisp.OnNeighborAdded(e.nic.id, e.neigh.Addr, e.neigh.LinkAddr, nextState, time.Now())
+ }
+}
+
+// dispatchChangeEventLocked signals to stack's NUD Dispatcher that the entry
+// has changed state or link-layer address.
+func (e *neighborEntry) dispatchChangeEventLocked(nextState NeighborState) {
+ if nudDisp := e.nic.stack.nudDisp; nudDisp != nil {
+ nudDisp.OnNeighborChanged(e.nic.id, e.neigh.Addr, e.neigh.LinkAddr, nextState, time.Now())
+ }
+}
+
+// dispatchRemoveEventLocked signals to stack's NUD Dispatcher that the entry
+// has been removed.
+func (e *neighborEntry) dispatchRemoveEventLocked() {
+ if nudDisp := e.nic.stack.nudDisp; nudDisp != nil {
+ nudDisp.OnNeighborRemoved(e.nic.id, e.neigh.Addr, e.neigh.LinkAddr, e.neigh.State, time.Now())
+ }
+}
+
+// setStateLocked transitions the entry to the specified state immediately.
+//
+// Follows the logic defined in RFC 4861 section 7.3.3.
+//
+// e.mu MUST be locked.
+func (e *neighborEntry) setStateLocked(next NeighborState) {
+ // Cancel the previously scheduled action, if there is one. Entries in
+ // Unknown, Stale, or Static state do not have scheduled actions.
+ if timer := e.job; timer != nil {
+ timer.Cancel()
+ }
+
+ prev := e.neigh.State
+ e.neigh.State = next
+ e.neigh.UpdatedAt = time.Now()
+ config := e.nudState.Config()
+
+ switch next {
+ case Incomplete:
+ var retryCounter uint32
+ var sendMulticastProbe func()
+
+ sendMulticastProbe = func() {
+ if retryCounter == config.MaxMulticastProbes {
+ // "If no Neighbor Advertisement is received after
+ // MAX_MULTICAST_SOLICIT solicitations, address resolution has failed.
+ // The sender MUST return ICMP destination unreachable indications with
+ // code 3 (Address Unreachable) for each packet queued awaiting address
+ // resolution." - RFC 4861 section 7.2.2
+ //
+ // There is no need to send an ICMP destination unreachable indication
+ // since the failure to resolve the address is expected to only occur
+ // on this node. Thus, redirecting traffic is currently not supported.
+ //
+ // "If the error occurs on a node other than the node originating the
+ // packet, an ICMP error message is generated. If the error occurs on
+ // the originating node, an implementation is not required to actually
+ // create and send an ICMP error packet to the source, as long as the
+ // upper-layer sender is notified through an appropriate mechanism
+ // (e.g. return value from a procedure call). Note, however, that an
+ // implementation may find it convenient in some cases to return errors
+ // to the sender by taking the offending packet, generating an ICMP
+ // error message, and then delivering it (locally) through the generic
+ // error-handling routines.' - RFC 4861 section 2.1
+ e.dispatchRemoveEventLocked()
+ e.setStateLocked(Failed)
+ return
+ }
+
+ if err := e.linkRes.LinkAddressRequest(e.neigh.Addr, e.neigh.LocalAddr, "", e.nic.linkEP); err != nil {
+ // There is no need to log the error here; the NUD implementation may
+ // assume a working link. A valid link should be the responsibility of
+ // the NIC/stack.LinkEndpoint.
+ e.dispatchRemoveEventLocked()
+ e.setStateLocked(Failed)
+ return
+ }
+
+ retryCounter++
+ e.job = e.nic.stack.newJob(&e.mu, sendMulticastProbe)
+ e.job.Schedule(config.RetransmitTimer)
+ }
+
+ sendMulticastProbe()
+
+ case Reachable:
+ e.job = e.nic.stack.newJob(&e.mu, func() {
+ e.dispatchChangeEventLocked(Stale)
+ e.setStateLocked(Stale)
+ })
+ e.job.Schedule(e.nudState.ReachableTime())
+
+ case Delay:
+ e.job = e.nic.stack.newJob(&e.mu, func() {
+ e.dispatchChangeEventLocked(Probe)
+ e.setStateLocked(Probe)
+ })
+ e.job.Schedule(config.DelayFirstProbeTime)
+
+ case Probe:
+ var retryCounter uint32
+ var sendUnicastProbe func()
+
+ sendUnicastProbe = func() {
+ if retryCounter == config.MaxUnicastProbes {
+ e.dispatchRemoveEventLocked()
+ e.setStateLocked(Failed)
+ return
+ }
+
+ if err := e.linkRes.LinkAddressRequest(e.neigh.Addr, e.neigh.LocalAddr, e.neigh.LinkAddr, e.nic.linkEP); err != nil {
+ e.dispatchRemoveEventLocked()
+ e.setStateLocked(Failed)
+ return
+ }
+
+ retryCounter++
+ if retryCounter == config.MaxUnicastProbes {
+ e.dispatchRemoveEventLocked()
+ e.setStateLocked(Failed)
+ return
+ }
+
+ e.job = e.nic.stack.newJob(&e.mu, sendUnicastProbe)
+ e.job.Schedule(config.RetransmitTimer)
+ }
+
+ sendUnicastProbe()
+
+ case Failed:
+ e.notifyWakersLocked()
+ e.job = e.nic.stack.newJob(&e.mu, func() {
+ e.nic.neigh.removeEntryLocked(e)
+ })
+ e.job.Schedule(config.UnreachableTime)
+
+ case Unknown, Stale, Static:
+ // Do nothing
+
+ default:
+ panic(fmt.Sprintf("Invalid state transition from %q to %q", prev, next))
+ }
+}
+
+// handlePacketQueuedLocked advances the state machine according to a packet
+// being queued for outgoing transmission.
+//
+// Follows the logic defined in RFC 4861 section 7.3.3.
+func (e *neighborEntry) handlePacketQueuedLocked() {
+ switch e.neigh.State {
+ case Unknown:
+ e.dispatchAddEventLocked(Incomplete)
+ e.setStateLocked(Incomplete)
+
+ case Stale:
+ e.dispatchChangeEventLocked(Delay)
+ e.setStateLocked(Delay)
+
+ case Incomplete, Reachable, Delay, Probe, Static, Failed:
+ // Do nothing
+
+ default:
+ panic(fmt.Sprintf("Invalid cache entry state: %s", e.neigh.State))
+ }
+}
+
+// handleProbeLocked processes an incoming neighbor probe (e.g. ARP request or
+// Neighbor Solicitation for ARP or NDP, respectively).
+//
+// Follows the logic defined in RFC 4861 section 7.2.3.
+func (e *neighborEntry) handleProbeLocked(remoteLinkAddr tcpip.LinkAddress) {
+ // Probes MUST be silently discarded if the target address is tentative, does
+ // not exist, or not bound to the NIC as per RFC 4861 section 7.2.3. These
+ // checks MUST be done by the NetworkEndpoint.
+
+ switch e.neigh.State {
+ case Unknown, Incomplete, Failed:
+ e.neigh.LinkAddr = remoteLinkAddr
+ e.dispatchAddEventLocked(Stale)
+ e.setStateLocked(Stale)
+ e.notifyWakersLocked()
+
+ case Reachable, Delay, Probe:
+ if e.neigh.LinkAddr != remoteLinkAddr {
+ e.neigh.LinkAddr = remoteLinkAddr
+ e.dispatchChangeEventLocked(Stale)
+ e.setStateLocked(Stale)
+ }
+
+ case Stale:
+ if e.neigh.LinkAddr != remoteLinkAddr {
+ e.neigh.LinkAddr = remoteLinkAddr
+ e.dispatchChangeEventLocked(Stale)
+ }
+
+ case Static:
+ // Do nothing
+
+ default:
+ panic(fmt.Sprintf("Invalid cache entry state: %s", e.neigh.State))
+ }
+}
+
+// handleConfirmationLocked processes an incoming neighbor confirmation
+// (e.g. ARP reply or Neighbor Advertisement for ARP or NDP, respectively).
+//
+// Follows the state machine defined by RFC 4861 section 7.2.5.
+//
+// TODO(gvisor.dev/issue/2277): To protect against ARP poisoning and other
+// attacks against NDP functions, Secure Neighbor Discovery (SEND) Protocol
+// should be deployed where preventing access to the broadcast segment might
+// not be possible. SEND uses RSA key pairs to produce Cryptographically
+// Generated Addresses (CGA), as defined in RFC 3972. This ensures that the
+// claimed source of an NDP message is the owner of the claimed address.
+func (e *neighborEntry) handleConfirmationLocked(linkAddr tcpip.LinkAddress, flags ReachabilityConfirmationFlags) {
+ switch e.neigh.State {
+ case Incomplete:
+ if len(linkAddr) == 0 {
+ // "If the link layer has addresses and no Target Link-Layer Address
+ // option is included, the receiving node SHOULD silently discard the
+ // received advertisement." - RFC 4861 section 7.2.5
+ break
+ }
+
+ e.neigh.LinkAddr = linkAddr
+ if flags.Solicited {
+ e.dispatchChangeEventLocked(Reachable)
+ e.setStateLocked(Reachable)
+ } else {
+ e.dispatchChangeEventLocked(Stale)
+ e.setStateLocked(Stale)
+ }
+ e.isRouter = flags.IsRouter
+ e.notifyWakersLocked()
+
+ // "Note that the Override flag is ignored if the entry is in the
+ // INCOMPLETE state." - RFC 4861 section 7.2.5
+
+ case Reachable, Stale, Delay, Probe:
+ sameLinkAddr := e.neigh.LinkAddr == linkAddr
+
+ if !sameLinkAddr {
+ if !flags.Override {
+ if e.neigh.State == Reachable {
+ e.dispatchChangeEventLocked(Stale)
+ e.setStateLocked(Stale)
+ }
+ break
+ }
+
+ e.neigh.LinkAddr = linkAddr
+
+ if !flags.Solicited {
+ if e.neigh.State != Stale {
+ e.dispatchChangeEventLocked(Stale)
+ e.setStateLocked(Stale)
+ } else {
+ // Notify the LinkAddr change, even though NUD state hasn't changed.
+ e.dispatchChangeEventLocked(e.neigh.State)
+ }
+ break
+ }
+ }
+
+ if flags.Solicited && (flags.Override || sameLinkAddr) {
+ if e.neigh.State != Reachable {
+ e.dispatchChangeEventLocked(Reachable)
+ }
+ // Set state to Reachable again to refresh timers.
+ e.setStateLocked(Reachable)
+ e.notifyWakersLocked()
+ }
+
+ if e.isRouter && !flags.IsRouter {
+ // "In those cases where the IsRouter flag changes from TRUE to FALSE as
+ // a result of this update, the node MUST remove that router from the
+ // Default Router List and update the Destination Cache entries for all
+ // destinations using that neighbor as a router as specified in Section
+ // 7.3.3. This is needed to detect when a node that is used as a router
+ // stops forwarding packets due to being configured as a host."
+ // - RFC 4861 section 7.2.5
+ e.nic.mu.Lock()
+ e.nic.mu.ndp.invalidateDefaultRouter(e.neigh.Addr)
+ e.nic.mu.Unlock()
+ }
+ e.isRouter = flags.IsRouter
+
+ case Unknown, Failed, Static:
+ // Do nothing
+
+ default:
+ panic(fmt.Sprintf("Invalid cache entry state: %s", e.neigh.State))
+ }
+}
+
+// handleUpperLevelConfirmationLocked processes an incoming upper-level protocol
+// (e.g. TCP acknowledgements) reachability confirmation.
+func (e *neighborEntry) handleUpperLevelConfirmationLocked() {
+ switch e.neigh.State {
+ case Reachable, Stale, Delay, Probe:
+ if e.neigh.State != Reachable {
+ e.dispatchChangeEventLocked(Reachable)
+ // Set state to Reachable again to refresh timers.
+ }
+ e.setStateLocked(Reachable)
+
+ case Unknown, Incomplete, Failed, Static:
+ // Do nothing
+
+ default:
+ panic(fmt.Sprintf("Invalid cache entry state: %s", e.neigh.State))
+ }
+}
diff --git a/pkg/tcpip/stack/neighbor_entry_test.go b/pkg/tcpip/stack/neighbor_entry_test.go
new file mode 100644
index 000000000..b769fb2fa
--- /dev/null
+++ b/pkg/tcpip/stack/neighbor_entry_test.go
@@ -0,0 +1,2870 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package stack
+
+import (
+ "fmt"
+ "math"
+ "math/rand"
+ "strings"
+ "sync"
+ "testing"
+ "time"
+
+ "github.com/google/go-cmp/cmp"
+ "github.com/google/go-cmp/cmp/cmpopts"
+ "gvisor.dev/gvisor/pkg/sleep"
+ "gvisor.dev/gvisor/pkg/tcpip"
+)
+
+const (
+ entryTestNetNumber tcpip.NetworkProtocolNumber = math.MaxUint32
+
+ entryTestNICID tcpip.NICID = 1
+ entryTestAddr1 = tcpip.Address("\x00\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01")
+ entryTestAddr2 = tcpip.Address("\x00\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02")
+
+ entryTestLinkAddr1 = tcpip.LinkAddress("\x0a\x00\x00\x00\x00\x01")
+ entryTestLinkAddr2 = tcpip.LinkAddress("\x0a\x00\x00\x00\x00\x02")
+
+ // entryTestNetDefaultMTU is the MTU, in bytes, used throughout the tests,
+ // except where another value is explicitly used. It is chosen to match the
+ // MTU of loopback interfaces on Linux systems.
+ entryTestNetDefaultMTU = 65536
+)
+
+// eventDiffOpts are the options passed to cmp.Diff to compare entry events.
+// The UpdatedAt field is ignored due to a lack of a deterministic method to
+// predict the time that an event will be dispatched.
+func eventDiffOpts() []cmp.Option {
+ return []cmp.Option{
+ cmpopts.IgnoreFields(testEntryEventInfo{}, "UpdatedAt"),
+ }
+}
+
+// eventDiffOptsWithSort is like eventDiffOpts but also includes an option to
+// sort slices of events for cases where ordering must be ignored.
+func eventDiffOptsWithSort() []cmp.Option {
+ return []cmp.Option{
+ cmpopts.IgnoreFields(testEntryEventInfo{}, "UpdatedAt"),
+ cmpopts.SortSlices(func(a, b testEntryEventInfo) bool {
+ return strings.Compare(string(a.Addr), string(b.Addr)) < 0
+ }),
+ }
+}
+
+// The following unit tests exercise every state transition and verify its
+// behavior with RFC 4681.
+//
+// | From | To | Cause | Action | Event |
+// | ========== | ========== | ========================================== | =============== | ======= |
+// | Unknown | Unknown | Confirmation w/ unknown address | | Added |
+// | Unknown | Incomplete | Packet queued to unknown address | Send probe | Added |
+// | Unknown | Stale | Probe w/ unknown address | | Added |
+// | Incomplete | Incomplete | Retransmit timer expired | Send probe | Changed |
+// | Incomplete | Reachable | Solicited confirmation | Notify wakers | Changed |
+// | Incomplete | Stale | Unsolicited confirmation | Notify wakers | Changed |
+// | Incomplete | Failed | Max probes sent without reply | Notify wakers | Removed |
+// | Reachable | Reachable | Confirmation w/ different isRouter flag | Update IsRouter | |
+// | Reachable | Stale | Reachable timer expired | | Changed |
+// | Reachable | Stale | Probe or confirmation w/ different address | | Changed |
+// | Stale | Reachable | Solicited override confirmation | Update LinkAddr | Changed |
+// | Stale | Stale | Override confirmation | Update LinkAddr | Changed |
+// | Stale | Stale | Probe w/ different address | Update LinkAddr | Changed |
+// | Stale | Delay | Packet sent | | Changed |
+// | Delay | Reachable | Upper-layer confirmation | | Changed |
+// | Delay | Reachable | Solicited override confirmation | Update LinkAddr | Changed |
+// | Delay | Stale | Probe or confirmation w/ different address | | Changed |
+// | Delay | Probe | Delay timer expired | Send probe | Changed |
+// | Probe | Reachable | Solicited override confirmation | Update LinkAddr | Changed |
+// | Probe | Reachable | Solicited confirmation w/ same address | Notify wakers | Changed |
+// | Probe | Stale | Probe or confirmation w/ different address | | Changed |
+// | Probe | Probe | Retransmit timer expired | Send probe | Changed |
+// | Probe | Failed | Max probes sent without reply | Notify wakers | Removed |
+// | Failed | | Unreachability timer expired | Delete entry | |
+
+type testEntryEventType uint8
+
+const (
+ entryTestAdded testEntryEventType = iota
+ entryTestChanged
+ entryTestRemoved
+)
+
+func (t testEntryEventType) String() string {
+ switch t {
+ case entryTestAdded:
+ return "add"
+ case entryTestChanged:
+ return "change"
+ case entryTestRemoved:
+ return "remove"
+ default:
+ return fmt.Sprintf("unknown (%d)", t)
+ }
+}
+
+// Fields are exported for use with cmp.Diff.
+type testEntryEventInfo struct {
+ EventType testEntryEventType
+ NICID tcpip.NICID
+ Addr tcpip.Address
+ LinkAddr tcpip.LinkAddress
+ State NeighborState
+ UpdatedAt time.Time
+}
+
+func (e testEntryEventInfo) String() string {
+ return fmt.Sprintf("%s event for NIC #%d, addr=%q, linkAddr=%q, state=%q", e.EventType, e.NICID, e.Addr, e.LinkAddr, e.State)
+}
+
+// testNUDDispatcher implements NUDDispatcher to validate the dispatching of
+// events upon certain NUD state machine events.
+type testNUDDispatcher struct {
+ mu sync.Mutex
+ events []testEntryEventInfo
+}
+
+var _ NUDDispatcher = (*testNUDDispatcher)(nil)
+
+func (d *testNUDDispatcher) queueEvent(e testEntryEventInfo) {
+ d.mu.Lock()
+ defer d.mu.Unlock()
+ d.events = append(d.events, e)
+}
+
+func (d *testNUDDispatcher) OnNeighborAdded(nicID tcpip.NICID, addr tcpip.Address, linkAddr tcpip.LinkAddress, state NeighborState, updatedAt time.Time) {
+ d.queueEvent(testEntryEventInfo{
+ EventType: entryTestAdded,
+ NICID: nicID,
+ Addr: addr,
+ LinkAddr: linkAddr,
+ State: state,
+ UpdatedAt: updatedAt,
+ })
+}
+
+func (d *testNUDDispatcher) OnNeighborChanged(nicID tcpip.NICID, addr tcpip.Address, linkAddr tcpip.LinkAddress, state NeighborState, updatedAt time.Time) {
+ d.queueEvent(testEntryEventInfo{
+ EventType: entryTestChanged,
+ NICID: nicID,
+ Addr: addr,
+ LinkAddr: linkAddr,
+ State: state,
+ UpdatedAt: updatedAt,
+ })
+}
+
+func (d *testNUDDispatcher) OnNeighborRemoved(nicID tcpip.NICID, addr tcpip.Address, linkAddr tcpip.LinkAddress, state NeighborState, updatedAt time.Time) {
+ d.queueEvent(testEntryEventInfo{
+ EventType: entryTestRemoved,
+ NICID: nicID,
+ Addr: addr,
+ LinkAddr: linkAddr,
+ State: state,
+ UpdatedAt: updatedAt,
+ })
+}
+
+type entryTestLinkResolver struct {
+ mu sync.Mutex
+ probes []entryTestProbeInfo
+}
+
+var _ LinkAddressResolver = (*entryTestLinkResolver)(nil)
+
+type entryTestProbeInfo struct {
+ RemoteAddress tcpip.Address
+ RemoteLinkAddress tcpip.LinkAddress
+ LocalAddress tcpip.Address
+}
+
+func (p entryTestProbeInfo) String() string {
+ return fmt.Sprintf("probe with RemoteAddress=%q, RemoteLinkAddress=%q, LocalAddress=%q", p.RemoteAddress, p.RemoteLinkAddress, p.LocalAddress)
+}
+
+// LinkAddressRequest sends a request for the LinkAddress of addr. Broadcasts
+// to the local network if linkAddr is the zero value.
+func (r *entryTestLinkResolver) LinkAddressRequest(addr, localAddr tcpip.Address, linkAddr tcpip.LinkAddress, linkEP LinkEndpoint) *tcpip.Error {
+ p := entryTestProbeInfo{
+ RemoteAddress: addr,
+ RemoteLinkAddress: linkAddr,
+ LocalAddress: localAddr,
+ }
+ r.mu.Lock()
+ defer r.mu.Unlock()
+ r.probes = append(r.probes, p)
+ return nil
+}
+
+// ResolveStaticAddress attempts to resolve address without sending requests.
+// It either resolves the name immediately or returns the empty LinkAddress.
+func (r *entryTestLinkResolver) ResolveStaticAddress(addr tcpip.Address) (tcpip.LinkAddress, bool) {
+ return "", false
+}
+
+// LinkAddressProtocol returns the network protocol of the addresses this
+// resolver can resolve.
+func (r *entryTestLinkResolver) LinkAddressProtocol() tcpip.NetworkProtocolNumber {
+ return entryTestNetNumber
+}
+
+func entryTestSetup(c NUDConfigurations) (*neighborEntry, *testNUDDispatcher, *entryTestLinkResolver, *fakeClock) {
+ clock := newFakeClock()
+ disp := testNUDDispatcher{}
+ nic := NIC{
+ id: entryTestNICID,
+ linkEP: nil, // entryTestLinkResolver doesn't use a LinkEndpoint
+ stack: &Stack{
+ clock: clock,
+ nudDisp: &disp,
+ },
+ }
+
+ rng := rand.New(rand.NewSource(time.Now().UnixNano()))
+ nudState := NewNUDState(c, rng)
+ linkRes := entryTestLinkResolver{}
+ entry := newNeighborEntry(&nic, entryTestAddr1 /* remoteAddr */, entryTestAddr2 /* localAddr */, nudState, &linkRes)
+
+ // Stub out ndpState to verify modification of default routers.
+ nic.mu.ndp = ndpState{
+ nic: &nic,
+ defaultRouters: make(map[tcpip.Address]defaultRouterState),
+ }
+
+ // Stub out the neighbor cache to verify deletion from the cache.
+ nic.neigh = &neighborCache{
+ nic: &nic,
+ state: nudState,
+ cache: make(map[tcpip.Address]*neighborEntry, neighborCacheSize),
+ }
+ nic.neigh.cache[entryTestAddr1] = entry
+
+ return entry, &disp, &linkRes, clock
+}
+
+// TestEntryInitiallyUnknown verifies that the state of a newly created
+// neighborEntry is Unknown.
+func TestEntryInitiallyUnknown(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, clock := entryTestSetup(c)
+
+ e.mu.Lock()
+ if got, want := e.neigh.State, Unknown; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ clock.advance(c.RetransmitTimer)
+
+ // No probes should have been sent.
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, []entryTestProbeInfo(nil))
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ // No events should have been dispatched.
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, []testEntryEventInfo(nil)); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+}
+
+func TestEntryUnknownToUnknownWhenConfirmationWithUnknownAddress(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, clock := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: false,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Unknown; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ clock.advance(time.Hour)
+
+ // No probes should have been sent.
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, []entryTestProbeInfo(nil))
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ // No events should have been dispatched.
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, []testEntryEventInfo(nil)); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+}
+
+func TestEntryUnknownToIncomplete(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, _ := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ if got, want := e.neigh.State, Incomplete; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ wantProbes := []entryTestProbeInfo{
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ }
+ {
+ nudDisp.mu.Lock()
+ diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...)
+ nudDisp.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ }
+}
+
+func TestEntryUnknownToStale(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, _ := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handleProbeLocked(entryTestLinkAddr1)
+ if got, want := e.neigh.State, Stale; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ // No probes should have been sent.
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, []entryTestProbeInfo(nil))
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+}
+
+func TestEntryIncompleteToIncompleteDoesNotChangeUpdatedAt(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ c.MaxMulticastProbes = 3
+ e, nudDisp, linkRes, clock := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ if got, want := e.neigh.State, Incomplete; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ updatedAt := e.neigh.UpdatedAt
+ e.mu.Unlock()
+
+ clock.advance(c.RetransmitTimer)
+
+ // UpdatedAt should remain the same during address resolution.
+ wantProbes := []entryTestProbeInfo{
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.probes = nil
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ e.mu.Lock()
+ if got, want := e.neigh.UpdatedAt, updatedAt; got != want {
+ t.Errorf("got e.neigh.UpdatedAt = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ clock.advance(c.RetransmitTimer)
+
+ // UpdatedAt should change after failing address resolution. Timing out after
+ // sending the last probe transitions the entry to Failed.
+ {
+ wantProbes := []entryTestProbeInfo{
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+ }
+
+ clock.advance(c.RetransmitTimer)
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestRemoved,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+
+ e.mu.Lock()
+ if got, notWant := e.neigh.UpdatedAt, updatedAt; got == notWant {
+ t.Errorf("expected e.neigh.UpdatedAt to change, got = %q", got)
+ }
+ e.mu.Unlock()
+}
+
+func TestEntryIncompleteToReachable(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, _ := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ if got, want := e.neigh.State, Incomplete; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: true,
+ Override: false,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Reachable; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ wantProbes := []entryTestProbeInfo{
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Reachable,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+}
+
+// TestEntryAddsAndClearsWakers verifies that wakers are added when
+// addWakerLocked is called and cleared when address resolution finishes. In
+// this case, address resolution will finish when transitioning from Incomplete
+// to Reachable.
+func TestEntryAddsAndClearsWakers(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, _ := entryTestSetup(c)
+
+ w := sleep.Waker{}
+ s := sleep.Sleeper{}
+ s.AddWaker(&w, 123)
+ defer s.Done()
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ if got := e.wakers; got != nil {
+ t.Errorf("got e.wakers = %v, want = nil", got)
+ }
+ e.addWakerLocked(&w)
+ if got, want := w.IsAsserted(), false; got != want {
+ t.Errorf("waker.IsAsserted() = %t, want = %t", got, want)
+ }
+ if e.wakers == nil {
+ t.Error("expected e.wakers to be non-nil")
+ }
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: true,
+ Override: false,
+ IsRouter: false,
+ })
+ if e.wakers != nil {
+ t.Errorf("got e.wakers = %v, want = nil", e.wakers)
+ }
+ if got, want := w.IsAsserted(), true; got != want {
+ t.Errorf("waker.IsAsserted() = %t, want = %t", got, want)
+ }
+ e.mu.Unlock()
+
+ wantProbes := []entryTestProbeInfo{
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Reachable,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+}
+
+func TestEntryIncompleteToReachableWithRouterFlag(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, _ := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ if got, want := e.neigh.State, Incomplete; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: true,
+ Override: false,
+ IsRouter: true,
+ })
+ if got, want := e.neigh.State, Reachable; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ if got, want := e.isRouter, true; got != want {
+ t.Errorf("got e.isRouter = %t, want = %t", got, want)
+ }
+ e.mu.Unlock()
+
+ wantProbes := []entryTestProbeInfo{
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ if diff := cmp.Diff(linkRes.probes, wantProbes); diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+ linkRes.mu.Unlock()
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Reachable,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+}
+
+func TestEntryIncompleteToStale(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, _ := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ if got, want := e.neigh.State, Incomplete; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: false,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Stale; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ wantProbes := []entryTestProbeInfo{
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+}
+
+func TestEntryIncompleteToFailed(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ c.MaxMulticastProbes = 3
+ e, nudDisp, linkRes, clock := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ if got, want := e.neigh.State, Incomplete; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ waitFor := c.RetransmitTimer * time.Duration(c.MaxMulticastProbes)
+ clock.advance(waitFor)
+
+ wantProbes := []entryTestProbeInfo{
+ // The Incomplete-to-Incomplete state transition is tested here by
+ // verifying that 3 reachability probes were sent.
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestRemoved,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+
+ e.mu.Lock()
+ if got, want := e.neigh.State, Failed; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+}
+
+type testLocker struct{}
+
+var _ sync.Locker = (*testLocker)(nil)
+
+func (*testLocker) Lock() {}
+func (*testLocker) Unlock() {}
+
+func TestEntryStaysReachableWhenConfirmationWithRouterFlag(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, _ := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: true,
+ Override: false,
+ IsRouter: true,
+ })
+ if got, want := e.neigh.State, Reachable; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ if got, want := e.isRouter, true; got != want {
+ t.Errorf("got e.isRouter = %t, want = %t", got, want)
+ }
+ e.nic.mu.ndp.defaultRouters[entryTestAddr1] = defaultRouterState{
+ invalidationJob: e.nic.stack.newJob(&testLocker{}, func() {}),
+ }
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: false,
+ IsRouter: false,
+ })
+ if got, want := e.isRouter, false; got != want {
+ t.Errorf("got e.isRouter = %t, want = %t", got, want)
+ }
+ if _, ok := e.nic.mu.ndp.defaultRouters[entryTestAddr1]; ok {
+ t.Errorf("unexpected defaultRouter for %s", entryTestAddr1)
+ }
+ e.mu.Unlock()
+
+ wantProbes := []entryTestProbeInfo{
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Reachable,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+
+ e.mu.Lock()
+ if got, want := e.neigh.State, Reachable; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+}
+
+func TestEntryStaysReachableWhenProbeWithSameAddress(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, _ := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: true,
+ Override: false,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Reachable; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.handleProbeLocked(entryTestLinkAddr1)
+ if got, want := e.neigh.State, Reachable; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ if got, want := e.neigh.LinkAddr, entryTestLinkAddr1; got != want {
+ t.Errorf("got e.neigh.LinkAddr = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ wantProbes := []entryTestProbeInfo{
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Reachable,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+}
+
+func TestEntryReachableToStaleWhenTimeout(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ // Eliminate random factors from ReachableTime computation so the transition
+ // from Stale to Reachable will only take BaseReachableTime duration.
+ c.MinRandomFactor = 1
+ c.MaxRandomFactor = 1
+
+ e, nudDisp, linkRes, clock := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: true,
+ Override: false,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Reachable; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ wantProbes := []entryTestProbeInfo{
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ clock.advance(c.BaseReachableTime)
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Reachable,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+
+ e.mu.Lock()
+ if got, want := e.neigh.State, Stale; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+}
+
+func TestEntryReachableToStaleWhenProbeWithDifferentAddress(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, _ := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: true,
+ Override: false,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Reachable; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.handleProbeLocked(entryTestLinkAddr2)
+ if got, want := e.neigh.State, Stale; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ wantProbes := []entryTestProbeInfo{
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Reachable,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr2,
+ State: Stale,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+
+ e.mu.Lock()
+ if got, want := e.neigh.State, Stale; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+}
+
+func TestEntryReachableToStaleWhenConfirmationWithDifferentAddress(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, _ := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: true,
+ Override: false,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Reachable; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.handleConfirmationLocked(entryTestLinkAddr2, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: false,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Stale; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ wantProbes := []entryTestProbeInfo{
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Reachable,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+
+ e.mu.Lock()
+ if got, want := e.neigh.State, Stale; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+}
+
+func TestEntryReachableToStaleWhenConfirmationWithDifferentAddressAndOverride(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, _ := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: true,
+ Override: false,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Reachable; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.handleConfirmationLocked(entryTestLinkAddr2, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: true,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Stale; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ wantProbes := []entryTestProbeInfo{
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Reachable,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr2,
+ State: Stale,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+
+ e.mu.Lock()
+ if got, want := e.neigh.State, Stale; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+}
+
+func TestEntryStaysStaleWhenProbeWithSameAddress(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, _ := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: false,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Stale; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.handleProbeLocked(entryTestLinkAddr1)
+ if got, want := e.neigh.State, Stale; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ if got, want := e.neigh.LinkAddr, entryTestLinkAddr1; got != want {
+ t.Errorf("got e.neigh.LinkAddr = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ wantProbes := []entryTestProbeInfo{
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+}
+
+func TestEntryStaleToReachableWhenSolicitedOverrideConfirmation(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, _ := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: false,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Stale; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.handleConfirmationLocked(entryTestLinkAddr2, ReachabilityConfirmationFlags{
+ Solicited: true,
+ Override: true,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Reachable; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ if got, want := e.neigh.LinkAddr, entryTestLinkAddr2; got != want {
+ t.Errorf("got e.neigh.LinkAddr = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ wantProbes := []entryTestProbeInfo{
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr2,
+ State: Reachable,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+}
+
+func TestEntryStaleToStaleWhenOverrideConfirmation(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, _ := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: false,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Stale; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.handleConfirmationLocked(entryTestLinkAddr2, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: true,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Stale; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ if got, want := e.neigh.LinkAddr, entryTestLinkAddr2; got != want {
+ t.Errorf("got e.neigh.LinkAddr = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ wantProbes := []entryTestProbeInfo{
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr2,
+ State: Stale,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+}
+
+func TestEntryStaleToStaleWhenProbeUpdateAddress(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, _ := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: false,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Stale; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.handleProbeLocked(entryTestLinkAddr2)
+ if got, want := e.neigh.State, Stale; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ if got, want := e.neigh.LinkAddr, entryTestLinkAddr2; got != want {
+ t.Errorf("got e.neigh.LinkAddr = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ wantProbes := []entryTestProbeInfo{
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr2,
+ State: Stale,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+}
+
+func TestEntryStaleToDelay(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, _ := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: false,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Stale; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.handlePacketQueuedLocked()
+ if got, want := e.neigh.State, Delay; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ wantProbes := []entryTestProbeInfo{
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Delay,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+}
+
+func TestEntryDelayToReachableWhenUpperLevelConfirmation(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ // Eliminate random factors from ReachableTime computation so the transition
+ // from Stale to Reachable will only take BaseReachableTime duration.
+ c.MinRandomFactor = 1
+ c.MaxRandomFactor = 1
+
+ e, nudDisp, linkRes, clock := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: false,
+ IsRouter: false,
+ })
+ e.handlePacketQueuedLocked()
+ if got, want := e.neigh.State, Delay; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.handleUpperLevelConfirmationLocked()
+ if got, want := e.neigh.State, Reachable; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ wantProbes := []entryTestProbeInfo{
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ clock.advance(c.BaseReachableTime)
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Delay,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Reachable,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+}
+
+func TestEntryDelayToReachableWhenSolicitedOverrideConfirmation(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ c.MaxMulticastProbes = 1
+ // Eliminate random factors from ReachableTime computation so the transition
+ // from Stale to Reachable will only take BaseReachableTime duration.
+ c.MinRandomFactor = 1
+ c.MaxRandomFactor = 1
+
+ e, nudDisp, linkRes, clock := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: false,
+ IsRouter: false,
+ })
+ e.handlePacketQueuedLocked()
+ if got, want := e.neigh.State, Delay; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.handleConfirmationLocked(entryTestLinkAddr2, ReachabilityConfirmationFlags{
+ Solicited: true,
+ Override: true,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Reachable; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ if got, want := e.neigh.LinkAddr, entryTestLinkAddr2; got != want {
+ t.Errorf("got e.neigh.LinkAddr = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ wantProbes := []entryTestProbeInfo{
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ clock.advance(c.BaseReachableTime)
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Delay,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr2,
+ State: Reachable,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr2,
+ State: Stale,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+}
+
+func TestEntryStaysDelayWhenOverrideConfirmationWithSameAddress(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, _ := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: false,
+ IsRouter: false,
+ })
+ e.handlePacketQueuedLocked()
+ if got, want := e.neigh.State, Delay; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: true,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Delay; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ if got, want := e.neigh.LinkAddr, entryTestLinkAddr1; got != want {
+ t.Errorf("got e.neigh.LinkAddr = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ wantProbes := []entryTestProbeInfo{
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Delay,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+}
+
+func TestEntryDelayToStaleWhenProbeWithDifferentAddress(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, _ := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: false,
+ IsRouter: false,
+ })
+ e.handlePacketQueuedLocked()
+ if got, want := e.neigh.State, Delay; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.handleProbeLocked(entryTestLinkAddr2)
+ if got, want := e.neigh.State, Stale; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ wantProbes := []entryTestProbeInfo{
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Delay,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr2,
+ State: Stale,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+}
+
+func TestEntryDelayToStaleWhenConfirmationWithDifferentAddress(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, _ := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: false,
+ IsRouter: false,
+ })
+ e.handlePacketQueuedLocked()
+ if got, want := e.neigh.State, Delay; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.handleConfirmationLocked(entryTestLinkAddr2, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: true,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Stale; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ wantProbes := []entryTestProbeInfo{
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Delay,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr2,
+ State: Stale,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+}
+
+func TestEntryDelayToProbe(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, clock := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: false,
+ IsRouter: false,
+ })
+ e.handlePacketQueuedLocked()
+ if got, want := e.neigh.State, Delay; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ clock.advance(c.DelayFirstProbeTime)
+
+ wantProbes := []entryTestProbeInfo{
+ // The first probe is caused by the Unknown-to-Incomplete transition.
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ // The second probe is caused by the Delay-to-Probe transition.
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: entryTestLinkAddr1,
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Delay,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Probe,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+
+ e.mu.Lock()
+ if got, want := e.neigh.State, Probe; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+}
+
+func TestEntryProbeToStaleWhenProbeWithDifferentAddress(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, clock := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: false,
+ IsRouter: false,
+ })
+ e.handlePacketQueuedLocked()
+ e.mu.Unlock()
+
+ clock.advance(c.DelayFirstProbeTime)
+
+ wantProbes := []entryTestProbeInfo{
+ // The first probe is caused by the Unknown-to-Incomplete transition.
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ // The second probe is caused by the Delay-to-Probe transition.
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: entryTestLinkAddr1,
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ e.mu.Lock()
+ if got, want := e.neigh.State, Probe; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.handleProbeLocked(entryTestLinkAddr2)
+ if got, want := e.neigh.State, Stale; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Delay,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Probe,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr2,
+ State: Stale,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+
+ e.mu.Lock()
+ if got, want := e.neigh.State, Stale; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+}
+
+func TestEntryProbeToStaleWhenConfirmationWithDifferentAddress(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, clock := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: false,
+ IsRouter: false,
+ })
+ e.handlePacketQueuedLocked()
+ e.mu.Unlock()
+
+ clock.advance(c.DelayFirstProbeTime)
+
+ wantProbes := []entryTestProbeInfo{
+ // The first probe is caused by the Unknown-to-Incomplete transition.
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ // The second probe is caused by the Delay-to-Probe transition.
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: entryTestLinkAddr1,
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ e.mu.Lock()
+ if got, want := e.neigh.State, Probe; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.handleConfirmationLocked(entryTestLinkAddr2, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: true,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Stale; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Delay,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Probe,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr2,
+ State: Stale,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+
+ e.mu.Lock()
+ if got, want := e.neigh.State, Stale; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+}
+
+func TestEntryStaysProbeWhenOverrideConfirmationWithSameAddress(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ e, nudDisp, linkRes, clock := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: false,
+ IsRouter: false,
+ })
+ e.handlePacketQueuedLocked()
+ e.mu.Unlock()
+
+ clock.advance(c.DelayFirstProbeTime)
+
+ wantProbes := []entryTestProbeInfo{
+ // The first probe is caused by the Unknown-to-Incomplete transition.
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ // The second probe is caused by the Delay-to-Probe transition.
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: entryTestLinkAddr1,
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ e.mu.Lock()
+ if got, want := e.neigh.State, Probe; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: true,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Probe; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ if got, want := e.neigh.LinkAddr, entryTestLinkAddr1; got != want {
+ t.Errorf("got e.neigh.LinkAddr = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Delay,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Probe,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+}
+
+// TestEntryUnknownToStaleToProbeToReachable exercises the following scenario:
+// 1. Probe is received
+// 2. Entry is created in Stale
+// 3. Packet is queued on the entry
+// 4. Entry transitions to Delay then Probe
+// 5. Probe is sent
+func TestEntryUnknownToStaleToProbeToReachable(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ // Eliminate random factors from ReachableTime computation so the transition
+ // from Probe to Reachable will only take BaseReachableTime duration.
+ c.MinRandomFactor = 1
+ c.MaxRandomFactor = 1
+
+ e, nudDisp, linkRes, clock := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handleProbeLocked(entryTestLinkAddr1)
+ e.handlePacketQueuedLocked()
+ e.mu.Unlock()
+
+ clock.advance(c.DelayFirstProbeTime)
+
+ wantProbes := []entryTestProbeInfo{
+ // Probe caused by the Delay-to-Probe transition
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: entryTestLinkAddr1,
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ e.mu.Lock()
+ if got, want := e.neigh.State, Probe; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.handleConfirmationLocked(entryTestLinkAddr2, ReachabilityConfirmationFlags{
+ Solicited: true,
+ Override: true,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Reachable; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ if got, want := e.neigh.LinkAddr, entryTestLinkAddr2; got != want {
+ t.Errorf("got e.neigh.LinkAddr = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ clock.advance(c.BaseReachableTime)
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Delay,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Probe,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr2,
+ State: Reachable,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr2,
+ State: Stale,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+}
+
+func TestEntryProbeToReachableWhenSolicitedOverrideConfirmation(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ // Eliminate random factors from ReachableTime computation so the transition
+ // from Stale to Reachable will only take BaseReachableTime duration.
+ c.MinRandomFactor = 1
+ c.MaxRandomFactor = 1
+
+ e, nudDisp, linkRes, clock := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: false,
+ IsRouter: false,
+ })
+ e.handlePacketQueuedLocked()
+ e.mu.Unlock()
+
+ clock.advance(c.DelayFirstProbeTime)
+
+ wantProbes := []entryTestProbeInfo{
+ // The first probe is caused by the Unknown-to-Incomplete transition.
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ // The second probe is caused by the Delay-to-Probe transition.
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: entryTestLinkAddr1,
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ e.mu.Lock()
+ if got, want := e.neigh.State, Probe; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.handleConfirmationLocked(entryTestLinkAddr2, ReachabilityConfirmationFlags{
+ Solicited: true,
+ Override: true,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Reachable; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ if got, want := e.neigh.LinkAddr, entryTestLinkAddr2; got != want {
+ t.Errorf("got e.neigh.LinkAddr = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ clock.advance(c.BaseReachableTime)
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Delay,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Probe,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr2,
+ State: Reachable,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr2,
+ State: Stale,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+}
+
+func TestEntryProbeToReachableWhenSolicitedConfirmationWithSameAddress(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ // Eliminate random factors from ReachableTime computation so the transition
+ // from Stale to Reachable will only take BaseReachableTime duration.
+ c.MinRandomFactor = 1
+ c.MaxRandomFactor = 1
+
+ e, nudDisp, linkRes, clock := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: false,
+ IsRouter: false,
+ })
+ e.handlePacketQueuedLocked()
+ e.mu.Unlock()
+
+ clock.advance(c.DelayFirstProbeTime)
+
+ wantProbes := []entryTestProbeInfo{
+ // The first probe is caused by the Unknown-to-Incomplete transition.
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ // The second probe is caused by the Delay-to-Probe transition.
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: entryTestLinkAddr1,
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ e.mu.Lock()
+ if got, want := e.neigh.State, Probe; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: true,
+ Override: false,
+ IsRouter: false,
+ })
+ if got, want := e.neigh.State, Reachable; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+
+ clock.advance(c.BaseReachableTime)
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Delay,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Probe,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Reachable,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+}
+
+func TestEntryProbeToFailed(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ c.MaxMulticastProbes = 3
+ c.MaxUnicastProbes = 3
+ e, nudDisp, linkRes, clock := entryTestSetup(c)
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: false,
+ IsRouter: false,
+ })
+ e.handlePacketQueuedLocked()
+ e.mu.Unlock()
+
+ waitFor := c.DelayFirstProbeTime + c.RetransmitTimer*time.Duration(c.MaxUnicastProbes)
+ clock.advance(waitFor)
+
+ wantProbes := []entryTestProbeInfo{
+ // The first probe is caused by the Unknown-to-Incomplete transition.
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ // The next three probe are caused by the Delay-to-Probe transition.
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: entryTestLinkAddr1,
+ LocalAddress: entryTestAddr2,
+ },
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: entryTestLinkAddr1,
+ LocalAddress: entryTestAddr2,
+ },
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: entryTestLinkAddr1,
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Delay,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Probe,
+ },
+ {
+ EventType: entryTestRemoved,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Probe,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+
+ e.mu.Lock()
+ if got, want := e.neigh.State, Failed; got != want {
+ t.Errorf("got e.neigh.State = %q, want = %q", got, want)
+ }
+ e.mu.Unlock()
+}
+
+func TestEntryFailedGetsDeleted(t *testing.T) {
+ c := DefaultNUDConfigurations()
+ c.MaxMulticastProbes = 3
+ c.MaxUnicastProbes = 3
+ e, nudDisp, linkRes, clock := entryTestSetup(c)
+
+ // Verify the cache contains the entry.
+ if _, ok := e.nic.neigh.cache[entryTestAddr1]; !ok {
+ t.Errorf("expected entry %q to exist in the neighbor cache", entryTestAddr1)
+ }
+
+ e.mu.Lock()
+ e.handlePacketQueuedLocked()
+ e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{
+ Solicited: false,
+ Override: false,
+ IsRouter: false,
+ })
+ e.handlePacketQueuedLocked()
+ e.mu.Unlock()
+
+ waitFor := c.DelayFirstProbeTime + c.RetransmitTimer*time.Duration(c.MaxUnicastProbes) + c.UnreachableTime
+ clock.advance(waitFor)
+
+ wantProbes := []entryTestProbeInfo{
+ // The first probe is caused by the Unknown-to-Incomplete transition.
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: tcpip.LinkAddress(""),
+ LocalAddress: entryTestAddr2,
+ },
+ // The next three probe are caused by the Delay-to-Probe transition.
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: entryTestLinkAddr1,
+ LocalAddress: entryTestAddr2,
+ },
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: entryTestLinkAddr1,
+ LocalAddress: entryTestAddr2,
+ },
+ {
+ RemoteAddress: entryTestAddr1,
+ RemoteLinkAddress: entryTestLinkAddr1,
+ LocalAddress: entryTestAddr2,
+ },
+ }
+ linkRes.mu.Lock()
+ diff := cmp.Diff(linkRes.probes, wantProbes)
+ linkRes.mu.Unlock()
+ if diff != "" {
+ t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff)
+ }
+
+ wantEvents := []testEntryEventInfo{
+ {
+ EventType: entryTestAdded,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: tcpip.LinkAddress(""),
+ State: Incomplete,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Stale,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Delay,
+ },
+ {
+ EventType: entryTestChanged,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Probe,
+ },
+ {
+ EventType: entryTestRemoved,
+ NICID: entryTestNICID,
+ Addr: entryTestAddr1,
+ LinkAddr: entryTestLinkAddr1,
+ State: Probe,
+ },
+ }
+ nudDisp.mu.Lock()
+ if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" {
+ t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff)
+ }
+ nudDisp.mu.Unlock()
+
+ // Verify the cache no longer contains the entry.
+ if _, ok := e.nic.neigh.cache[entryTestAddr1]; ok {
+ t.Errorf("entry %q should have been deleted from the neighbor cache", entryTestAddr1)
+ }
+}
diff --git a/pkg/tcpip/stack/neighborstate_string.go b/pkg/tcpip/stack/neighborstate_string.go
new file mode 100644
index 000000000..aa7311ec6
--- /dev/null
+++ b/pkg/tcpip/stack/neighborstate_string.go
@@ -0,0 +1,44 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Code generated by "stringer -type NeighborState"; DO NOT EDIT.
+
+package stack
+
+import "strconv"
+
+func _() {
+ // An "invalid array index" compiler error signifies that the constant values have changed.
+ // Re-run the stringer command to generate them again.
+ var x [1]struct{}
+ _ = x[Unknown-0]
+ _ = x[Incomplete-1]
+ _ = x[Reachable-2]
+ _ = x[Stale-3]
+ _ = x[Delay-4]
+ _ = x[Probe-5]
+ _ = x[Static-6]
+ _ = x[Failed-7]
+}
+
+const _NeighborState_name = "UnknownIncompleteReachableStaleDelayProbeStaticFailed"
+
+var _NeighborState_index = [...]uint8{0, 7, 17, 26, 31, 36, 41, 47, 53}
+
+func (i NeighborState) String() string {
+ if i >= NeighborState(len(_NeighborState_index)-1) {
+ return "NeighborState(" + strconv.FormatInt(int64(i), 10) + ")"
+ }
+ return _NeighborState_name[_NeighborState_index[i]:_NeighborState_index[i+1]]
+}
diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go
index ab6798aa6..e74d2562a 100644
--- a/pkg/tcpip/stack/nic.go
+++ b/pkg/tcpip/stack/nic.go
@@ -15,48 +15,66 @@
package stack
import (
- "strings"
- "sync"
+ "fmt"
+ "math/rand"
+ "reflect"
+ "sort"
"sync/atomic"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/tcpip/header"
)
+var ipv4BroadcastAddr = tcpip.ProtocolAddress{
+ Protocol: header.IPv4ProtocolNumber,
+ AddressWithPrefix: tcpip.AddressWithPrefix{
+ Address: header.IPv4Broadcast,
+ PrefixLen: 8 * header.IPv4AddressSize,
+ },
+}
+
// NIC represents a "network interface card" to which the networking stack is
// attached.
type NIC struct {
- stack *Stack
- id tcpip.NICID
- name string
- linkEP LinkEndpoint
- loopback bool
-
- mu sync.RWMutex
- spoofing bool
- promiscuous bool
- 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
+ stack *Stack
+ id tcpip.NICID
+ name string
+ linkEP LinkEndpoint
+ context NICContext
+
+ stats NICStats
+ neigh *neighborCache
+ networkEndpoints map[tcpip.NetworkProtocolNumber]NetworkEndpoint
+
+ mu struct {
+ sync.RWMutex
+ enabled bool
+ spoofing bool
+ promiscuous bool
+ primary map[tcpip.NetworkProtocolNumber][]*referencedNetworkEndpoint
+ endpoints map[NetworkEndpointID]*referencedNetworkEndpoint
+ mcastJoins map[NetworkEndpointID]uint32
+ // packetEPs is protected by mu, but the contained PacketEndpoint
+ // values are not.
+ packetEPs map[tcpip.NetworkProtocolNumber][]PacketEndpoint
+ ndp ndpState
+ }
}
// NICStats includes transmitted and received stats.
type NICStats struct {
Tx DirectionStats
Rx DirectionStats
+
+ DisabledRx DirectionStats
+}
+
+func makeNICStats() NICStats {
+ var s NICStats
+ tcpip.InitStatCounters(reflect.ValueOf(&s).Elem())
+ return s
}
// DirectionStats includes packet and byte counts.
@@ -85,59 +103,170 @@ const (
)
// 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 {
+func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, ctx NICContext) *NIC {
// TODO(b/141011931): Validate a LinkEndpoint (ep) is valid. For
// example, make sure that the link address it provides is a valid
// unicast ethernet address.
+
+ // TODO(b/143357959): RFC 8200 section 5 requires that IPv6 endpoints
+ // observe an MTU of at least 1280 bytes. Ensure that this requirement
+ // of IPv6 is supported on this endpoint's LinkEndpoint.
+
nic := &NIC{
- stack: stack,
- id: id,
- name: name,
- linkEP: ep,
- loopback: loopback,
- primary: make(map[tcpip.NetworkProtocolNumber][]*referencedNetworkEndpoint),
- endpoints: make(map[NetworkEndpointID]*referencedNetworkEndpoint),
- mcastJoins: make(map[NetworkEndpointID]int32),
- packetEPs: make(map[tcpip.NetworkProtocolNumber][]PacketEndpoint),
- stats: NICStats{
- Tx: DirectionStats{
- Packets: &tcpip.StatCounter{},
- Bytes: &tcpip.StatCounter{},
- },
- Rx: DirectionStats{
- Packets: &tcpip.StatCounter{},
- Bytes: &tcpip.StatCounter{},
- },
- },
- ndp: ndpState{
- configs: stack.ndpConfigs,
- dad: make(map[tcpip.Address]dadState),
- },
- }
- nic.ndp.nic = nic
+ stack: stack,
+ id: id,
+ name: name,
+ linkEP: ep,
+ context: ctx,
+ stats: makeNICStats(),
+ networkEndpoints: make(map[tcpip.NetworkProtocolNumber]NetworkEndpoint),
+ }
+ nic.mu.primary = make(map[tcpip.NetworkProtocolNumber][]*referencedNetworkEndpoint)
+ nic.mu.endpoints = make(map[NetworkEndpointID]*referencedNetworkEndpoint)
+ nic.mu.mcastJoins = make(map[NetworkEndpointID]uint32)
+ nic.mu.packetEPs = make(map[tcpip.NetworkProtocolNumber][]PacketEndpoint)
+ nic.mu.ndp = ndpState{
+ nic: nic,
+ configs: stack.ndpConfigs,
+ dad: make(map[tcpip.Address]dadState),
+ defaultRouters: make(map[tcpip.Address]defaultRouterState),
+ onLinkPrefixes: make(map[tcpip.Subnet]onLinkPrefixState),
+ slaacPrefixes: make(map[tcpip.Subnet]slaacPrefixState),
+ }
+ nic.mu.ndp.initializeTempAddrState()
// Register supported packet endpoint protocols.
for _, netProto := range header.Ethertypes {
- nic.packetEPs[netProto] = []PacketEndpoint{}
+ nic.mu.packetEPs[netProto] = []PacketEndpoint{}
}
for _, netProto := range stack.networkProtocols {
- nic.packetEPs[netProto.Number()] = []PacketEndpoint{}
+ netNum := netProto.Number()
+ nic.mu.packetEPs[netNum] = nil
+ nic.networkEndpoints[netNum] = netProto.NewEndpoint(id, stack, nic, ep, stack)
+ }
+
+ // Check for Neighbor Unreachability Detection support.
+ if ep.Capabilities()&CapabilityResolutionRequired != 0 && len(stack.linkAddrResolvers) != 0 {
+ rng := rand.New(rand.NewSource(stack.clock.NowNanoseconds()))
+ nic.neigh = &neighborCache{
+ nic: nic,
+ state: NewNUDState(stack.nudConfigs, rng),
+ cache: make(map[tcpip.Address]*neighborEntry, neighborCacheSize),
+ }
}
+ nic.linkEP.Attach(nic)
+
return nic
}
-// enable enables the NIC. enable will attach the link to its LinkEndpoint and
-// join the IPv6 All-Nodes Multicast address (ff02::1).
+// enabled returns true if n is enabled.
+func (n *NIC) enabled() bool {
+ n.mu.RLock()
+ enabled := n.mu.enabled
+ n.mu.RUnlock()
+ return enabled
+}
+
+// disable disables n.
+//
+// It undoes the work done by enable.
+func (n *NIC) disable() *tcpip.Error {
+ n.mu.RLock()
+ enabled := n.mu.enabled
+ n.mu.RUnlock()
+ if !enabled {
+ return nil
+ }
+
+ n.mu.Lock()
+ err := n.disableLocked()
+ n.mu.Unlock()
+ return err
+}
+
+// disableLocked disables n.
+//
+// It undoes the work done by enable.
+//
+// n MUST be locked.
+func (n *NIC) disableLocked() *tcpip.Error {
+ if !n.mu.enabled {
+ return nil
+ }
+
+ // TODO(gvisor.dev/issue/1491): Should Routes that are currently bound to n be
+ // invalidated? Currently, Routes will continue to work when a NIC is enabled
+ // again, and applications may not know that the underlying NIC was ever
+ // disabled.
+
+ if _, ok := n.stack.networkProtocols[header.IPv6ProtocolNumber]; ok {
+ n.mu.ndp.stopSolicitingRouters()
+ n.mu.ndp.cleanupState(false /* hostOnly */)
+
+ // Stop DAD for all the unicast IPv6 endpoints that are in the
+ // permanentTentative state.
+ for _, r := range n.mu.endpoints {
+ if addr := r.address(); r.getKind() == permanentTentative && header.IsV6UnicastAddress(addr) {
+ n.mu.ndp.stopDuplicateAddressDetection(addr)
+ }
+ }
+
+ // The NIC may have already left the multicast group.
+ if err := n.leaveGroupLocked(header.IPv6AllNodesMulticastAddress, false /* force */); err != nil && err != tcpip.ErrBadLocalAddress {
+ return err
+ }
+ }
+
+ if _, ok := n.stack.networkProtocols[header.IPv4ProtocolNumber]; ok {
+ // The NIC may have already left the multicast group.
+ if err := n.leaveGroupLocked(header.IPv4AllSystems, false /* force */); err != nil && err != tcpip.ErrBadLocalAddress {
+ return err
+ }
+
+ // The address may have already been removed.
+ if err := n.removePermanentAddressLocked(ipv4BroadcastAddr.AddressWithPrefix.Address); err != nil && err != tcpip.ErrBadLocalAddress {
+ return err
+ }
+ }
+
+ n.mu.enabled = false
+ return nil
+}
+
+// enable enables n.
+//
+// If the stack has IPv6 enabled, enable will join the IPv6 All-Nodes Multicast
+// address (ff02::1), start DAD for permanent addresses, and start soliciting
+// routers if the stack is not operating as a router. If the stack is also
+// configured to auto-generate a link-local address, one will be generated.
func (n *NIC) enable() *tcpip.Error {
- n.attachLinkEndpoint()
+ n.mu.RLock()
+ enabled := n.mu.enabled
+ n.mu.RUnlock()
+ if enabled {
+ return nil
+ }
+
+ n.mu.Lock()
+ defer n.mu.Unlock()
+
+ if n.mu.enabled {
+ return nil
+ }
+
+ n.mu.enabled = true
// Create an endpoint to receive broadcast packets on this interface.
if _, ok := n.stack.networkProtocols[header.IPv4ProtocolNumber]; ok {
- if err := n.AddAddress(tcpip.ProtocolAddress{
- Protocol: header.IPv4ProtocolNumber,
- AddressWithPrefix: tcpip.AddressWithPrefix{header.IPv4Broadcast, 8 * header.IPv4AddressSize},
- }, NeverPrimaryEndpoint); err != nil {
+ if _, err := n.addAddressLocked(ipv4BroadcastAddr, NeverPrimaryEndpoint, permanent, static, false /* deprecated */); err != nil {
+ return err
+ }
+
+ // As per RFC 1122 section 3.3.7, all hosts should join the all-hosts
+ // multicast group. Note, the IANA calls the all-hosts multicast group the
+ // all-systems multicast group.
+ if err := n.joinGroupLocked(header.IPv4ProtocolNumber, header.IPv4AllSystems); err != nil {
return err
}
}
@@ -159,77 +288,298 @@ func (n *NIC) enable() *tcpip.Error {
return nil
}
- n.mu.Lock()
- defer n.mu.Unlock()
-
+ // Join the All-Nodes multicast group before starting DAD as responses to DAD
+ // (NDP NS) messages may be sent to the All-Nodes multicast group if the
+ // source address of the NDP NS is the unspecified address, as per RFC 4861
+ // section 7.2.4.
if err := n.joinGroupLocked(header.IPv6ProtocolNumber, header.IPv6AllNodesMulticastAddress); err != nil {
return err
}
- if !n.stack.autoGenIPv6LinkLocal {
- return nil
+ // Perform DAD on the all the unicast IPv6 endpoints that are in the permanent
+ // state.
+ //
+ // Addresses may have aleady completed DAD but in the time since the NIC was
+ // last enabled, other devices may have acquired the same addresses.
+ for _, r := range n.mu.endpoints {
+ addr := r.address()
+ if k := r.getKind(); (k != permanent && k != permanentTentative) || !header.IsV6UnicastAddress(addr) {
+ continue
+ }
+
+ r.setKind(permanentTentative)
+ if err := n.mu.ndp.startDuplicateAddressDetection(addr, r); err != nil {
+ return err
+ }
}
- l2addr := n.linkEP.LinkAddress()
+ // Do not auto-generate an IPv6 link-local address for loopback devices.
+ if n.stack.autoGenIPv6LinkLocal && !n.isLoopback() {
+ // The valid and preferred lifetime is infinite for the auto-generated
+ // link-local address.
+ n.mu.ndp.doSLAAC(header.IPv6LinkLocalPrefix.Subnet(), header.NDPInfiniteLifetime, header.NDPInfiniteLifetime)
+ }
- // Only attempt to generate the link-local address if we have a
- // valid MAC address.
+ // If we are operating as a router, then do not solicit routers since we
+ // won't process the RAs anyways.
//
- // TODO(b/141011931): Validate a LinkEndpoint's link address
- // (provided by LinkEndpoint.LinkAddress) before reaching this
- // point.
- if !header.IsValidUnicastEthernetAddress(l2addr) {
- return nil
+ // Routers do not process Router Advertisements (RA) the same way a host
+ // does. That is, routers do not learn from RAs (e.g. on-link prefixes
+ // and default routers). Therefore, soliciting RAs from other routers on
+ // a link is unnecessary for routers.
+ if !n.stack.Forwarding(header.IPv6ProtocolNumber) {
+ n.mu.ndp.startSolicitingRouters()
}
- addr := header.LinkLocalAddr(l2addr)
+ return nil
+}
- _, err := n.addPermanentAddressLocked(tcpip.ProtocolAddress{
- Protocol: header.IPv6ProtocolNumber,
- AddressWithPrefix: tcpip.AddressWithPrefix{
- Address: addr,
- PrefixLen: header.IPv6LinkLocalPrefix.PrefixLen,
- },
- }, CanBePrimaryEndpoint)
+// remove detaches NIC from the link endpoint, and marks existing referenced
+// network endpoints expired. This guarantees no packets between this NIC and
+// the network stack.
+func (n *NIC) remove() *tcpip.Error {
+ n.mu.Lock()
+ defer n.mu.Unlock()
+
+ n.disableLocked()
+
+ // TODO(b/151378115): come up with a better way to pick an error than the
+ // first one.
+ var err *tcpip.Error
+
+ // Forcefully leave multicast groups.
+ for nid := range n.mu.mcastJoins {
+ if tempErr := n.leaveGroupLocked(nid.LocalAddress, true /* force */); tempErr != nil && err == nil {
+ err = tempErr
+ }
+ }
+
+ // Remove permanent and permanentTentative addresses, so no packet goes out.
+ for nid, ref := range n.mu.endpoints {
+ switch ref.getKind() {
+ case permanentTentative, permanent:
+ if tempErr := n.removePermanentAddressLocked(nid.LocalAddress); tempErr != nil && err == nil {
+ err = tempErr
+ }
+ }
+ }
+
+ // Release any resources the network endpoint may hold.
+ for _, ep := range n.networkEndpoints {
+ ep.Close()
+ }
+
+ // Detach from link endpoint, so no packet comes in.
+ n.linkEP.Attach(nil)
return err
}
-// attachLinkEndpoint attaches the NIC to the endpoint, which will enable it
-// to start delivering packets.
-func (n *NIC) attachLinkEndpoint() {
- n.linkEP.Attach(n)
+// becomeIPv6Router transitions n into an IPv6 router.
+//
+// When transitioning into an IPv6 router, host-only state (NDP discovered
+// routers, discovered on-link prefixes, and auto-generated addresses) will
+// be cleaned up/invalidated and NDP router solicitations will be stopped.
+func (n *NIC) becomeIPv6Router() {
+ n.mu.Lock()
+ defer n.mu.Unlock()
+
+ n.mu.ndp.cleanupState(true /* hostOnly */)
+ n.mu.ndp.stopSolicitingRouters()
+}
+
+// becomeIPv6Host transitions n into an IPv6 host.
+//
+// When transitioning into an IPv6 host, NDP router solicitations will be
+// started.
+func (n *NIC) becomeIPv6Host() {
+ n.mu.Lock()
+ defer n.mu.Unlock()
+
+ n.mu.ndp.startSolicitingRouters()
}
// setPromiscuousMode enables or disables promiscuous mode.
func (n *NIC) setPromiscuousMode(enable bool) {
n.mu.Lock()
- n.promiscuous = enable
+ n.mu.promiscuous = enable
n.mu.Unlock()
}
func (n *NIC) isPromiscuousMode() bool {
n.mu.RLock()
- rv := n.promiscuous
+ rv := n.mu.promiscuous
n.mu.RUnlock()
return rv
}
+func (n *NIC) isLoopback() bool {
+ return n.linkEP.Capabilities()&CapabilityLoopback != 0
+}
+
// setSpoofing enables or disables address spoofing.
func (n *NIC) setSpoofing(enable bool) {
n.mu.Lock()
- n.spoofing = enable
+ n.mu.spoofing = enable
n.mu.Unlock()
}
-// primaryEndpoint returns the primary endpoint of n for the given network
-// protocol.
-func (n *NIC) primaryEndpoint(protocol tcpip.NetworkProtocolNumber) *referencedNetworkEndpoint {
+// primaryEndpoint will return the first non-deprecated endpoint if such an
+// endpoint exists for the given protocol and remoteAddr. If no non-deprecated
+// endpoint exists, the first deprecated endpoint will be returned.
+//
+// If an IPv6 primary endpoint is requested, Source Address Selection (as
+// defined by RFC 6724 section 5) will be performed.
+func (n *NIC) primaryEndpoint(protocol tcpip.NetworkProtocolNumber, remoteAddr tcpip.Address) *referencedNetworkEndpoint {
+ if protocol == header.IPv6ProtocolNumber && remoteAddr != "" {
+ return n.primaryIPv6Endpoint(remoteAddr)
+ }
+
n.mu.RLock()
defer n.mu.RUnlock()
- for _, r := range n.primary[protocol] {
- if r.isValidForOutgoing() && r.tryIncRef() {
+ var deprecatedEndpoint *referencedNetworkEndpoint
+ for _, r := range n.mu.primary[protocol] {
+ if !r.isValidForOutgoingRLocked() {
+ continue
+ }
+
+ if !r.deprecated {
+ if r.tryIncRef() {
+ // r is not deprecated, so return it immediately.
+ //
+ // If we kept track of a deprecated endpoint, decrement its reference
+ // count since it was incremented when we decided to keep track of it.
+ if deprecatedEndpoint != nil {
+ deprecatedEndpoint.decRefLocked()
+ deprecatedEndpoint = nil
+ }
+
+ return r
+ }
+ } else if deprecatedEndpoint == nil && r.tryIncRef() {
+ // We prefer an endpoint that is not deprecated, but we keep track of r in
+ // case n doesn't have any non-deprecated endpoints.
+ //
+ // If we end up finding a more preferred endpoint, r's reference count
+ // will be decremented when such an endpoint is found.
+ deprecatedEndpoint = r
+ }
+ }
+
+ // n doesn't have any valid non-deprecated endpoints, so return
+ // deprecatedEndpoint (which may be nil if n doesn't have any valid deprecated
+ // endpoints either).
+ return deprecatedEndpoint
+}
+
+// ipv6AddrCandidate is an IPv6 candidate for Source Address Selection (RFC
+// 6724 section 5).
+type ipv6AddrCandidate struct {
+ ref *referencedNetworkEndpoint
+ scope header.IPv6AddressScope
+}
+
+// primaryIPv6Endpoint returns an IPv6 endpoint following Source Address
+// Selection (RFC 6724 section 5).
+//
+// Note, only rules 1-3 and 7 are followed.
+//
+// remoteAddr must be a valid IPv6 address.
+func (n *NIC) primaryIPv6Endpoint(remoteAddr tcpip.Address) *referencedNetworkEndpoint {
+ n.mu.RLock()
+ ref := n.primaryIPv6EndpointRLocked(remoteAddr)
+ n.mu.RUnlock()
+ return ref
+}
+
+// primaryIPv6EndpointLocked returns an IPv6 endpoint following Source Address
+// Selection (RFC 6724 section 5).
+//
+// Note, only rules 1-3 and 7 are followed.
+//
+// remoteAddr must be a valid IPv6 address.
+//
+// n.mu MUST be read locked.
+func (n *NIC) primaryIPv6EndpointRLocked(remoteAddr tcpip.Address) *referencedNetworkEndpoint {
+ primaryAddrs := n.mu.primary[header.IPv6ProtocolNumber]
+
+ if len(primaryAddrs) == 0 {
+ return nil
+ }
+
+ // Create a candidate set of available addresses we can potentially use as a
+ // source address.
+ cs := make([]ipv6AddrCandidate, 0, len(primaryAddrs))
+ for _, r := range primaryAddrs {
+ // If r is not valid for outgoing connections, it is not a valid endpoint.
+ if !r.isValidForOutgoingRLocked() {
+ continue
+ }
+
+ addr := r.address()
+ scope, err := header.ScopeForIPv6Address(addr)
+ if err != nil {
+ // Should never happen as we got r from the primary IPv6 endpoint list and
+ // ScopeForIPv6Address only returns an error if addr is not an IPv6
+ // address.
+ panic(fmt.Sprintf("header.ScopeForIPv6Address(%s): %s", addr, err))
+ }
+
+ cs = append(cs, ipv6AddrCandidate{
+ ref: r,
+ scope: scope,
+ })
+ }
+
+ remoteScope, err := header.ScopeForIPv6Address(remoteAddr)
+ if err != nil {
+ // primaryIPv6Endpoint should never be called with an invalid IPv6 address.
+ panic(fmt.Sprintf("header.ScopeForIPv6Address(%s): %s", remoteAddr, err))
+ }
+
+ // Sort the addresses as per RFC 6724 section 5 rules 1-3.
+ //
+ // TODO(b/146021396): Implement rules 4-8 of RFC 6724 section 5.
+ sort.Slice(cs, func(i, j int) bool {
+ sa := cs[i]
+ sb := cs[j]
+
+ // Prefer same address as per RFC 6724 section 5 rule 1.
+ if sa.ref.address() == remoteAddr {
+ return true
+ }
+ if sb.ref.address() == remoteAddr {
+ return false
+ }
+
+ // Prefer appropriate scope as per RFC 6724 section 5 rule 2.
+ if sa.scope < sb.scope {
+ return sa.scope >= remoteScope
+ } else if sb.scope < sa.scope {
+ return sb.scope < remoteScope
+ }
+
+ // Avoid deprecated addresses as per RFC 6724 section 5 rule 3.
+ if saDep, sbDep := sa.ref.deprecated, sb.ref.deprecated; saDep != sbDep {
+ // If sa is not deprecated, it is preferred over sb.
+ return sbDep
+ }
+
+ // Prefer temporary addresses as per RFC 6724 section 5 rule 7.
+ if saTemp, sbTemp := sa.ref.configType == slaacTemp, sb.ref.configType == slaacTemp; saTemp != sbTemp {
+ return saTemp
+ }
+
+ // sa and sb are equal, return the endpoint that is closest to the front of
+ // the primary endpoint list.
+ return i < j
+ })
+
+ // Return the most preferred address that can have its reference count
+ // incremented.
+ for _, c := range cs {
+ if r := c.ref; r.tryIncRef() {
return r
}
}
@@ -237,62 +587,87 @@ func (n *NIC) primaryEndpoint(protocol tcpip.NetworkProtocolNumber) *referencedN
return nil
}
+// hasPermanentAddrLocked returns true if n has a permanent (including currently
+// tentative) address, addr.
+func (n *NIC) hasPermanentAddrLocked(addr tcpip.Address) bool {
+ ref, ok := n.mu.endpoints[NetworkEndpointID{addr}]
+
+ if !ok {
+ return false
+ }
+
+ kind := ref.getKind()
+
+ return kind == permanent || kind == permanentTentative
+}
+
+type getRefBehaviour int
+
+const (
+ // spoofing indicates that the NIC's spoofing flag should be observed when
+ // getting a NIC's referenced network endpoint.
+ spoofing getRefBehaviour = iota
+
+ // promiscuous indicates that the NIC's promiscuous flag should be observed
+ // when getting a NIC's referenced network endpoint.
+ promiscuous
+)
+
func (n *NIC) getRef(protocol tcpip.NetworkProtocolNumber, dst tcpip.Address) *referencedNetworkEndpoint {
- return n.getRefOrCreateTemp(protocol, dst, CanBePrimaryEndpoint, n.promiscuous)
+ return n.getRefOrCreateTemp(protocol, dst, CanBePrimaryEndpoint, promiscuous)
}
// findEndpoint finds the endpoint, if any, with the given address.
func (n *NIC) findEndpoint(protocol tcpip.NetworkProtocolNumber, address tcpip.Address, peb PrimaryEndpointBehavior) *referencedNetworkEndpoint {
- return n.getRefOrCreateTemp(protocol, address, peb, n.spoofing)
+ return n.getRefOrCreateTemp(protocol, address, peb, spoofing)
}
// getRefEpOrCreateTemp returns the referenced network endpoint for the given
-// protocol and address. If none exists a temporary one may be created if
-// we are in promiscuous mode or spoofing.
-func (n *NIC) getRefOrCreateTemp(protocol tcpip.NetworkProtocolNumber, address tcpip.Address, peb PrimaryEndpointBehavior, spoofingOrPromiscuous bool) *referencedNetworkEndpoint {
- id := NetworkEndpointID{address}
-
+// protocol and address.
+//
+// If none exists a temporary one may be created if we are in promiscuous mode
+// or spoofing. Promiscuous mode will only be checked if promiscuous is true.
+// Similarly, spoofing will only be checked if spoofing is true.
+//
+// If the address is the IPv4 broadcast address for an endpoint's network, that
+// endpoint will be returned.
+func (n *NIC) getRefOrCreateTemp(protocol tcpip.NetworkProtocolNumber, address tcpip.Address, peb PrimaryEndpointBehavior, tempRef getRefBehaviour) *referencedNetworkEndpoint {
n.mu.RLock()
- if ref, ok := n.endpoints[id]; ok {
+ var spoofingOrPromiscuous bool
+ switch tempRef {
+ case spoofing:
+ spoofingOrPromiscuous = n.mu.spoofing
+ case promiscuous:
+ spoofingOrPromiscuous = n.mu.promiscuous
+ }
+
+ if ref, ok := n.mu.endpoints[NetworkEndpointID{address}]; ok {
// An endpoint with this id exists, check if it can be used and return it.
- switch ref.getKind() {
- case permanentExpired:
- if !spoofingOrPromiscuous {
- n.mu.RUnlock()
- return nil
- }
- fallthrough
- case temporary, permanent:
- if ref.tryIncRef() {
- n.mu.RUnlock()
- return ref
- }
+ if !ref.isAssignedRLocked(spoofingOrPromiscuous) {
+ n.mu.RUnlock()
+ return nil
+ }
+
+ if ref.tryIncRef() {
+ n.mu.RUnlock()
+ return ref
}
}
- // A usable reference was not found, create a temporary one if requested by
- // the caller or if the address is found in the NIC's subnets.
- createTempEP := spoofingOrPromiscuous
- if !createTempEP {
- for _, sn := range n.addressRanges {
- // Skip the subnet address.
- if address == sn.ID() {
- continue
- }
- // For now just skip the broadcast address, until we support it.
- // FIXME(b/137608825): Add support for sending/receiving directed
- // (subnet) broadcast.
- if address == sn.Broadcast() {
- continue
- }
- if sn.Contains(address) {
- createTempEP = true
- break
- }
+ // Check if address is a broadcast address for the endpoint's network.
+ //
+ // Only IPv4 has a notion of broadcast addresses.
+ if protocol == header.IPv4ProtocolNumber {
+ if ref := n.getRefForBroadcastRLocked(address); ref != nil {
+ n.mu.RUnlock()
+ return ref
}
}
+ // A usable reference was not found, create a temporary one if requested by
+ // the caller or if the address is found in the NIC's subnets.
+ createTempEP := spoofingOrPromiscuous
n.mu.RUnlock()
if !createTempEP {
@@ -303,11 +678,44 @@ func (n *NIC) getRefOrCreateTemp(protocol tcpip.NetworkProtocolNumber, address t
// endpoint, create a new "temporary" endpoint. It will only exist while
// there's a route through it.
n.mu.Lock()
- if ref, ok := n.endpoints[id]; ok {
+ ref := n.getRefOrCreateTempLocked(protocol, address, peb)
+ n.mu.Unlock()
+ return ref
+}
+
+// getRefForBroadcastLocked returns an endpoint where address is the IPv4
+// broadcast address for the endpoint's network.
+//
+// n.mu MUST be read locked.
+func (n *NIC) getRefForBroadcastRLocked(address tcpip.Address) *referencedNetworkEndpoint {
+ for _, ref := range n.mu.endpoints {
+ // Only IPv4 has a notion of broadcast addresses.
+ if ref.protocol != header.IPv4ProtocolNumber {
+ continue
+ }
+
+ addr := ref.addrWithPrefix()
+ subnet := addr.Subnet()
+ if subnet.IsBroadcast(address) && ref.tryIncRef() {
+ return ref
+ }
+ }
+
+ return nil
+}
+
+/// getRefOrCreateTempLocked returns an existing endpoint for address or creates
+/// and returns a temporary endpoint.
+//
+// If the address is the IPv4 broadcast address for an endpoint's network, that
+// endpoint will be returned.
+//
+// n.mu must be write locked.
+func (n *NIC) getRefOrCreateTempLocked(protocol tcpip.NetworkProtocolNumber, address tcpip.Address, peb PrimaryEndpointBehavior) *referencedNetworkEndpoint {
+ if ref, ok := n.mu.endpoints[NetworkEndpointID{address}]; ok {
// No need to check the type as we are ok with expired endpoints at this
// point.
if ref.tryIncRef() {
- n.mu.Unlock()
return ref
}
// tryIncRef failing means the endpoint is scheduled to be removed once the
@@ -316,10 +724,18 @@ func (n *NIC) getRefOrCreateTemp(protocol tcpip.NetworkProtocolNumber, address t
n.removeEndpointLocked(ref)
}
+ // Check if address is a broadcast address for an endpoint's network.
+ //
+ // Only IPv4 has a notion of broadcast addresses.
+ if protocol == header.IPv4ProtocolNumber {
+ if ref := n.getRefForBroadcastRLocked(address); ref != nil {
+ return ref
+ }
+ }
+
// Add a new temporary endpoint.
netProto, ok := n.stack.networkProtocols[protocol]
if !ok {
- n.mu.Unlock()
return nil
}
ref, _ := n.addAddressLocked(tcpip.ProtocolAddress{
@@ -328,26 +744,40 @@ func (n *NIC) getRefOrCreateTemp(protocol tcpip.NetworkProtocolNumber, address t
Address: address,
PrefixLen: netProto.DefaultPrefixLen(),
},
- }, peb, temporary)
-
- n.mu.Unlock()
+ }, peb, temporary, static, false)
return ref
}
-func (n *NIC) addPermanentAddressLocked(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior) (*referencedNetworkEndpoint, *tcpip.Error) {
- id := NetworkEndpointID{protocolAddress.AddressWithPrefix.Address}
- if ref, ok := n.endpoints[id]; ok {
+// addAddressLocked adds a new protocolAddress to n.
+//
+// If n already has the address in a non-permanent state, and the kind given is
+// permanent, that address will be promoted in place and its properties set to
+// the properties provided. Otherwise, it returns tcpip.ErrDuplicateAddress.
+func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior, kind networkEndpointKind, configType networkEndpointConfigType, deprecated bool) (*referencedNetworkEndpoint, *tcpip.Error) {
+ // TODO(b/141022673): Validate IP addresses before adding them.
+
+ // Sanity check.
+ id := NetworkEndpointID{LocalAddress: protocolAddress.AddressWithPrefix.Address}
+ if ref, ok := n.mu.endpoints[id]; ok {
+ // Endpoint already exists.
+ if kind != permanent {
+ return nil, tcpip.ErrDuplicateAddress
+ }
switch ref.getKind() {
case permanentTentative, permanent:
// The NIC already have a permanent endpoint with that address.
return nil, tcpip.ErrDuplicateAddress
case permanentExpired, temporary:
- // Promote the endpoint to become permanent and respect
- // the new peb.
+ // Promote the endpoint to become permanent and respect the new peb,
+ // configType and deprecated status.
if ref.tryIncRef() {
+ // TODO(b/147748385): Perform Duplicate Address Detection when promoting
+ // an IPv6 endpoint to permanent.
ref.setKind(permanent)
+ ref.deprecated = deprecated
+ ref.configType = configType
- refs := n.primary[ref.protocol]
+ refs := n.mu.primary[ref.protocol]
for i, r := range refs {
if r == ref {
switch peb {
@@ -357,9 +787,9 @@ func (n *NIC) addPermanentAddressLocked(protocolAddress tcpip.ProtocolAddress, p
if i == 0 {
return ref, nil
}
- n.primary[r.protocol] = append(refs[:i], refs[i+1:]...)
+ n.mu.primary[r.protocol] = append(refs[:i], refs[i+1:]...)
case NeverPrimaryEndpoint:
- n.primary[r.protocol] = append(refs[:i], refs[i+1:]...)
+ n.mu.primary[r.protocol] = append(refs[:i], refs[i+1:]...)
return ref, nil
}
}
@@ -377,44 +807,30 @@ func (n *NIC) addPermanentAddressLocked(protocolAddress tcpip.ProtocolAddress, p
}
}
- return n.addAddressLocked(protocolAddress, peb, permanent)
-}
-
-func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior, kind networkEndpointKind) (*referencedNetworkEndpoint, *tcpip.Error) {
- // TODO(b/141022673): Validate IP address before adding them.
-
- // Sanity check.
- id := NetworkEndpointID{protocolAddress.AddressWithPrefix.Address}
- if _, ok := n.endpoints[id]; ok {
- // Endpoint already exists.
- return nil, tcpip.ErrDuplicateAddress
- }
-
- netProto, ok := n.stack.networkProtocols[protocolAddress.Protocol]
+ ep, ok := n.networkEndpoints[protocolAddress.Protocol]
if !ok {
return nil, tcpip.ErrUnknownProtocol
}
- // Create the new network endpoint.
- ep, err := netProto.NewEndpoint(n.id, protocolAddress.AddressWithPrefix, n.stack, n, n.linkEP)
- if err != nil {
- return nil, err
- }
-
isIPv6Unicast := protocolAddress.Protocol == header.IPv6ProtocolNumber && header.IsV6UnicastAddress(protocolAddress.AddressWithPrefix.Address)
// If the address is an IPv6 address and it is a permanent address,
- // mark it as tentative so it goes through the DAD process.
+ // mark it as tentative so it goes through the DAD process if the NIC is
+ // enabled. If the NIC is not enabled, DAD will be started when the NIC is
+ // enabled.
if isIPv6Unicast && kind == permanent {
kind = permanentTentative
}
ref := &referencedNetworkEndpoint{
- refs: 1,
- ep: ep,
- nic: n,
- protocol: protocolAddress.Protocol,
- kind: kind,
+ refs: 1,
+ addr: protocolAddress.AddressWithPrefix,
+ ep: ep,
+ nic: n,
+ protocol: protocolAddress.Protocol,
+ kind: kind,
+ configType: configType,
+ deprecated: deprecated,
}
// Set up cache if link address resolution exists for this protocol.
@@ -433,13 +849,13 @@ func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb Primar
}
}
- n.endpoints[id] = ref
+ n.mu.endpoints[id] = ref
n.insertPrimaryEndpointLocked(ref, peb)
- // If we are adding a tentative IPv6 address, start DAD.
- if isIPv6Unicast && kind == permanentTentative {
- if err := n.ndp.startDuplicateAddressDetection(protocolAddress.AddressWithPrefix.Address, ref); err != nil {
+ // If we are adding a tentative IPv6 address, start DAD if the NIC is enabled.
+ if isIPv6Unicast && kind == permanentTentative && n.mu.enabled {
+ if err := n.mu.ndp.startDuplicateAddressDetection(protocolAddress.AddressWithPrefix.Address, ref); err != nil {
return nil, err
}
}
@@ -452,7 +868,7 @@ func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb Primar
func (n *NIC) AddAddress(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior) *tcpip.Error {
// Add the endpoint.
n.mu.Lock()
- _, err := n.addPermanentAddressLocked(protocolAddress, peb)
+ _, err := n.addAddressLocked(protocolAddress, peb, permanent, static, false /* deprecated */)
n.mu.Unlock()
return err
@@ -464,22 +880,18 @@ func (n *NIC) AllAddresses() []tcpip.ProtocolAddress {
n.mu.RLock()
defer n.mu.RUnlock()
- addrs := make([]tcpip.ProtocolAddress, 0, len(n.endpoints))
- for nid, ref := range n.endpoints {
+ addrs := make([]tcpip.ProtocolAddress, 0, len(n.mu.endpoints))
+ for _, ref := range n.mu.endpoints {
// Don't include tentative, expired or temporary endpoints to
// avoid confusion and prevent the caller from using those.
switch ref.getKind() {
- case permanentTentative, permanentExpired, temporary:
- // TODO(b/140898488): Should tentative addresses be
- // returned?
+ case permanentExpired, temporary:
continue
}
+
addrs = append(addrs, tcpip.ProtocolAddress{
- Protocol: ref.protocol,
- AddressWithPrefix: tcpip.AddressWithPrefix{
- Address: nid.LocalAddress,
- PrefixLen: ref.ep.PrefixLen(),
- },
+ Protocol: ref.protocol,
+ AddressWithPrefix: ref.addrWithPrefix(),
})
}
return addrs
@@ -491,7 +903,7 @@ func (n *NIC) PrimaryAddresses() []tcpip.ProtocolAddress {
defer n.mu.RUnlock()
var addrs []tcpip.ProtocolAddress
- for proto, list := range n.primary {
+ for proto, list := range n.mu.primary {
for _, ref := range list {
// Don't include tentative, expired or tempory endpoints
// to avoid confusion and prevent the caller from using
@@ -502,59 +914,51 @@ func (n *NIC) PrimaryAddresses() []tcpip.ProtocolAddress {
}
addrs = append(addrs, tcpip.ProtocolAddress{
- Protocol: proto,
- AddressWithPrefix: tcpip.AddressWithPrefix{
- Address: ref.ep.ID().LocalAddress,
- PrefixLen: ref.ep.PrefixLen(),
- },
+ Protocol: proto,
+ AddressWithPrefix: ref.addrWithPrefix(),
})
}
}
return addrs
}
-// AddAddressRange adds a range of addresses to n, so that it starts accepting
-// packets targeted at the given addresses and network protocol. The range is
-// given by a subnet address, and all addresses contained in the subnet are
-// used except for the subnet address itself and the subnet's broadcast
-// address.
-func (n *NIC) AddAddressRange(protocol tcpip.NetworkProtocolNumber, subnet tcpip.Subnet) {
- n.mu.Lock()
- n.addressRanges = append(n.addressRanges, subnet)
- n.mu.Unlock()
-}
+// primaryAddress returns the primary address associated with this NIC.
+//
+// primaryAddress will return the first non-deprecated address if such an
+// address exists. If no non-deprecated address exists, the first deprecated
+// address will be returned.
+func (n *NIC) primaryAddress(proto tcpip.NetworkProtocolNumber) tcpip.AddressWithPrefix {
+ n.mu.RLock()
+ defer n.mu.RUnlock()
-// RemoveAddressRange removes the given address range from n.
-func (n *NIC) RemoveAddressRange(subnet tcpip.Subnet) {
- n.mu.Lock()
+ list, ok := n.mu.primary[proto]
+ if !ok {
+ return tcpip.AddressWithPrefix{}
+ }
- // Use the same underlying array.
- tmp := n.addressRanges[:0]
- for _, sub := range n.addressRanges {
- if sub != subnet {
- tmp = append(tmp, sub)
+ var deprecatedEndpoint *referencedNetworkEndpoint
+ for _, ref := range list {
+ // Don't include tentative, expired or tempory endpoints to avoid confusion
+ // and prevent the caller from using those.
+ switch ref.getKind() {
+ case permanentTentative, permanentExpired, temporary:
+ continue
}
- }
- n.addressRanges = tmp
- n.mu.Unlock()
-}
+ if !ref.deprecated {
+ return ref.addrWithPrefix()
+ }
-// Subnets returns the Subnets associated with this NIC.
-func (n *NIC) AddressRanges() []tcpip.Subnet {
- n.mu.RLock()
- defer n.mu.RUnlock()
- sns := make([]tcpip.Subnet, 0, len(n.addressRanges)+len(n.endpoints))
- for nid := range n.endpoints {
- sn, err := tcpip.NewSubnet(nid.LocalAddress, tcpip.AddressMask(strings.Repeat("\xff", len(nid.LocalAddress))))
- if err != nil {
- // This should never happen as the mask has been carefully crafted to
- // match the address.
- panic("Invalid endpoint subnet: " + err.Error())
+ if deprecatedEndpoint == nil {
+ deprecatedEndpoint = ref
}
- sns = append(sns, sn)
}
- return append(sns, n.addressRanges...)
+
+ if deprecatedEndpoint != nil {
+ return deprecatedEndpoint.addrWithPrefix()
+ }
+
+ return tcpip.AddressWithPrefix{}
}
// insertPrimaryEndpointLocked adds r to n's primary endpoint list as required
@@ -564,21 +968,21 @@ func (n *NIC) AddressRanges() []tcpip.Subnet {
func (n *NIC) insertPrimaryEndpointLocked(r *referencedNetworkEndpoint, peb PrimaryEndpointBehavior) {
switch peb {
case CanBePrimaryEndpoint:
- n.primary[r.protocol] = append(n.primary[r.protocol], r)
+ n.mu.primary[r.protocol] = append(n.mu.primary[r.protocol], r)
case FirstPrimaryEndpoint:
- n.primary[r.protocol] = append([]*referencedNetworkEndpoint{r}, n.primary[r.protocol]...)
+ n.mu.primary[r.protocol] = append([]*referencedNetworkEndpoint{r}, n.mu.primary[r.protocol]...)
}
}
func (n *NIC) removeEndpointLocked(r *referencedNetworkEndpoint) {
- id := *r.ep.ID()
+ id := NetworkEndpointID{LocalAddress: r.address()}
// Nothing to do if the reference has already been replaced with a different
// one. This happens in the case where 1) this endpoint's ref count hit zero
// and was waiting (on the lock) to be removed and 2) the same address was
// re-added in the meantime by removing this endpoint from the list and
// adding a new one.
- if n.endpoints[id] != r {
+ if n.mu.endpoints[id] != r {
return
}
@@ -586,16 +990,15 @@ func (n *NIC) removeEndpointLocked(r *referencedNetworkEndpoint) {
panic("Reference count dropped to zero before being removed")
}
- delete(n.endpoints, id)
- refs := n.primary[r.protocol]
+ delete(n.mu.endpoints, id)
+ refs := n.mu.primary[r.protocol]
for i, ref := range refs {
if ref == r {
- n.primary[r.protocol] = append(refs[:i], refs[i+1:]...)
+ n.mu.primary[r.protocol] = append(refs[:i], refs[i+1:]...)
+ refs[len(refs)-1] = nil
break
}
}
-
- r.ep.Close()
}
func (n *NIC) removeEndpoint(r *referencedNetworkEndpoint) {
@@ -605,7 +1008,7 @@ func (n *NIC) removeEndpoint(r *referencedNetworkEndpoint) {
}
func (n *NIC) removePermanentAddressLocked(addr tcpip.Address) *tcpip.Error {
- r, ok := n.endpoints[NetworkEndpointID{addr}]
+ r, ok := n.mu.endpoints[NetworkEndpointID{addr}]
if !ok {
return tcpip.ErrBadLocalAddress
}
@@ -615,26 +1018,45 @@ func (n *NIC) removePermanentAddressLocked(addr tcpip.Address) *tcpip.Error {
return tcpip.ErrBadLocalAddress
}
- isIPv6Unicast := r.protocol == header.IPv6ProtocolNumber && header.IsV6UnicastAddress(addr)
-
- // If we are removing a tentative IPv6 unicast address, stop DAD.
- if isIPv6Unicast && kind == permanentTentative {
- n.ndp.stopDuplicateAddressDetection(addr)
+ switch r.protocol {
+ case header.IPv6ProtocolNumber:
+ return n.removePermanentIPv6EndpointLocked(r, true /* allowSLAACInvalidation */)
+ default:
+ r.expireLocked()
+ return nil
}
+}
- r.setKind(permanentExpired)
- if !r.decRefLocked() {
- // The endpoint still has references to it.
- return nil
+func (n *NIC) removePermanentIPv6EndpointLocked(r *referencedNetworkEndpoint, allowSLAACInvalidation bool) *tcpip.Error {
+ addr := r.addrWithPrefix()
+
+ isIPv6Unicast := header.IsV6UnicastAddress(addr.Address)
+
+ if isIPv6Unicast {
+ n.mu.ndp.stopDuplicateAddressDetection(addr.Address)
+
+ // If we are removing an address generated via SLAAC, cleanup
+ // its SLAAC resources and notify the integrator.
+ switch r.configType {
+ case slaac:
+ n.mu.ndp.cleanupSLAACAddrResourcesAndNotify(addr, allowSLAACInvalidation)
+ case slaacTemp:
+ n.mu.ndp.cleanupTempSLAACAddrResourcesAndNotify(addr, allowSLAACInvalidation)
+ }
}
+ r.expireLocked()
+
// At this point the endpoint is deleted.
// If we are removing an IPv6 unicast address, leave the solicited-node
// multicast address.
+ //
+ // We ignore the tcpip.ErrBadLocalAddress error because the solicited-node
+ // multicast group may be left by user action.
if isIPv6Unicast {
- snmc := header.SolicitedNodeAddr(addr)
- if err := n.leaveGroupLocked(snmc); err != nil {
+ snmc := header.SolicitedNodeAddr(addr.Address)
+ if err := n.leaveGroupLocked(snmc, false /* force */); err != nil && err != tcpip.ErrBadLocalAddress {
return err
}
}
@@ -668,23 +1090,23 @@ func (n *NIC) joinGroupLocked(protocol tcpip.NetworkProtocolNumber, addr tcpip.A
// outlined in RFC 3810 section 5.
id := NetworkEndpointID{addr}
- joins := n.mcastJoins[id]
+ joins := n.mu.mcastJoins[id]
if joins == 0 {
netProto, ok := n.stack.networkProtocols[protocol]
if !ok {
return tcpip.ErrUnknownProtocol
}
- if _, err := n.addPermanentAddressLocked(tcpip.ProtocolAddress{
+ if _, err := n.addAddressLocked(tcpip.ProtocolAddress{
Protocol: protocol,
AddressWithPrefix: tcpip.AddressWithPrefix{
Address: addr,
PrefixLen: netProto.DefaultPrefixLen(),
},
- }, NeverPrimaryEndpoint); err != nil {
+ }, NeverPrimaryEndpoint, permanent, static, false /* deprecated */); err != nil {
return err
}
}
- n.mcastJoins[id] = joins + 1
+ n.mu.mcastJoins[id] = joins + 1
return nil
}
@@ -694,48 +1116,75 @@ func (n *NIC) leaveGroup(addr tcpip.Address) *tcpip.Error {
n.mu.Lock()
defer n.mu.Unlock()
- return n.leaveGroupLocked(addr)
+ return n.leaveGroupLocked(addr, false /* force */)
}
// leaveGroupLocked decrements the count for the given multicast address, and
// when it reaches zero removes the endpoint for this address. n MUST be locked
// before leaveGroupLocked is called.
-func (n *NIC) leaveGroupLocked(addr tcpip.Address) *tcpip.Error {
+//
+// If force is true, then the count for the multicast addres is ignored and the
+// endpoint will be removed immediately.
+func (n *NIC) leaveGroupLocked(addr tcpip.Address, force bool) *tcpip.Error {
id := NetworkEndpointID{addr}
- joins := n.mcastJoins[id]
- switch joins {
- case 0:
+ joins, ok := n.mu.mcastJoins[id]
+ if !ok {
// There are no joins with this address on this NIC.
return tcpip.ErrBadLocalAddress
- case 1:
- // This is the last one, clean up.
- if err := n.removePermanentAddressLocked(addr); err != nil {
- return err
- }
}
- n.mcastJoins[id] = joins - 1
+
+ joins--
+ if force || joins == 0 {
+ // There are no outstanding joins or we are forced to leave, clean up.
+ delete(n.mu.mcastJoins, id)
+ return n.removePermanentAddressLocked(addr)
+ }
+
+ n.mu.mcastJoins[id] = joins
return nil
}
-func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, localLinkAddr, remotelinkAddr tcpip.LinkAddress, ref *referencedNetworkEndpoint, vv buffer.VectorisedView) {
+// isInGroup returns true if n has joined the multicast group addr.
+func (n *NIC) isInGroup(addr tcpip.Address) bool {
+ n.mu.RLock()
+ joins := n.mu.mcastJoins[NetworkEndpointID{addr}]
+ n.mu.RUnlock()
+
+ return joins != 0
+}
+
+func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, localLinkAddr, remotelinkAddr tcpip.LinkAddress, ref *referencedNetworkEndpoint, pkt *PacketBuffer) {
r := makeRoute(protocol, dst, src, localLinkAddr, ref, false /* handleLocal */, false /* multicastLoop */)
r.RemoteLinkAddress = remotelinkAddr
- ref.ep.HandlePacket(&r, vv)
+
+ ref.ep.HandlePacket(&r, pkt)
ref.decRef()
}
// DeliverNetworkPacket finds the appropriate network protocol endpoint and
// hands the packet over for further processing. This function is called when
-// the NIC receives a packet from the physical interface.
+// the NIC receives a packet from the link endpoint.
// 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, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView, linkHeader buffer.View) {
+func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) {
+ n.mu.RLock()
+ enabled := n.mu.enabled
+ // If the NIC is not yet enabled, don't receive any packets.
+ if !enabled {
+ n.mu.RUnlock()
+
+ n.stats.DisabledRx.Packets.Increment()
+ n.stats.DisabledRx.Bytes.IncrementBy(uint64(pkt.Data.Size()))
+ return
+ }
+
n.stats.Rx.Packets.Increment()
- n.stats.Rx.Bytes.IncrementBy(uint64(vv.Size()))
+ n.stats.Rx.Bytes.IncrementBy(uint64(pkt.Data.Size()))
netProto, ok := n.stack.networkProtocols[protocol]
if !ok {
+ n.mu.RUnlock()
n.stack.stats.UnknownProtocolRcvdPackets.Increment()
return
}
@@ -747,32 +1196,59 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link
}
// 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]...)
- }
+ packetEPs := n.mu.packetEPs[protocol]
+ // Add any other packet sockets that maybe listening for all protocols.
+ packetEPs = append(packetEPs, n.mu.packetEPs[header.EthernetProtocolAll]...)
n.mu.RUnlock()
for _, ep := range packetEPs {
- ep.HandlePacket(n.id, local, protocol, vv, linkHeader)
+ p := pkt.Clone()
+ p.PktType = tcpip.PacketHost
+ ep.HandlePacket(n.id, local, protocol, p)
}
if netProto.Number() == header.IPv4ProtocolNumber || netProto.Number() == header.IPv6ProtocolNumber {
n.stack.stats.IP.PacketsReceived.Increment()
}
- if len(vv.First()) < netProto.MinimumPacketSize() {
+ // Parse headers.
+ transProtoNum, hasTransportHdr, ok := netProto.Parse(pkt)
+ if !ok {
+ // The packet is too small to contain a network header.
n.stack.stats.MalformedRcvdPackets.Increment()
return
}
+ if hasTransportHdr {
+ // Parse the transport header if present.
+ if state, ok := n.stack.transportProtocols[transProtoNum]; ok {
+ state.proto.Parse(pkt)
+ }
+ }
+
+ src, dst := netProto.ParseAddresses(pkt.NetworkHeader().View())
+
+ if n.stack.handleLocal && !n.isLoopback() && n.getRef(protocol, src) != nil {
+ // The source address is one of our own, so we never should have gotten a
+ // packet like this unless handleLocal is false. Loopback also calls this
+ // function even though the packets didn't come from the physical interface
+ // so don't drop those.
+ n.stack.stats.IP.InvalidSourceAddressesReceived.Increment()
+ return
+ }
- src, dst := netProto.ParseAddresses(vv.First())
+ // TODO(gvisor.dev/issue/170): Not supporting iptables for IPv6 yet.
+ // Loopback traffic skips the prerouting chain.
+ if protocol == header.IPv4ProtocolNumber && !n.isLoopback() {
+ // iptables filtering.
+ ipt := n.stack.IPTables()
+ address := n.primaryAddress(protocol)
+ if ok := ipt.Check(Prerouting, pkt, nil, nil, address.Address, ""); !ok {
+ // iptables is telling us to drop the packet.
+ return
+ }
+ }
if ref := n.getRef(protocol, dst); ref != nil {
- handlePacket(protocol, dst, src, linkEP.LinkAddress(), remote, ref, vv)
+ handlePacket(protocol, dst, src, n.linkEP.LinkAddress(), remote, ref, pkt)
return
}
@@ -783,54 +1259,98 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link
if n.stack.Forwarding(protocol) {
r, err := n.stack.FindRoute(0, "", dst, protocol, false /* multicastLoop */)
if err != nil {
- n.stack.stats.IP.InvalidAddressesReceived.Increment()
+ n.stack.stats.IP.InvalidDestinationAddressesReceived.Increment()
return
}
- defer r.Release()
-
- r.LocalLinkAddress = n.linkEP.LinkAddress()
- r.RemoteLinkAddress = remote
// Found a NIC.
n := r.ref.nic
n.mu.RLock()
- ref, ok := n.endpoints[NetworkEndpointID{dst}]
- ok = ok && ref.isValidForOutgoing() && ref.tryIncRef()
+ ref, ok := n.mu.endpoints[NetworkEndpointID{dst}]
+ ok = ok && ref.isValidForOutgoingRLocked() && ref.tryIncRef()
n.mu.RUnlock()
if ok {
+ r.LocalLinkAddress = n.linkEP.LinkAddress()
+ r.RemoteLinkAddress = remote
r.RemoteAddress = src
// TODO(b/123449044): Update the source NIC as well.
- ref.ep.HandlePacket(&r, vv)
+ ref.ep.HandlePacket(&r, pkt)
ref.decRef()
- } else {
- // n doesn't have a destination endpoint.
- // Send the packet out of n.
- // If we want to send the packet to a link-layer,
- // we have to reserve space for an Ethernet header.
- hdr := buffer.NewPrependableFromView(vv.First(), int(n.linkEP.MaxHeaderLength()))
- vv.RemoveFirst()
-
- // TODO(gvisor.dev/issue/1085): According to the RFC, we must decrease the TTL field for ipv4/ipv6.
- // TODO(b/128629022): use route.WritePacket.
- if err := n.linkEP.WritePacket(&r, nil /* gso */, hdr, vv, protocol); err != nil {
- r.Stats().IP.OutgoingPacketErrors.Increment()
- } else {
- n.stats.Tx.Packets.Increment()
- n.stats.Tx.Bytes.IncrementBy(uint64(hdr.UsedLength() + vv.Size()))
+ r.Release()
+ return
+ }
+
+ // n doesn't have a destination endpoint.
+ // Send the packet out of n.
+ // TODO(b/128629022): move this logic to route.WritePacket.
+ // TODO(gvisor.dev/issue/1085): According to the RFC, we must decrease the TTL field for ipv4/ipv6.
+ if ch, err := r.Resolve(nil); err != nil {
+ if err == tcpip.ErrWouldBlock {
+ n.stack.forwarder.enqueue(ch, n, &r, protocol, pkt)
+ // forwarder will release route.
+ return
}
+ n.stack.stats.IP.InvalidDestinationAddressesReceived.Increment()
+ r.Release()
+ return
}
+
+ // The link-address resolution finished immediately.
+ n.forwardPacket(&r, protocol, pkt)
+ r.Release()
return
}
// If a packet socket handled the packet, don't treat it as invalid.
if len(packetEPs) == 0 {
- n.stack.stats.IP.InvalidAddressesReceived.Increment()
+ n.stack.stats.IP.InvalidDestinationAddressesReceived.Increment()
+ }
+}
+
+// DeliverOutboundPacket implements NetworkDispatcher.DeliverOutboundPacket.
+func (n *NIC) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) {
+ n.mu.RLock()
+ // We do not deliver to protocol specific packet endpoints as on Linux
+ // only ETH_P_ALL endpoints get outbound packets.
+ // Add any other packet sockets that maybe listening for all protocols.
+ packetEPs := n.mu.packetEPs[header.EthernetProtocolAll]
+ n.mu.RUnlock()
+ for _, ep := range packetEPs {
+ p := pkt.Clone()
+ p.PktType = tcpip.PacketOutgoing
+ // Add the link layer header as outgoing packets are intercepted
+ // before the link layer header is created.
+ n.linkEP.AddHeader(local, remote, protocol, p)
+ ep.HandlePacket(n.id, local, protocol, p)
}
}
+func (n *NIC) forwardPacket(r *Route, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) {
+ // TODO(b/143425874) Decrease the TTL field in forwarded packets.
+
+ // pkt may have set its header and may not have enough headroom for link-layer
+ // header for the other link to prepend. Here we create a new packet to
+ // forward.
+ fwdPkt := NewPacketBuffer(PacketBufferOptions{
+ ReserveHeaderBytes: int(n.linkEP.MaxHeaderLength()),
+ Data: buffer.NewVectorisedView(pkt.Size(), pkt.Views()),
+ })
+
+ // WritePacket takes ownership of fwdPkt, calculate numBytes first.
+ numBytes := fwdPkt.Size()
+
+ if err := n.linkEP.WritePacket(r, nil /* gso */, protocol, fwdPkt); err != nil {
+ r.Stats().IP.OutgoingPacketErrors.Increment()
+ return
+ }
+
+ n.stats.Tx.Packets.Increment()
+ n.stats.Tx.Bytes.IncrementBy(uint64(numBytes))
+}
+
// DeliverTransportPacket delivers the packets to the appropriate transport
// protocol endpoint.
-func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, netHeader buffer.View, vv buffer.VectorisedView) {
+func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt *PacketBuffer) {
state, ok := n.stack.transportProtocols[protocol]
if !ok {
n.stack.stats.UnknownProtocolRcvdPackets.Increment()
@@ -842,41 +1362,60 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN
// Raw socket packets are delivered based solely on the transport
// protocol number. We do not inspect the payload to ensure it's
// validly formed.
- n.stack.demux.deliverRawPacket(r, protocol, netHeader, vv)
+ n.stack.demux.deliverRawPacket(r, protocol, pkt)
+
+ // TransportHeader is empty only when pkt is an ICMP packet or was reassembled
+ // from fragments.
+ if pkt.TransportHeader().View().IsEmpty() {
+ // TODO(gvisor.dev/issue/170): ICMP packets don't have their TransportHeader
+ // fields set yet, parse it here. See icmp/protocol.go:protocol.Parse for a
+ // full explanation.
+ if protocol == header.ICMPv4ProtocolNumber || protocol == header.ICMPv6ProtocolNumber {
+ // ICMP packets may be longer, but until icmp.Parse is implemented, here
+ // we parse it using the minimum size.
+ if _, ok := pkt.TransportHeader().Consume(transProto.MinimumPacketSize()); !ok {
+ n.stack.stats.MalformedRcvdPackets.Increment()
+ return
+ }
+ } else {
+ // This is either a bad packet or was re-assembled from fragments.
+ transProto.Parse(pkt)
+ }
+ }
- if len(vv.First()) < transProto.MinimumPacketSize() {
+ if pkt.TransportHeader().View().Size() < transProto.MinimumPacketSize() {
n.stack.stats.MalformedRcvdPackets.Increment()
return
}
- srcPort, dstPort, err := transProto.ParsePorts(vv.First())
+ srcPort, dstPort, err := transProto.ParsePorts(pkt.TransportHeader().View())
if err != nil {
n.stack.stats.MalformedRcvdPackets.Increment()
return
}
id := TransportEndpointID{dstPort, r.LocalAddress, srcPort, r.RemoteAddress}
- if n.stack.demux.deliverPacket(r, protocol, netHeader, vv, id) {
+ if n.stack.demux.deliverPacket(r, protocol, pkt, id) {
return
}
// Try to deliver to per-stack default handler.
if state.defaultHandler != nil {
- if state.defaultHandler(r, id, netHeader, vv) {
+ if state.defaultHandler(r, id, pkt) {
return
}
}
// We could not find an appropriate destination for this packet, so
// deliver it to the global handler.
- if !transProto.HandleUnknownDestinationPacket(r, id, netHeader, vv) {
+ if !transProto.HandleUnknownDestinationPacket(r, id, pkt) {
n.stack.stats.MalformedRcvdPackets.Increment()
}
}
// DeliverTransportControlPacket delivers control packets to the appropriate
// transport protocol endpoint.
-func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, vv buffer.VectorisedView) {
+func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, pkt *PacketBuffer) {
state, ok := n.stack.transportProtocols[trans]
if !ok {
return
@@ -887,17 +1426,18 @@ func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcp
// ICMPv4 only guarantees that 8 bytes of the transport protocol will
// be present in the payload. We know that the ports are within the
// first 8 bytes for all known transport protocols.
- if len(vv.First()) < 8 {
+ transHeader, ok := pkt.Data.PullUp(8)
+ if !ok {
return
}
- srcPort, dstPort, err := transProto.ParsePorts(vv.First())
+ srcPort, dstPort, err := transProto.ParsePorts(transHeader)
if err != nil {
return
}
id := TransportEndpointID{srcPort, local, dstPort, remote}
- if n.stack.demux.deliverControlPacket(n, net, trans, typ, extra, vv, id) {
+ if n.stack.demux.deliverControlPacket(n, net, trans, typ, extra, pkt, id) {
return
}
}
@@ -907,18 +1447,31 @@ func (n *NIC) ID() tcpip.NICID {
return n.id
}
+// Name returns the name of n.
+func (n *NIC) Name() string {
+ return n.name
+}
+
// Stack returns the instance of the Stack that owns this NIC.
func (n *NIC) Stack() *Stack {
return n.stack
}
+// LinkEndpoint returns the link endpoint of n.
+func (n *NIC) LinkEndpoint() LinkEndpoint {
+ return n.linkEP
+}
+
// isAddrTentative returns true if addr is tentative on n.
//
// Note that if addr is not associated with n, then this function will return
// false. It will only return true if the address is associated with the NIC
// AND it is tentative.
func (n *NIC) isAddrTentative(addr tcpip.Address) bool {
- ref, ok := n.endpoints[NetworkEndpointID{addr}]
+ n.mu.RLock()
+ defer n.mu.RUnlock()
+
+ ref, ok := n.mu.endpoints[NetworkEndpointID{addr}]
if !ok {
return false
}
@@ -926,15 +1479,17 @@ func (n *NIC) isAddrTentative(addr tcpip.Address) bool {
return ref.getKind() == permanentTentative
}
-// dupTentativeAddrDetected attempts to inform n that a tentative addr
-// is a duplicate on a link.
+// dupTentativeAddrDetected attempts to inform n that a tentative addr is a
+// duplicate on a link.
//
-// dupTentativeAddrDetected will delete the tentative address if it exists.
+// dupTentativeAddrDetected will remove the tentative address if it exists. If
+// the address was generated via SLAAC, an attempt will be made to generate a
+// new address.
func (n *NIC) dupTentativeAddrDetected(addr tcpip.Address) *tcpip.Error {
n.mu.Lock()
defer n.mu.Unlock()
- ref, ok := n.endpoints[NetworkEndpointID{addr}]
+ ref, ok := n.mu.endpoints[NetworkEndpointID{addr}]
if !ok {
return tcpip.ErrBadAddress
}
@@ -943,7 +1498,24 @@ func (n *NIC) dupTentativeAddrDetected(addr tcpip.Address) *tcpip.Error {
return tcpip.ErrInvalidEndpointState
}
- return n.removePermanentAddressLocked(addr)
+ // If the address is a SLAAC address, do not invalidate its SLAAC prefix as a
+ // new address will be generated for it.
+ if err := n.removePermanentIPv6EndpointLocked(ref, false /* allowSLAACInvalidation */); err != nil {
+ return err
+ }
+
+ prefix := ref.addrWithPrefix().Subnet()
+
+ switch ref.configType {
+ case slaac:
+ n.mu.ndp.regenerateSLAACAddr(prefix)
+ case slaacTemp:
+ // Do not reset the generation attempts counter for the prefix as the
+ // temporary address is being regenerated in response to a DAD conflict.
+ n.mu.ndp.regenerateTempSLAACAddr(prefix, false /* resetGenAttempts */)
+ }
+
+ return nil
}
// setNDPConfigs sets the NDP configurations for n.
@@ -954,10 +1526,39 @@ func (n *NIC) setNDPConfigs(c NDPConfigurations) {
c.validate()
n.mu.Lock()
- n.ndp.configs = c
+ n.mu.ndp.configs = c
n.mu.Unlock()
}
+// NUDConfigs gets the NUD configurations for n.
+func (n *NIC) NUDConfigs() (NUDConfigurations, *tcpip.Error) {
+ if n.neigh == nil {
+ return NUDConfigurations{}, tcpip.ErrNotSupported
+ }
+ return n.neigh.config(), nil
+}
+
+// setNUDConfigs sets the NUD configurations for n.
+//
+// Note, if c contains invalid NUD configuration values, it will be fixed to
+// use default values for the erroneous values.
+func (n *NIC) setNUDConfigs(c NUDConfigurations) *tcpip.Error {
+ if n.neigh == nil {
+ return tcpip.ErrNotSupported
+ }
+ c.resetInvalidFields()
+ n.neigh.setConfig(c)
+ return nil
+}
+
+// handleNDPRA handles an NDP Router Advertisement message that arrived on n.
+func (n *NIC) handleNDPRA(ip tcpip.Address, ra header.NDPRouterAdvert) {
+ n.mu.Lock()
+ defer n.mu.Unlock()
+
+ n.mu.ndp.handleRA(ip, ra)
+}
+
type networkEndpointKind int32
const (
@@ -977,7 +1578,7 @@ const (
// removing the permanent address from the NIC.
permanent
- // An expired permanent endoint is a permanent endoint that had its address
+ // An expired permanent endpoint is a permanent endpoint that had its address
// removed from the NIC, and it is waiting to be removed once no more routes
// hold a reference to it. This is achieved by decreasing its reference count
// by 1. If its address is re-added before the endpoint is removed, its type
@@ -997,11 +1598,11 @@ func (n *NIC) registerPacketEndpoint(netProto tcpip.NetworkProtocolNumber, ep Pa
n.mu.Lock()
defer n.mu.Unlock()
- eps, ok := n.packetEPs[netProto]
+ eps, ok := n.mu.packetEPs[netProto]
if !ok {
return tcpip.ErrNotSupported
}
- n.packetEPs[netProto] = append(eps, ep)
+ n.mu.packetEPs[netProto] = append(eps, ep)
return nil
}
@@ -1010,21 +1611,40 @@ func (n *NIC) unregisterPacketEndpoint(netProto tcpip.NetworkProtocolNumber, ep
n.mu.Lock()
defer n.mu.Unlock()
- eps, ok := n.packetEPs[netProto]
+ eps, ok := n.mu.packetEPs[netProto]
if !ok {
return
}
for i, epOther := range eps {
if epOther == ep {
- n.packetEPs[netProto] = append(eps[:i], eps[i+1:]...)
+ n.mu.packetEPs[netProto] = append(eps[:i], eps[i+1:]...)
return
}
}
}
+type networkEndpointConfigType int32
+
+const (
+ // A statically configured endpoint is an address that was added by
+ // some user-specified action (adding an explicit address, joining a
+ // multicast group).
+ static networkEndpointConfigType = iota
+
+ // A SLAAC configured endpoint is an IPv6 endpoint that was added by
+ // SLAAC as per RFC 4862 section 5.5.3.
+ slaac
+
+ // A temporary SLAAC configured endpoint is an IPv6 endpoint that was added by
+ // SLAAC as per RFC 4941. Temporary SLAAC addresses are short-lived and are
+ // not expected to be valid (or preferred) forever; hence the term temporary.
+ slaacTemp
+)
+
type referencedNetworkEndpoint struct {
ep NetworkEndpoint
+ addr tcpip.AddressWithPrefix
nic *NIC
protocol tcpip.NetworkProtocolNumber
@@ -1038,6 +1658,24 @@ type referencedNetworkEndpoint struct {
// networkEndpointKind must only be accessed using {get,set}Kind().
kind networkEndpointKind
+
+ // configType is the method that was used to configure this endpoint.
+ // This must never change except during endpoint creation and promotion to
+ // permanent.
+ configType networkEndpointConfigType
+
+ // deprecated indicates whether or not the endpoint should be considered
+ // deprecated. That is, when deprecated is true, other endpoints that are not
+ // deprecated should be preferred.
+ deprecated bool
+}
+
+func (r *referencedNetworkEndpoint) address() tcpip.Address {
+ return r.addr.Address
+}
+
+func (r *referencedNetworkEndpoint) addrWithPrefix() tcpip.AddressWithPrefix {
+ return r.addr
}
func (r *referencedNetworkEndpoint) getKind() networkEndpointKind {
@@ -1049,17 +1687,44 @@ func (r *referencedNetworkEndpoint) setKind(kind networkEndpointKind) {
}
// isValidForOutgoing returns true if the endpoint can be used to send out a
-// packet. It requires the endpoint to not be marked expired (i.e., its address
-// has been removed), or the NIC to be in spoofing mode.
+// packet. It requires the endpoint to not be marked expired (i.e., its address)
+// has been removed) unless the NIC is in spoofing mode, or temporary.
func (r *referencedNetworkEndpoint) isValidForOutgoing() bool {
- return r.getKind() != permanentExpired || r.nic.spoofing
+ r.nic.mu.RLock()
+ defer r.nic.mu.RUnlock()
+
+ return r.isValidForOutgoingRLocked()
}
-// isValidForIncoming returns true if the endpoint can accept an incoming
-// packet. It requires the endpoint to not be marked expired (i.e., its address
-// has been removed), or the NIC to be in promiscuous mode.
-func (r *referencedNetworkEndpoint) isValidForIncoming() bool {
- return r.getKind() != permanentExpired || r.nic.promiscuous
+// isValidForOutgoingRLocked is the same as isValidForOutgoing but requires
+// r.nic.mu to be read locked.
+func (r *referencedNetworkEndpoint) isValidForOutgoingRLocked() bool {
+ if !r.nic.mu.enabled {
+ return false
+ }
+
+ return r.isAssignedRLocked(r.nic.mu.spoofing)
+}
+
+// isAssignedRLocked returns true if r is considered to be assigned to the NIC.
+//
+// r.nic.mu must be read locked.
+func (r *referencedNetworkEndpoint) isAssignedRLocked(spoofingOrPromiscuous bool) bool {
+ switch r.getKind() {
+ case permanentTentative:
+ return false
+ case permanentExpired:
+ return spoofingOrPromiscuous
+ default:
+ return true
+ }
+}
+
+// expireLocked decrements the reference count and marks the permanent endpoint
+// as expired.
+func (r *referencedNetworkEndpoint) expireLocked() {
+ r.setKind(permanentExpired)
+ r.decRefLocked()
}
// decRef decrements the ref count and cleans up the endpoint once it reaches
@@ -1071,14 +1736,11 @@ func (r *referencedNetworkEndpoint) decRef() {
}
// decRefLocked is the same as decRef but assumes that the NIC.mu mutex is
-// locked. Returns true if the endpoint was removed.
-func (r *referencedNetworkEndpoint) decRefLocked() bool {
+// locked.
+func (r *referencedNetworkEndpoint) decRefLocked() {
if atomic.AddInt32(&r.refs, -1) == 0 {
r.nic.removeEndpointLocked(r)
- return true
}
-
- return false
}
// incRef increments the ref count. It must only be called when the caller is
diff --git a/pkg/tcpip/stack/nic_test.go b/pkg/tcpip/stack/nic_test.go
new file mode 100644
index 000000000..d312a79eb
--- /dev/null
+++ b/pkg/tcpip/stack/nic_test.go
@@ -0,0 +1,316 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package stack
+
+import (
+ "math"
+ "testing"
+ "time"
+
+ "gvisor.dev/gvisor/pkg/tcpip"
+ "gvisor.dev/gvisor/pkg/tcpip/buffer"
+ "gvisor.dev/gvisor/pkg/tcpip/header"
+)
+
+var _ LinkEndpoint = (*testLinkEndpoint)(nil)
+
+// A LinkEndpoint that throws away outgoing packets.
+//
+// We use this instead of the channel endpoint as the channel package depends on
+// the stack package which this test lives in, causing a cyclic dependency.
+type testLinkEndpoint struct {
+ dispatcher NetworkDispatcher
+}
+
+// Attach implements LinkEndpoint.Attach.
+func (e *testLinkEndpoint) Attach(dispatcher NetworkDispatcher) {
+ e.dispatcher = dispatcher
+}
+
+// IsAttached implements LinkEndpoint.IsAttached.
+func (e *testLinkEndpoint) IsAttached() bool {
+ return e.dispatcher != nil
+}
+
+// MTU implements LinkEndpoint.MTU.
+func (*testLinkEndpoint) MTU() uint32 {
+ return math.MaxUint16
+}
+
+// Capabilities implements LinkEndpoint.Capabilities.
+func (*testLinkEndpoint) Capabilities() LinkEndpointCapabilities {
+ return CapabilityResolutionRequired
+}
+
+// MaxHeaderLength implements LinkEndpoint.MaxHeaderLength.
+func (*testLinkEndpoint) MaxHeaderLength() uint16 {
+ return 0
+}
+
+// LinkAddress returns the link address of this endpoint.
+func (*testLinkEndpoint) LinkAddress() tcpip.LinkAddress {
+ return ""
+}
+
+// Wait implements LinkEndpoint.Wait.
+func (*testLinkEndpoint) Wait() {}
+
+// WritePacket implements LinkEndpoint.WritePacket.
+func (e *testLinkEndpoint) WritePacket(*Route, *GSO, tcpip.NetworkProtocolNumber, *PacketBuffer) *tcpip.Error {
+ return nil
+}
+
+// WritePackets implements LinkEndpoint.WritePackets.
+func (e *testLinkEndpoint) WritePackets(*Route, *GSO, PacketBufferList, tcpip.NetworkProtocolNumber) (int, *tcpip.Error) {
+ // Our tests don't use this so we don't support it.
+ return 0, tcpip.ErrNotSupported
+}
+
+// WriteRawPacket implements LinkEndpoint.WriteRawPacket.
+func (e *testLinkEndpoint) WriteRawPacket(buffer.VectorisedView) *tcpip.Error {
+ // Our tests don't use this so we don't support it.
+ return tcpip.ErrNotSupported
+}
+
+// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType.
+func (*testLinkEndpoint) ARPHardwareType() header.ARPHardwareType {
+ panic("not implemented")
+}
+
+// AddHeader implements stack.LinkEndpoint.AddHeader.
+func (e *testLinkEndpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) {
+ panic("not implemented")
+}
+
+var _ NetworkEndpoint = (*testIPv6Endpoint)(nil)
+
+// An IPv6 NetworkEndpoint that throws away outgoing packets.
+//
+// We use this instead of ipv6.endpoint because the ipv6 package depends on
+// the stack package which this test lives in, causing a cyclic dependency.
+type testIPv6Endpoint struct {
+ nicID tcpip.NICID
+ linkEP LinkEndpoint
+ protocol *testIPv6Protocol
+}
+
+// DefaultTTL implements NetworkEndpoint.DefaultTTL.
+func (*testIPv6Endpoint) DefaultTTL() uint8 {
+ return 0
+}
+
+// MTU implements NetworkEndpoint.MTU.
+func (e *testIPv6Endpoint) MTU() uint32 {
+ return e.linkEP.MTU() - header.IPv6MinimumSize
+}
+
+// Capabilities implements NetworkEndpoint.Capabilities.
+func (e *testIPv6Endpoint) Capabilities() LinkEndpointCapabilities {
+ return e.linkEP.Capabilities()
+}
+
+// MaxHeaderLength implements NetworkEndpoint.MaxHeaderLength.
+func (e *testIPv6Endpoint) MaxHeaderLength() uint16 {
+ return e.linkEP.MaxHeaderLength() + header.IPv6MinimumSize
+}
+
+// WritePacket implements NetworkEndpoint.WritePacket.
+func (*testIPv6Endpoint) WritePacket(*Route, *GSO, NetworkHeaderParams, *PacketBuffer) *tcpip.Error {
+ return nil
+}
+
+// WritePackets implements NetworkEndpoint.WritePackets.
+func (*testIPv6Endpoint) WritePackets(*Route, *GSO, PacketBufferList, NetworkHeaderParams) (int, *tcpip.Error) {
+ // Our tests don't use this so we don't support it.
+ return 0, tcpip.ErrNotSupported
+}
+
+// WriteHeaderIncludedPacket implements
+// NetworkEndpoint.WriteHeaderIncludedPacket.
+func (*testIPv6Endpoint) WriteHeaderIncludedPacket(*Route, *PacketBuffer) *tcpip.Error {
+ // Our tests don't use this so we don't support it.
+ return tcpip.ErrNotSupported
+}
+
+// NICID implements NetworkEndpoint.NICID.
+func (e *testIPv6Endpoint) NICID() tcpip.NICID {
+ return e.nicID
+}
+
+// HandlePacket implements NetworkEndpoint.HandlePacket.
+func (*testIPv6Endpoint) HandlePacket(*Route, *PacketBuffer) {
+}
+
+// Close implements NetworkEndpoint.Close.
+func (*testIPv6Endpoint) Close() {}
+
+// NetworkProtocolNumber implements NetworkEndpoint.NetworkProtocolNumber.
+func (*testIPv6Endpoint) NetworkProtocolNumber() tcpip.NetworkProtocolNumber {
+ return header.IPv6ProtocolNumber
+}
+
+var _ NetworkProtocol = (*testIPv6Protocol)(nil)
+
+// An IPv6 NetworkProtocol that supports the bare minimum to make a stack
+// believe it supports IPv6.
+//
+// We use this instead of ipv6.protocol because the ipv6 package depends on
+// the stack package which this test lives in, causing a cyclic dependency.
+type testIPv6Protocol struct{}
+
+// Number implements NetworkProtocol.Number.
+func (*testIPv6Protocol) Number() tcpip.NetworkProtocolNumber {
+ return header.IPv6ProtocolNumber
+}
+
+// MinimumPacketSize implements NetworkProtocol.MinimumPacketSize.
+func (*testIPv6Protocol) MinimumPacketSize() int {
+ return header.IPv6MinimumSize
+}
+
+// DefaultPrefixLen implements NetworkProtocol.DefaultPrefixLen.
+func (*testIPv6Protocol) DefaultPrefixLen() int {
+ return header.IPv6AddressSize * 8
+}
+
+// ParseAddresses implements NetworkProtocol.ParseAddresses.
+func (*testIPv6Protocol) ParseAddresses(v buffer.View) (src, dst tcpip.Address) {
+ h := header.IPv6(v)
+ return h.SourceAddress(), h.DestinationAddress()
+}
+
+// NewEndpoint implements NetworkProtocol.NewEndpoint.
+func (p *testIPv6Protocol) NewEndpoint(nicID tcpip.NICID, _ LinkAddressCache, _ TransportDispatcher, linkEP LinkEndpoint, _ *Stack) NetworkEndpoint {
+ return &testIPv6Endpoint{
+ nicID: nicID,
+ linkEP: linkEP,
+ protocol: p,
+ }
+}
+
+// SetOption implements NetworkProtocol.SetOption.
+func (*testIPv6Protocol) SetOption(interface{}) *tcpip.Error {
+ return nil
+}
+
+// Option implements NetworkProtocol.Option.
+func (*testIPv6Protocol) Option(interface{}) *tcpip.Error {
+ return nil
+}
+
+// Close implements NetworkProtocol.Close.
+func (*testIPv6Protocol) Close() {}
+
+// Wait implements NetworkProtocol.Wait.
+func (*testIPv6Protocol) Wait() {}
+
+// Parse implements NetworkProtocol.Parse.
+func (*testIPv6Protocol) Parse(*PacketBuffer) (tcpip.TransportProtocolNumber, bool, bool) {
+ return 0, false, false
+}
+
+var _ LinkAddressResolver = (*testIPv6Protocol)(nil)
+
+// LinkAddressProtocol implements LinkAddressResolver.
+func (*testIPv6Protocol) LinkAddressProtocol() tcpip.NetworkProtocolNumber {
+ return header.IPv6ProtocolNumber
+}
+
+// LinkAddressRequest implements LinkAddressResolver.
+func (*testIPv6Protocol) LinkAddressRequest(_, _ tcpip.Address, _ tcpip.LinkAddress, _ LinkEndpoint) *tcpip.Error {
+ return nil
+}
+
+// ResolveStaticAddress implements LinkAddressResolver.
+func (*testIPv6Protocol) ResolveStaticAddress(addr tcpip.Address) (tcpip.LinkAddress, bool) {
+ if header.IsV6MulticastAddress(addr) {
+ return header.EthernetAddressFromMulticastIPv6Address(addr), true
+ }
+ return "", false
+}
+
+// Test the race condition where a NIC is removed and an RS timer fires at the
+// same time.
+func TestRemoveNICWhileHandlingRSTimer(t *testing.T) {
+ const (
+ nicID = 1
+
+ maxRtrSolicitations = 5
+ )
+
+ e := testLinkEndpoint{}
+ s := New(Options{
+ NetworkProtocols: []NetworkProtocol{&testIPv6Protocol{}},
+ NDPConfigs: NDPConfigurations{
+ MaxRtrSolicitations: maxRtrSolicitations,
+ RtrSolicitationInterval: minimumRtrSolicitationInterval,
+ },
+ })
+
+ if err := s.CreateNIC(nicID, &e); err != nil {
+ t.Fatalf("s.CreateNIC(%d, _) = %s", nicID, err)
+ }
+
+ s.mu.Lock()
+ // Wait for the router solicitation timer to fire and block trying to obtain
+ // the stack lock when doing link address resolution.
+ time.Sleep(minimumRtrSolicitationInterval * 2)
+ if err := s.removeNICLocked(nicID); err != nil {
+ t.Fatalf("s.removeNICLocked(%d) = %s", nicID, err)
+ }
+ s.mu.Unlock()
+}
+
+func TestDisabledRxStatsWhenNICDisabled(t *testing.T) {
+ // When the NIC is disabled, the only field that matters is the stats field.
+ // This test is limited to stats counter checks.
+ nic := NIC{
+ stats: makeNICStats(),
+ }
+
+ if got := nic.stats.DisabledRx.Packets.Value(); got != 0 {
+ t.Errorf("got DisabledRx.Packets = %d, want = 0", got)
+ }
+ if got := nic.stats.DisabledRx.Bytes.Value(); got != 0 {
+ t.Errorf("got DisabledRx.Bytes = %d, want = 0", got)
+ }
+ if got := nic.stats.Rx.Packets.Value(); got != 0 {
+ t.Errorf("got Rx.Packets = %d, want = 0", got)
+ }
+ if got := nic.stats.Rx.Bytes.Value(); got != 0 {
+ t.Errorf("got Rx.Bytes = %d, want = 0", got)
+ }
+
+ if t.Failed() {
+ t.FailNow()
+ }
+
+ nic.DeliverNetworkPacket("", "", 0, NewPacketBuffer(PacketBufferOptions{
+ Data: buffer.View([]byte{1, 2, 3, 4}).ToVectorisedView(),
+ }))
+
+ if got := nic.stats.DisabledRx.Packets.Value(); got != 1 {
+ t.Errorf("got DisabledRx.Packets = %d, want = 1", got)
+ }
+ if got := nic.stats.DisabledRx.Bytes.Value(); got != 4 {
+ t.Errorf("got DisabledRx.Bytes = %d, want = 4", got)
+ }
+ if got := nic.stats.Rx.Packets.Value(); got != 0 {
+ t.Errorf("got Rx.Packets = %d, want = 0", got)
+ }
+ if got := nic.stats.Rx.Bytes.Value(); got != 0 {
+ t.Errorf("got Rx.Bytes = %d, want = 0", got)
+ }
+}
diff --git a/pkg/tcpip/stack/nud.go b/pkg/tcpip/stack/nud.go
new file mode 100644
index 000000000..e1ec15487
--- /dev/null
+++ b/pkg/tcpip/stack/nud.go
@@ -0,0 +1,466 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package stack
+
+import (
+ "math"
+ "sync"
+ "time"
+
+ "gvisor.dev/gvisor/pkg/tcpip"
+)
+
+const (
+ // defaultBaseReachableTime is the default base duration for computing the
+ // random reachable time.
+ //
+ // Reachable time is the duration for which a neighbor is considered
+ // reachable after a positive reachability confirmation is received. It is a
+ // function of a uniformly distributed random value between the minimum and
+ // maximum random factors, multiplied by the base reachable time. Using a
+ // random component eliminates the possibility that Neighbor Unreachability
+ // Detection messages will synchronize with each other.
+ //
+ // Default taken from REACHABLE_TIME of RFC 4861 section 10.
+ defaultBaseReachableTime = 30 * time.Second
+
+ // minimumBaseReachableTime is the minimum base duration for computing the
+ // random reachable time.
+ //
+ // Minimum = 1ms
+ minimumBaseReachableTime = time.Millisecond
+
+ // defaultMinRandomFactor is the default minimum value of the random factor
+ // used for computing reachable time.
+ //
+ // Default taken from MIN_RANDOM_FACTOR of RFC 4861 section 10.
+ defaultMinRandomFactor = 0.5
+
+ // defaultMaxRandomFactor is the default maximum value of the random factor
+ // used for computing reachable time.
+ //
+ // The default value depends on the value of MinRandomFactor.
+ // If MinRandomFactor is less than MAX_RANDOM_FACTOR of RFC 4861 section 10,
+ // the value from the RFC will be used; otherwise, the default is
+ // MinRandomFactor multiplied by three.
+ defaultMaxRandomFactor = 1.5
+
+ // defaultRetransmitTimer is the default amount of time to wait between
+ // sending reachability probes.
+ //
+ // Default taken from RETRANS_TIMER of RFC 4861 section 10.
+ defaultRetransmitTimer = time.Second
+
+ // minimumRetransmitTimer is the minimum amount of time to wait between
+ // sending reachability probes.
+ //
+ // Note, RFC 4861 does not impose a minimum Retransmit Timer, but we do here
+ // to make sure the messages are not sent all at once. We also come to this
+ // value because in the RetransmitTimer field of a Router Advertisement, a
+ // value of 0 means unspecified, so the smallest valid value is 1. Note, the
+ // unit of the RetransmitTimer field in the Router Advertisement is
+ // milliseconds.
+ minimumRetransmitTimer = time.Millisecond
+
+ // defaultDelayFirstProbeTime is the default duration to wait for a
+ // non-Neighbor-Discovery related protocol to reconfirm reachability after
+ // entering the DELAY state. After this time, a reachability probe will be
+ // sent and the entry will transition to the PROBE state.
+ //
+ // Default taken from DELAY_FIRST_PROBE_TIME of RFC 4861 section 10.
+ defaultDelayFirstProbeTime = 5 * time.Second
+
+ // defaultMaxMulticastProbes is the default number of reachabililty probes
+ // to send before concluding negative reachability and deleting the neighbor
+ // entry from the INCOMPLETE state.
+ //
+ // Default taken from MAX_MULTICAST_SOLICIT of RFC 4861 section 10.
+ defaultMaxMulticastProbes = 3
+
+ // defaultMaxUnicastProbes is the default number of reachability probes to
+ // send before concluding retransmission from within the PROBE state should
+ // cease and the entry SHOULD be deleted.
+ //
+ // Default taken from MAX_UNICASE_SOLICIT of RFC 4861 section 10.
+ defaultMaxUnicastProbes = 3
+
+ // defaultMaxAnycastDelayTime is the default time in which the stack SHOULD
+ // delay sending a response for a random time between 0 and this time, if the
+ // target address is an anycast address.
+ //
+ // Default taken from MAX_ANYCAST_DELAY_TIME of RFC 4861 section 10.
+ defaultMaxAnycastDelayTime = time.Second
+
+ // defaultMaxReachbilityConfirmations is the default amount of unsolicited
+ // reachability confirmation messages a node MAY send to all-node multicast
+ // address when it determines its link-layer address has changed.
+ //
+ // Default taken from MAX_NEIGHBOR_ADVERTISEMENT of RFC 4861 section 10.
+ defaultMaxReachbilityConfirmations = 3
+
+ // defaultUnreachableTime is the default duration for how long an entry will
+ // remain in the FAILED state before being removed from the neighbor cache.
+ //
+ // Note, there is no equivalent protocol constant defined in RFC 4861. It
+ // leaves the specifics of any garbage collection mechanism up to the
+ // implementation.
+ defaultUnreachableTime = 5 * time.Second
+)
+
+// NUDDispatcher is the interface integrators of netstack must implement to
+// receive and handle NUD related events.
+type NUDDispatcher interface {
+ // OnNeighborAdded will be called when a new entry is added to a NIC's (with
+ // ID nicID) neighbor table.
+ //
+ // This function is permitted to block indefinitely without interfering with
+ // the stack's operation.
+ //
+ // May be called concurrently.
+ OnNeighborAdded(nicID tcpip.NICID, ipAddr tcpip.Address, linkAddr tcpip.LinkAddress, state NeighborState, updatedAt time.Time)
+
+ // OnNeighborChanged will be called when an entry in a NIC's (with ID nicID)
+ // neighbor table changes state and/or link address.
+ //
+ // This function is permitted to block indefinitely without interfering with
+ // the stack's operation.
+ //
+ // May be called concurrently.
+ OnNeighborChanged(nicID tcpip.NICID, ipAddr tcpip.Address, linkAddr tcpip.LinkAddress, state NeighborState, updatedAt time.Time)
+
+ // OnNeighborRemoved will be called when an entry is removed from a NIC's
+ // (with ID nicID) neighbor table.
+ //
+ // This function is permitted to block indefinitely without interfering with
+ // the stack's operation.
+ //
+ // May be called concurrently.
+ OnNeighborRemoved(nicID tcpip.NICID, ipAddr tcpip.Address, linkAddr tcpip.LinkAddress, state NeighborState, updatedAt time.Time)
+}
+
+// ReachabilityConfirmationFlags describes the flags used within a reachability
+// confirmation (e.g. ARP reply or Neighbor Advertisement for ARP or NDP,
+// respectively).
+type ReachabilityConfirmationFlags struct {
+ // Solicited indicates that the advertisement was sent in response to a
+ // reachability probe.
+ Solicited bool
+
+ // Override indicates that the reachability confirmation should override an
+ // existing neighbor cache entry and update the cached link-layer address.
+ // When Override is not set the confirmation will not update a cached
+ // link-layer address, but will update an existing neighbor cache entry for
+ // which no link-layer address is known.
+ Override bool
+
+ // IsRouter indicates that the sender is a router.
+ IsRouter bool
+}
+
+// NUDHandler communicates external events to the Neighbor Unreachability
+// Detection state machine, which is implemented per-interface. This is used by
+// network endpoints to inform the Neighbor Cache of probes and confirmations.
+type NUDHandler interface {
+ // HandleProbe processes an incoming neighbor probe (e.g. ARP request or
+ // Neighbor Solicitation for ARP or NDP, respectively). Validation of the
+ // probe needs to be performed before calling this function since the
+ // Neighbor Cache doesn't have access to view the NIC's assigned addresses.
+ HandleProbe(remoteAddr, localAddr tcpip.Address, protocol tcpip.NetworkProtocolNumber, remoteLinkAddr tcpip.LinkAddress, linkRes LinkAddressResolver)
+
+ // HandleConfirmation processes an incoming neighbor confirmation (e.g. ARP
+ // reply or Neighbor Advertisement for ARP or NDP, respectively).
+ HandleConfirmation(addr tcpip.Address, linkAddr tcpip.LinkAddress, flags ReachabilityConfirmationFlags)
+
+ // HandleUpperLevelConfirmation processes an incoming upper-level protocol
+ // (e.g. TCP acknowledgements) reachability confirmation.
+ HandleUpperLevelConfirmation(addr tcpip.Address)
+}
+
+// NUDConfigurations is the NUD configurations for the netstack. This is used
+// by the neighbor cache to operate the NUD state machine on each device in the
+// local network.
+type NUDConfigurations struct {
+ // BaseReachableTime is the base duration for computing the random reachable
+ // time.
+ //
+ // Reachable time is the duration for which a neighbor is considered
+ // reachable after a positive reachability confirmation is received. It is a
+ // function of uniformly distributed random value between minRandomFactor and
+ // maxRandomFactor multiplied by baseReachableTime. Using a random component
+ // eliminates the possibility that Neighbor Unreachability Detection messages
+ // will synchronize with each other.
+ //
+ // After this time, a neighbor entry will transition from REACHABLE to STALE
+ // state.
+ //
+ // Must be greater than 0.
+ BaseReachableTime time.Duration
+
+ // LearnBaseReachableTime enables learning BaseReachableTime during runtime
+ // from the neighbor discovery protocol, if supported.
+ //
+ // TODO(gvisor.dev/issue/2240): Implement this NUD configuration option.
+ LearnBaseReachableTime bool
+
+ // MinRandomFactor is the minimum value of the random factor used for
+ // computing reachable time.
+ //
+ // See BaseReachbleTime for more information on computing the reachable time.
+ //
+ // Must be greater than 0.
+ MinRandomFactor float32
+
+ // MaxRandomFactor is the maximum value of the random factor used for
+ // computing reachabile time.
+ //
+ // See BaseReachbleTime for more information on computing the reachable time.
+ //
+ // Must be great than or equal to MinRandomFactor.
+ MaxRandomFactor float32
+
+ // RetransmitTimer is the duration between retransmission of reachability
+ // probes in the PROBE state.
+ RetransmitTimer time.Duration
+
+ // LearnRetransmitTimer enables learning RetransmitTimer during runtime from
+ // the neighbor discovery protocol, if supported.
+ //
+ // TODO(gvisor.dev/issue/2241): Implement this NUD configuration option.
+ LearnRetransmitTimer bool
+
+ // DelayFirstProbeTime is the duration to wait for a non-Neighbor-Discovery
+ // related protocol to reconfirm reachability after entering the DELAY state.
+ // After this time, a reachability probe will be sent and the entry will
+ // transition to the PROBE state.
+ //
+ // Must be greater than 0.
+ DelayFirstProbeTime time.Duration
+
+ // MaxMulticastProbes is the number of reachability probes to send before
+ // concluding negative reachability and deleting the neighbor entry from the
+ // INCOMPLETE state.
+ //
+ // Must be greater than 0.
+ MaxMulticastProbes uint32
+
+ // MaxUnicastProbes is the number of reachability probes to send before
+ // concluding retransmission from within the PROBE state should cease and
+ // entry SHOULD be deleted.
+ //
+ // Must be greater than 0.
+ MaxUnicastProbes uint32
+
+ // MaxAnycastDelayTime is the time in which the stack SHOULD delay sending a
+ // response for a random time between 0 and this time, if the target address
+ // is an anycast address.
+ //
+ // TODO(gvisor.dev/issue/2242): Use this option when sending solicited
+ // neighbor confirmations to anycast addresses and proxying neighbor
+ // confirmations.
+ MaxAnycastDelayTime time.Duration
+
+ // MaxReachabilityConfirmations is the number of unsolicited reachability
+ // confirmation messages a node MAY send to all-node multicast address when
+ // it determines its link-layer address has changed.
+ //
+ // TODO(gvisor.dev/issue/2246): Discuss if implementation of this NUD
+ // configuration option is necessary.
+ MaxReachabilityConfirmations uint32
+
+ // UnreachableTime describes how long an entry will remain in the FAILED
+ // state before being removed from the neighbor cache.
+ UnreachableTime time.Duration
+}
+
+// DefaultNUDConfigurations returns a NUDConfigurations populated with default
+// values defined by RFC 4861 section 10.
+func DefaultNUDConfigurations() NUDConfigurations {
+ return NUDConfigurations{
+ BaseReachableTime: defaultBaseReachableTime,
+ LearnBaseReachableTime: true,
+ MinRandomFactor: defaultMinRandomFactor,
+ MaxRandomFactor: defaultMaxRandomFactor,
+ RetransmitTimer: defaultRetransmitTimer,
+ LearnRetransmitTimer: true,
+ DelayFirstProbeTime: defaultDelayFirstProbeTime,
+ MaxMulticastProbes: defaultMaxMulticastProbes,
+ MaxUnicastProbes: defaultMaxUnicastProbes,
+ MaxAnycastDelayTime: defaultMaxAnycastDelayTime,
+ MaxReachabilityConfirmations: defaultMaxReachbilityConfirmations,
+ UnreachableTime: defaultUnreachableTime,
+ }
+}
+
+// resetInvalidFields modifies an invalid NDPConfigurations with valid values.
+// If invalid values are present in c, the corresponding default values will be
+// used instead. This is needed to check, and conditionally fix, user-specified
+// NUDConfigurations.
+func (c *NUDConfigurations) resetInvalidFields() {
+ if c.BaseReachableTime < minimumBaseReachableTime {
+ c.BaseReachableTime = defaultBaseReachableTime
+ }
+ if c.MinRandomFactor <= 0 {
+ c.MinRandomFactor = defaultMinRandomFactor
+ }
+ if c.MaxRandomFactor < c.MinRandomFactor {
+ c.MaxRandomFactor = calcMaxRandomFactor(c.MinRandomFactor)
+ }
+ if c.RetransmitTimer < minimumRetransmitTimer {
+ c.RetransmitTimer = defaultRetransmitTimer
+ }
+ if c.DelayFirstProbeTime == 0 {
+ c.DelayFirstProbeTime = defaultDelayFirstProbeTime
+ }
+ if c.MaxMulticastProbes == 0 {
+ c.MaxMulticastProbes = defaultMaxMulticastProbes
+ }
+ if c.MaxUnicastProbes == 0 {
+ c.MaxUnicastProbes = defaultMaxUnicastProbes
+ }
+ if c.UnreachableTime == 0 {
+ c.UnreachableTime = defaultUnreachableTime
+ }
+}
+
+// calcMaxRandomFactor calculates the maximum value of the random factor used
+// for computing reachable time. This function is necessary for when the
+// default specified in RFC 4861 section 10 is less than the current
+// MinRandomFactor.
+//
+// Assumes minRandomFactor is positive since validation of the minimum value
+// should come before the validation of the maximum.
+func calcMaxRandomFactor(minRandomFactor float32) float32 {
+ if minRandomFactor > defaultMaxRandomFactor {
+ return minRandomFactor * 3
+ }
+ return defaultMaxRandomFactor
+}
+
+// A Rand is a source of random numbers.
+type Rand interface {
+ // Float32 returns, as a float32, a pseudo-random number in [0.0,1.0).
+ Float32() float32
+}
+
+// NUDState stores states needed for calculating reachable time.
+type NUDState struct {
+ rng Rand
+
+ // mu protects the fields below.
+ //
+ // It is necessary for NUDState to handle its own locking since neighbor
+ // entries may access the NUD state from within the goroutine spawned by
+ // time.AfterFunc(). This goroutine may run concurrently with the main
+ // process for controlling the neighbor cache and would otherwise introduce
+ // race conditions if NUDState was not locked properly.
+ mu sync.RWMutex
+
+ config NUDConfigurations
+
+ // reachableTime is the duration to wait for a REACHABLE entry to
+ // transition into STALE after inactivity. This value is calculated with
+ // the algorithm defined in RFC 4861 section 6.3.2.
+ reachableTime time.Duration
+
+ expiration time.Time
+ prevBaseReachableTime time.Duration
+ prevMinRandomFactor float32
+ prevMaxRandomFactor float32
+}
+
+// NewNUDState returns new NUDState using c as configuration and the specified
+// random number generator for use in recomputing ReachableTime.
+func NewNUDState(c NUDConfigurations, rng Rand) *NUDState {
+ s := &NUDState{
+ rng: rng,
+ }
+ s.config = c
+ return s
+}
+
+// Config returns the NUD configuration.
+func (s *NUDState) Config() NUDConfigurations {
+ s.mu.RLock()
+ defer s.mu.RUnlock()
+ return s.config
+}
+
+// SetConfig replaces the existing NUD configurations with c.
+func (s *NUDState) SetConfig(c NUDConfigurations) {
+ s.mu.Lock()
+ defer s.mu.Unlock()
+ s.config = c
+}
+
+// ReachableTime returns the duration to wait for a REACHABLE entry to
+// transition into STALE after inactivity. This value is recalculated for new
+// values of BaseReachableTime, MinRandomFactor, and MaxRandomFactor using the
+// algorithm defined in RFC 4861 section 6.3.2.
+func (s *NUDState) ReachableTime() time.Duration {
+ s.mu.Lock()
+ defer s.mu.Unlock()
+
+ if time.Now().After(s.expiration) ||
+ s.config.BaseReachableTime != s.prevBaseReachableTime ||
+ s.config.MinRandomFactor != s.prevMinRandomFactor ||
+ s.config.MaxRandomFactor != s.prevMaxRandomFactor {
+ return s.recomputeReachableTimeLocked()
+ }
+ return s.reachableTime
+}
+
+// recomputeReachableTimeLocked forces a recalculation of ReachableTime using
+// the algorithm defined in RFC 4861 section 6.3.2.
+//
+// This SHOULD automatically be invoked during certain situations, as per
+// RFC 4861 section 6.3.4:
+//
+// If the received Reachable Time value is non-zero, the host SHOULD set its
+// BaseReachableTime variable to the received value. If the new value
+// differs from the previous value, the host SHOULD re-compute a new random
+// ReachableTime value. ReachableTime is computed as a uniformly
+// distributed random value between MIN_RANDOM_FACTOR and MAX_RANDOM_FACTOR
+// times the BaseReachableTime. Using a random component eliminates the
+// possibility that Neighbor Unreachability Detection messages will
+// synchronize with each other.
+//
+// In most cases, the advertised Reachable Time value will be the same in
+// consecutive Router Advertisements, and a host's BaseReachableTime rarely
+// changes. In such cases, an implementation SHOULD ensure that a new
+// random value gets re-computed at least once every few hours.
+//
+// s.mu MUST be locked for writing.
+func (s *NUDState) recomputeReachableTimeLocked() time.Duration {
+ s.prevBaseReachableTime = s.config.BaseReachableTime
+ s.prevMinRandomFactor = s.config.MinRandomFactor
+ s.prevMaxRandomFactor = s.config.MaxRandomFactor
+
+ randomFactor := s.config.MinRandomFactor + s.rng.Float32()*(s.config.MaxRandomFactor-s.config.MinRandomFactor)
+
+ // Check for overflow, given that minRandomFactor and maxRandomFactor are
+ // guaranteed to be positive numbers.
+ if float32(math.MaxInt64)/randomFactor < float32(s.config.BaseReachableTime) {
+ s.reachableTime = time.Duration(math.MaxInt64)
+ } else if randomFactor == 1 {
+ // Avoid loss of precision when a large base reachable time is used.
+ s.reachableTime = s.config.BaseReachableTime
+ } else {
+ reachableTime := int64(float32(s.config.BaseReachableTime) * randomFactor)
+ s.reachableTime = time.Duration(reachableTime)
+ }
+
+ s.expiration = time.Now().Add(2 * time.Hour)
+ return s.reachableTime
+}
diff --git a/pkg/tcpip/stack/nud_test.go b/pkg/tcpip/stack/nud_test.go
new file mode 100644
index 000000000..2494ee610
--- /dev/null
+++ b/pkg/tcpip/stack/nud_test.go
@@ -0,0 +1,795 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package stack_test
+
+import (
+ "math"
+ "testing"
+ "time"
+
+ "gvisor.dev/gvisor/pkg/tcpip"
+ "gvisor.dev/gvisor/pkg/tcpip/link/channel"
+ "gvisor.dev/gvisor/pkg/tcpip/network/ipv6"
+ "gvisor.dev/gvisor/pkg/tcpip/stack"
+)
+
+const (
+ defaultBaseReachableTime = 30 * time.Second
+ minimumBaseReachableTime = time.Millisecond
+ defaultMinRandomFactor = 0.5
+ defaultMaxRandomFactor = 1.5
+ defaultRetransmitTimer = time.Second
+ minimumRetransmitTimer = time.Millisecond
+ defaultDelayFirstProbeTime = 5 * time.Second
+ defaultMaxMulticastProbes = 3
+ defaultMaxUnicastProbes = 3
+ defaultMaxAnycastDelayTime = time.Second
+ defaultMaxReachbilityConfirmations = 3
+ defaultUnreachableTime = 5 * time.Second
+
+ defaultFakeRandomNum = 0.5
+)
+
+// fakeRand is a deterministic random number generator.
+type fakeRand struct {
+ num float32
+}
+
+var _ stack.Rand = (*fakeRand)(nil)
+
+func (f *fakeRand) Float32() float32 {
+ return f.num
+}
+
+// TestSetNUDConfigurationFailsForBadNICID tests to make sure we get an error if
+// we attempt to update NUD configurations using an invalid NICID.
+func TestSetNUDConfigurationFailsForBadNICID(t *testing.T) {
+ s := stack.New(stack.Options{
+ // A neighbor cache is required to store NUDConfigurations. The networking
+ // stack will only allocate neighbor caches if a protocol providing link
+ // address resolution is specified (e.g. ARP or IPv6).
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ })
+
+ // No NIC with ID 1 yet.
+ config := stack.NUDConfigurations{}
+ if err := s.SetNUDConfigurations(1, config); err != tcpip.ErrUnknownNICID {
+ t.Fatalf("got s.SetNDPConfigurations(1, %+v) = %v, want = %s", config, err, tcpip.ErrUnknownNICID)
+ }
+}
+
+// TestNUDConfigurationFailsForNotSupported tests to make sure we get a
+// NotSupported error if we attempt to retrieve NUD configurations when the
+// stack doesn't support NUD.
+//
+// The stack will report to not support NUD if a neighbor cache for a given NIC
+// is not allocated. The networking stack will only allocate neighbor caches if
+// a protocol providing link address resolution is specified (e.g. ARP, IPv6).
+func TestNUDConfigurationFailsForNotSupported(t *testing.T) {
+ const nicID = 1
+
+ e := channel.New(0, 1280, linkAddr1)
+ e.LinkEPCapabilities |= stack.CapabilityResolutionRequired
+
+ s := stack.New(stack.Options{
+ NUDConfigs: stack.DefaultNUDConfigurations(),
+ })
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+ if _, err := s.NUDConfigurations(nicID); err != tcpip.ErrNotSupported {
+ t.Fatalf("got s.NDPConfigurations(%d) = %v, want = %s", nicID, err, tcpip.ErrNotSupported)
+ }
+}
+
+// TestNUDConfigurationFailsForNotSupported tests to make sure we get a
+// NotSupported error if we attempt to set NUD configurations when the stack
+// doesn't support NUD.
+//
+// The stack will report to not support NUD if a neighbor cache for a given NIC
+// is not allocated. The networking stack will only allocate neighbor caches if
+// a protocol providing link address resolution is specified (e.g. ARP, IPv6).
+func TestSetNUDConfigurationFailsForNotSupported(t *testing.T) {
+ const nicID = 1
+
+ e := channel.New(0, 1280, linkAddr1)
+ e.LinkEPCapabilities |= stack.CapabilityResolutionRequired
+
+ s := stack.New(stack.Options{
+ NUDConfigs: stack.DefaultNUDConfigurations(),
+ })
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+
+ config := stack.NUDConfigurations{}
+ if err := s.SetNUDConfigurations(nicID, config); err != tcpip.ErrNotSupported {
+ t.Fatalf("got s.SetNDPConfigurations(%d, %+v) = %v, want = %s", nicID, config, err, tcpip.ErrNotSupported)
+ }
+}
+
+// TestDefaultNUDConfigurationIsValid verifies that calling
+// resetInvalidFields() on the result of DefaultNUDConfigurations() does not
+// change anything. DefaultNUDConfigurations() should return a valid
+// NUDConfigurations.
+func TestDefaultNUDConfigurations(t *testing.T) {
+ const nicID = 1
+
+ e := channel.New(0, 1280, linkAddr1)
+ e.LinkEPCapabilities |= stack.CapabilityResolutionRequired
+
+ s := stack.New(stack.Options{
+ // A neighbor cache is required to store NUDConfigurations. The networking
+ // stack will only allocate neighbor caches if a protocol providing link
+ // address resolution is specified (e.g. ARP or IPv6).
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NUDConfigs: stack.DefaultNUDConfigurations(),
+ })
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+ c, err := s.NUDConfigurations(nicID)
+ if err != nil {
+ t.Fatalf("got stack.NUDConfigurations(%d) = %s", nicID, err)
+ }
+ if got, want := c, stack.DefaultNUDConfigurations(); got != want {
+ t.Errorf("got stack.NUDConfigurations(%d) = %+v, want = %+v", nicID, got, want)
+ }
+}
+
+func TestNUDConfigurationsBaseReachableTime(t *testing.T) {
+ tests := []struct {
+ name string
+ baseReachableTime time.Duration
+ want time.Duration
+ }{
+ // Invalid cases
+ {
+ name: "EqualToZero",
+ baseReachableTime: 0,
+ want: defaultBaseReachableTime,
+ },
+ // Valid cases
+ {
+ name: "MoreThanZero",
+ baseReachableTime: time.Millisecond,
+ want: time.Millisecond,
+ },
+ {
+ name: "MoreThanDefaultBaseReachableTime",
+ baseReachableTime: 2 * defaultBaseReachableTime,
+ want: 2 * defaultBaseReachableTime,
+ },
+ }
+
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ const nicID = 1
+
+ c := stack.DefaultNUDConfigurations()
+ c.BaseReachableTime = test.baseReachableTime
+
+ e := channel.New(0, 1280, linkAddr1)
+ e.LinkEPCapabilities |= stack.CapabilityResolutionRequired
+
+ s := stack.New(stack.Options{
+ // A neighbor cache is required to store NUDConfigurations. The
+ // networking stack will only allocate neighbor caches if a protocol
+ // providing link address resolution is specified (e.g. ARP or IPv6).
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NUDConfigs: c,
+ })
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+ sc, err := s.NUDConfigurations(nicID)
+ if err != nil {
+ t.Fatalf("got stack.NUDConfigurations(%d) = %s", nicID, err)
+ }
+ if got := sc.BaseReachableTime; got != test.want {
+ t.Errorf("got BaseReachableTime = %q, want = %q", got, test.want)
+ }
+ })
+ }
+}
+
+func TestNUDConfigurationsMinRandomFactor(t *testing.T) {
+ tests := []struct {
+ name string
+ minRandomFactor float32
+ want float32
+ }{
+ // Invalid cases
+ {
+ name: "LessThanZero",
+ minRandomFactor: -1,
+ want: defaultMinRandomFactor,
+ },
+ {
+ name: "EqualToZero",
+ minRandomFactor: 0,
+ want: defaultMinRandomFactor,
+ },
+ // Valid cases
+ {
+ name: "MoreThanZero",
+ minRandomFactor: 1,
+ want: 1,
+ },
+ }
+
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ const nicID = 1
+
+ c := stack.DefaultNUDConfigurations()
+ c.MinRandomFactor = test.minRandomFactor
+
+ e := channel.New(0, 1280, linkAddr1)
+ e.LinkEPCapabilities |= stack.CapabilityResolutionRequired
+
+ s := stack.New(stack.Options{
+ // A neighbor cache is required to store NUDConfigurations. The
+ // networking stack will only allocate neighbor caches if a protocol
+ // providing link address resolution is specified (e.g. ARP or IPv6).
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NUDConfigs: c,
+ })
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+ sc, err := s.NUDConfigurations(nicID)
+ if err != nil {
+ t.Fatalf("got stack.NUDConfigurations(%d) = %s", nicID, err)
+ }
+ if got := sc.MinRandomFactor; got != test.want {
+ t.Errorf("got MinRandomFactor = %f, want = %f", got, test.want)
+ }
+ })
+ }
+}
+
+func TestNUDConfigurationsMaxRandomFactor(t *testing.T) {
+ tests := []struct {
+ name string
+ minRandomFactor float32
+ maxRandomFactor float32
+ want float32
+ }{
+ // Invalid cases
+ {
+ name: "LessThanZero",
+ minRandomFactor: defaultMinRandomFactor,
+ maxRandomFactor: -1,
+ want: defaultMaxRandomFactor,
+ },
+ {
+ name: "EqualToZero",
+ minRandomFactor: defaultMinRandomFactor,
+ maxRandomFactor: 0,
+ want: defaultMaxRandomFactor,
+ },
+ {
+ name: "LessThanMinRandomFactor",
+ minRandomFactor: defaultMinRandomFactor,
+ maxRandomFactor: defaultMinRandomFactor * 0.99,
+ want: defaultMaxRandomFactor,
+ },
+ {
+ name: "MoreThanMinRandomFactorWhenMinRandomFactorIsLargerThanMaxRandomFactorDefault",
+ minRandomFactor: defaultMaxRandomFactor * 2,
+ maxRandomFactor: defaultMaxRandomFactor,
+ want: defaultMaxRandomFactor * 6,
+ },
+ // Valid cases
+ {
+ name: "EqualToMinRandomFactor",
+ minRandomFactor: defaultMinRandomFactor,
+ maxRandomFactor: defaultMinRandomFactor,
+ want: defaultMinRandomFactor,
+ },
+ {
+ name: "MoreThanMinRandomFactor",
+ minRandomFactor: defaultMinRandomFactor,
+ maxRandomFactor: defaultMinRandomFactor * 1.1,
+ want: defaultMinRandomFactor * 1.1,
+ },
+ }
+
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ const nicID = 1
+
+ c := stack.DefaultNUDConfigurations()
+ c.MinRandomFactor = test.minRandomFactor
+ c.MaxRandomFactor = test.maxRandomFactor
+
+ e := channel.New(0, 1280, linkAddr1)
+ e.LinkEPCapabilities |= stack.CapabilityResolutionRequired
+
+ s := stack.New(stack.Options{
+ // A neighbor cache is required to store NUDConfigurations. The
+ // networking stack will only allocate neighbor caches if a protocol
+ // providing link address resolution is specified (e.g. ARP or IPv6).
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NUDConfigs: c,
+ })
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+ sc, err := s.NUDConfigurations(nicID)
+ if err != nil {
+ t.Fatalf("got stack.NUDConfigurations(%d) = %s", nicID, err)
+ }
+ if got := sc.MaxRandomFactor; got != test.want {
+ t.Errorf("got MaxRandomFactor = %f, want = %f", got, test.want)
+ }
+ })
+ }
+}
+
+func TestNUDConfigurationsRetransmitTimer(t *testing.T) {
+ tests := []struct {
+ name string
+ retransmitTimer time.Duration
+ want time.Duration
+ }{
+ // Invalid cases
+ {
+ name: "EqualToZero",
+ retransmitTimer: 0,
+ want: defaultRetransmitTimer,
+ },
+ {
+ name: "LessThanMinimumRetransmitTimer",
+ retransmitTimer: minimumRetransmitTimer - time.Nanosecond,
+ want: defaultRetransmitTimer,
+ },
+ // Valid cases
+ {
+ name: "EqualToMinimumRetransmitTimer",
+ retransmitTimer: minimumRetransmitTimer,
+ want: minimumBaseReachableTime,
+ },
+ {
+ name: "LargetThanMinimumRetransmitTimer",
+ retransmitTimer: 2 * minimumBaseReachableTime,
+ want: 2 * minimumBaseReachableTime,
+ },
+ }
+
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ const nicID = 1
+
+ c := stack.DefaultNUDConfigurations()
+ c.RetransmitTimer = test.retransmitTimer
+
+ e := channel.New(0, 1280, linkAddr1)
+ e.LinkEPCapabilities |= stack.CapabilityResolutionRequired
+
+ s := stack.New(stack.Options{
+ // A neighbor cache is required to store NUDConfigurations. The
+ // networking stack will only allocate neighbor caches if a protocol
+ // providing link address resolution is specified (e.g. ARP or IPv6).
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NUDConfigs: c,
+ })
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+ sc, err := s.NUDConfigurations(nicID)
+ if err != nil {
+ t.Fatalf("got stack.NUDConfigurations(%d) = %s", nicID, err)
+ }
+ if got := sc.RetransmitTimer; got != test.want {
+ t.Errorf("got RetransmitTimer = %q, want = %q", got, test.want)
+ }
+ })
+ }
+}
+
+func TestNUDConfigurationsDelayFirstProbeTime(t *testing.T) {
+ tests := []struct {
+ name string
+ delayFirstProbeTime time.Duration
+ want time.Duration
+ }{
+ // Invalid cases
+ {
+ name: "EqualToZero",
+ delayFirstProbeTime: 0,
+ want: defaultDelayFirstProbeTime,
+ },
+ // Valid cases
+ {
+ name: "MoreThanZero",
+ delayFirstProbeTime: time.Millisecond,
+ want: time.Millisecond,
+ },
+ }
+
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ const nicID = 1
+
+ c := stack.DefaultNUDConfigurations()
+ c.DelayFirstProbeTime = test.delayFirstProbeTime
+
+ e := channel.New(0, 1280, linkAddr1)
+ e.LinkEPCapabilities |= stack.CapabilityResolutionRequired
+
+ s := stack.New(stack.Options{
+ // A neighbor cache is required to store NUDConfigurations. The
+ // networking stack will only allocate neighbor caches if a protocol
+ // providing link address resolution is specified (e.g. ARP or IPv6).
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NUDConfigs: c,
+ })
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+ sc, err := s.NUDConfigurations(nicID)
+ if err != nil {
+ t.Fatalf("got stack.NUDConfigurations(%d) = %s", nicID, err)
+ }
+ if got := sc.DelayFirstProbeTime; got != test.want {
+ t.Errorf("got DelayFirstProbeTime = %q, want = %q", got, test.want)
+ }
+ })
+ }
+}
+
+func TestNUDConfigurationsMaxMulticastProbes(t *testing.T) {
+ tests := []struct {
+ name string
+ maxMulticastProbes uint32
+ want uint32
+ }{
+ // Invalid cases
+ {
+ name: "EqualToZero",
+ maxMulticastProbes: 0,
+ want: defaultMaxMulticastProbes,
+ },
+ // Valid cases
+ {
+ name: "MoreThanZero",
+ maxMulticastProbes: 1,
+ want: 1,
+ },
+ }
+
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ const nicID = 1
+
+ c := stack.DefaultNUDConfigurations()
+ c.MaxMulticastProbes = test.maxMulticastProbes
+
+ e := channel.New(0, 1280, linkAddr1)
+ e.LinkEPCapabilities |= stack.CapabilityResolutionRequired
+
+ s := stack.New(stack.Options{
+ // A neighbor cache is required to store NUDConfigurations. The
+ // networking stack will only allocate neighbor caches if a protocol
+ // providing link address resolution is specified (e.g. ARP or IPv6).
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NUDConfigs: c,
+ })
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+ sc, err := s.NUDConfigurations(nicID)
+ if err != nil {
+ t.Fatalf("got stack.NUDConfigurations(%d) = %s", nicID, err)
+ }
+ if got := sc.MaxMulticastProbes; got != test.want {
+ t.Errorf("got MaxMulticastProbes = %q, want = %q", got, test.want)
+ }
+ })
+ }
+}
+
+func TestNUDConfigurationsMaxUnicastProbes(t *testing.T) {
+ tests := []struct {
+ name string
+ maxUnicastProbes uint32
+ want uint32
+ }{
+ // Invalid cases
+ {
+ name: "EqualToZero",
+ maxUnicastProbes: 0,
+ want: defaultMaxUnicastProbes,
+ },
+ // Valid cases
+ {
+ name: "MoreThanZero",
+ maxUnicastProbes: 1,
+ want: 1,
+ },
+ }
+
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ const nicID = 1
+
+ c := stack.DefaultNUDConfigurations()
+ c.MaxUnicastProbes = test.maxUnicastProbes
+
+ e := channel.New(0, 1280, linkAddr1)
+ e.LinkEPCapabilities |= stack.CapabilityResolutionRequired
+
+ s := stack.New(stack.Options{
+ // A neighbor cache is required to store NUDConfigurations. The
+ // networking stack will only allocate neighbor caches if a protocol
+ // providing link address resolution is specified (e.g. ARP or IPv6).
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NUDConfigs: c,
+ })
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+ sc, err := s.NUDConfigurations(nicID)
+ if err != nil {
+ t.Fatalf("got stack.NUDConfigurations(%d) = %s", nicID, err)
+ }
+ if got := sc.MaxUnicastProbes; got != test.want {
+ t.Errorf("got MaxUnicastProbes = %q, want = %q", got, test.want)
+ }
+ })
+ }
+}
+
+func TestNUDConfigurationsUnreachableTime(t *testing.T) {
+ tests := []struct {
+ name string
+ unreachableTime time.Duration
+ want time.Duration
+ }{
+ // Invalid cases
+ {
+ name: "EqualToZero",
+ unreachableTime: 0,
+ want: defaultUnreachableTime,
+ },
+ // Valid cases
+ {
+ name: "MoreThanZero",
+ unreachableTime: time.Millisecond,
+ want: time.Millisecond,
+ },
+ }
+
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ const nicID = 1
+
+ c := stack.DefaultNUDConfigurations()
+ c.UnreachableTime = test.unreachableTime
+
+ e := channel.New(0, 1280, linkAddr1)
+ e.LinkEPCapabilities |= stack.CapabilityResolutionRequired
+
+ s := stack.New(stack.Options{
+ // A neighbor cache is required to store NUDConfigurations. The
+ // networking stack will only allocate neighbor caches if a protocol
+ // providing link address resolution is specified (e.g. ARP or IPv6).
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NUDConfigs: c,
+ })
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+ sc, err := s.NUDConfigurations(nicID)
+ if err != nil {
+ t.Fatalf("got stack.NUDConfigurations(%d) = %s", nicID, err)
+ }
+ if got := sc.UnreachableTime; got != test.want {
+ t.Errorf("got UnreachableTime = %q, want = %q", got, test.want)
+ }
+ })
+ }
+}
+
+// TestNUDStateReachableTime verifies the correctness of the ReachableTime
+// computation.
+func TestNUDStateReachableTime(t *testing.T) {
+ tests := []struct {
+ name string
+ baseReachableTime time.Duration
+ minRandomFactor float32
+ maxRandomFactor float32
+ want time.Duration
+ }{
+ {
+ name: "AllZeros",
+ baseReachableTime: 0,
+ minRandomFactor: 0,
+ maxRandomFactor: 0,
+ want: 0,
+ },
+ {
+ name: "ZeroMaxRandomFactor",
+ baseReachableTime: time.Second,
+ minRandomFactor: 0,
+ maxRandomFactor: 0,
+ want: 0,
+ },
+ {
+ name: "ZeroMinRandomFactor",
+ baseReachableTime: time.Second,
+ minRandomFactor: 0,
+ maxRandomFactor: 1,
+ want: time.Duration(defaultFakeRandomNum * float32(time.Second)),
+ },
+ {
+ name: "FractionalRandomFactor",
+ baseReachableTime: time.Duration(math.MaxInt64),
+ minRandomFactor: 0.001,
+ maxRandomFactor: 0.002,
+ want: time.Duration((0.001 + (0.001 * defaultFakeRandomNum)) * float32(math.MaxInt64)),
+ },
+ {
+ name: "MinAndMaxRandomFactorsEqual",
+ baseReachableTime: time.Second,
+ minRandomFactor: 1,
+ maxRandomFactor: 1,
+ want: time.Second,
+ },
+ {
+ name: "MinAndMaxRandomFactorsDifferent",
+ baseReachableTime: time.Second,
+ minRandomFactor: 1,
+ maxRandomFactor: 2,
+ want: time.Duration((1.0 + defaultFakeRandomNum) * float32(time.Second)),
+ },
+ {
+ name: "MaxInt64",
+ baseReachableTime: time.Duration(math.MaxInt64),
+ minRandomFactor: 1,
+ maxRandomFactor: 1,
+ want: time.Duration(math.MaxInt64),
+ },
+ {
+ name: "Overflow",
+ baseReachableTime: time.Duration(math.MaxInt64),
+ minRandomFactor: 1.5,
+ maxRandomFactor: 1.5,
+ want: time.Duration(math.MaxInt64),
+ },
+ {
+ name: "DoubleOverflow",
+ baseReachableTime: time.Duration(math.MaxInt64),
+ minRandomFactor: 2.5,
+ maxRandomFactor: 2.5,
+ want: time.Duration(math.MaxInt64),
+ },
+ }
+
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ c := stack.NUDConfigurations{
+ BaseReachableTime: test.baseReachableTime,
+ MinRandomFactor: test.minRandomFactor,
+ MaxRandomFactor: test.maxRandomFactor,
+ }
+ // A fake random number generator is used to ensure deterministic
+ // results.
+ rng := fakeRand{
+ num: defaultFakeRandomNum,
+ }
+ s := stack.NewNUDState(c, &rng)
+ if got, want := s.ReachableTime(), test.want; got != want {
+ t.Errorf("got ReachableTime = %q, want = %q", got, want)
+ }
+ })
+ }
+}
+
+// TestNUDStateRecomputeReachableTime exercises the ReachableTime function
+// twice to verify recomputation of reachable time when the min random factor,
+// max random factor, or base reachable time changes.
+func TestNUDStateRecomputeReachableTime(t *testing.T) {
+ const defaultBase = time.Second
+ const defaultMin = 2.0 * defaultMaxRandomFactor
+ const defaultMax = 3.0 * defaultMaxRandomFactor
+
+ tests := []struct {
+ name string
+ baseReachableTime time.Duration
+ minRandomFactor float32
+ maxRandomFactor float32
+ want time.Duration
+ }{
+ {
+ name: "BaseReachableTime",
+ baseReachableTime: 2 * defaultBase,
+ minRandomFactor: defaultMin,
+ maxRandomFactor: defaultMax,
+ want: time.Duration((defaultMin + (defaultMax-defaultMin)*defaultFakeRandomNum) * float32(2*defaultBase)),
+ },
+ {
+ name: "MinRandomFactor",
+ baseReachableTime: defaultBase,
+ minRandomFactor: defaultMax,
+ maxRandomFactor: defaultMax,
+ want: time.Duration(defaultMax * float32(defaultBase)),
+ },
+ {
+ name: "MaxRandomFactor",
+ baseReachableTime: defaultBase,
+ minRandomFactor: defaultMin,
+ maxRandomFactor: defaultMin,
+ want: time.Duration(defaultMin * float32(defaultBase)),
+ },
+ {
+ name: "BothRandomFactor",
+ baseReachableTime: defaultBase,
+ minRandomFactor: 2 * defaultMin,
+ maxRandomFactor: 2 * defaultMax,
+ want: time.Duration((2*defaultMin + (2*defaultMax-2*defaultMin)*defaultFakeRandomNum) * float32(defaultBase)),
+ },
+ {
+ name: "BaseReachableTimeAndBothRandomFactors",
+ baseReachableTime: 2 * defaultBase,
+ minRandomFactor: 2 * defaultMin,
+ maxRandomFactor: 2 * defaultMax,
+ want: time.Duration((2*defaultMin + (2*defaultMax-2*defaultMin)*defaultFakeRandomNum) * float32(2*defaultBase)),
+ },
+ }
+
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ c := stack.DefaultNUDConfigurations()
+ c.BaseReachableTime = defaultBase
+ c.MinRandomFactor = defaultMin
+ c.MaxRandomFactor = defaultMax
+
+ // A fake random number generator is used to ensure deterministic
+ // results.
+ rng := fakeRand{
+ num: defaultFakeRandomNum,
+ }
+ s := stack.NewNUDState(c, &rng)
+ old := s.ReachableTime()
+
+ if got, want := s.ReachableTime(), old; got != want {
+ t.Errorf("got ReachableTime = %q, want = %q", got, want)
+ }
+
+ // Check for recomputation when changing the min random factor, the max
+ // random factor, the base reachability time, or any permutation of those
+ // three options.
+ c.BaseReachableTime = test.baseReachableTime
+ c.MinRandomFactor = test.minRandomFactor
+ c.MaxRandomFactor = test.maxRandomFactor
+ s.SetConfig(c)
+
+ if got, want := s.ReachableTime(), test.want; got != want {
+ t.Errorf("got ReachableTime = %q, want = %q", got, want)
+ }
+
+ // Verify that ReachableTime isn't recomputed when none of the
+ // configuration options change. The random factor is changed so that if
+ // a recompution were to occur, ReachableTime would change.
+ rng.num = defaultFakeRandomNum / 2.0
+ if got, want := s.ReachableTime(), test.want; got != want {
+ t.Errorf("got ReachableTime = %q, want = %q", got, want)
+ }
+ })
+ }
+}
diff --git a/pkg/tcpip/stack/packet_buffer.go b/pkg/tcpip/stack/packet_buffer.go
new file mode 100644
index 000000000..17b8beebb
--- /dev/null
+++ b/pkg/tcpip/stack/packet_buffer.go
@@ -0,0 +1,299 @@
+// Copyright 2019 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at //
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package stack
+
+import (
+ "fmt"
+
+ "gvisor.dev/gvisor/pkg/sync"
+ "gvisor.dev/gvisor/pkg/tcpip"
+ "gvisor.dev/gvisor/pkg/tcpip/buffer"
+)
+
+type headerType int
+
+const (
+ linkHeader headerType = iota
+ networkHeader
+ transportHeader
+ numHeaderType
+)
+
+// PacketBufferOptions specifies options for PacketBuffer creation.
+type PacketBufferOptions struct {
+ // ReserveHeaderBytes is the number of bytes to reserve for headers. Total
+ // number of bytes pushed onto the headers must not exceed this value.
+ ReserveHeaderBytes int
+
+ // Data is the initial unparsed data for the new packet. If set, it will be
+ // owned by the new packet.
+ Data buffer.VectorisedView
+}
+
+// A PacketBuffer contains all the data of a network packet.
+//
+// As a PacketBuffer traverses up the stack, it may be necessary to pass it to
+// multiple endpoints.
+//
+// The whole packet is expected to be a series of bytes in the following order:
+// LinkHeader, NetworkHeader, TransportHeader, and Data. Any of them can be
+// empty. Use of PacketBuffer in any other order is unsupported.
+//
+// PacketBuffer must be created with NewPacketBuffer.
+type PacketBuffer struct {
+ _ sync.NoCopy
+
+ // PacketBufferEntry is used to build an intrusive list of
+ // PacketBuffers.
+ PacketBufferEntry
+
+ // Data holds the payload of the packet.
+ //
+ // For inbound packets, Data is initially the whole packet. Then gets moved to
+ // headers via PacketHeader.Consume, when the packet is being parsed.
+ //
+ // For outbound packets, Data is the innermost layer, defined by the protocol.
+ // Headers are pushed in front of it via PacketHeader.Push.
+ //
+ // The bytes backing Data are immutable, a.k.a. users shouldn't write to its
+ // backing storage.
+ Data buffer.VectorisedView
+
+ // headers stores metadata about each header.
+ headers [numHeaderType]headerInfo
+
+ // header is the internal storage for outbound packets. Headers will be pushed
+ // (prepended) on this storage as the packet is being constructed.
+ //
+ // TODO(gvisor.dev/issue/2404): Switch to an implementation that header and
+ // data are held in the same underlying buffer storage.
+ header buffer.Prependable
+
+ // NetworkProtocol is only valid when NetworkHeader is set.
+ // TODO(gvisor.dev/issue/3574): Remove the separately passed protocol
+ // numbers in registration APIs that take a PacketBuffer.
+ NetworkProtocolNumber tcpip.NetworkProtocolNumber
+
+ // Hash is the transport layer hash of this packet. A value of zero
+ // indicates no valid hash has been set.
+ Hash uint32
+
+ // Owner is implemented by task to get the uid and gid.
+ // Only set for locally generated packets.
+ Owner tcpip.PacketOwner
+
+ // The following fields are only set by the qdisc layer when the packet
+ // is added to a queue.
+ EgressRoute *Route
+ GSOOptions *GSO
+
+ // NatDone indicates if the packet has been manipulated as per NAT
+ // iptables rule.
+ NatDone bool
+
+ // PktType indicates the SockAddrLink.PacketType of the packet as defined in
+ // https://www.man7.org/linux/man-pages/man7/packet.7.html.
+ PktType tcpip.PacketType
+}
+
+// NewPacketBuffer creates a new PacketBuffer with opts.
+func NewPacketBuffer(opts PacketBufferOptions) *PacketBuffer {
+ pk := &PacketBuffer{
+ Data: opts.Data,
+ }
+ if opts.ReserveHeaderBytes != 0 {
+ pk.header = buffer.NewPrependable(opts.ReserveHeaderBytes)
+ }
+ return pk
+}
+
+// ReservedHeaderBytes returns the number of bytes initially reserved for
+// headers.
+func (pk *PacketBuffer) ReservedHeaderBytes() int {
+ return pk.header.UsedLength() + pk.header.AvailableLength()
+}
+
+// AvailableHeaderBytes returns the number of bytes currently available for
+// headers. This is relevant to PacketHeader.Push method only.
+func (pk *PacketBuffer) AvailableHeaderBytes() int {
+ return pk.header.AvailableLength()
+}
+
+// LinkHeader returns the handle to link-layer header.
+func (pk *PacketBuffer) LinkHeader() PacketHeader {
+ return PacketHeader{
+ pk: pk,
+ typ: linkHeader,
+ }
+}
+
+// NetworkHeader returns the handle to network-layer header.
+func (pk *PacketBuffer) NetworkHeader() PacketHeader {
+ return PacketHeader{
+ pk: pk,
+ typ: networkHeader,
+ }
+}
+
+// TransportHeader returns the handle to transport-layer header.
+func (pk *PacketBuffer) TransportHeader() PacketHeader {
+ return PacketHeader{
+ pk: pk,
+ typ: transportHeader,
+ }
+}
+
+// HeaderSize returns the total size of all headers in bytes.
+func (pk *PacketBuffer) HeaderSize() int {
+ // Note for inbound packets (Consume called), headers are not stored in
+ // pk.header. Thus, calculation of size of each header is needed.
+ var size int
+ for i := range pk.headers {
+ size += len(pk.headers[i].buf)
+ }
+ return size
+}
+
+// Size returns the size of packet in bytes.
+func (pk *PacketBuffer) Size() int {
+ return pk.HeaderSize() + pk.Data.Size()
+}
+
+// Views returns the underlying storage of the whole packet.
+func (pk *PacketBuffer) Views() []buffer.View {
+ // Optimization for outbound packets that headers are in pk.header.
+ useHeader := true
+ for i := range pk.headers {
+ if !canUseHeader(&pk.headers[i]) {
+ useHeader = false
+ break
+ }
+ }
+
+ dataViews := pk.Data.Views()
+
+ var vs []buffer.View
+ if useHeader {
+ vs = make([]buffer.View, 0, 1+len(dataViews))
+ vs = append(vs, pk.header.View())
+ } else {
+ vs = make([]buffer.View, 0, len(pk.headers)+len(dataViews))
+ for i := range pk.headers {
+ if v := pk.headers[i].buf; len(v) > 0 {
+ vs = append(vs, v)
+ }
+ }
+ }
+ return append(vs, dataViews...)
+}
+
+func canUseHeader(h *headerInfo) bool {
+ // h.offset will be negative if the header was pushed in to prependable
+ // portion, or doesn't matter when it's empty.
+ return len(h.buf) == 0 || h.offset < 0
+}
+
+func (pk *PacketBuffer) push(typ headerType, size int) buffer.View {
+ h := &pk.headers[typ]
+ if h.buf != nil {
+ panic(fmt.Sprintf("push must not be called twice: type %s", typ))
+ }
+ h.buf = buffer.View(pk.header.Prepend(size))
+ h.offset = -pk.header.UsedLength()
+ return h.buf
+}
+
+func (pk *PacketBuffer) consume(typ headerType, size int) (v buffer.View, consumed bool) {
+ h := &pk.headers[typ]
+ if h.buf != nil {
+ panic(fmt.Sprintf("consume must not be called twice: type %s", typ))
+ }
+ v, ok := pk.Data.PullUp(size)
+ if !ok {
+ return
+ }
+ pk.Data.TrimFront(size)
+ h.buf = v
+ return h.buf, true
+}
+
+// Clone makes a shallow copy of pk.
+//
+// Clone should be called in such cases so that no modifications is done to
+// underlying packet payload.
+func (pk *PacketBuffer) Clone() *PacketBuffer {
+ newPk := &PacketBuffer{
+ PacketBufferEntry: pk.PacketBufferEntry,
+ Data: pk.Data.Clone(nil),
+ headers: pk.headers,
+ header: pk.header,
+ Hash: pk.Hash,
+ Owner: pk.Owner,
+ EgressRoute: pk.EgressRoute,
+ GSOOptions: pk.GSOOptions,
+ NetworkProtocolNumber: pk.NetworkProtocolNumber,
+ NatDone: pk.NatDone,
+ }
+ return newPk
+}
+
+// headerInfo stores metadata about a header in a packet.
+type headerInfo struct {
+ // buf is the memorized slice for both prepended and consumed header.
+ // When header is prepended, buf serves as memorized value, which is a slice
+ // of pk.header. When header is consumed, buf is the slice pulled out from
+ // pk.Data, which is the only place to hold this header.
+ buf buffer.View
+
+ // offset will be a negative number denoting the offset where this header is
+ // from the end of pk.header, if it is prepended. Otherwise, zero.
+ offset int
+}
+
+// PacketHeader is a handle object to a header in the underlying packet.
+type PacketHeader struct {
+ pk *PacketBuffer
+ typ headerType
+}
+
+// View returns the underlying storage of h.
+func (h PacketHeader) View() buffer.View {
+ return h.pk.headers[h.typ].buf
+}
+
+// Push pushes size bytes in the front of its residing packet, and returns the
+// backing storage. Callers may only call one of Push or Consume once on each
+// header in the lifetime of the underlying packet.
+func (h PacketHeader) Push(size int) buffer.View {
+ return h.pk.push(h.typ, size)
+}
+
+// Consume moves the first size bytes of the unparsed data portion in the packet
+// to h, and returns the backing storage. In the case of data is shorter than
+// size, consumed will be false, and the state of h will not be affected.
+// Callers may only call one of Push or Consume once on each header in the
+// lifetime of the underlying packet.
+func (h PacketHeader) Consume(size int) (v buffer.View, consumed bool) {
+ return h.pk.consume(h.typ, size)
+}
+
+// PayloadSince returns packet payload starting from and including a particular
+// header. This method isn't optimized and should be used in test only.
+func PayloadSince(h PacketHeader) buffer.View {
+ var v buffer.View
+ for _, hinfo := range h.pk.headers[h.typ:] {
+ v = append(v, hinfo.buf...)
+ }
+ return append(v, h.pk.Data.ToView()...)
+}
diff --git a/pkg/tcpip/stack/packet_buffer_test.go b/pkg/tcpip/stack/packet_buffer_test.go
new file mode 100644
index 000000000..c6fa8da5f
--- /dev/null
+++ b/pkg/tcpip/stack/packet_buffer_test.go
@@ -0,0 +1,397 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at //
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package stack
+
+import (
+ "bytes"
+ "testing"
+
+ "gvisor.dev/gvisor/pkg/tcpip/buffer"
+)
+
+func TestPacketHeaderPush(t *testing.T) {
+ for _, test := range []struct {
+ name string
+ reserved int
+ link []byte
+ network []byte
+ transport []byte
+ data []byte
+ }{
+ {
+ name: "construct empty packet",
+ },
+ {
+ name: "construct link header only packet",
+ reserved: 60,
+ link: makeView(10),
+ },
+ {
+ name: "construct link and network header only packet",
+ reserved: 60,
+ link: makeView(10),
+ network: makeView(20),
+ },
+ {
+ name: "construct header only packet",
+ reserved: 60,
+ link: makeView(10),
+ network: makeView(20),
+ transport: makeView(30),
+ },
+ {
+ name: "construct data only packet",
+ data: makeView(40),
+ },
+ {
+ name: "construct L3 packet",
+ reserved: 60,
+ network: makeView(20),
+ transport: makeView(30),
+ data: makeView(40),
+ },
+ {
+ name: "construct L2 packet",
+ reserved: 60,
+ link: makeView(10),
+ network: makeView(20),
+ transport: makeView(30),
+ data: makeView(40),
+ },
+ } {
+ t.Run(test.name, func(t *testing.T) {
+ pk := NewPacketBuffer(PacketBufferOptions{
+ ReserveHeaderBytes: test.reserved,
+ // Make a copy of data to make sure our truth data won't be taint by
+ // PacketBuffer.
+ Data: buffer.NewViewFromBytes(test.data).ToVectorisedView(),
+ })
+
+ allHdrSize := len(test.link) + len(test.network) + len(test.transport)
+
+ // Check the initial values for packet.
+ checkInitialPacketBuffer(t, pk, PacketBufferOptions{
+ ReserveHeaderBytes: test.reserved,
+ Data: buffer.View(test.data).ToVectorisedView(),
+ })
+
+ // Push headers.
+ if v := test.transport; len(v) > 0 {
+ copy(pk.TransportHeader().Push(len(v)), v)
+ }
+ if v := test.network; len(v) > 0 {
+ copy(pk.NetworkHeader().Push(len(v)), v)
+ }
+ if v := test.link; len(v) > 0 {
+ copy(pk.LinkHeader().Push(len(v)), v)
+ }
+
+ // Check the after values for packet.
+ if got, want := pk.ReservedHeaderBytes(), test.reserved; got != want {
+ t.Errorf("After pk.ReservedHeaderBytes() = %d, want %d", got, want)
+ }
+ if got, want := pk.AvailableHeaderBytes(), test.reserved-allHdrSize; got != want {
+ t.Errorf("After pk.AvailableHeaderBytes() = %d, want %d", got, want)
+ }
+ if got, want := pk.HeaderSize(), allHdrSize; got != want {
+ t.Errorf("After pk.HeaderSize() = %d, want %d", got, want)
+ }
+ if got, want := pk.Size(), allHdrSize+len(test.data); got != want {
+ t.Errorf("After pk.Size() = %d, want %d", got, want)
+ }
+ checkViewEqual(t, "After pk.Data.Views()", concatViews(pk.Data.Views()...), test.data)
+ checkViewEqual(t, "After pk.Views()", concatViews(pk.Views()...),
+ concatViews(test.link, test.network, test.transport, test.data))
+ // Check the after values for each header.
+ checkPacketHeader(t, "After pk.LinkHeader", pk.LinkHeader(), test.link)
+ checkPacketHeader(t, "After pk.NetworkHeader", pk.NetworkHeader(), test.network)
+ checkPacketHeader(t, "After pk.TransportHeader", pk.TransportHeader(), test.transport)
+ // Check the after values for PayloadSince.
+ checkViewEqual(t, "After PayloadSince(LinkHeader)",
+ PayloadSince(pk.LinkHeader()),
+ concatViews(test.link, test.network, test.transport, test.data))
+ checkViewEqual(t, "After PayloadSince(NetworkHeader)",
+ PayloadSince(pk.NetworkHeader()),
+ concatViews(test.network, test.transport, test.data))
+ checkViewEqual(t, "After PayloadSince(TransportHeader)",
+ PayloadSince(pk.TransportHeader()),
+ concatViews(test.transport, test.data))
+ })
+ }
+}
+
+func TestPacketHeaderConsume(t *testing.T) {
+ for _, test := range []struct {
+ name string
+ data []byte
+ link int
+ network int
+ transport int
+ }{
+ {
+ name: "parse L2 packet",
+ data: concatViews(makeView(10), makeView(20), makeView(30), makeView(40)),
+ link: 10,
+ network: 20,
+ transport: 30,
+ },
+ {
+ name: "parse L3 packet",
+ data: concatViews(makeView(20), makeView(30), makeView(40)),
+ network: 20,
+ transport: 30,
+ },
+ } {
+ t.Run(test.name, func(t *testing.T) {
+ pk := NewPacketBuffer(PacketBufferOptions{
+ // Make a copy of data to make sure our truth data won't be taint by
+ // PacketBuffer.
+ Data: buffer.NewViewFromBytes(test.data).ToVectorisedView(),
+ })
+
+ // Check the initial values for packet.
+ checkInitialPacketBuffer(t, pk, PacketBufferOptions{
+ Data: buffer.View(test.data).ToVectorisedView(),
+ })
+
+ // Consume headers.
+ if size := test.link; size > 0 {
+ if _, ok := pk.LinkHeader().Consume(size); !ok {
+ t.Fatalf("pk.LinkHeader().Consume() = false, want true")
+ }
+ }
+ if size := test.network; size > 0 {
+ if _, ok := pk.NetworkHeader().Consume(size); !ok {
+ t.Fatalf("pk.NetworkHeader().Consume() = false, want true")
+ }
+ }
+ if size := test.transport; size > 0 {
+ if _, ok := pk.TransportHeader().Consume(size); !ok {
+ t.Fatalf("pk.TransportHeader().Consume() = false, want true")
+ }
+ }
+
+ allHdrSize := test.link + test.network + test.transport
+
+ // Check the after values for packet.
+ if got, want := pk.ReservedHeaderBytes(), 0; got != want {
+ t.Errorf("After pk.ReservedHeaderBytes() = %d, want %d", got, want)
+ }
+ if got, want := pk.AvailableHeaderBytes(), 0; got != want {
+ t.Errorf("After pk.AvailableHeaderBytes() = %d, want %d", got, want)
+ }
+ if got, want := pk.HeaderSize(), allHdrSize; got != want {
+ t.Errorf("After pk.HeaderSize() = %d, want %d", got, want)
+ }
+ if got, want := pk.Size(), len(test.data); got != want {
+ t.Errorf("After pk.Size() = %d, want %d", got, want)
+ }
+ // After state of pk.
+ var (
+ link = test.data[:test.link]
+ network = test.data[test.link:][:test.network]
+ transport = test.data[test.link+test.network:][:test.transport]
+ payload = test.data[allHdrSize:]
+ )
+ checkViewEqual(t, "After pk.Data.Views()", concatViews(pk.Data.Views()...), payload)
+ checkViewEqual(t, "After pk.Views()", concatViews(pk.Views()...), test.data)
+ // Check the after values for each header.
+ checkPacketHeader(t, "After pk.LinkHeader", pk.LinkHeader(), link)
+ checkPacketHeader(t, "After pk.NetworkHeader", pk.NetworkHeader(), network)
+ checkPacketHeader(t, "After pk.TransportHeader", pk.TransportHeader(), transport)
+ // Check the after values for PayloadSince.
+ checkViewEqual(t, "After PayloadSince(LinkHeader)",
+ PayloadSince(pk.LinkHeader()),
+ concatViews(link, network, transport, payload))
+ checkViewEqual(t, "After PayloadSince(NetworkHeader)",
+ PayloadSince(pk.NetworkHeader()),
+ concatViews(network, transport, payload))
+ checkViewEqual(t, "After PayloadSince(TransportHeader)",
+ PayloadSince(pk.TransportHeader()),
+ concatViews(transport, payload))
+ })
+ }
+}
+
+func TestPacketHeaderConsumeDataTooShort(t *testing.T) {
+ data := makeView(10)
+
+ pk := NewPacketBuffer(PacketBufferOptions{
+ // Make a copy of data to make sure our truth data won't be taint by
+ // PacketBuffer.
+ Data: buffer.NewViewFromBytes(data).ToVectorisedView(),
+ })
+
+ // Consume should fail if pkt.Data is too short.
+ if _, ok := pk.LinkHeader().Consume(11); ok {
+ t.Fatalf("pk.LinkHeader().Consume() = _, true; want _, false")
+ }
+ if _, ok := pk.NetworkHeader().Consume(11); ok {
+ t.Fatalf("pk.NetworkHeader().Consume() = _, true; want _, false")
+ }
+ if _, ok := pk.TransportHeader().Consume(11); ok {
+ t.Fatalf("pk.TransportHeader().Consume() = _, true; want _, false")
+ }
+
+ // Check packet should look the same as initial packet.
+ checkInitialPacketBuffer(t, pk, PacketBufferOptions{
+ Data: buffer.View(data).ToVectorisedView(),
+ })
+}
+
+func TestPacketHeaderPushCalledAtMostOnce(t *testing.T) {
+ const headerSize = 10
+
+ pk := NewPacketBuffer(PacketBufferOptions{
+ ReserveHeaderBytes: headerSize * int(numHeaderType),
+ })
+
+ for _, h := range []PacketHeader{
+ pk.TransportHeader(),
+ pk.NetworkHeader(),
+ pk.LinkHeader(),
+ } {
+ t.Run("PushedTwice/"+h.typ.String(), func(t *testing.T) {
+ h.Push(headerSize)
+
+ defer func() { recover() }()
+ h.Push(headerSize)
+ t.Fatal("Second push should have panicked")
+ })
+ }
+}
+
+func TestPacketHeaderConsumeCalledAtMostOnce(t *testing.T) {
+ const headerSize = 10
+
+ pk := NewPacketBuffer(PacketBufferOptions{
+ Data: makeView(headerSize * int(numHeaderType)).ToVectorisedView(),
+ })
+
+ for _, h := range []PacketHeader{
+ pk.LinkHeader(),
+ pk.NetworkHeader(),
+ pk.TransportHeader(),
+ } {
+ t.Run("ConsumedTwice/"+h.typ.String(), func(t *testing.T) {
+ if _, ok := h.Consume(headerSize); !ok {
+ t.Fatal("First consume should succeed")
+ }
+
+ defer func() { recover() }()
+ h.Consume(headerSize)
+ t.Fatal("Second consume should have panicked")
+ })
+ }
+}
+
+func TestPacketHeaderPushThenConsumePanics(t *testing.T) {
+ const headerSize = 10
+
+ pk := NewPacketBuffer(PacketBufferOptions{
+ ReserveHeaderBytes: headerSize * int(numHeaderType),
+ })
+
+ for _, h := range []PacketHeader{
+ pk.TransportHeader(),
+ pk.NetworkHeader(),
+ pk.LinkHeader(),
+ } {
+ t.Run(h.typ.String(), func(t *testing.T) {
+ h.Push(headerSize)
+
+ defer func() { recover() }()
+ h.Consume(headerSize)
+ t.Fatal("Consume should have panicked")
+ })
+ }
+}
+
+func TestPacketHeaderConsumeThenPushPanics(t *testing.T) {
+ const headerSize = 10
+
+ pk := NewPacketBuffer(PacketBufferOptions{
+ Data: makeView(headerSize * int(numHeaderType)).ToVectorisedView(),
+ })
+
+ for _, h := range []PacketHeader{
+ pk.LinkHeader(),
+ pk.NetworkHeader(),
+ pk.TransportHeader(),
+ } {
+ t.Run(h.typ.String(), func(t *testing.T) {
+ h.Consume(headerSize)
+
+ defer func() { recover() }()
+ h.Push(headerSize)
+ t.Fatal("Push should have panicked")
+ })
+ }
+}
+
+func checkInitialPacketBuffer(t *testing.T, pk *PacketBuffer, opts PacketBufferOptions) {
+ t.Helper()
+ reserved := opts.ReserveHeaderBytes
+ if got, want := pk.ReservedHeaderBytes(), reserved; got != want {
+ t.Errorf("Initial pk.ReservedHeaderBytes() = %d, want %d", got, want)
+ }
+ if got, want := pk.AvailableHeaderBytes(), reserved; got != want {
+ t.Errorf("Initial pk.AvailableHeaderBytes() = %d, want %d", got, want)
+ }
+ if got, want := pk.HeaderSize(), 0; got != want {
+ t.Errorf("Initial pk.HeaderSize() = %d, want %d", got, want)
+ }
+ data := opts.Data.ToView()
+ if got, want := pk.Size(), len(data); got != want {
+ t.Errorf("Initial pk.Size() = %d, want %d", got, want)
+ }
+ checkViewEqual(t, "Initial pk.Data.Views()", concatViews(pk.Data.Views()...), data)
+ checkViewEqual(t, "Initial pk.Views()", concatViews(pk.Views()...), data)
+ // Check the initial values for each header.
+ checkPacketHeader(t, "Initial pk.LinkHeader", pk.LinkHeader(), nil)
+ checkPacketHeader(t, "Initial pk.NetworkHeader", pk.NetworkHeader(), nil)
+ checkPacketHeader(t, "Initial pk.TransportHeader", pk.TransportHeader(), nil)
+ // Check the initial valies for PayloadSince.
+ checkViewEqual(t, "Initial PayloadSince(LinkHeader)",
+ PayloadSince(pk.LinkHeader()), data)
+ checkViewEqual(t, "Initial PayloadSince(NetworkHeader)",
+ PayloadSince(pk.NetworkHeader()), data)
+ checkViewEqual(t, "Initial PayloadSince(TransportHeader)",
+ PayloadSince(pk.TransportHeader()), data)
+}
+
+func checkPacketHeader(t *testing.T, name string, h PacketHeader, want []byte) {
+ t.Helper()
+ checkViewEqual(t, name+".View()", h.View(), want)
+}
+
+func checkViewEqual(t *testing.T, what string, got, want buffer.View) {
+ t.Helper()
+ if !bytes.Equal(got, want) {
+ t.Errorf("%s = %x, want %x", what, got, want)
+ }
+}
+
+func makeView(size int) buffer.View {
+ b := byte(size)
+ return bytes.Repeat([]byte{b}, size)
+}
+
+func concatViews(views ...buffer.View) buffer.View {
+ var all buffer.View
+ for _, v := range views {
+ all = append(all, v...)
+ }
+ return all
+}
diff --git a/pkg/tcpip/stack/rand.go b/pkg/tcpip/stack/rand.go
new file mode 100644
index 000000000..421fb5c15
--- /dev/null
+++ b/pkg/tcpip/stack/rand.go
@@ -0,0 +1,40 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package stack
+
+import (
+ mathrand "math/rand"
+
+ "gvisor.dev/gvisor/pkg/sync"
+)
+
+// lockedRandomSource provides a threadsafe rand.Source.
+type lockedRandomSource struct {
+ mu sync.Mutex
+ src mathrand.Source
+}
+
+func (r *lockedRandomSource) Int63() (n int64) {
+ r.mu.Lock()
+ n = r.src.Int63()
+ r.mu.Unlock()
+ return n
+}
+
+func (r *lockedRandomSource) Seed(seed int64) {
+ r.mu.Lock()
+ r.src.Seed(seed)
+ r.mu.Unlock()
+}
diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go
index 0869fb084..aca2f77f8 100644
--- a/pkg/tcpip/stack/registration.go
+++ b/pkg/tcpip/stack/registration.go
@@ -18,6 +18,7 @@ import (
"gvisor.dev/gvisor/pkg/sleep"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
+ "gvisor.dev/gvisor/pkg/tcpip/header"
"gvisor.dev/gvisor/pkg/waiter"
)
@@ -51,8 +52,11 @@ type TransportEndpointID struct {
type ControlType int
// The following are the allowed values for ControlType values.
+// TODO(http://gvisor.dev/issue/3210): Support time exceeded messages.
const (
- ControlPacketTooBig ControlType = iota
+ ControlNetworkUnreachable ControlType = iota
+ ControlNoRoute
+ ControlPacketTooBig
ControlPortUnreachable
ControlUnknown
)
@@ -60,13 +64,34 @@ const (
// TransportEndpoint is the interface that needs to be implemented by transport
// protocol (e.g., tcp, udp) endpoints that can handle packets.
type TransportEndpoint interface {
+ // UniqueID returns an unique ID for this transport endpoint.
+ UniqueID() uint64
+
// HandlePacket is called by the stack when new packets arrive to
- // this transport endpoint.
- HandlePacket(r *Route, id TransportEndpointID, vv buffer.VectorisedView)
+ // this transport endpoint. It sets pkt.TransportHeader.
+ //
+ // HandlePacket takes ownership of pkt.
+ HandlePacket(r *Route, id TransportEndpointID, pkt *PacketBuffer)
- // HandleControlPacket is called by the stack when new control (e.g.,
+ // HandleControlPacket is called by the stack when new control (e.g.
// ICMP) packets arrive to this transport endpoint.
- HandleControlPacket(id TransportEndpointID, typ ControlType, extra uint32, vv buffer.VectorisedView)
+ // HandleControlPacket takes ownership of pkt.
+ HandleControlPacket(id TransportEndpointID, typ ControlType, extra uint32, pkt *PacketBuffer)
+
+ // Abort initiates an expedited endpoint teardown. It puts the endpoint
+ // in a closed state and frees all resources associated with it. This
+ // cleanup may happen asynchronously. Wait can be used to block on this
+ // asynchronous cleanup.
+ Abort()
+
+ // Wait waits for any worker goroutines owned by the endpoint to stop.
+ //
+ // An endpoint can be requested to stop its worker goroutines by calling
+ // its Close method.
+ //
+ // Wait will not block if the endpoint hasn't started any goroutines
+ // yet, even if it might later.
+ Wait()
}
// RawTransportEndpoint is the interface that needs to be implemented by raw
@@ -77,7 +102,9 @@ type RawTransportEndpoint interface {
// HandlePacket is called by the stack when new packets arrive to
// this transport endpoint. The packet contains all data from the link
// layer up.
- HandlePacket(r *Route, netHeader buffer.View, packet buffer.VectorisedView)
+ //
+ // HandlePacket takes ownership of pkt.
+ HandlePacket(r *Route, pkt *PacketBuffer)
}
// PacketEndpoint is the interface that needs to be implemented by packet
@@ -93,7 +120,9 @@ type PacketEndpoint interface {
//
// 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)
+ //
+ // HandlePacket takes ownership of pkt.
+ HandlePacket(nicID tcpip.NICID, addr tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, pkt *PacketBuffer)
}
// TransportProtocol is the interface that needs to be implemented by transport
@@ -123,7 +152,9 @@ type TransportProtocol interface {
//
// The return value indicates whether the packet was well-formed (for
// stats purposes only).
- HandleUnknownDestinationPacket(r *Route, id TransportEndpointID, netHeader buffer.View, vv buffer.VectorisedView) bool
+ //
+ // HandleUnknownDestinationPacket takes ownership of pkt.
+ HandleUnknownDestinationPacket(r *Route, id TransportEndpointID, pkt *PacketBuffer) bool
// SetOption allows enabling/disabling protocol specific features.
// SetOption returns an error if the option is not supported or the
@@ -134,6 +165,18 @@ type TransportProtocol interface {
// Option returns an error if the option is not supported or the
// provided option value is invalid.
Option(option interface{}) *tcpip.Error
+
+ // Close requests that any worker goroutines owned by the protocol
+ // stop.
+ Close()
+
+ // Wait waits for any worker goroutines owned by the protocol to stop.
+ Wait()
+
+ // Parse sets pkt.TransportHeader and trims pkt.Data appropriately. It does
+ // neither and returns false if pkt.Data is too small, i.e. pkt.Data.Size() <
+ // MinimumPacketSize()
+ Parse(pkt *PacketBuffer) (ok bool)
}
// TransportDispatcher contains the methods used by the network stack to deliver
@@ -141,13 +184,21 @@ type TransportProtocol interface {
// the network layer.
type TransportDispatcher interface {
// DeliverTransportPacket delivers packets to the appropriate
- // transport protocol endpoint. It also returns the network layer
- // header for the enpoint to inspect or pass up the stack.
- DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, netHeader buffer.View, vv buffer.VectorisedView)
+ // transport protocol endpoint.
+ //
+ // pkt.NetworkHeader must be set before calling DeliverTransportPacket.
+ //
+ // DeliverTransportPacket takes ownership of pkt.
+ DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt *PacketBuffer)
// DeliverTransportControlPacket delivers control packets to the
// appropriate transport protocol endpoint.
- DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, vv buffer.VectorisedView)
+ //
+ // pkt.NetworkHeader must be set before calling
+ // DeliverTransportControlPacket.
+ //
+ // DeliverTransportControlPacket takes ownership of pkt.
+ DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, pkt *PacketBuffer)
}
// PacketLooping specifies where an outbound packet should be sent.
@@ -198,32 +249,34 @@ type NetworkEndpoint interface {
MaxHeaderLength() uint16
// WritePacket writes a packet to the given destination address and
- // protocol.
- WritePacket(r *Route, gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params NetworkHeaderParams, loop PacketLooping) *tcpip.Error
+ // protocol. It takes ownership of pkt. pkt.TransportHeader must have
+ // already been set.
+ WritePacket(r *Route, gso *GSO, params NetworkHeaderParams, pkt *PacketBuffer) *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)
+ // protocol. pkts must not be zero length. It takes ownership of pkts and
+ // underlying packets.
+ WritePackets(r *Route, gso *GSO, pkts PacketBufferList, params NetworkHeaderParams) (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
-
- // ID returns the network protocol endpoint ID.
- ID() *NetworkEndpointID
-
- // PrefixLen returns the network endpoint's subnet prefix length in bits.
- PrefixLen() int
+ // header to the given destination address. It takes ownership of pkt.
+ WriteHeaderIncludedPacket(r *Route, pkt *PacketBuffer) *tcpip.Error
// NICID returns the id of the NIC this endpoint belongs to.
NICID() tcpip.NICID
// HandlePacket is called by the link layer when new packets arrive to
- // this network endpoint.
- HandlePacket(r *Route, vv buffer.VectorisedView)
+ // this network endpoint. It sets pkt.NetworkHeader.
+ //
+ // HandlePacket takes ownership of pkt.
+ HandlePacket(r *Route, pkt *PacketBuffer)
// Close is called when the endpoint is reomved from a stack.
Close()
+
+ // NetworkProtocolNumber returns the tcpip.NetworkProtocolNumber for
+ // this endpoint.
+ NetworkProtocolNumber() tcpip.NetworkProtocolNumber
}
// NetworkProtocol is the interface that needs to be implemented by network
@@ -240,12 +293,12 @@ type NetworkProtocol interface {
// DefaultPrefixLen returns the protocol's default prefix length.
DefaultPrefixLen() int
- // ParsePorts returns the source and destination addresses stored in a
+ // ParseAddresses returns the source and destination addresses stored in a
// packet of this protocol.
ParseAddresses(v buffer.View) (src, dst tcpip.Address)
// NewEndpoint creates a new endpoint of this protocol.
- NewEndpoint(nicid tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache LinkAddressCache, dispatcher TransportDispatcher, sender LinkEndpoint) (NetworkEndpoint, *tcpip.Error)
+ NewEndpoint(nicID tcpip.NICID, linkAddrCache LinkAddressCache, dispatcher TransportDispatcher, sender LinkEndpoint, st *Stack) NetworkEndpoint
// SetOption allows enabling/disabling protocol specific features.
// SetOption returns an error if the option is not supported or the
@@ -256,16 +309,45 @@ type NetworkProtocol interface {
// Option returns an error if the option is not supported or the
// provided option value is invalid.
Option(option interface{}) *tcpip.Error
+
+ // Close requests that any worker goroutines owned by the protocol
+ // stop.
+ Close()
+
+ // Wait waits for any worker goroutines owned by the protocol to stop.
+ Wait()
+
+ // Parse sets pkt.NetworkHeader and trims pkt.Data appropriately. It
+ // returns:
+ // - The encapsulated protocol, if present.
+ // - Whether there is an encapsulated transport protocol payload (e.g. ARP
+ // does not encapsulate anything).
+ // - Whether pkt.Data was large enough to parse and set pkt.NetworkHeader.
+ Parse(pkt *PacketBuffer) (proto tcpip.TransportProtocolNumber, hasTransportHdr bool, ok bool)
}
// NetworkDispatcher contains the methods used by the network stack to deliver
-// packets to the appropriate network endpoint after it has been handled by
-// the data link layer.
+// inbound/outbound packets to the appropriate network/packet(if any) endpoints.
type NetworkDispatcher interface {
// DeliverNetworkPacket finds the appropriate network protocol endpoint
- // and hands the packet over for further processing. 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)
+ // and hands the packet over for further processing.
+ //
+ // pkt.LinkHeader may or may not be set before calling
+ // DeliverNetworkPacket. Some packets do not have link headers (e.g.
+ // packets sent via loopback), and won't have the field set.
+ //
+ // DeliverNetworkPacket takes ownership of pkt.
+ DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer)
+
+ // DeliverOutboundPacket is called by link layer when a packet is being
+ // sent out.
+ //
+ // pkt.LinkHeader may or may not be set before calling
+ // DeliverOutboundPacket. Some packets do not have link headers (e.g.
+ // packets sent via loopback), and won't have the field set.
+ //
+ // DeliverOutboundPacket takes ownership of pkt.
+ DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer)
}
// LinkEndpointCapabilities is the type associated with the capabilities
@@ -296,7 +378,9 @@ const (
// LinkEndpoint is the interface implemented by data link layer protocols (e.g.,
// ethernet, loopback, raw) and used by network layer protocols to send packets
-// out through the implementer's data link endpoint.
+// out through the implementer's data link endpoint. When a link header exists,
+// it sets each PacketBuffer's LinkHeader field before passing it up the
+// stack.
type LinkEndpoint interface {
// MTU is the maximum transmission unit for this endpoint. This is
// usually dictated by the backing physical network; when such a
@@ -318,28 +402,33 @@ type LinkEndpoint interface {
// link endpoint.
LinkAddress() tcpip.LinkAddress
- // WritePacket writes a packet with the given protocol through the given
- // route.
+ // WritePacket writes a packet with the given protocol through the
+ // given route. It takes ownership of pkt. pkt.NetworkHeader and
+ // pkt.TransportHeader must have already been set.
//
// To participate in transparent bridging, a LinkEndpoint implementation
// should call eth.Encode with header.EthernetFields.SrcAddr set to
// r.LocalLinkAddress if it is provided.
- WritePacket(r *Route, gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error
+ WritePacket(r *Route, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) *tcpip.Error
// WritePackets writes packets with the given protocol through the
- // given route.
+ // given route. pkts must not be zero length. It takes ownership of pkts and
+ // underlying packets.
//
// 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)
+ WritePackets(r *Route, gso *GSO, pkts PacketBufferList, 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
+ // should already have an ethernet header. It takes ownership of vv.
+ WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error
// Attach attaches the data link layer endpoint to the network-layer
// dispatcher of the stack.
+ //
+ // Attach will be called with a nil dispatcher if the receiver's associated
+ // NIC is being removed.
Attach(dispatcher NetworkDispatcher)
// IsAttached returns whether a NetworkDispatcher is attached to the
@@ -354,6 +443,15 @@ type LinkEndpoint interface {
// Wait will not block if the endpoint hasn't started any goroutines
// yet, even if it might later.
Wait()
+
+ // ARPHardwareType returns the ARPHRD_TYPE of the link endpoint.
+ //
+ // See:
+ // https://github.com/torvalds/linux/blob/aa0c9086b40c17a7ad94425b3b70dd1fdd7497bf/include/uapi/linux/if_arp.h#L30
+ ARPHardwareType() header.ARPHardwareType
+
+ // AddHeader adds a link layer header to pkt if required.
+ AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer)
}
// InjectableLinkEndpoint is a LinkEndpoint where inbound packets are
@@ -362,7 +460,7 @@ type InjectableLinkEndpoint interface {
LinkEndpoint
// InjectInbound injects an inbound packet.
- InjectInbound(protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView)
+ InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer)
// InjectOutbound writes a fully formed outbound packet directly to the
// link.
@@ -374,12 +472,13 @@ type InjectableLinkEndpoint interface {
// A LinkAddressResolver is an extension to a NetworkProtocol that
// can resolve link addresses.
type LinkAddressResolver interface {
- // LinkAddressRequest sends a request for the LinkAddress of addr.
- // The request is sent on linkEP with localAddr as the source.
+ // LinkAddressRequest sends a request for the LinkAddress of addr. Broadcasts
+ // the request on the local network if remoteLinkAddr is the zero value. The
+ // request is sent on linkEP with localAddr as the source.
//
// A valid response will cause the discovery protocol's network
// endpoint to call AddLinkAddress.
- LinkAddressRequest(addr, localAddr tcpip.Address, linkEP LinkEndpoint) *tcpip.Error
+ LinkAddressRequest(addr, localAddr tcpip.Address, remoteLinkAddr tcpip.LinkAddress, linkEP LinkEndpoint) *tcpip.Error
// ResolveStaticAddress attempts to resolve address without sending
// requests. It either resolves the name immediately or returns the
@@ -397,10 +496,10 @@ type LinkAddressResolver interface {
type LinkAddressCache interface {
// CheckLocalAddress determines if the given local address exists, and if it
// does not exist.
- CheckLocalAddress(nicid tcpip.NICID, protocol tcpip.NetworkProtocolNumber, addr tcpip.Address) tcpip.NICID
+ CheckLocalAddress(nicID tcpip.NICID, protocol tcpip.NetworkProtocolNumber, addr tcpip.Address) tcpip.NICID
// AddLinkAddress adds a link address to the cache.
- AddLinkAddress(nicid tcpip.NICID, addr tcpip.Address, linkAddr tcpip.LinkAddress)
+ AddLinkAddress(nicID tcpip.NICID, addr tcpip.Address, linkAddr tcpip.LinkAddress)
// GetLinkAddress looks up the cache to translate address to link address (e.g. IP -> MAC).
// If the LinkEndpoint requests address resolution and there is a LinkAddressResolver
@@ -411,10 +510,10 @@ type LinkAddressCache interface {
// If address resolution is required, ErrNoLinkAddress and a notification channel is
// returned for the top level caller to block. Channel is closed once address resolution
// is complete (success or not).
- GetLinkAddress(nicid tcpip.NICID, addr, localAddr tcpip.Address, protocol tcpip.NetworkProtocolNumber, w *sleep.Waker) (tcpip.LinkAddress, <-chan struct{}, *tcpip.Error)
+ GetLinkAddress(nicID tcpip.NICID, addr, localAddr tcpip.Address, protocol tcpip.NetworkProtocolNumber, w *sleep.Waker) (tcpip.LinkAddress, <-chan struct{}, *tcpip.Error)
// RemoveWaker removes a waker that has been added in GetLinkAddress().
- RemoveWaker(nicid tcpip.NICID, addr tcpip.Address, waker *sleep.Waker)
+ RemoveWaker(nicID tcpip.NICID, addr tcpip.Address, waker *sleep.Waker)
}
// RawFactory produces endpoints for writing various types of raw packets.
diff --git a/pkg/tcpip/stack/route.go b/pkg/tcpip/stack/route.go
index 1a0a51b57..e267bebb0 100644
--- a/pkg/tcpip/stack/route.go
+++ b/pkg/tcpip/stack/route.go
@@ -17,7 +17,6 @@ package stack
import (
"gvisor.dev/gvisor/pkg/sleep"
"gvisor.dev/gvisor/pkg/tcpip"
- "gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/tcpip/header"
)
@@ -49,6 +48,10 @@ type Route struct {
// Loop controls where WritePacket should send packets.
Loop PacketLooping
+
+ // directedBroadcast indicates whether this route is sending a directed
+ // broadcast packet.
+ directedBroadcast bool
}
// makeRoute initializes a new route. It takes ownership of the provided
@@ -107,6 +110,12 @@ func (r *Route) GSOMaxSize() uint32 {
return 0
}
+// ResolveWith immediately resolves a route with the specified remote link
+// address.
+func (r *Route) ResolveWith(addr tcpip.LinkAddress) {
+ r.RemoteLinkAddress = addr
+}
+
// Resolve attempts to resolve the link address if necessary. Returns ErrWouldBlock in
// case address resolution requires blocking, e.g. wait for ARP reply. Waker is
// notified when address resolution is complete (success or not).
@@ -114,6 +123,8 @@ func (r *Route) GSOMaxSize() uint32 {
// If address resolution is required, ErrNoLinkAddress and a notification channel is
// returned for the top level caller to block. Channel is closed once address resolution
// is complete (success or not).
+//
+// The NIC r uses must not be locked.
func (r *Route) Resolve(waker *sleep.Waker) (<-chan struct{}, *tcpip.Error) {
if !r.IsResolutionRequired() {
// Nothing to do if there is no cache (which does the resolution on cache miss) or
@@ -149,77 +160,72 @@ func (r *Route) RemoveWaker(waker *sleep.Waker) {
// IsResolutionRequired returns true if Resolve() must be called to resolve
// the link address before the this route can be written to.
+//
+// The NIC r uses must not be locked.
func (r *Route) IsResolutionRequired() bool {
return r.ref.isValidForOutgoing() && r.ref.linkCache != nil && r.RemoteLinkAddress == ""
}
// WritePacket writes the packet through the given route.
-func (r *Route) WritePacket(gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params NetworkHeaderParams) *tcpip.Error {
+func (r *Route) WritePacket(gso *GSO, params NetworkHeaderParams, pkt *PacketBuffer) *tcpip.Error {
if !r.ref.isValidForOutgoing() {
return tcpip.ErrInvalidEndpointState
}
- err := r.ref.ep.WritePacket(r, gso, hdr, payload, params, r.Loop)
+ // WritePacket takes ownership of pkt, calculate numBytes first.
+ numBytes := pkt.Size()
+
+ err := r.ref.ep.WritePacket(r, gso, params, pkt)
if err != nil {
r.Stats().IP.OutgoingPacketErrors.Increment()
} else {
r.ref.nic.stats.Tx.Packets.Increment()
- r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(hdr.UsedLength() + payload.Size()))
+ r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(numBytes))
}
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) {
+// WritePackets writes a list of n packets through the given route and returns
+// the number of packets written.
+func (r *Route) WritePackets(gso *GSO, pkts PacketBufferList, 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)
+ // WritePackets takes ownership of pkt, calculate length first.
+ numPkts := pkts.Len()
+
+ n, err := r.ref.ep.WritePackets(r, gso, pkts, params)
if err != nil {
- r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(len(hdrs) - n))
+ r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(numPkts - 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
+
+ writtenBytes := 0
+ for i, pb := 0, pkts.Front(); i < n && pb != nil; i, pb = i+1, pb.Next() {
+ writtenBytes += pb.Size()
}
- r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(payloadSize))
+
+ r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(writtenBytes))
return n, err
}
// WriteHeaderIncludedPacket writes a packet already containing a network
// header through the given route.
-func (r *Route) WriteHeaderIncludedPacket(payload buffer.VectorisedView) *tcpip.Error {
+func (r *Route) WriteHeaderIncludedPacket(pkt *PacketBuffer) *tcpip.Error {
if !r.ref.isValidForOutgoing() {
return tcpip.ErrInvalidEndpointState
}
- if err := r.ref.ep.WriteHeaderIncludedPacket(r, payload, r.Loop); err != nil {
+ // WriteHeaderIncludedPacket takes ownership of pkt, calculate numBytes first.
+ numBytes := pkt.Data.Size()
+
+ if err := r.ref.ep.WriteHeaderIncludedPacket(r, pkt); err != nil {
r.Stats().IP.OutgoingPacketErrors.Increment()
return err
}
r.ref.nic.stats.Tx.Packets.Increment()
- r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(payload.Size()))
+ r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(numBytes))
return nil
}
@@ -233,6 +239,12 @@ func (r *Route) MTU() uint32 {
return r.ref.ep.MTU()
}
+// NetworkProtocolNumber returns the NetworkProtocolNumber of the underlying
+// network endpoint.
+func (r *Route) NetworkProtocolNumber() tcpip.NetworkProtocolNumber {
+ return r.ref.ep.NetworkProtocolNumber()
+}
+
// Release frees all resources associated with the route.
func (r *Route) Release() {
if r.ref != nil {
@@ -244,7 +256,9 @@ func (r *Route) Release() {
// Clone Clone a route such that the original one can be released and the new
// one will remain valid.
func (r *Route) Clone() Route {
- r.ref.incRef()
+ if r.ref != nil {
+ r.ref.incRef()
+ }
return *r
}
@@ -269,3 +283,36 @@ func (r *Route) MakeLoopedRoute() Route {
func (r *Route) Stack() *Stack {
return r.ref.stack()
}
+
+// IsOutboundBroadcast returns true if the route is for an outbound broadcast
+// packet.
+func (r *Route) IsOutboundBroadcast() bool {
+ // Only IPv4 has a notion of broadcast.
+ return r.directedBroadcast || r.RemoteAddress == header.IPv4Broadcast
+}
+
+// IsInboundBroadcast returns true if the route is for an inbound broadcast
+// packet.
+func (r *Route) IsInboundBroadcast() bool {
+ // Only IPv4 has a notion of broadcast.
+ if r.LocalAddress == header.IPv4Broadcast {
+ return true
+ }
+
+ addr := r.ref.addrWithPrefix()
+ subnet := addr.Subnet()
+ return subnet.IsBroadcast(r.LocalAddress)
+}
+
+// ReverseRoute returns new route with given source and destination address.
+func (r *Route) ReverseRoute(src tcpip.Address, dst tcpip.Address) Route {
+ return Route{
+ NetProto: r.NetProto,
+ LocalAddress: dst,
+ LocalLinkAddress: r.RemoteLinkAddress,
+ RemoteAddress: src,
+ RemoteLinkAddress: r.LocalLinkAddress,
+ ref: r.ref,
+ Loop: r.Loop,
+ }
+}
diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go
index 71e0618f4..814b3e94a 100644
--- a/pkg/tcpip/stack/stack.go
+++ b/pkg/tcpip/stack/stack.go
@@ -20,19 +20,20 @@
package stack
import (
+ "bytes"
"encoding/binary"
"math"
- "sync"
+ mathrand "math/rand"
"sync/atomic"
"time"
"golang.org/x/time/rate"
"gvisor.dev/gvisor/pkg/rand"
"gvisor.dev/gvisor/pkg/sleep"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/tcpip/header"
- "gvisor.dev/gvisor/pkg/tcpip/iptables"
"gvisor.dev/gvisor/pkg/tcpip/ports"
"gvisor.dev/gvisor/pkg/tcpip/seqnum"
"gvisor.dev/gvisor/pkg/waiter"
@@ -54,7 +55,7 @@ const (
// fakeNetNumber is used as a protocol number in tests.
//
// This constant should match fakeNetNumber in stack_test.go.
- fakeNetNumber tcpip.NetworkProtocolNumber = math.MaxUint32
+ fakeNetNumber tcpip.NetworkProtocolNumber = math.MaxUint32
)
type forwardingFlag uint32
@@ -77,8 +78,7 @@ func getForwardingFlag(protocol tcpip.NetworkProtocolNumber) forwardingFlag {
case header.IPv6ProtocolNumber:
flag = forwardingIPv6
case fakeNetNumber:
- // This network protocol number is used in stack_test to test
- // packet forwarding.
+ // This network protocol number is used to test packet forwarding.
flag = forwardingFake
default:
// We only support forwarding for IPv4 and IPv6.
@@ -88,7 +88,7 @@ func getForwardingFlag(protocol tcpip.NetworkProtocolNumber) forwardingFlag {
type transportProtocolState struct {
proto TransportProtocol
- defaultHandler func(r *Route, id TransportEndpointID, netHeader buffer.View, vv buffer.VectorisedView) bool
+ defaultHandler func(r *Route, id TransportEndpointID, pkt *PacketBuffer) bool
}
// TCPProbeFunc is the expected function type for a TCP probe function to be
@@ -109,6 +109,16 @@ type TCPCubicState struct {
WEst float64
}
+// TCPRACKState is used to hold a copy of the internal RACK state when the
+// TCPProbeFunc is invoked.
+type TCPRACKState struct {
+ XmitTime time.Time
+ EndSequence seqnum.Value
+ FACK seqnum.Value
+ RTT time.Duration
+ Reord bool
+}
+
// TCPEndpointID is the unique 4 tuple that identifies a given endpoint.
type TCPEndpointID struct {
// LocalPort is the local port associated with the endpoint.
@@ -248,6 +258,9 @@ type TCPSenderState struct {
// Cubic holds the state related to CUBIC congestion control.
Cubic TCPCubicState
+
+ // RACKState holds the state related to RACK loss detection algorithm.
+ RACKState TCPRACKState
}
// TCPSACKInfo holds TCP SACK related information for a given TCP endpoint.
@@ -271,11 +284,11 @@ type RcvBufAutoTuneParams struct {
// was started.
MeasureTime time.Time
- // CopiedBytes is the number of bytes copied to user space since
+ // CopiedBytes is the number of bytes copied to userspace since
// this measure began.
CopiedBytes int
- // PrevCopiedBytes is the number of bytes copied to user space in
+ // PrevCopiedBytes is the number of bytes copied to userspace in
// the previous RTT period.
PrevCopiedBytes int
@@ -382,6 +395,45 @@ type ResumableEndpoint interface {
Resume(*Stack)
}
+// uniqueIDGenerator is a default unique ID generator.
+type uniqueIDGenerator uint64
+
+func (u *uniqueIDGenerator) UniqueID() uint64 {
+ return atomic.AddUint64((*uint64)(u), 1)
+}
+
+// NICNameFromID is a function that returns a stable name for the specified NIC,
+// even if different NIC IDs are used to refer to the same NIC in different
+// program runs. It is used when generating opaque interface identifiers (IIDs).
+// If the NIC was created with a name, it will be passed to NICNameFromID.
+//
+// NICNameFromID SHOULD return unique NIC names so unique opaque IIDs are
+// generated for the same prefix on differnt NICs.
+type NICNameFromID func(tcpip.NICID, string) string
+
+// OpaqueInterfaceIdentifierOptions holds the options related to the generation
+// of opaque interface indentifiers (IIDs) as defined by RFC 7217.
+type OpaqueInterfaceIdentifierOptions struct {
+ // NICNameFromID is a function that returns a stable name for a specified NIC,
+ // even if the NIC ID changes over time.
+ //
+ // Must be specified to generate the opaque IID.
+ NICNameFromID NICNameFromID
+
+ // SecretKey is a pseudo-random number used as the secret key when generating
+ // opaque IIDs as defined by RFC 7217. The key SHOULD be at least
+ // header.OpaqueIIDSecretKeyMinBytes bytes and MUST follow minimum randomness
+ // requirements for security as outlined by RFC 4086. SecretKey MUST NOT
+ // change between program runs, unless explicitly changed.
+ //
+ // OpaqueInterfaceIdentifierOptions takes ownership of SecretKey. SecretKey
+ // MUST NOT be modified after Stack is created.
+ //
+ // May be nil, but a nil value is highly discouraged to maintain
+ // some level of randomness between nodes.
+ SecretKey []byte
+}
+
// Stack is a networking stack, with all supported protocols, NICs, and route
// table.
type Stack struct {
@@ -399,12 +451,17 @@ type Stack struct {
linkAddrCache *linkAddrCache
- mu sync.RWMutex
- nics map[tcpip.NICID]*NIC
+ mu sync.RWMutex
+ nics map[tcpip.NICID]*NIC
// forwarding contains the enable bits for packet forwarding for different
// network protocols.
- forwarding uint32
+ forwarding struct {
+ sync.RWMutex
+ flag forwardingFlag
+ }
+
+ cleanupEndpoints map[TransportEndpoint]struct{}
// route is the route table passed in by the user via SetRouteTable(),
// it is used by FindRoute() to build a route for a specific
@@ -424,7 +481,8 @@ type Stack struct {
handleLocal bool
// tables are the iptables packet filtering and manipulation rules.
- tables iptables.IPTables
+ // TODO(gvisor.dev/issue/170): S/R this field.
+ tables *IPTables
// resumableEndpoints is a list of endpoints that need to be resumed if the
// stack is being restored.
@@ -434,23 +492,62 @@ type Stack struct {
// by the stack.
icmpRateLimiter *ICMPRateLimiter
- // portSeed is a one-time random value initialized at stack startup
+ // seed is a one-time random value initialized at stack startup
// and is used to seed the TCP port picking on active connections
//
// TODO(gvisor.dev/issue/940): S/R this field.
- portSeed uint32
+ seed uint32
// ndpConfigs is the default NDP configurations used by interfaces.
ndpConfigs NDPConfigurations
+ // nudConfigs is the default NUD configurations used by interfaces.
+ nudConfigs NUDConfigurations
+
// autoGenIPv6LinkLocal determines whether or not the stack will attempt
- // to auto-generate an IPv6 link-local address for newly enabled NICs.
- // See the AutoGenIPv6LinkLocal field of Options for more details.
+ // to auto-generate an IPv6 link-local address for newly enabled non-loopback
+ // 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
+
+ // nudDisp is the NUD event dispatcher that is used to send the netstack
+ // integrator NUD related events.
+ nudDisp NUDDispatcher
+
+ // uniqueIDGenerator is a generator of unique identifiers.
+ uniqueIDGenerator UniqueID
+
+ // opaqueIIDOpts hold the options for generating opaque interface identifiers
+ // (IIDs) as outlined by RFC 7217.
+ opaqueIIDOpts OpaqueInterfaceIdentifierOptions
+
+ // tempIIDSeed is used to seed the initial temporary interface identifier
+ // history value used to generate IIDs for temporary SLAAC addresses.
+ tempIIDSeed []byte
+
+ // forwarder holds the packets that wait for their link-address resolutions
+ // to complete, and forwards them when each resolution is done.
+ forwarder *forwardQueue
+
+ // randomGenerator is an injectable pseudo random generator that can be
+ // used when a random number is required.
+ randomGenerator *mathrand.Rand
+
+ // sendBufferSize holds the min/default/max send buffer sizes for
+ // endpoints other than TCP.
+ sendBufferSize SendBufferSizeOption
+
+ // receiveBufferSize holds the min/default/max receive buffer sizes for
+ // endpoints other than TCP.
+ receiveBufferSize ReceiveBufferSizeOption
+}
+
+// UniqueID is an abstract generator of unique identifiers.
+type UniqueID interface {
+ UniqueID() uint64
}
// Options contains optional Stack configuration.
@@ -474,6 +571,9 @@ type Options struct {
// stack (false).
HandleLocal bool
+ // UniqueID is an optional generator of unique identifiers.
+ UniqueID UniqueID
+
// NDPConfigs is the default NDP configurations used by interfaces.
//
// By default, NDPConfigs will have a zero value for its
@@ -481,13 +581,18 @@ type Options struct {
// 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.
+ // NUDConfigs is the default NUD configurations used by interfaces.
+ NUDConfigs NUDConfigurations
+
+ // AutoGenIPv6LinkLocal determines whether or not the stack will attempt to
+ // auto-generate an IPv6 link-local address for newly enabled non-loopback
+ // NICs.
+ //
// 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.
+ // 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.
@@ -497,9 +602,39 @@ type Options struct {
// receive NDP related events.
NDPDisp NDPDispatcher
+ // NUDDisp is the NUD event dispatcher that an integrator can provide to
+ // receive NUD related events.
+ NUDDisp NUDDispatcher
+
// RawFactory produces raw endpoints. Raw endpoints are enabled only if
// this is non-nil.
RawFactory RawFactory
+
+ // OpaqueIIDOpts hold the options for generating opaque interface
+ // identifiers (IIDs) as outlined by RFC 7217.
+ OpaqueIIDOpts OpaqueInterfaceIdentifierOptions
+
+ // RandSource is an optional source to use to generate random
+ // numbers. If omitted it defaults to a Source seeded by the data
+ // returned by rand.Read().
+ //
+ // RandSource must be thread-safe.
+ RandSource mathrand.Source
+
+ // TempIIDSeed is used to seed the initial temporary interface identifier
+ // history value used to generate IIDs for temporary SLAAC addresses.
+ //
+ // Temporary SLAAC adresses are short-lived addresses which are unpredictable
+ // and random from the perspective of other nodes on the network. It is
+ // recommended that the seed be a random byte buffer of at least
+ // header.IIDSize bytes to make sure that temporary SLAAC addresses are
+ // sufficiently random. It should follow minimum randomness requirements for
+ // security as outlined by RFC 4086.
+ //
+ // Note: using a nil value, the same seed across netstack program runs, or a
+ // seed that is too small would reduce randomness and increase predictability,
+ // defeating the purpose of temporary SLAAC addresses.
+ TempIIDSeed []byte
}
// TransportEndpointInfo holds useful information about a transport endpoint
@@ -526,6 +661,51 @@ type TransportEndpointInfo struct {
RegisterNICID tcpip.NICID
}
+// AddrNetProtoLocked unwraps the specified address if it is a V4-mapped V6
+// address and returns the network protocol number to be used to communicate
+// with the specified address. It returns an error if the passed address is
+// incompatible with the receiver.
+//
+// Preconditon: the parent endpoint mu must be held while calling this method.
+func (e *TransportEndpointInfo) AddrNetProtoLocked(addr tcpip.FullAddress, v6only bool) (tcpip.FullAddress, tcpip.NetworkProtocolNumber, *tcpip.Error) {
+ netProto := e.NetProto
+ switch len(addr.Addr) {
+ case header.IPv4AddressSize:
+ netProto = header.IPv4ProtocolNumber
+ case header.IPv6AddressSize:
+ if header.IsV4MappedAddress(addr.Addr) {
+ netProto = header.IPv4ProtocolNumber
+ addr.Addr = addr.Addr[header.IPv6AddressSize-header.IPv4AddressSize:]
+ if addr.Addr == header.IPv4Any {
+ addr.Addr = ""
+ }
+ }
+ }
+
+ switch len(e.ID.LocalAddress) {
+ case header.IPv4AddressSize:
+ if len(addr.Addr) == header.IPv6AddressSize {
+ return tcpip.FullAddress{}, 0, tcpip.ErrInvalidEndpointState
+ }
+ case header.IPv6AddressSize:
+ if len(addr.Addr) == header.IPv4AddressSize {
+ return tcpip.FullAddress{}, 0, tcpip.ErrNetworkUnreachable
+ }
+ }
+
+ switch {
+ case netProto == e.NetProto:
+ case netProto == header.IPv4ProtocolNumber && e.NetProto == header.IPv6ProtocolNumber:
+ if v6only {
+ return tcpip.FullAddress{}, 0, tcpip.ErrNoRoute
+ }
+ default:
+ return tcpip.FullAddress{}, 0, tcpip.ErrInvalidEndpointState
+ }
+
+ return addr, netProto, nil
+}
+
// IsEndpointInfo is an empty method to implement the tcpip.EndpointInfo
// marker interface.
func (*TransportEndpointInfo) IsEndpointInfo() {}
@@ -546,24 +726,56 @@ func New(opts Options) *Stack {
clock = &tcpip.StdClock{}
}
+ if opts.UniqueID == nil {
+ opts.UniqueID = new(uniqueIDGenerator)
+ }
+
+ randSrc := opts.RandSource
+ if randSrc == nil {
+ // Source provided by mathrand.NewSource is not thread-safe so
+ // we wrap it in a simple thread-safe version.
+ randSrc = &lockedRandomSource{src: mathrand.NewSource(generateRandInt64())}
+ }
+
// Make sure opts.NDPConfigs contains valid values only.
opts.NDPConfigs.validate()
+ opts.NUDConfigs.resetInvalidFields()
+
s := &Stack{
transportProtocols: make(map[tcpip.TransportProtocolNumber]*transportProtocolState),
networkProtocols: make(map[tcpip.NetworkProtocolNumber]NetworkProtocol),
linkAddrResolvers: make(map[tcpip.NetworkProtocolNumber]LinkAddressResolver),
nics: make(map[tcpip.NICID]*NIC),
+ cleanupEndpoints: make(map[TransportEndpoint]struct{}),
linkAddrCache: newLinkAddrCache(ageLimit, resolutionTimeout, resolutionAttempts),
PortManager: ports.NewPortManager(),
clock: clock,
stats: opts.Stats.FillIn(),
handleLocal: opts.HandleLocal,
+ tables: DefaultTables(),
icmpRateLimiter: NewICMPRateLimiter(),
- portSeed: generateRandUint32(),
+ seed: generateRandUint32(),
ndpConfigs: opts.NDPConfigs,
+ nudConfigs: opts.NUDConfigs,
autoGenIPv6LinkLocal: opts.AutoGenIPv6LinkLocal,
+ uniqueIDGenerator: opts.UniqueID,
ndpDisp: opts.NDPDisp,
+ nudDisp: opts.NUDDisp,
+ opaqueIIDOpts: opts.OpaqueIIDOpts,
+ tempIIDSeed: opts.TempIIDSeed,
+ forwarder: newForwardQueue(),
+ randomGenerator: mathrand.New(randSrc),
+ sendBufferSize: SendBufferSizeOption{
+ Min: MinBufferSize,
+ Default: DefaultBufferSize,
+ Max: DefaultMaxBufferSize,
+ },
+ receiveBufferSize: ReceiveBufferSizeOption{
+ Min: MinBufferSize,
+ Default: DefaultBufferSize,
+ Max: DefaultMaxBufferSize,
+ },
}
// Add specified network protocols.
@@ -590,6 +802,16 @@ func New(opts Options) *Stack {
return s
}
+// newJob returns a tcpip.Job using the Stack clock.
+func (s *Stack) newJob(l sync.Locker, f func()) *tcpip.Job {
+ return tcpip.NewJob(s.clock, l, f)
+}
+
+// UniqueID returns a unique identifier.
+func (s *Stack) UniqueID() uint64 {
+ return s.uniqueIDGenerator.UniqueID()
+}
+
// SetNetworkProtocolOption allows configuring individual protocol level
// options. This method returns an error if the protocol is not supported or
// option is not supported by the protocol implementation or the provided value
@@ -651,16 +873,17 @@ func (s *Stack) TransportProtocolOption(transport tcpip.TransportProtocolNumber,
//
// It must be called only during initialization of the stack. Changing it as the
// stack is operating is not supported.
-func (s *Stack) SetTransportProtocolHandler(p tcpip.TransportProtocolNumber, h func(*Route, TransportEndpointID, buffer.View, buffer.VectorisedView) bool) {
+func (s *Stack) SetTransportProtocolHandler(p tcpip.TransportProtocolNumber, h func(*Route, TransportEndpointID, *PacketBuffer) bool) {
state := s.transportProtocols[p]
if state != nil {
state.defaultHandler = h
}
}
-// NowNanoseconds implements tcpip.Clock.NowNanoseconds.
-func (s *Stack) NowNanoseconds() int64 {
- return s.clock.NowNanoseconds()
+// Clock returns the Stack's clock for retrieving the current time and
+// scheduling work.
+func (s *Stack) Clock() tcpip.Clock {
+ return s.clock
}
// Stats returns a mutable copy of the current stats.
@@ -673,30 +896,55 @@ func (s *Stack) Stats() tcpip.Stats {
// SetForwarding enables or disables packet forwarding between NICs.
func (s *Stack) SetForwarding(protocol tcpip.NetworkProtocolNumber, enable bool) {
+ s.forwarding.Lock()
+ defer s.forwarding.Unlock()
+
+ // If this stack does not support the protocol, do nothing.
+ if _, ok := s.networkProtocols[protocol]; !ok {
+ return
+ }
+
flag := getForwardingFlag(protocol)
- for {
- forwarding := forwardingFlag(atomic.LoadUint32(&s.forwarding))
- var newValue forwardingFlag
+
+ // If the forwarding value for this protocol hasn't changed then do
+ // nothing.
+ if s.forwarding.flag&getForwardingFlag(protocol) != 0 == enable {
+ return
+ }
+
+ var newValue forwardingFlag
+ if enable {
+ newValue = s.forwarding.flag | flag
+ } else {
+ newValue = s.forwarding.flag & ^flag
+ }
+ s.forwarding.flag = newValue
+
+ // Enable or disable NDP for IPv6.
+ if protocol == header.IPv6ProtocolNumber {
if enable {
- newValue = forwarding | flag
+ for _, nic := range s.nics {
+ nic.becomeIPv6Router()
+ }
} else {
- newValue = forwarding & ^flag
- }
- if atomic.CompareAndSwapUint32(&s.forwarding, uint32(forwarding), uint32(newValue)) {
- break
+ for _, nic := range s.nics {
+ nic.becomeIPv6Host()
+ }
}
}
}
// Forwarding returns if packet forwarding between NICs is enabled.
func (s *Stack) Forwarding(protocol tcpip.NetworkProtocolNumber) bool {
- flag := getForwardingFlag(protocol)
- forwarding := forwardingFlag(atomic.LoadUint32(&s.forwarding))
- return forwarding & flag != 0
+ s.forwarding.RLock()
+ defer s.forwarding.RUnlock()
+ return s.forwarding.flag&getForwardingFlag(protocol) != 0
}
// SetRouteTable assigns the route table to be used by this stack. It
// specifies which NIC to use for given destination address ranges.
+//
+// This method takes ownership of the table.
func (s *Stack) SetRouteTable(table []tcpip.Route) {
s.mu.Lock()
defer s.mu.Unlock()
@@ -711,6 +959,13 @@ func (s *Stack) GetRouteTable() []tcpip.Route {
return append([]tcpip.Route(nil), s.routeTable...)
}
+// AddRoute appends a route to the route table.
+func (s *Stack) AddRoute(route tcpip.Route) {
+ s.mu.Lock()
+ defer s.mu.Unlock()
+ s.routeTable = append(s.routeTable, route)
+}
+
// NewEndpoint creates a new transport layer endpoint of the given protocol.
func (s *Stack) NewEndpoint(transport tcpip.TransportProtocolNumber, network tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) {
t, ok := s.transportProtocols[transport]
@@ -751,9 +1006,32 @@ func (s *Stack) NewPacketEndpoint(cooked bool, netProto tcpip.NetworkProtocolNum
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 {
+// NICContext is an opaque pointer used to store client-supplied NIC metadata.
+type NICContext interface{}
+
+// NICOptions specifies the configuration of a NIC as it is being created.
+// The zero value creates an enabled, unnamed NIC.
+type NICOptions struct {
+ // Name specifies the name of the NIC.
+ Name string
+
+ // Disabled specifies whether to avoid calling Attach on the passed
+ // LinkEndpoint.
+ Disabled bool
+
+ // Context specifies user-defined data that will be returned in stack.NICInfo
+ // for the NIC. Clients of this library can use it to add metadata that
+ // should be tracked alongside a NIC, to avoid having to keep a
+ // map[tcpip.NICID]metadata mirroring stack.Stack's nic map.
+ Context NICContext
+}
+
+// CreateNICWithOptions creates a NIC with the provided id, LinkEndpoint, and
+// NICOptions. See the documentation on type NICOptions for details on how
+// NICs can be configured.
+//
+// LinkEndpoint.Attach will be called to bind ep with a NetworkDispatcher.
+func (s *Stack) CreateNICWithOptions(id tcpip.NICID, ep LinkEndpoint, opts NICOptions) *tcpip.Error {
s.mu.Lock()
defer s.mu.Unlock()
@@ -762,44 +1040,40 @@ func (s *Stack) createNIC(id tcpip.NICID, name string, ep LinkEndpoint, enabled,
return tcpip.ErrDuplicateNICID
}
- n := newNIC(s, id, name, ep, loopback)
+ // Make sure name is unique, unless unnamed.
+ if opts.Name != "" {
+ for _, n := range s.nics {
+ if n.Name() == opts.Name {
+ return tcpip.ErrDuplicateNICID
+ }
+ }
+ }
+ n := newNIC(s, id, opts.Name, ep, opts.Context)
s.nics[id] = n
- if enabled {
+ if !opts.Disabled {
return n.enable()
}
return nil
}
-// CreateNIC creates a NIC with the provided id and link-layer endpoint.
+// CreateNIC creates a NIC with the provided id and LinkEndpoint and calls
+// LinkEndpoint.Attach to bind ep with a NetworkDispatcher.
func (s *Stack) CreateNIC(id tcpip.NICID, ep LinkEndpoint) *tcpip.Error {
- return s.createNIC(id, "", ep, true, false)
+ return s.CreateNICWithOptions(id, ep, NICOptions{})
}
-// CreateNamedNIC creates a NIC with the provided id and link-layer endpoint,
-// and a human-readable name.
-func (s *Stack) CreateNamedNIC(id tcpip.NICID, name string, ep LinkEndpoint) *tcpip.Error {
- return s.createNIC(id, name, ep, true, false)
-}
-
-// CreateNamedLoopbackNIC creates a NIC with the provided id and link-layer
-// endpoint, and a human-readable name.
-func (s *Stack) CreateNamedLoopbackNIC(id tcpip.NICID, name string, ep LinkEndpoint) *tcpip.Error {
- return s.createNIC(id, name, ep, true, true)
-}
-
-// CreateDisabledNIC creates a NIC with the provided id and link-layer endpoint,
-// but leave it disable. Stack.EnableNIC must be called before the link-layer
-// endpoint starts delivering packets to it.
-func (s *Stack) CreateDisabledNIC(id tcpip.NICID, ep LinkEndpoint) *tcpip.Error {
- return s.createNIC(id, "", ep, false, false)
-}
-
-// CreateDisabledNamedNIC is a combination of CreateNamedNIC and
-// CreateDisabledNIC.
-func (s *Stack) CreateDisabledNamedNIC(id tcpip.NICID, name string, ep LinkEndpoint) *tcpip.Error {
- return s.createNIC(id, name, ep, false, false)
+// GetNICByName gets the NIC specified by name.
+func (s *Stack) GetNICByName(name string) (*NIC, bool) {
+ s.mu.RLock()
+ defer s.mu.RUnlock()
+ for _, nic := range s.nics {
+ if nic.Name() == name {
+ return nic, true
+ }
+ }
+ return nil, false
}
// EnableNIC enables the given NIC so that the link-layer endpoint can start
@@ -808,36 +1082,72 @@ func (s *Stack) EnableNIC(id tcpip.NICID) *tcpip.Error {
s.mu.RLock()
defer s.mu.RUnlock()
- nic := s.nics[id]
- if nic == nil {
+ nic, ok := s.nics[id]
+ if !ok {
return tcpip.ErrUnknownNICID
}
return nic.enable()
}
-// CheckNIC checks if a NIC is usable.
-func (s *Stack) CheckNIC(id tcpip.NICID) bool {
+// DisableNIC disables the given NIC.
+func (s *Stack) DisableNIC(id tcpip.NICID) *tcpip.Error {
s.mu.RLock()
+ defer s.mu.RUnlock()
+
nic, ok := s.nics[id]
- s.mu.RUnlock()
- if ok {
- return nic.linkEP.IsAttached()
+ if !ok {
+ return tcpip.ErrUnknownNICID
}
- return false
+
+ return nic.disable()
}
-// NICSubnets returns a map of NICIDs to their associated subnets.
-func (s *Stack) NICAddressRanges() map[tcpip.NICID][]tcpip.Subnet {
+// CheckNIC checks if a NIC is usable.
+func (s *Stack) CheckNIC(id tcpip.NICID) bool {
s.mu.RLock()
defer s.mu.RUnlock()
- nics := map[tcpip.NICID][]tcpip.Subnet{}
+ nic, ok := s.nics[id]
+ if !ok {
+ return false
+ }
- for id, nic := range s.nics {
- nics[id] = append(nics[id], nic.AddressRanges()...)
+ return nic.enabled()
+}
+
+// RemoveNIC removes NIC and all related routes from the network stack.
+func (s *Stack) RemoveNIC(id tcpip.NICID) *tcpip.Error {
+ s.mu.Lock()
+ defer s.mu.Unlock()
+
+ return s.removeNICLocked(id)
+}
+
+// removeNICLocked removes NIC and all related routes from the network stack.
+//
+// s.mu must be locked.
+func (s *Stack) removeNICLocked(id tcpip.NICID) *tcpip.Error {
+ nic, ok := s.nics[id]
+ if !ok {
+ return tcpip.ErrUnknownNICID
}
- return nics
+ delete(s.nics, id)
+
+ // Remove routes in-place. n tracks the number of routes written.
+ n := 0
+ for i, r := range s.routeTable {
+ s.routeTable[i] = tcpip.Route{}
+ if r.NIC != id {
+ // Keep this route.
+ s.routeTable[n] = r
+ n++
+ }
+ }
+
+ s.routeTable = s.routeTable[:n]
+
+ return nic.remove()
}
// NICInfo captures the name and addresses assigned to a NIC.
@@ -853,6 +1163,23 @@ type NICInfo struct {
MTU uint32
Stats NICStats
+
+ // Context is user-supplied data optionally supplied in CreateNICWithOptions.
+ // See type NICOptions for more details.
+ Context NICContext
+
+ // ARPHardwareType holds the ARP Hardware type of the NIC. This is the
+ // value sent in haType field of an ARP Request sent by this NIC and the
+ // value expected in the haType field of an ARP response.
+ ARPHardwareType header.ARPHardwareType
+}
+
+// HasNIC returns true if the NICID is defined in the stack.
+func (s *Stack) HasNIC(id tcpip.NICID) bool {
+ s.mu.RLock()
+ _, ok := s.nics[id]
+ s.mu.RUnlock()
+ return ok
}
// NICInfo returns a map of NICIDs to their associated information.
@@ -864,9 +1191,9 @@ func (s *Stack) NICInfo() map[tcpip.NICID]NICInfo {
for id, nic := range s.nics {
flags := NICStateFlags{
Up: true, // Netstack interfaces are always up.
- Running: nic.linkEP.IsAttached(),
+ Running: nic.enabled(),
Promiscuous: nic.isPromiscuousMode(),
- Loopback: nic.linkEP.Capabilities()&CapabilityLoopback != 0,
+ Loopback: nic.isLoopback(),
}
nics[id] = NICInfo{
Name: nic.name,
@@ -875,6 +1202,8 @@ func (s *Stack) NICInfo() map[tcpip.NICID]NICInfo {
Flags: flags,
MTU: nic.linkEP.MTU(),
Stats: nic.stats,
+ Context: nic.context,
+ ARPHardwareType: nic.linkEP.ARPHardwareType(),
}
}
return nics
@@ -936,35 +1265,6 @@ func (s *Stack) AddProtocolAddressWithOptions(id tcpip.NICID, protocolAddress tc
return nic.AddAddress(protocolAddress, peb)
}
-// AddAddressRange adds a range of addresses to the specified NIC. The range is
-// given by a subnet address, and all addresses contained in the subnet are
-// used except for the subnet address itself and the subnet's broadcast
-// address.
-func (s *Stack) AddAddressRange(id tcpip.NICID, protocol tcpip.NetworkProtocolNumber, subnet tcpip.Subnet) *tcpip.Error {
- s.mu.RLock()
- defer s.mu.RUnlock()
-
- if nic, ok := s.nics[id]; ok {
- nic.AddAddressRange(protocol, subnet)
- return nil
- }
-
- return tcpip.ErrUnknownNICID
-}
-
-// RemoveAddressRange removes the range of addresses from the specified NIC.
-func (s *Stack) RemoveAddressRange(id tcpip.NICID, subnet tcpip.Subnet) *tcpip.Error {
- s.mu.RLock()
- defer s.mu.RUnlock()
-
- if nic, ok := s.nics[id]; ok {
- nic.RemoveAddressRange(subnet)
- return nil
- }
-
- return tcpip.ErrUnknownNICID
-}
-
// RemoveAddress removes an existing network-layer address from the specified
// NIC.
func (s *Stack) RemoveAddress(id tcpip.NICID, addr tcpip.Address) *tcpip.Error {
@@ -991,9 +1291,11 @@ func (s *Stack) AllAddresses() map[tcpip.NICID][]tcpip.ProtocolAddress {
return nics
}
-// GetMainNICAddress returns the first primary address and prefix for the given
-// NIC and protocol. Returns an error if the NIC doesn't exist and an empty
-// value if the NIC doesn't have a primary address for the given protocol.
+// GetMainNICAddress returns the first non-deprecated primary address and prefix
+// for the given NIC and protocol. If no non-deprecated primary address exists,
+// a deprecated primary address and prefix will be returned. Returns an error if
+// the NIC doesn't exist and an empty value if the NIC doesn't have a primary
+// address for the given protocol.
func (s *Stack) GetMainNICAddress(id tcpip.NICID, protocol tcpip.NetworkProtocolNumber) (tcpip.AddressWithPrefix, *tcpip.Error) {
s.mu.RLock()
defer s.mu.RUnlock()
@@ -1003,17 +1305,12 @@ func (s *Stack) GetMainNICAddress(id tcpip.NICID, protocol tcpip.NetworkProtocol
return tcpip.AddressWithPrefix{}, tcpip.ErrUnknownNICID
}
- for _, a := range nic.PrimaryAddresses() {
- if a.Protocol == protocol {
- return a.AddressWithPrefix, nil
- }
- }
- return tcpip.AddressWithPrefix{}, nil
+ return nic.primaryAddress(protocol), nil
}
-func (s *Stack) getRefEP(nic *NIC, localAddr tcpip.Address, netProto tcpip.NetworkProtocolNumber) (ref *referencedNetworkEndpoint) {
+func (s *Stack) getRefEP(nic *NIC, localAddr, remoteAddr tcpip.Address, netProto tcpip.NetworkProtocolNumber) (ref *referencedNetworkEndpoint) {
if len(localAddr) == 0 {
- return nic.primaryEndpoint(netProto)
+ return nic.primaryEndpoint(netProto, remoteAddr)
}
return nic.findEndpoint(netProto, localAddr, CanBePrimaryEndpoint)
}
@@ -1024,13 +1321,13 @@ func (s *Stack) FindRoute(id tcpip.NICID, localAddr, remoteAddr tcpip.Address, n
s.mu.RLock()
defer s.mu.RUnlock()
- isBroadcast := remoteAddr == header.IPv4Broadcast
+ isLocalBroadcast := remoteAddr == header.IPv4Broadcast
isMulticast := header.IsV4MulticastAddress(remoteAddr) || header.IsV6MulticastAddress(remoteAddr)
- needRoute := !(isBroadcast || isMulticast || header.IsV6LinkLocalAddress(remoteAddr))
+ needRoute := !(isLocalBroadcast || isMulticast || header.IsV6LinkLocalAddress(remoteAddr))
if id != 0 && !needRoute {
- if nic, ok := s.nics[id]; ok {
- if ref := s.getRefEP(nic, localAddr, netProto); ref != nil {
- return makeRoute(netProto, ref.ep.ID().LocalAddress, remoteAddr, nic.linkEP.LinkAddress(), ref, s.handleLocal && !nic.loopback, multicastLoop && !nic.loopback), nil
+ if nic, ok := s.nics[id]; ok && nic.enabled() {
+ if ref := s.getRefEP(nic, localAddr, remoteAddr, netProto); ref != nil {
+ return makeRoute(netProto, ref.address(), remoteAddr, nic.linkEP.LinkAddress(), ref, s.handleLocal && !nic.isLoopback(), multicastLoop && !nic.isLoopback()), nil
}
}
} else {
@@ -1038,18 +1335,25 @@ func (s *Stack) FindRoute(id tcpip.NICID, localAddr, remoteAddr tcpip.Address, n
if (id != 0 && id != route.NIC) || (len(remoteAddr) != 0 && !route.Destination.Contains(remoteAddr)) {
continue
}
- if nic, ok := s.nics[route.NIC]; ok {
- if ref := s.getRefEP(nic, localAddr, netProto); ref != nil {
+ if nic, ok := s.nics[route.NIC]; ok && nic.enabled() {
+ if ref := s.getRefEP(nic, localAddr, remoteAddr, netProto); ref != nil {
if len(remoteAddr) == 0 {
// If no remote address was provided, then the route
// provided will refer to the link local address.
- remoteAddr = ref.ep.ID().LocalAddress
+ remoteAddr = ref.address()
}
- r := makeRoute(netProto, ref.ep.ID().LocalAddress, remoteAddr, nic.linkEP.LinkAddress(), ref, s.handleLocal && !nic.loopback, multicastLoop && !nic.loopback)
- if needRoute {
- r.NextHop = route.Gateway
+ r := makeRoute(netProto, ref.address(), remoteAddr, nic.linkEP.LinkAddress(), ref, s.handleLocal && !nic.isLoopback(), multicastLoop && !nic.isLoopback())
+ r.directedBroadcast = route.Destination.IsBroadcast(remoteAddr)
+
+ if len(route.Gateway) > 0 {
+ if needRoute {
+ r.NextHop = route.Gateway
+ }
+ } else if r.directedBroadcast {
+ r.RemoteLinkAddress = header.EthernetBroadcastAddress
}
+
return r, nil
}
}
@@ -1073,13 +1377,13 @@ func (s *Stack) CheckNetworkProtocol(protocol tcpip.NetworkProtocolNumber) bool
// CheckLocalAddress determines if the given local address exists, and if it
// does, returns the id of the NIC it's bound to. Returns 0 if the address
// does not exist.
-func (s *Stack) CheckLocalAddress(nicid tcpip.NICID, protocol tcpip.NetworkProtocolNumber, addr tcpip.Address) tcpip.NICID {
+func (s *Stack) CheckLocalAddress(nicID tcpip.NICID, protocol tcpip.NetworkProtocolNumber, addr tcpip.Address) tcpip.NICID {
s.mu.RLock()
defer s.mu.RUnlock()
// If a NIC is specified, we try to find the address there only.
- if nicid != 0 {
- nic := s.nics[nicid]
+ if nicID != 0 {
+ nic := s.nics[nicID]
if nic == nil {
return 0
}
@@ -1138,35 +1442,35 @@ func (s *Stack) SetSpoofing(nicID tcpip.NICID, enable bool) *tcpip.Error {
}
// AddLinkAddress adds a link address to the stack link cache.
-func (s *Stack) AddLinkAddress(nicid tcpip.NICID, addr tcpip.Address, linkAddr tcpip.LinkAddress) {
- fullAddr := tcpip.FullAddress{NIC: nicid, Addr: addr}
+func (s *Stack) AddLinkAddress(nicID tcpip.NICID, addr tcpip.Address, linkAddr tcpip.LinkAddress) {
+ fullAddr := tcpip.FullAddress{NIC: nicID, Addr: addr}
s.linkAddrCache.add(fullAddr, linkAddr)
// TODO: provide a way for a transport endpoint to receive a signal
// that AddLinkAddress for a particular address has been called.
}
// GetLinkAddress implements LinkAddressCache.GetLinkAddress.
-func (s *Stack) GetLinkAddress(nicid tcpip.NICID, addr, localAddr tcpip.Address, protocol tcpip.NetworkProtocolNumber, waker *sleep.Waker) (tcpip.LinkAddress, <-chan struct{}, *tcpip.Error) {
+func (s *Stack) GetLinkAddress(nicID tcpip.NICID, addr, localAddr tcpip.Address, protocol tcpip.NetworkProtocolNumber, waker *sleep.Waker) (tcpip.LinkAddress, <-chan struct{}, *tcpip.Error) {
s.mu.RLock()
- nic := s.nics[nicid]
+ nic := s.nics[nicID]
if nic == nil {
s.mu.RUnlock()
return "", nil, tcpip.ErrUnknownNICID
}
s.mu.RUnlock()
- fullAddr := tcpip.FullAddress{NIC: nicid, Addr: addr}
+ fullAddr := tcpip.FullAddress{NIC: nicID, Addr: addr}
linkRes := s.linkAddrResolvers[protocol]
return s.linkAddrCache.get(fullAddr, linkRes, localAddr, nic.linkEP, waker)
}
// RemoveWaker implements LinkAddressCache.RemoveWaker.
-func (s *Stack) RemoveWaker(nicid tcpip.NICID, addr tcpip.Address, waker *sleep.Waker) {
+func (s *Stack) RemoveWaker(nicID tcpip.NICID, addr tcpip.Address, waker *sleep.Waker) {
s.mu.RLock()
defer s.mu.RUnlock()
- if nic := s.nics[nicid]; nic == nil {
- fullAddr := tcpip.FullAddress{NIC: nicid, Addr: addr}
+ if nic := s.nics[nicID]; nic == nil {
+ fullAddr := tcpip.FullAddress{NIC: nicID, Addr: addr}
s.linkAddrCache.removeWaker(fullAddr, waker)
}
}
@@ -1175,14 +1479,45 @@ func (s *Stack) RemoveWaker(nicid tcpip.NICID, addr tcpip.Address, waker *sleep.
// transport dispatcher. Received packets that match the provided id will be
// delivered to the given endpoint; specifying a nic is optional, but
// nic-specific IDs have precedence over global ones.
-func (s *Stack) RegisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error {
- return s.demux.registerEndpoint(netProtos, protocol, id, ep, reusePort, bindToDevice)
+func (s *Stack) RegisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, flags ports.Flags, bindToDevice tcpip.NICID) *tcpip.Error {
+ return s.demux.registerEndpoint(netProtos, protocol, id, ep, flags, bindToDevice)
+}
+
+// CheckRegisterTransportEndpoint checks if an endpoint can be registered with
+// the stack transport dispatcher.
+func (s *Stack) CheckRegisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, flags ports.Flags, bindToDevice tcpip.NICID) *tcpip.Error {
+ return s.demux.checkEndpoint(netProtos, protocol, id, flags, bindToDevice)
}
// UnregisterTransportEndpoint removes the endpoint with the given id from the
// stack transport dispatcher.
-func (s *Stack) UnregisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, bindToDevice tcpip.NICID) {
- s.demux.unregisterEndpoint(netProtos, protocol, id, ep, bindToDevice)
+func (s *Stack) UnregisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, flags ports.Flags, bindToDevice tcpip.NICID) {
+ s.demux.unregisterEndpoint(netProtos, protocol, id, ep, flags, bindToDevice)
+}
+
+// StartTransportEndpointCleanup removes the endpoint with the given id from
+// the stack transport dispatcher. It also transitions it to the cleanup stage.
+func (s *Stack) StartTransportEndpointCleanup(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, flags ports.Flags, bindToDevice tcpip.NICID) {
+ s.mu.Lock()
+ defer s.mu.Unlock()
+
+ s.cleanupEndpoints[ep] = struct{}{}
+
+ s.demux.unregisterEndpoint(netProtos, protocol, id, ep, flags, bindToDevice)
+}
+
+// CompleteTransportEndpointCleanup removes the endpoint from the cleanup
+// stage.
+func (s *Stack) CompleteTransportEndpointCleanup(ep TransportEndpoint) {
+ s.mu.Lock()
+ delete(s.cleanupEndpoints, ep)
+ s.mu.Unlock()
+}
+
+// FindTransportEndpoint finds an endpoint that most closely matches the provided
+// id. If no endpoint is found it returns nil.
+func (s *Stack) FindTransportEndpoint(netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, id TransportEndpointID, r *Route) TransportEndpoint {
+ return s.demux.findTransportEndpoint(netProto, transProto, id, r)
}
// RegisterRawTransportEndpoint registers the given endpoint with the stack
@@ -1206,6 +1541,81 @@ func (s *Stack) RegisterRestoredEndpoint(e ResumableEndpoint) {
s.mu.Unlock()
}
+// RegisteredEndpoints returns all endpoints which are currently registered.
+func (s *Stack) RegisteredEndpoints() []TransportEndpoint {
+ s.mu.Lock()
+ defer s.mu.Unlock()
+ var es []TransportEndpoint
+ for _, e := range s.demux.protocol {
+ es = append(es, e.transportEndpoints()...)
+ }
+ return es
+}
+
+// CleanupEndpoints returns endpoints currently in the cleanup state.
+func (s *Stack) CleanupEndpoints() []TransportEndpoint {
+ s.mu.Lock()
+ es := make([]TransportEndpoint, 0, len(s.cleanupEndpoints))
+ for e := range s.cleanupEndpoints {
+ es = append(es, e)
+ }
+ s.mu.Unlock()
+ return es
+}
+
+// RestoreCleanupEndpoints adds endpoints to cleanup tracking. This is useful
+// for restoring a stack after a save.
+func (s *Stack) RestoreCleanupEndpoints(es []TransportEndpoint) {
+ s.mu.Lock()
+ for _, e := range es {
+ s.cleanupEndpoints[e] = struct{}{}
+ }
+ s.mu.Unlock()
+}
+
+// Close closes all currently registered transport endpoints.
+//
+// Endpoints created or modified during this call may not get closed.
+func (s *Stack) Close() {
+ for _, e := range s.RegisteredEndpoints() {
+ e.Abort()
+ }
+ for _, p := range s.transportProtocols {
+ p.proto.Close()
+ }
+ for _, p := range s.networkProtocols {
+ p.Close()
+ }
+}
+
+// Wait waits for all transport and link endpoints to halt their worker
+// goroutines.
+//
+// Endpoints created or modified during this call may not get waited on.
+//
+// Note that link endpoints must be stopped via an implementation specific
+// mechanism.
+func (s *Stack) Wait() {
+ for _, e := range s.RegisteredEndpoints() {
+ e.Wait()
+ }
+ for _, e := range s.CleanupEndpoints() {
+ e.Wait()
+ }
+ for _, p := range s.transportProtocols {
+ p.proto.Wait()
+ }
+ for _, p := range s.networkProtocols {
+ p.Wait()
+ }
+
+ s.mu.RLock()
+ defer s.mu.RUnlock()
+ for _, n := range s.nics {
+ n.linkEP.Wait()
+ }
+}
+
// Resume restarts the stack after a restore. This must be called after the
// entire system has been restored.
func (s *Stack) Resume() {
@@ -1280,9 +1690,9 @@ func (s *Stack) unregisterPacketEndpointLocked(nicID tcpip.NICID, netProto tcpip
// 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 {
+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]
+ nic, ok := s.nics[nicID]
s.mu.Unlock()
if !ok {
return tcpip.ErrUnknownDevice
@@ -1296,10 +1706,10 @@ func (s *Stack) WritePacket(nicid tcpip.NICID, dst tcpip.LinkAddress, netProto t
}
fakeHeader := make(header.Ethernet, header.EthernetMinimumSize)
fakeHeader.Encode(&ethFields)
- ethHeader := buffer.View(fakeHeader).ToVectorisedView()
- ethHeader.Append(payload)
+ vv := buffer.View(fakeHeader).ToVectorisedView()
+ vv.Append(payload)
- if err := nic.linkEP.WriteRawPacket(ethHeader); err != nil {
+ if err := nic.linkEP.WriteRawPacket(vv); err != nil {
return err
}
@@ -1308,9 +1718,9 @@ func (s *Stack) WritePacket(nicid tcpip.NICID, dst tcpip.LinkAddress, netProto t
// WriteRawPacket writes data directly to the specified NIC without adding any
// headers.
-func (s *Stack) WriteRawPacket(nicid tcpip.NICID, payload buffer.VectorisedView) *tcpip.Error {
+func (s *Stack) WriteRawPacket(nicID tcpip.NICID, payload buffer.VectorisedView) *tcpip.Error {
s.mu.Lock()
- nic, ok := s.nics[nicid]
+ nic, ok := s.nics[nicID]
s.mu.Unlock()
if !ok {
return tcpip.ErrUnknownDevice
@@ -1403,14 +1813,21 @@ func (s *Stack) LeaveGroup(protocol tcpip.NetworkProtocolNumber, nicID tcpip.NIC
return tcpip.ErrUnknownNICID
}
-// IPTables returns the stack's iptables.
-func (s *Stack) IPTables() iptables.IPTables {
- return s.tables
+// IsInGroup returns true if the NIC with ID nicID has joined the multicast
+// group multicastAddr.
+func (s *Stack) IsInGroup(nicID tcpip.NICID, multicastAddr tcpip.Address) (bool, *tcpip.Error) {
+ s.mu.RLock()
+ defer s.mu.RUnlock()
+
+ if nic, ok := s.nics[nicID]; ok {
+ return nic.isInGroup(multicastAddr), nil
+ }
+ return false, tcpip.ErrUnknownNICID
}
-// SetIPTables sets the stack's iptables.
-func (s *Stack) SetIPTables(ipt iptables.IPTables) {
- s.tables = ipt
+// IPTables returns the stack's iptables.
+func (s *Stack) IPTables() *IPTables {
+ return s.tables
}
// ICMPLimit returns the maximum number of ICMP messages that can be sent
@@ -1489,16 +1906,66 @@ func (s *Stack) SetNDPConfigurations(id tcpip.NICID, c NDPConfigurations) *tcpip
}
nic.setNDPConfigs(c)
+ return nil
+}
+
+// NUDConfigurations gets the per-interface NUD configurations.
+func (s *Stack) NUDConfigurations(id tcpip.NICID) (NUDConfigurations, *tcpip.Error) {
+ s.mu.RLock()
+ nic, ok := s.nics[id]
+ s.mu.RUnlock()
+
+ if !ok {
+ return NUDConfigurations{}, tcpip.ErrUnknownNICID
+ }
+
+ return nic.NUDConfigs()
+}
+
+// SetNUDConfigurations sets the per-interface NUD configurations.
+//
+// Note, if c contains invalid NUD configuration values, it will be fixed to
+// use default values for the erroneous values.
+func (s *Stack) SetNUDConfigurations(id tcpip.NICID, c NUDConfigurations) *tcpip.Error {
+ s.mu.RLock()
+ nic, ok := s.nics[id]
+ s.mu.RUnlock()
+
+ if !ok {
+ return tcpip.ErrUnknownNICID
+ }
+
+ return nic.setNUDConfigs(c)
+}
+
+// HandleNDPRA provides a NIC with ID id a validated NDP Router Advertisement
+// message that it needs to handle.
+func (s *Stack) HandleNDPRA(id tcpip.NICID, ip tcpip.Address, ra header.NDPRouterAdvert) *tcpip.Error {
+ s.mu.Lock()
+ defer s.mu.Unlock()
+
+ nic, ok := s.nics[id]
+ if !ok {
+ return tcpip.ErrUnknownNICID
+ }
+
+ nic.handleNDPRA(ip, ra)
return nil
}
-// PortSeed returns a 32 bit value that can be used as a seed value for port
-// picking.
+// Seed returns a 32 bit value that can be used as a seed value for port
+// picking, ISN generation etc.
//
// NOTE: The seed is generated once during stack initialization only.
-func (s *Stack) PortSeed() uint32 {
- return s.portSeed
+func (s *Stack) Seed() uint32 {
+ return s.seed
+}
+
+// Rand returns a reference to a pseudo random generator that can be used
+// to generate random numbers as required.
+func (s *Stack) Rand() *mathrand.Rand {
+ return s.randomGenerator
}
func generateRandUint32() uint32 {
@@ -1508,3 +1975,49 @@ func generateRandUint32() uint32 {
}
return binary.LittleEndian.Uint32(b)
}
+
+func generateRandInt64() int64 {
+ b := make([]byte, 8)
+ if _, err := rand.Read(b); err != nil {
+ panic(err)
+ }
+ buf := bytes.NewReader(b)
+ var v int64
+ if err := binary.Read(buf, binary.LittleEndian, &v); err != nil {
+ panic(err)
+ }
+ return v
+}
+
+// FindNetworkEndpoint returns the network endpoint for the given address.
+func (s *Stack) FindNetworkEndpoint(netProto tcpip.NetworkProtocolNumber, address tcpip.Address) (NetworkEndpoint, *tcpip.Error) {
+ s.mu.RLock()
+ defer s.mu.RUnlock()
+
+ for _, nic := range s.nics {
+ id := NetworkEndpointID{address}
+
+ if ref, ok := nic.mu.endpoints[id]; ok {
+ nic.mu.RLock()
+ defer nic.mu.RUnlock()
+
+ // An endpoint with this id exists, check if it can be
+ // used and return it.
+ return ref.ep, nil
+ }
+ }
+ return nil, tcpip.ErrBadAddress
+}
+
+// FindNICNameFromID returns the name of the nic for the given NICID.
+func (s *Stack) FindNICNameFromID(id tcpip.NICID) string {
+ s.mu.RLock()
+ defer s.mu.RUnlock()
+
+ nic, ok := s.nics[id]
+ if !ok {
+ return ""
+ }
+
+ return nic.Name()
+}
diff --git a/pkg/tcpip/stack/stack_options.go b/pkg/tcpip/stack/stack_options.go
new file mode 100644
index 000000000..0b093e6c5
--- /dev/null
+++ b/pkg/tcpip/stack/stack_options.go
@@ -0,0 +1,106 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package stack
+
+import "gvisor.dev/gvisor/pkg/tcpip"
+
+const (
+ // MinBufferSize is the smallest size of a receive or send buffer.
+ MinBufferSize = 4 << 10 // 4 KiB
+
+ // DefaultBufferSize is the default size of the send/recv buffer for a
+ // transport endpoint.
+ DefaultBufferSize = 212 << 10 // 212 KiB
+
+ // DefaultMaxBufferSize is the default maximum permitted size of a
+ // send/receive buffer.
+ DefaultMaxBufferSize = 4 << 20 // 4 MiB
+)
+
+// SendBufferSizeOption is used by stack.(Stack*).Option/SetOption to
+// get/set the default, min and max send buffer sizes.
+type SendBufferSizeOption struct {
+ Min int
+ Default int
+ Max int
+}
+
+// ReceiveBufferSizeOption is used by stack.(Stack*).Option/SetOption to
+// get/set the default, min and max receive buffer sizes.
+type ReceiveBufferSizeOption struct {
+ Min int
+ Default int
+ Max int
+}
+
+// SetOption allows setting stack wide options.
+func (s *Stack) SetOption(option interface{}) *tcpip.Error {
+ switch v := option.(type) {
+ case SendBufferSizeOption:
+ // Make sure we don't allow lowering the buffer below minimum
+ // required for stack to work.
+ if v.Min < MinBufferSize {
+ return tcpip.ErrInvalidOptionValue
+ }
+
+ if v.Default < v.Min || v.Default > v.Max {
+ return tcpip.ErrInvalidOptionValue
+ }
+
+ s.mu.Lock()
+ s.sendBufferSize = v
+ s.mu.Unlock()
+ return nil
+
+ case ReceiveBufferSizeOption:
+ // Make sure we don't allow lowering the buffer below minimum
+ // required for stack to work.
+ if v.Min < MinBufferSize {
+ return tcpip.ErrInvalidOptionValue
+ }
+
+ if v.Default < v.Min || v.Default > v.Max {
+ return tcpip.ErrInvalidOptionValue
+ }
+
+ s.mu.Lock()
+ s.receiveBufferSize = v
+ s.mu.Unlock()
+ return nil
+
+ default:
+ return tcpip.ErrUnknownProtocolOption
+ }
+}
+
+// Option allows retrieving stack wide options.
+func (s *Stack) Option(option interface{}) *tcpip.Error {
+ switch v := option.(type) {
+ case *SendBufferSizeOption:
+ s.mu.RLock()
+ *v = s.sendBufferSize
+ s.mu.RUnlock()
+ return nil
+
+ case *ReceiveBufferSizeOption:
+ s.mu.RLock()
+ *v = s.receiveBufferSize
+ s.mu.RUnlock()
+ return nil
+
+ default:
+ return tcpip.ErrUnknownProtocolOption
+ }
+}
diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go
index ef3d1beb0..f168be402 100644
--- a/pkg/tcpip/stack/stack_test.go
+++ b/pkg/tcpip/stack/stack_test.go
@@ -21,18 +21,24 @@ import (
"bytes"
"fmt"
"math"
+ "net"
"sort"
- "strings"
"testing"
"time"
"github.com/google/go-cmp/cmp"
+ "github.com/google/go-cmp/cmp/cmpopts"
+ "gvisor.dev/gvisor/pkg/rand"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/tcpip/header"
"gvisor.dev/gvisor/pkg/tcpip/link/channel"
+ "gvisor.dev/gvisor/pkg/tcpip/link/loopback"
+ "gvisor.dev/gvisor/pkg/tcpip/network/arp"
+ "gvisor.dev/gvisor/pkg/tcpip/network/ipv4"
"gvisor.dev/gvisor/pkg/tcpip/network/ipv6"
"gvisor.dev/gvisor/pkg/tcpip/stack"
+ "gvisor.dev/gvisor/pkg/tcpip/transport/udp"
)
const (
@@ -51,6 +57,10 @@ const (
// where another value is explicitly used. It is chosen to match the MTU
// of loopback interfaces on linux systems.
defaultMTU = 65536
+
+ dstAddrOffset = 0
+ srcAddrOffset = 1
+ protocolNumberOffset = 2
)
// fakeNetworkEndpoint is a network-layer protocol endpoint. It counts sent and
@@ -61,9 +71,7 @@ const (
// use the first three: destination address, source address, and transport
// protocol. They're all one byte fields to simplify parsing.
type fakeNetworkEndpoint struct {
- nicid tcpip.NICID
- id stack.NetworkEndpointID
- prefixLen int
+ nicID tcpip.NICID
proto *fakeNetworkProtocol
dispatcher stack.TransportDispatcher
ep stack.LinkEndpoint
@@ -74,43 +82,35 @@ func (f *fakeNetworkEndpoint) MTU() uint32 {
}
func (f *fakeNetworkEndpoint) NICID() tcpip.NICID {
- return f.nicid
-}
-
-func (f *fakeNetworkEndpoint) PrefixLen() int {
- return f.prefixLen
+ return f.nicID
}
func (*fakeNetworkEndpoint) DefaultTTL() uint8 {
return 123
}
-func (f *fakeNetworkEndpoint) ID() *stack.NetworkEndpointID {
- return &f.id
-}
-
-func (f *fakeNetworkEndpoint) HandlePacket(r *stack.Route, vv buffer.VectorisedView) {
+func (f *fakeNetworkEndpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) {
// Increment the received packet count in the protocol descriptor.
- f.proto.packetCount[int(f.id.LocalAddress[0])%len(f.proto.packetCount)]++
-
- // Consume the network header.
- b := vv.First()
- vv.TrimFront(fakeNetHeaderLen)
+ f.proto.packetCount[int(r.LocalAddress[0])%len(f.proto.packetCount)]++
// Handle control packets.
- if b[2] == uint8(fakeControlProtocol) {
- nb := vv.First()
- if len(nb) < fakeNetHeaderLen {
+ if pkt.NetworkHeader().View()[protocolNumberOffset] == uint8(fakeControlProtocol) {
+ nb, ok := pkt.Data.PullUp(fakeNetHeaderLen)
+ if !ok {
return
}
-
- vv.TrimFront(fakeNetHeaderLen)
- f.dispatcher.DeliverTransportControlPacket(tcpip.Address(nb[1:2]), tcpip.Address(nb[0:1]), fakeNetNumber, tcpip.TransportProtocolNumber(nb[2]), stack.ControlPortUnreachable, 0, vv)
+ pkt.Data.TrimFront(fakeNetHeaderLen)
+ f.dispatcher.DeliverTransportControlPacket(
+ tcpip.Address(nb[srcAddrOffset:srcAddrOffset+1]),
+ tcpip.Address(nb[dstAddrOffset:dstAddrOffset+1]),
+ fakeNetNumber,
+ tcpip.TransportProtocolNumber(nb[protocolNumberOffset]),
+ stack.ControlPortUnreachable, 0, pkt)
return
}
// Dispatch the packet to the transport protocol.
- f.dispatcher.DeliverTransportPacket(r, tcpip.TransportProtocolNumber(b[2]), buffer.View([]byte{}), vv)
+ f.dispatcher.DeliverTransportPacket(r, tcpip.TransportProtocolNumber(pkt.NetworkHeader().View()[protocolNumberOffset]), pkt)
}
func (f *fakeNetworkEndpoint) MaxHeaderLength() uint16 {
@@ -125,37 +125,37 @@ func (f *fakeNetworkEndpoint) Capabilities() stack.LinkEndpointCapabilities {
return f.ep.Capabilities()
}
-func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) *tcpip.Error {
+func (f *fakeNetworkEndpoint) NetworkProtocolNumber() tcpip.NetworkProtocolNumber {
+ return f.proto.Number()
+}
+
+func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt *stack.PacketBuffer) *tcpip.Error {
// Increment the sent packet count in the protocol descriptor.
f.proto.sendPacketCount[int(r.RemoteAddress[0])%len(f.proto.sendPacketCount)]++
// Add the protocol's header to the packet and send it to the link
// endpoint.
- b := hdr.Prepend(fakeNetHeaderLen)
- b[0] = r.RemoteAddress[0]
- b[1] = f.id.LocalAddress[0]
- b[2] = byte(params.Protocol)
-
- if loop&stack.PacketLoop != 0 {
- views := make([]buffer.View, 1, 1+len(payload.Views()))
- views[0] = hdr.View()
- views = append(views, payload.Views()...)
- vv := buffer.NewVectorisedView(len(views[0])+payload.Size(), views)
- f.HandlePacket(r, vv)
- }
- if loop&stack.PacketOut == 0 {
+ hdr := pkt.NetworkHeader().Push(fakeNetHeaderLen)
+ hdr[dstAddrOffset] = r.RemoteAddress[0]
+ hdr[srcAddrOffset] = r.LocalAddress[0]
+ hdr[protocolNumberOffset] = byte(params.Protocol)
+
+ if r.Loop&stack.PacketLoop != 0 {
+ f.HandlePacket(r, pkt)
+ }
+ if r.Loop&stack.PacketOut == 0 {
return nil
}
- return f.ep.WritePacket(r, gso, hdr, payload, fakeNetNumber)
+ return f.ep.WritePacket(r, gso, fakeNetNumber, pkt)
}
// WritePackets implements stack.LinkEndpoint.WritePackets.
-func (f *fakeNetworkEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) {
+func (f *fakeNetworkEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, params stack.NetworkHeaderParams) (int, *tcpip.Error) {
panic("not implemented")
}
-func (*fakeNetworkEndpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error {
+func (*fakeNetworkEndpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBuffer) *tcpip.Error {
return tcpip.ErrNotSupported
}
@@ -197,18 +197,16 @@ func (f *fakeNetworkProtocol) PacketCount(intfAddr byte) int {
}
func (*fakeNetworkProtocol) ParseAddresses(v buffer.View) (src, dst tcpip.Address) {
- return tcpip.Address(v[1:2]), tcpip.Address(v[0:1])
+ return tcpip.Address(v[srcAddrOffset : srcAddrOffset+1]), tcpip.Address(v[dstAddrOffset : dstAddrOffset+1])
}
-func (f *fakeNetworkProtocol) NewEndpoint(nicid tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, ep stack.LinkEndpoint) (stack.NetworkEndpoint, *tcpip.Error) {
+func (f *fakeNetworkProtocol) NewEndpoint(nicID tcpip.NICID, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, ep stack.LinkEndpoint, _ *stack.Stack) stack.NetworkEndpoint {
return &fakeNetworkEndpoint{
- nicid: nicid,
- id: stack.NetworkEndpointID{LocalAddress: addrWithPrefix.Address},
- prefixLen: addrWithPrefix.PrefixLen,
+ nicID: nicID,
proto: f,
dispatcher: dispatcher,
ep: ep,
- }, nil
+ }
}
func (f *fakeNetworkProtocol) SetOption(option interface{}) *tcpip.Error {
@@ -233,10 +231,53 @@ func (f *fakeNetworkProtocol) Option(option interface{}) *tcpip.Error {
}
}
+// Close implements TransportProtocol.Close.
+func (*fakeNetworkProtocol) Close() {}
+
+// Wait implements TransportProtocol.Wait.
+func (*fakeNetworkProtocol) Wait() {}
+
+// Parse implements TransportProtocol.Parse.
+func (*fakeNetworkProtocol) Parse(pkt *stack.PacketBuffer) (tcpip.TransportProtocolNumber, bool, bool) {
+ hdr, ok := pkt.NetworkHeader().Consume(fakeNetHeaderLen)
+ if !ok {
+ return 0, false, false
+ }
+ return tcpip.TransportProtocolNumber(hdr[protocolNumberOffset]), true, true
+}
+
func fakeNetFactory() stack.NetworkProtocol {
return &fakeNetworkProtocol{}
}
+// linkEPWithMockedAttach is a stack.LinkEndpoint that tests can use to verify
+// that LinkEndpoint.Attach was called.
+type linkEPWithMockedAttach struct {
+ stack.LinkEndpoint
+ attached bool
+}
+
+// Attach implements stack.LinkEndpoint.Attach.
+func (l *linkEPWithMockedAttach) Attach(d stack.NetworkDispatcher) {
+ l.LinkEndpoint.Attach(d)
+ l.attached = d != nil
+}
+
+func (l *linkEPWithMockedAttach) isAttached() bool {
+ return l.attached
+}
+
+// Checks to see if list contains an address.
+func containsAddr(list []tcpip.ProtocolAddress, item tcpip.ProtocolAddress) bool {
+ for _, i := range list {
+ if i == item {
+ return true
+ }
+ }
+
+ return false
+}
+
func TestNetworkReceive(t *testing.T) {
// Create a stack with the fake network protocol, one nic, and two
// addresses attached to it: 1 & 2.
@@ -261,8 +302,10 @@ func TestNetworkReceive(t *testing.T) {
buf := buffer.NewView(30)
// Make sure packet with wrong address is not delivered.
- buf[0] = 3
- ep.Inject(fakeNetNumber, buf.ToVectorisedView())
+ buf[dstAddrOffset] = 3
+ ep.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ }))
if fakeNet.packetCount[1] != 0 {
t.Errorf("packetCount[1] = %d, want %d", fakeNet.packetCount[1], 0)
}
@@ -271,8 +314,10 @@ func TestNetworkReceive(t *testing.T) {
}
// Make sure packet is delivered to first endpoint.
- buf[0] = 1
- ep.Inject(fakeNetNumber, buf.ToVectorisedView())
+ buf[dstAddrOffset] = 1
+ ep.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ }))
if fakeNet.packetCount[1] != 1 {
t.Errorf("packetCount[1] = %d, want %d", fakeNet.packetCount[1], 1)
}
@@ -281,8 +326,10 @@ func TestNetworkReceive(t *testing.T) {
}
// Make sure packet is delivered to second endpoint.
- buf[0] = 2
- ep.Inject(fakeNetNumber, buf.ToVectorisedView())
+ buf[dstAddrOffset] = 2
+ ep.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ }))
if fakeNet.packetCount[1] != 1 {
t.Errorf("packetCount[1] = %d, want %d", fakeNet.packetCount[1], 1)
}
@@ -291,7 +338,9 @@ func TestNetworkReceive(t *testing.T) {
}
// Make sure packet is not delivered if protocol number is wrong.
- ep.Inject(fakeNetNumber-1, buf.ToVectorisedView())
+ ep.InjectInbound(fakeNetNumber-1, stack.NewPacketBuffer(stack.PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ }))
if fakeNet.packetCount[1] != 1 {
t.Errorf("packetCount[1] = %d, want %d", fakeNet.packetCount[1], 1)
}
@@ -301,7 +350,9 @@ func TestNetworkReceive(t *testing.T) {
// Make sure packet that is too small is dropped.
buf.CapLength(2)
- ep.Inject(fakeNetNumber, buf.ToVectorisedView())
+ ep.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ }))
if fakeNet.packetCount[1] != 1 {
t.Errorf("packetCount[1] = %d, want %d", fakeNet.packetCount[1], 1)
}
@@ -320,8 +371,10 @@ func sendTo(s *stack.Stack, addr tcpip.Address, payload buffer.View) *tcpip.Erro
}
func send(r stack.Route, payload buffer.View) *tcpip.Error {
- hdr := buffer.NewPrependable(int(r.MaxHeaderLength()))
- return r.WritePacket(nil /* gso */, hdr, payload.ToVectorisedView(), stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS})
+ return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}, stack.NewPacketBuffer(stack.PacketBufferOptions{
+ ReserveHeaderBytes: int(r.MaxHeaderLength()),
+ Data: payload.ToVectorisedView(),
+ }))
}
func testSendTo(t *testing.T, s *stack.Stack, addr tcpip.Address, ep *channel.Endpoint, payload buffer.View) {
@@ -376,7 +429,9 @@ func testFailingRecv(t *testing.T, fakeNet *fakeNetworkProtocol, localAddrByte b
func testRecvInternal(t *testing.T, fakeNet *fakeNetworkProtocol, localAddrByte byte, ep *channel.Endpoint, buf buffer.View, want int) {
t.Helper()
- ep.Inject(fakeNetNumber, buf.ToVectorisedView())
+ ep.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ }))
if got := fakeNet.PacketCount(localAddrByte); got != want {
t.Errorf("receive packet count: got = %d, want %d", got, want)
}
@@ -493,6 +548,340 @@ func testNoRoute(t *testing.T, s *stack.Stack, nic tcpip.NICID, srcAddr, dstAddr
}
}
+// TestAttachToLinkEndpointImmediately tests that a LinkEndpoint is attached to
+// a NetworkDispatcher when the NIC is created.
+func TestAttachToLinkEndpointImmediately(t *testing.T) {
+ const nicID = 1
+
+ tests := []struct {
+ name string
+ nicOpts stack.NICOptions
+ }{
+ {
+ name: "Create enabled NIC",
+ nicOpts: stack.NICOptions{Disabled: false},
+ },
+ {
+ name: "Create disabled NIC",
+ nicOpts: stack.NICOptions{Disabled: true},
+ },
+ }
+
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()},
+ })
+
+ e := linkEPWithMockedAttach{
+ LinkEndpoint: loopback.New(),
+ }
+
+ if err := s.CreateNICWithOptions(nicID, &e, test.nicOpts); err != nil {
+ t.Fatalf("CreateNICWithOptions(%d, _, %+v) = %s", nicID, test.nicOpts, err)
+ }
+ if !e.isAttached() {
+ t.Fatal("link endpoint not attached to a network dispatcher")
+ }
+ })
+ }
+}
+
+func TestDisableUnknownNIC(t *testing.T) {
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()},
+ })
+
+ if err := s.DisableNIC(1); err != tcpip.ErrUnknownNICID {
+ t.Fatalf("got s.DisableNIC(1) = %v, want = %s", err, tcpip.ErrUnknownNICID)
+ }
+}
+
+func TestDisabledNICsNICInfoAndCheckNIC(t *testing.T) {
+ const nicID = 1
+
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()},
+ })
+
+ e := loopback.New()
+ nicOpts := stack.NICOptions{Disabled: true}
+ if err := s.CreateNICWithOptions(nicID, e, nicOpts); err != nil {
+ t.Fatalf("CreateNICWithOptions(%d, _, %+v) = %s", nicID, nicOpts, err)
+ }
+
+ checkNIC := func(enabled bool) {
+ t.Helper()
+
+ allNICInfo := s.NICInfo()
+ nicInfo, ok := allNICInfo[nicID]
+ if !ok {
+ t.Errorf("entry for %d missing from allNICInfo = %+v", nicID, allNICInfo)
+ } else if nicInfo.Flags.Running != enabled {
+ t.Errorf("got nicInfo.Flags.Running = %t, want = %t", nicInfo.Flags.Running, enabled)
+ }
+
+ if got := s.CheckNIC(nicID); got != enabled {
+ t.Errorf("got s.CheckNIC(%d) = %t, want = %t", nicID, got, enabled)
+ }
+ }
+
+ // NIC should initially report itself as disabled.
+ checkNIC(false)
+
+ if err := s.EnableNIC(nicID); err != nil {
+ t.Fatalf("s.EnableNIC(%d): %s", nicID, err)
+ }
+ checkNIC(true)
+
+ // If the NIC is not reporting a correct enabled status, we cannot trust the
+ // next check so end the test here.
+ if t.Failed() {
+ t.FailNow()
+ }
+
+ if err := s.DisableNIC(nicID); err != nil {
+ t.Fatalf("s.DisableNIC(%d): %s", nicID, err)
+ }
+ checkNIC(false)
+}
+
+func TestRemoveUnknownNIC(t *testing.T) {
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()},
+ })
+
+ if err := s.RemoveNIC(1); err != tcpip.ErrUnknownNICID {
+ t.Fatalf("got s.RemoveNIC(1) = %v, want = %s", err, tcpip.ErrUnknownNICID)
+ }
+}
+
+func TestRemoveNIC(t *testing.T) {
+ const nicID = 1
+
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()},
+ })
+
+ e := linkEPWithMockedAttach{
+ LinkEndpoint: loopback.New(),
+ }
+ if err := s.CreateNIC(nicID, &e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+
+ // NIC should be present in NICInfo and attached to a NetworkDispatcher.
+ allNICInfo := s.NICInfo()
+ if _, ok := allNICInfo[nicID]; !ok {
+ t.Errorf("entry for %d missing from allNICInfo = %+v", nicID, allNICInfo)
+ }
+ if !e.isAttached() {
+ t.Fatal("link endpoint not attached to a network dispatcher")
+ }
+
+ // Removing a NIC should remove it from NICInfo and e should be detached from
+ // the NetworkDispatcher.
+ if err := s.RemoveNIC(nicID); err != nil {
+ t.Fatalf("s.RemoveNIC(%d): %s", nicID, err)
+ }
+ if nicInfo, ok := s.NICInfo()[nicID]; ok {
+ t.Errorf("got unexpected NICInfo entry for deleted NIC %d = %+v", nicID, nicInfo)
+ }
+ if e.isAttached() {
+ t.Error("link endpoint for removed NIC still attached to a network dispatcher")
+ }
+}
+
+func TestRouteWithDownNIC(t *testing.T) {
+ tests := []struct {
+ name string
+ downFn func(s *stack.Stack, nicID tcpip.NICID) *tcpip.Error
+ upFn func(s *stack.Stack, nicID tcpip.NICID) *tcpip.Error
+ }{
+ {
+ name: "Disabled NIC",
+ downFn: (*stack.Stack).DisableNIC,
+ upFn: (*stack.Stack).EnableNIC,
+ },
+
+ // Once a NIC is removed, it cannot be brought up.
+ {
+ name: "Removed NIC",
+ downFn: (*stack.Stack).RemoveNIC,
+ },
+ }
+
+ const unspecifiedNIC = 0
+ const nicID1 = 1
+ const nicID2 = 2
+ const addr1 = tcpip.Address("\x01")
+ const addr2 = tcpip.Address("\x02")
+ const nic1Dst = tcpip.Address("\x05")
+ const nic2Dst = tcpip.Address("\x06")
+
+ setup := func(t *testing.T) (*stack.Stack, *channel.Endpoint, *channel.Endpoint) {
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()},
+ })
+
+ ep1 := channel.New(1, defaultMTU, "")
+ if err := s.CreateNIC(nicID1, ep1); err != nil {
+ t.Fatalf("CreateNIC(%d, _): %s", nicID1, err)
+ }
+
+ if err := s.AddAddress(nicID1, fakeNetNumber, addr1); err != nil {
+ t.Fatalf("AddAddress(%d, %d, %s): %s", nicID1, fakeNetNumber, addr1, err)
+ }
+
+ ep2 := channel.New(1, defaultMTU, "")
+ if err := s.CreateNIC(nicID2, ep2); err != nil {
+ t.Fatalf("CreateNIC(%d, _): %s", nicID2, err)
+ }
+
+ if err := s.AddAddress(nicID2, fakeNetNumber, addr2); err != nil {
+ t.Fatalf("AddAddress(%d, %d, %s): %s", nicID2, fakeNetNumber, addr2, err)
+ }
+
+ // Set a route table that sends all packets with odd destination
+ // addresses through the first NIC, and all even destination address
+ // through the second one.
+ {
+ subnet0, err := tcpip.NewSubnet("\x00", "\x01")
+ if err != nil {
+ t.Fatal(err)
+ }
+ subnet1, err := tcpip.NewSubnet("\x01", "\x01")
+ if err != nil {
+ t.Fatal(err)
+ }
+ s.SetRouteTable([]tcpip.Route{
+ {Destination: subnet1, Gateway: "\x00", NIC: nicID1},
+ {Destination: subnet0, Gateway: "\x00", NIC: nicID2},
+ })
+ }
+
+ return s, ep1, ep2
+ }
+
+ // Tests that routes through a down NIC are not used when looking up a route
+ // for a destination.
+ t.Run("Find", func(t *testing.T) {
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ s, _, _ := setup(t)
+
+ // Test routes to odd address.
+ testRoute(t, s, unspecifiedNIC, "", "\x05", addr1)
+ testRoute(t, s, unspecifiedNIC, addr1, "\x05", addr1)
+ testRoute(t, s, nicID1, addr1, "\x05", addr1)
+
+ // Test routes to even address.
+ testRoute(t, s, unspecifiedNIC, "", "\x06", addr2)
+ testRoute(t, s, unspecifiedNIC, addr2, "\x06", addr2)
+ testRoute(t, s, nicID2, addr2, "\x06", addr2)
+
+ // Bringing NIC1 down should result in no routes to odd addresses. Routes to
+ // even addresses should continue to be available as NIC2 is still up.
+ if err := test.downFn(s, nicID1); err != nil {
+ t.Fatalf("test.downFn(_, %d): %s", nicID1, err)
+ }
+ testNoRoute(t, s, unspecifiedNIC, "", nic1Dst)
+ testNoRoute(t, s, unspecifiedNIC, addr1, nic1Dst)
+ testNoRoute(t, s, nicID1, addr1, nic1Dst)
+ testRoute(t, s, unspecifiedNIC, "", nic2Dst, addr2)
+ testRoute(t, s, unspecifiedNIC, addr2, nic2Dst, addr2)
+ testRoute(t, s, nicID2, addr2, nic2Dst, addr2)
+
+ // Bringing NIC2 down should result in no routes to even addresses. No
+ // route should be available to any address as routes to odd addresses
+ // were made unavailable by bringing NIC1 down above.
+ if err := test.downFn(s, nicID2); err != nil {
+ t.Fatalf("test.downFn(_, %d): %s", nicID2, err)
+ }
+ testNoRoute(t, s, unspecifiedNIC, "", nic1Dst)
+ testNoRoute(t, s, unspecifiedNIC, addr1, nic1Dst)
+ testNoRoute(t, s, nicID1, addr1, nic1Dst)
+ testNoRoute(t, s, unspecifiedNIC, "", nic2Dst)
+ testNoRoute(t, s, unspecifiedNIC, addr2, nic2Dst)
+ testNoRoute(t, s, nicID2, addr2, nic2Dst)
+
+ if upFn := test.upFn; upFn != nil {
+ // Bringing NIC1 up should make routes to odd addresses available
+ // again. Routes to even addresses should continue to be unavailable
+ // as NIC2 is still down.
+ if err := upFn(s, nicID1); err != nil {
+ t.Fatalf("test.upFn(_, %d): %s", nicID1, err)
+ }
+ testRoute(t, s, unspecifiedNIC, "", nic1Dst, addr1)
+ testRoute(t, s, unspecifiedNIC, addr1, nic1Dst, addr1)
+ testRoute(t, s, nicID1, addr1, nic1Dst, addr1)
+ testNoRoute(t, s, unspecifiedNIC, "", nic2Dst)
+ testNoRoute(t, s, unspecifiedNIC, addr2, nic2Dst)
+ testNoRoute(t, s, nicID2, addr2, nic2Dst)
+ }
+ })
+ }
+ })
+
+ // Tests that writing a packet using a Route through a down NIC fails.
+ t.Run("WritePacket", func(t *testing.T) {
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ s, ep1, ep2 := setup(t)
+
+ r1, err := s.FindRoute(nicID1, addr1, nic1Dst, fakeNetNumber, false /* multicastLoop */)
+ if err != nil {
+ t.Errorf("FindRoute(%d, %s, %s, %d, false): %s", nicID1, addr1, nic1Dst, fakeNetNumber, err)
+ }
+ defer r1.Release()
+
+ r2, err := s.FindRoute(nicID2, addr2, nic2Dst, fakeNetNumber, false /* multicastLoop */)
+ if err != nil {
+ t.Errorf("FindRoute(%d, %s, %s, %d, false): %s", nicID2, addr2, nic2Dst, fakeNetNumber, err)
+ }
+ defer r2.Release()
+
+ // If we failed to get routes r1 or r2, we cannot proceed with the test.
+ if t.Failed() {
+ t.FailNow()
+ }
+
+ buf := buffer.View([]byte{1})
+ testSend(t, r1, ep1, buf)
+ testSend(t, r2, ep2, buf)
+
+ // Writes with Routes that use NIC1 after being brought down should fail.
+ if err := test.downFn(s, nicID1); err != nil {
+ t.Fatalf("test.downFn(_, %d): %s", nicID1, err)
+ }
+ testFailingSend(t, r1, ep1, buf, tcpip.ErrInvalidEndpointState)
+ testSend(t, r2, ep2, buf)
+
+ // Writes with Routes that use NIC2 after being brought down should fail.
+ if err := test.downFn(s, nicID2); err != nil {
+ t.Fatalf("test.downFn(_, %d): %s", nicID2, err)
+ }
+ testFailingSend(t, r1, ep1, buf, tcpip.ErrInvalidEndpointState)
+ testFailingSend(t, r2, ep2, buf, tcpip.ErrInvalidEndpointState)
+
+ if upFn := test.upFn; upFn != nil {
+ // Writes with Routes that use NIC1 after being brought up should
+ // succeed.
+ //
+ // TODO(gvisor.dev/issue/1491): Should we instead completely
+ // invalidate all Routes that were bound to a NIC that was brought
+ // down at some point?
+ if err := upFn(s, nicID1); err != nil {
+ t.Fatalf("test.upFn(_, %d): %s", nicID1, err)
+ }
+ testSend(t, r1, ep1, buf)
+ testFailingSend(t, r2, ep2, buf, tcpip.ErrInvalidEndpointState)
+ }
+ })
+ }
+ })
+}
+
func TestRoutes(t *testing.T) {
// Create a stack with the fake network protocol, two nics, and two
// addresses per nic, the first nic has odd address, the second one has
@@ -602,7 +991,7 @@ func TestAddressRemoval(t *testing.T) {
buf := buffer.NewView(30)
// Send and receive packets, and verify they are received.
- buf[0] = localAddrByte
+ buf[dstAddrOffset] = localAddrByte
testRecv(t, fakeNet, localAddrByte, ep, buf)
testSendTo(t, s, remoteAddr, ep, nil)
@@ -652,7 +1041,7 @@ func TestAddressRemovalWithRouteHeld(t *testing.T) {
}
// Send and receive packets, and verify they are received.
- buf[0] = localAddrByte
+ buf[dstAddrOffset] = localAddrByte
testRecv(t, fakeNet, localAddrByte, ep, buf)
testSend(t, r, ep, nil)
testSendTo(t, s, remoteAddr, ep, nil)
@@ -671,11 +1060,11 @@ func TestAddressRemovalWithRouteHeld(t *testing.T) {
}
}
-func verifyAddress(t *testing.T, s *stack.Stack, nicid tcpip.NICID, addr tcpip.Address) {
+func verifyAddress(t *testing.T, s *stack.Stack, nicID tcpip.NICID, addr tcpip.Address) {
t.Helper()
- info, ok := s.NICInfo()[nicid]
+ info, ok := s.NICInfo()[nicID]
if !ok {
- t.Fatalf("NICInfo() failed to find nicid=%d", nicid)
+ t.Fatalf("NICInfo() failed to find nicID=%d", nicID)
}
if len(addr) == 0 {
// No address given, verify that there is no address assigned to the NIC.
@@ -708,7 +1097,7 @@ func TestEndpointExpiration(t *testing.T) {
localAddrByte byte = 0x01
remoteAddr tcpip.Address = "\x03"
noAddr tcpip.Address = ""
- nicid tcpip.NICID = 1
+ nicID tcpip.NICID = 1
)
localAddr := tcpip.Address([]byte{localAddrByte})
@@ -720,7 +1109,7 @@ func TestEndpointExpiration(t *testing.T) {
})
ep := channel.New(10, defaultMTU, "")
- if err := s.CreateNIC(nicid, ep); err != nil {
+ if err := s.CreateNIC(nicID, ep); err != nil {
t.Fatal("CreateNIC failed:", err)
}
@@ -734,16 +1123,16 @@ func TestEndpointExpiration(t *testing.T) {
fakeNet := s.NetworkProtocolInstance(fakeNetNumber).(*fakeNetworkProtocol)
buf := buffer.NewView(30)
- buf[0] = localAddrByte
+ buf[dstAddrOffset] = localAddrByte
if promiscuous {
- if err := s.SetPromiscuousMode(nicid, true); err != nil {
+ if err := s.SetPromiscuousMode(nicID, true); err != nil {
t.Fatal("SetPromiscuousMode failed:", err)
}
}
if spoofing {
- if err := s.SetSpoofing(nicid, true); err != nil {
+ if err := s.SetSpoofing(nicID, true); err != nil {
t.Fatal("SetSpoofing failed:", err)
}
}
@@ -751,7 +1140,7 @@ func TestEndpointExpiration(t *testing.T) {
// 1. No Address yet, send should only work for spoofing, receive for
// promiscuous mode.
//-----------------------
- verifyAddress(t, s, nicid, noAddr)
+ verifyAddress(t, s, nicID, noAddr)
if promiscuous {
testRecv(t, fakeNet, localAddrByte, ep, buf)
} else {
@@ -766,20 +1155,20 @@ func TestEndpointExpiration(t *testing.T) {
// 2. Add Address, everything should work.
//-----------------------
- if err := s.AddAddress(nicid, fakeNetNumber, localAddr); err != nil {
+ if err := s.AddAddress(nicID, fakeNetNumber, localAddr); err != nil {
t.Fatal("AddAddress failed:", err)
}
- verifyAddress(t, s, nicid, localAddr)
+ verifyAddress(t, s, nicID, localAddr)
testRecv(t, fakeNet, localAddrByte, ep, buf)
testSendTo(t, s, remoteAddr, ep, nil)
// 3. Remove the address, send should only work for spoofing, receive
// for promiscuous mode.
//-----------------------
- if err := s.RemoveAddress(nicid, localAddr); err != nil {
+ if err := s.RemoveAddress(nicID, localAddr); err != nil {
t.Fatal("RemoveAddress failed:", err)
}
- verifyAddress(t, s, nicid, noAddr)
+ verifyAddress(t, s, nicID, noAddr)
if promiscuous {
testRecv(t, fakeNet, localAddrByte, ep, buf)
} else {
@@ -794,10 +1183,10 @@ func TestEndpointExpiration(t *testing.T) {
// 4. Add Address back, everything should work again.
//-----------------------
- if err := s.AddAddress(nicid, fakeNetNumber, localAddr); err != nil {
+ if err := s.AddAddress(nicID, fakeNetNumber, localAddr); err != nil {
t.Fatal("AddAddress failed:", err)
}
- verifyAddress(t, s, nicid, localAddr)
+ verifyAddress(t, s, nicID, localAddr)
testRecv(t, fakeNet, localAddrByte, ep, buf)
testSendTo(t, s, remoteAddr, ep, nil)
@@ -815,10 +1204,10 @@ func TestEndpointExpiration(t *testing.T) {
// 6. Remove the address. Send should only work for spoofing, receive
// for promiscuous mode.
//-----------------------
- if err := s.RemoveAddress(nicid, localAddr); err != nil {
+ if err := s.RemoveAddress(nicID, localAddr); err != nil {
t.Fatal("RemoveAddress failed:", err)
}
- verifyAddress(t, s, nicid, noAddr)
+ verifyAddress(t, s, nicID, noAddr)
if promiscuous {
testRecv(t, fakeNet, localAddrByte, ep, buf)
} else {
@@ -834,10 +1223,10 @@ func TestEndpointExpiration(t *testing.T) {
// 7. Add Address back, everything should work again.
//-----------------------
- if err := s.AddAddress(nicid, fakeNetNumber, localAddr); err != nil {
+ if err := s.AddAddress(nicID, fakeNetNumber, localAddr); err != nil {
t.Fatal("AddAddress failed:", err)
}
- verifyAddress(t, s, nicid, localAddr)
+ verifyAddress(t, s, nicID, localAddr)
testRecv(t, fakeNet, localAddrByte, ep, buf)
testSendTo(t, s, remoteAddr, ep, nil)
testSend(t, r, ep, nil)
@@ -845,17 +1234,17 @@ func TestEndpointExpiration(t *testing.T) {
// 8. Remove the route, sendTo/recv should still work.
//-----------------------
r.Release()
- verifyAddress(t, s, nicid, localAddr)
+ verifyAddress(t, s, nicID, localAddr)
testRecv(t, fakeNet, localAddrByte, ep, buf)
testSendTo(t, s, remoteAddr, ep, nil)
// 9. Remove the address. Send should only work for spoofing, receive
// for promiscuous mode.
//-----------------------
- if err := s.RemoveAddress(nicid, localAddr); err != nil {
+ if err := s.RemoveAddress(nicID, localAddr); err != nil {
t.Fatal("RemoveAddress failed:", err)
}
- verifyAddress(t, s, nicid, noAddr)
+ verifyAddress(t, s, nicID, noAddr)
if promiscuous {
testRecv(t, fakeNet, localAddrByte, ep, buf)
} else {
@@ -897,7 +1286,7 @@ func TestPromiscuousMode(t *testing.T) {
// Write a packet, and check that it doesn't get delivered as we don't
// have a matching endpoint.
const localAddrByte byte = 0x01
- buf[0] = localAddrByte
+ buf[dstAddrOffset] = localAddrByte
testFailingRecv(t, fakeNet, localAddrByte, ep, buf)
// Set promiscuous mode, then check that packet is delivered.
@@ -1071,19 +1460,19 @@ func TestOutgoingBroadcastWithEmptyRouteTable(t *testing.T) {
protoAddr := tcpip.ProtocolAddress{Protocol: fakeNetNumber, AddressWithPrefix: tcpip.AddressWithPrefix{header.IPv4Any, 0}}
if err := s.AddProtocolAddress(1, protoAddr); err != nil {
- t.Fatalf("AddProtocolAddress(1, %s) failed: %s", protoAddr, err)
+ t.Fatalf("AddProtocolAddress(1, %v) failed: %v", protoAddr, err)
}
r, err := s.FindRoute(1, header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, false /* multicastLoop */)
if err != nil {
- t.Fatalf("FindRoute(1, %s, %s, %d) failed: %s", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err)
+ t.Fatalf("FindRoute(1, %v, %v, %d) failed: %v", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err)
}
if err := verifyRoute(r, stack.Route{LocalAddress: header.IPv4Any, RemoteAddress: header.IPv4Broadcast}); err != nil {
- t.Errorf("FindRoute(1, %s, %s, %d) returned unexpected Route: %s)", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err)
+ t.Errorf("FindRoute(1, %v, %v, %d) returned unexpected Route: %v", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err)
}
// If the NIC doesn't exist, it won't work.
if _, err := s.FindRoute(2, header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, false /* multicastLoop */); err != tcpip.ErrNetworkUnreachable {
- t.Fatalf("got FindRoute(2, %s, %s, %d) = %s want = %s", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err, tcpip.ErrNetworkUnreachable)
+ t.Fatalf("got FindRoute(2, %v, %v, %d) = %v want = %v", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err, tcpip.ErrNetworkUnreachable)
}
}
@@ -1109,12 +1498,12 @@ func TestOutgoingBroadcastWithRouteTable(t *testing.T) {
}
nic1ProtoAddr := tcpip.ProtocolAddress{fakeNetNumber, nic1Addr}
if err := s.AddProtocolAddress(1, nic1ProtoAddr); err != nil {
- t.Fatalf("AddProtocolAddress(1, %s) failed: %s", nic1ProtoAddr, err)
+ t.Fatalf("AddProtocolAddress(1, %v) failed: %v", nic1ProtoAddr, err)
}
nic2ProtoAddr := tcpip.ProtocolAddress{fakeNetNumber, nic2Addr}
if err := s.AddProtocolAddress(2, nic2ProtoAddr); err != nil {
- t.Fatalf("AddAddress(2, %s) failed: %s", nic2ProtoAddr, err)
+ t.Fatalf("AddAddress(2, %v) failed: %v", nic2ProtoAddr, err)
}
// Set the initial route table.
@@ -1129,10 +1518,10 @@ func TestOutgoingBroadcastWithRouteTable(t *testing.T) {
// When an interface is given, the route for a broadcast goes through it.
r, err := s.FindRoute(1, nic1Addr.Address, header.IPv4Broadcast, fakeNetNumber, false /* multicastLoop */)
if err != nil {
- t.Fatalf("FindRoute(1, %s, %s, %d) failed: %s", nic1Addr.Address, header.IPv4Broadcast, fakeNetNumber, err)
+ t.Fatalf("FindRoute(1, %v, %v, %d) failed: %v", nic1Addr.Address, header.IPv4Broadcast, fakeNetNumber, err)
}
if err := verifyRoute(r, stack.Route{LocalAddress: nic1Addr.Address, RemoteAddress: header.IPv4Broadcast}); err != nil {
- t.Errorf("FindRoute(1, %s, %s, %d) returned unexpected Route: %s)", nic1Addr.Address, header.IPv4Broadcast, fakeNetNumber, err)
+ t.Errorf("FindRoute(1, %v, %v, %d) returned unexpected Route: %v", nic1Addr.Address, header.IPv4Broadcast, fakeNetNumber, err)
}
// When an interface is not given, it consults the route table.
@@ -1254,149 +1643,6 @@ func TestMulticastOrIPv6LinkLocalNeedsNoRoute(t *testing.T) {
}
}
-// Add a range of addresses, then check that a packet is delivered.
-func TestAddressRangeAcceptsMatchingPacket(t *testing.T) {
- s := stack.New(stack.Options{
- NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()},
- })
-
- ep := channel.New(10, defaultMTU, "")
- if err := s.CreateNIC(1, ep); err != nil {
- t.Fatal("CreateNIC failed:", err)
- }
-
- {
- subnet, err := tcpip.NewSubnet("\x00", "\x00")
- if err != nil {
- t.Fatal(err)
- }
- s.SetRouteTable([]tcpip.Route{{Destination: subnet, Gateway: "\x00", NIC: 1}})
- }
-
- fakeNet := s.NetworkProtocolInstance(fakeNetNumber).(*fakeNetworkProtocol)
-
- buf := buffer.NewView(30)
-
- const localAddrByte byte = 0x01
- buf[0] = localAddrByte
- subnet, err := tcpip.NewSubnet(tcpip.Address("\x00"), tcpip.AddressMask("\xF0"))
- if err != nil {
- t.Fatal("NewSubnet failed:", err)
- }
- if err := s.AddAddressRange(1, fakeNetNumber, subnet); err != nil {
- t.Fatal("AddAddressRange failed:", err)
- }
-
- testRecv(t, fakeNet, localAddrByte, ep, buf)
-}
-
-func testNicForAddressRange(t *testing.T, nicID tcpip.NICID, s *stack.Stack, subnet tcpip.Subnet, rangeExists bool) {
- t.Helper()
-
- // Loop over all addresses and check them.
- numOfAddresses := 1 << uint(8-subnet.Prefix())
- if numOfAddresses < 1 || numOfAddresses > 255 {
- t.Fatalf("got numOfAddresses = %d, want = [1 .. 255] (subnet=%s)", numOfAddresses, subnet)
- }
-
- addrBytes := []byte(subnet.ID())
- for i := 0; i < numOfAddresses; i++ {
- addr := tcpip.Address(addrBytes)
- wantNicID := nicID
- // The subnet and broadcast addresses are skipped.
- if !rangeExists || addr == subnet.ID() || addr == subnet.Broadcast() {
- wantNicID = 0
- }
- if gotNicID := s.CheckLocalAddress(0, fakeNetNumber, addr); gotNicID != wantNicID {
- t.Errorf("got CheckLocalAddress(0, %d, %s) = %d, want = %d", fakeNetNumber, addr, gotNicID, wantNicID)
- }
- addrBytes[0]++
- }
-
- // Trying the next address should always fail since it is outside the range.
- if gotNicID := s.CheckLocalAddress(0, fakeNetNumber, tcpip.Address(addrBytes)); gotNicID != 0 {
- t.Errorf("got CheckLocalAddress(0, %d, %s) = %d, want = %d", fakeNetNumber, tcpip.Address(addrBytes), gotNicID, 0)
- }
-}
-
-// Set a range of addresses, then remove it again, and check at each step that
-// CheckLocalAddress returns the correct NIC for each address or zero if not
-// existent.
-func TestCheckLocalAddressForSubnet(t *testing.T) {
- const nicID tcpip.NICID = 1
- s := stack.New(stack.Options{
- NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()},
- })
-
- ep := channel.New(10, defaultMTU, "")
- if err := s.CreateNIC(nicID, ep); err != nil {
- t.Fatal("CreateNIC failed:", err)
- }
-
- {
- subnet, err := tcpip.NewSubnet("\x00", "\x00")
- if err != nil {
- t.Fatal(err)
- }
- s.SetRouteTable([]tcpip.Route{{Destination: subnet, Gateway: "\x00", NIC: nicID}})
- }
-
- subnet, err := tcpip.NewSubnet(tcpip.Address("\xa0"), tcpip.AddressMask("\xf0"))
- if err != nil {
- t.Fatal("NewSubnet failed:", err)
- }
-
- testNicForAddressRange(t, nicID, s, subnet, false /* rangeExists */)
-
- if err := s.AddAddressRange(nicID, fakeNetNumber, subnet); err != nil {
- t.Fatal("AddAddressRange failed:", err)
- }
-
- testNicForAddressRange(t, nicID, s, subnet, true /* rangeExists */)
-
- if err := s.RemoveAddressRange(nicID, subnet); err != nil {
- t.Fatal("RemoveAddressRange failed:", err)
- }
-
- testNicForAddressRange(t, nicID, s, subnet, false /* rangeExists */)
-}
-
-// Set a range of addresses, then send a packet to a destination outside the
-// range and then check it doesn't get delivered.
-func TestAddressRangeRejectsNonmatchingPacket(t *testing.T) {
- s := stack.New(stack.Options{
- NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()},
- })
-
- ep := channel.New(10, defaultMTU, "")
- if err := s.CreateNIC(1, ep); err != nil {
- t.Fatal("CreateNIC failed:", err)
- }
-
- {
- subnet, err := tcpip.NewSubnet("\x00", "\x00")
- if err != nil {
- t.Fatal(err)
- }
- s.SetRouteTable([]tcpip.Route{{Destination: subnet, Gateway: "\x00", NIC: 1}})
- }
-
- fakeNet := s.NetworkProtocolInstance(fakeNetNumber).(*fakeNetworkProtocol)
-
- buf := buffer.NewView(30)
-
- const localAddrByte byte = 0x01
- buf[0] = localAddrByte
- subnet, err := tcpip.NewSubnet(tcpip.Address("\x10"), tcpip.AddressMask("\xF0"))
- if err != nil {
- t.Fatal("NewSubnet failed:", err)
- }
- if err := s.AddAddressRange(1, fakeNetNumber, subnet); err != nil {
- t.Fatal("AddAddressRange failed:", err)
- }
- testFailingRecv(t, fakeNet, localAddrByte, ep, buf)
-}
-
func TestNetworkOptions(t *testing.T) {
s := stack.New(stack.Options{
NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()},
@@ -1440,56 +1686,6 @@ func TestNetworkOptions(t *testing.T) {
}
}
-func stackContainsAddressRange(s *stack.Stack, id tcpip.NICID, addrRange tcpip.Subnet) bool {
- ranges, ok := s.NICAddressRanges()[id]
- if !ok {
- return false
- }
- for _, r := range ranges {
- if r == addrRange {
- return true
- }
- }
- return false
-}
-
-func TestAddresRangeAddRemove(t *testing.T) {
- s := stack.New(stack.Options{
- NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()},
- })
- ep := channel.New(10, defaultMTU, "")
- if err := s.CreateNIC(1, ep); err != nil {
- t.Fatal("CreateNIC failed:", err)
- }
-
- addr := tcpip.Address("\x01\x01\x01\x01")
- mask := tcpip.AddressMask(strings.Repeat("\xff", len(addr)))
- addrRange, err := tcpip.NewSubnet(addr, mask)
- if err != nil {
- t.Fatal("NewSubnet failed:", err)
- }
-
- if got, want := stackContainsAddressRange(s, 1, addrRange), false; got != want {
- t.Fatalf("got stackContainsAddressRange(...) = %t, want = %t", got, want)
- }
-
- if err := s.AddAddressRange(1, fakeNetNumber, addrRange); err != nil {
- t.Fatal("AddAddressRange failed:", err)
- }
-
- if got, want := stackContainsAddressRange(s, 1, addrRange), true; got != want {
- t.Fatalf("got stackContainsAddressRange(...) = %t, want = %t", got, want)
- }
-
- if err := s.RemoveAddressRange(1, addrRange); err != nil {
- t.Fatal("RemoveAddressRange failed:", err)
- }
-
- if got, want := stackContainsAddressRange(s, 1, addrRange), false; got != want {
- t.Fatalf("got stackContainsAddressRange(...) = %t, want = %t", got, want)
- }
-}
-
func TestGetMainNICAddressAddPrimaryNonPrimary(t *testing.T) {
for _, addrLen := range []int{4, 16} {
t.Run(fmt.Sprintf("addrLen=%d", addrLen), func(t *testing.T) {
@@ -1648,12 +1844,12 @@ func verifyAddresses(t *testing.T, expectedAddresses, gotAddresses []tcpip.Proto
}
func TestAddAddress(t *testing.T) {
- const nicid = 1
+ const nicID = 1
s := stack.New(stack.Options{
NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()},
})
ep := channel.New(10, defaultMTU, "")
- if err := s.CreateNIC(nicid, ep); err != nil {
+ if err := s.CreateNIC(nicID, ep); err != nil {
t.Fatal("CreateNIC failed:", err)
}
@@ -1661,7 +1857,7 @@ func TestAddAddress(t *testing.T) {
expectedAddresses := make([]tcpip.ProtocolAddress, 0, 2)
for _, addrLen := range []int{4, 16} {
address := addrGen.next(addrLen)
- if err := s.AddAddress(nicid, fakeNetNumber, address); err != nil {
+ if err := s.AddAddress(nicID, fakeNetNumber, address); err != nil {
t.Fatalf("AddAddress(address=%s) failed: %s", address, err)
}
expectedAddresses = append(expectedAddresses, tcpip.ProtocolAddress{
@@ -1670,17 +1866,17 @@ func TestAddAddress(t *testing.T) {
})
}
- gotAddresses := s.AllAddresses()[nicid]
+ gotAddresses := s.AllAddresses()[nicID]
verifyAddresses(t, expectedAddresses, gotAddresses)
}
func TestAddProtocolAddress(t *testing.T) {
- const nicid = 1
+ const nicID = 1
s := stack.New(stack.Options{
NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()},
})
ep := channel.New(10, defaultMTU, "")
- if err := s.CreateNIC(nicid, ep); err != nil {
+ if err := s.CreateNIC(nicID, ep); err != nil {
t.Fatal("CreateNIC failed:", err)
}
@@ -1697,24 +1893,24 @@ func TestAddProtocolAddress(t *testing.T) {
PrefixLen: prefixLen,
},
}
- if err := s.AddProtocolAddress(nicid, protocolAddress); err != nil {
+ if err := s.AddProtocolAddress(nicID, protocolAddress); err != nil {
t.Errorf("AddProtocolAddress(%+v) failed: %s", protocolAddress, err)
}
expectedAddresses = append(expectedAddresses, protocolAddress)
}
}
- gotAddresses := s.AllAddresses()[nicid]
+ gotAddresses := s.AllAddresses()[nicID]
verifyAddresses(t, expectedAddresses, gotAddresses)
}
func TestAddAddressWithOptions(t *testing.T) {
- const nicid = 1
+ const nicID = 1
s := stack.New(stack.Options{
NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()},
})
ep := channel.New(10, defaultMTU, "")
- if err := s.CreateNIC(nicid, ep); err != nil {
+ if err := s.CreateNIC(nicID, ep); err != nil {
t.Fatal("CreateNIC failed:", err)
}
@@ -1725,7 +1921,7 @@ func TestAddAddressWithOptions(t *testing.T) {
for _, addrLen := range addrLenRange {
for _, behavior := range behaviorRange {
address := addrGen.next(addrLen)
- if err := s.AddAddressWithOptions(nicid, fakeNetNumber, address, behavior); err != nil {
+ if err := s.AddAddressWithOptions(nicID, fakeNetNumber, address, behavior); err != nil {
t.Fatalf("AddAddressWithOptions(address=%s, behavior=%d) failed: %s", address, behavior, err)
}
expectedAddresses = append(expectedAddresses, tcpip.ProtocolAddress{
@@ -1735,17 +1931,17 @@ func TestAddAddressWithOptions(t *testing.T) {
}
}
- gotAddresses := s.AllAddresses()[nicid]
+ gotAddresses := s.AllAddresses()[nicID]
verifyAddresses(t, expectedAddresses, gotAddresses)
}
func TestAddProtocolAddressWithOptions(t *testing.T) {
- const nicid = 1
+ const nicID = 1
s := stack.New(stack.Options{
NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()},
})
ep := channel.New(10, defaultMTU, "")
- if err := s.CreateNIC(nicid, ep); err != nil {
+ if err := s.CreateNIC(nicID, ep); err != nil {
t.Fatal("CreateNIC failed:", err)
}
@@ -1764,7 +1960,7 @@ func TestAddProtocolAddressWithOptions(t *testing.T) {
PrefixLen: prefixLen,
},
}
- if err := s.AddProtocolAddressWithOptions(nicid, protocolAddress, behavior); err != nil {
+ if err := s.AddProtocolAddressWithOptions(nicID, protocolAddress, behavior); err != nil {
t.Fatalf("AddProtocolAddressWithOptions(%+v, %d) failed: %s", protocolAddress, behavior, err)
}
expectedAddresses = append(expectedAddresses, protocolAddress)
@@ -1772,10 +1968,95 @@ func TestAddProtocolAddressWithOptions(t *testing.T) {
}
}
- gotAddresses := s.AllAddresses()[nicid]
+ gotAddresses := s.AllAddresses()[nicID]
verifyAddresses(t, expectedAddresses, gotAddresses)
}
+func TestCreateNICWithOptions(t *testing.T) {
+ type callArgsAndExpect struct {
+ nicID tcpip.NICID
+ opts stack.NICOptions
+ err *tcpip.Error
+ }
+
+ tests := []struct {
+ desc string
+ calls []callArgsAndExpect
+ }{
+ {
+ desc: "DuplicateNICID",
+ calls: []callArgsAndExpect{
+ {
+ nicID: tcpip.NICID(1),
+ opts: stack.NICOptions{Name: "eth1"},
+ err: nil,
+ },
+ {
+ nicID: tcpip.NICID(1),
+ opts: stack.NICOptions{Name: "eth2"},
+ err: tcpip.ErrDuplicateNICID,
+ },
+ },
+ },
+ {
+ desc: "DuplicateName",
+ calls: []callArgsAndExpect{
+ {
+ nicID: tcpip.NICID(1),
+ opts: stack.NICOptions{Name: "lo"},
+ err: nil,
+ },
+ {
+ nicID: tcpip.NICID(2),
+ opts: stack.NICOptions{Name: "lo"},
+ err: tcpip.ErrDuplicateNICID,
+ },
+ },
+ },
+ {
+ desc: "Unnamed",
+ calls: []callArgsAndExpect{
+ {
+ nicID: tcpip.NICID(1),
+ opts: stack.NICOptions{},
+ err: nil,
+ },
+ {
+ nicID: tcpip.NICID(2),
+ opts: stack.NICOptions{},
+ err: nil,
+ },
+ },
+ },
+ {
+ desc: "UnnamedDuplicateNICID",
+ calls: []callArgsAndExpect{
+ {
+ nicID: tcpip.NICID(1),
+ opts: stack.NICOptions{},
+ err: nil,
+ },
+ {
+ nicID: tcpip.NICID(1),
+ opts: stack.NICOptions{},
+ err: tcpip.ErrDuplicateNICID,
+ },
+ },
+ },
+ }
+ for _, test := range tests {
+ t.Run(test.desc, func(t *testing.T) {
+ s := stack.New(stack.Options{})
+ ep := channel.New(0, 0, tcpip.LinkAddress("\x00\x00\x00\x00\x00\x00"))
+ for _, call := range test.calls {
+ if got, want := s.CreateNICWithOptions(call.nicID, ep, call.opts), call.err; got != want {
+ t.Fatalf("CreateNICWithOptions(%v, _, %+v) = %v, want %v", call.nicID, call.opts, got, want)
+ }
+ }
+ })
+ }
+}
+
func TestNICStats(t *testing.T) {
s := stack.New(stack.Options{
NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()},
@@ -1798,7 +2079,9 @@ func TestNICStats(t *testing.T) {
// Send a packet to address 1.
buf := buffer.NewView(30)
- ep1.Inject(fakeNetNumber, buf.ToVectorisedView())
+ ep1.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ }))
if got, want := s.NICInfo()[1].Stats.Rx.Packets.Value(), uint64(1); got != want {
t.Errorf("got Rx.Packets.Value() = %d, want = %d", got, want)
}
@@ -1823,150 +2106,386 @@ func TestNICStats(t *testing.T) {
}
func TestNICForwarding(t *testing.T) {
- // Create a stack with the fake network protocol, two NICs, each with
- // an address.
- s := stack.New(stack.Options{
- NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()},
- })
- s.SetForwarding(fakeNetNumber, true)
+ const nicID1 = 1
+ const nicID2 = 2
+ const dstAddr = tcpip.Address("\x03")
- ep1 := channel.New(10, defaultMTU, "")
- if err := s.CreateNIC(1, ep1); err != nil {
- t.Fatal("CreateNIC #1 failed:", err)
- }
- if err := s.AddAddress(1, fakeNetNumber, "\x01"); err != nil {
- t.Fatal("AddAddress #1 failed:", err)
+ tests := []struct {
+ name string
+ headerLen uint16
+ }{
+ {
+ name: "Zero header length",
+ },
+ {
+ name: "Non-zero header length",
+ headerLen: 16,
+ },
}
- ep2 := channel.New(10, defaultMTU, "")
- if err := s.CreateNIC(2, ep2); err != nil {
- t.Fatal("CreateNIC #2 failed:", err)
- }
- if err := s.AddAddress(2, fakeNetNumber, "\x02"); err != nil {
- t.Fatal("AddAddress #2 failed:", err)
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()},
+ })
+ s.SetForwarding(fakeNetNumber, true)
+
+ ep1 := channel.New(10, defaultMTU, "")
+ if err := s.CreateNIC(nicID1, ep1); err != nil {
+ t.Fatalf("CreateNIC(%d, _): %s", nicID1, err)
+ }
+ if err := s.AddAddress(nicID1, fakeNetNumber, "\x01"); err != nil {
+ t.Fatalf("AddAddress(%d, %d, 0x01): %s", nicID1, fakeNetNumber, err)
+ }
+
+ ep2 := channelLinkWithHeaderLength{
+ Endpoint: channel.New(10, defaultMTU, ""),
+ headerLength: test.headerLen,
+ }
+ if err := s.CreateNIC(nicID2, &ep2); err != nil {
+ t.Fatalf("CreateNIC(%d, _): %s", nicID2, err)
+ }
+ if err := s.AddAddress(nicID2, fakeNetNumber, "\x02"); err != nil {
+ t.Fatalf("AddAddress(%d, %d, 0x02): %s", nicID2, fakeNetNumber, err)
+ }
+
+ // Route all packets to dstAddr to NIC 2.
+ {
+ subnet, err := tcpip.NewSubnet(dstAddr, "\xff")
+ if err != nil {
+ t.Fatal(err)
+ }
+ s.SetRouteTable([]tcpip.Route{{Destination: subnet, Gateway: "\x00", NIC: nicID2}})
+ }
+
+ // Send a packet to dstAddr.
+ buf := buffer.NewView(30)
+ buf[dstAddrOffset] = dstAddr[0]
+ ep1.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ }))
+
+ pkt, ok := ep2.Read()
+ if !ok {
+ t.Fatal("packet not forwarded")
+ }
+
+ // Test that the link's MaxHeaderLength is honoured.
+ if capacity, want := pkt.Pkt.AvailableHeaderBytes(), int(test.headerLen); capacity != want {
+ t.Errorf("got LinkHeader.AvailableLength() = %d, want = %d", capacity, want)
+ }
+
+ // Test that forwarding increments Tx stats correctly.
+ if got, want := s.NICInfo()[nicID2].Stats.Tx.Packets.Value(), uint64(1); got != want {
+ t.Errorf("got Tx.Packets.Value() = %d, want = %d", got, want)
+ }
+
+ if got, want := s.NICInfo()[nicID2].Stats.Tx.Bytes.Value(), uint64(len(buf)); got != want {
+ t.Errorf("got Tx.Bytes.Value() = %d, want = %d", got, want)
+ }
+ })
}
+}
- // Route all packets to address 3 to NIC 2.
- {
- subnet, err := tcpip.NewSubnet("\x03", "\xff")
- if err != nil {
- t.Fatal(err)
- }
- s.SetRouteTable([]tcpip.Route{{Destination: subnet, Gateway: "\x00", NIC: 2}})
+// TestNICContextPreservation tests that you can read out via stack.NICInfo the
+// Context data you pass via NICContext.Context in stack.CreateNICWithOptions.
+func TestNICContextPreservation(t *testing.T) {
+ var ctx *int
+ tests := []struct {
+ name string
+ opts stack.NICOptions
+ want stack.NICContext
+ }{
+ {
+ "context_set",
+ stack.NICOptions{Context: ctx},
+ ctx,
+ },
+ {
+ "context_not_set",
+ stack.NICOptions{},
+ nil,
+ },
+ }
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ s := stack.New(stack.Options{})
+ id := tcpip.NICID(1)
+ ep := channel.New(0, 0, tcpip.LinkAddress("\x00\x00\x00\x00\x00\x00"))
+ if err := s.CreateNICWithOptions(id, ep, test.opts); err != nil {
+ t.Fatalf("got stack.CreateNICWithOptions(%d, %+v, %+v) = %s, want nil", id, ep, test.opts, err)
+ }
+ nicinfos := s.NICInfo()
+ nicinfo, ok := nicinfos[id]
+ if !ok {
+ t.Fatalf("got nicinfos[%d] = _, %t, want _, true; nicinfos = %+v", id, ok, nicinfos)
+ }
+ if got, want := nicinfo.Context == test.want, true; got != want {
+ t.Fatalf("got nicinfo.Context == ctx = %t, want %t; nicinfo.Context = %p, ctx = %p", got, want, nicinfo.Context, test.want)
+ }
+ })
}
+}
- // Send a packet to address 3.
- buf := buffer.NewView(30)
- buf[0] = 3
- ep1.Inject(fakeNetNumber, buf.ToVectorisedView())
+// TestNICAutoGenLinkLocalAddr tests the auto-generation of IPv6 link-local
+// addresses.
+func TestNICAutoGenLinkLocalAddr(t *testing.T) {
+ const nicID = 1
- select {
- case <-ep2.C:
- default:
- t.Fatal("Packet not forwarded")
+ var secretKey [header.OpaqueIIDSecretKeyMinBytes]byte
+ n, err := rand.Read(secretKey[:])
+ if err != nil {
+ t.Fatalf("rand.Read(_): %s", err)
}
-
- // Test that forwarding increments Tx stats correctly.
- if got, want := s.NICInfo()[2].Stats.Tx.Packets.Value(), uint64(1); got != want {
- t.Errorf("got Tx.Packets.Value() = %d, want = %d", got, want)
+ if n != header.OpaqueIIDSecretKeyMinBytes {
+ t.Fatalf("expected rand.Read to read %d bytes, read %d bytes", header.OpaqueIIDSecretKeyMinBytes, n)
}
- if got, want := s.NICInfo()[2].Stats.Tx.Bytes.Value(), uint64(len(buf)); got != want {
- t.Errorf("got Tx.Bytes.Value() = %d, want = %d", got, want)
+ nicNameFunc := func(_ tcpip.NICID, name string) string {
+ return name
}
-}
-// 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
+ name string
+ nicName string
+ autoGen bool
+ linkAddr tcpip.LinkAddress
+ iidOpts stack.OpaqueInterfaceIdentifierOptions
+ shouldGen bool
+ expectedAddr tcpip.Address
}{
{
- "Disabled",
- false,
- linkAddr1,
- false,
+ name: "Disabled",
+ nicName: "nic1",
+ autoGen: false,
+ linkAddr: linkAddr1,
+ shouldGen: false,
+ },
+ {
+ name: "Disabled without OIID options",
+ nicName: "nic1",
+ autoGen: false,
+ linkAddr: linkAddr1,
+ iidOpts: stack.OpaqueInterfaceIdentifierOptions{
+ NICNameFromID: nicNameFunc,
+ SecretKey: secretKey[:],
+ },
+ shouldGen: false,
+ },
+
+ // Tests for EUI64 based addresses.
+ {
+ name: "EUI64 Enabled",
+ autoGen: true,
+ linkAddr: linkAddr1,
+ shouldGen: true,
+ expectedAddr: header.LinkLocalAddr(linkAddr1),
+ },
+ {
+ name: "EUI64 Empty MAC",
+ autoGen: true,
+ shouldGen: false,
},
{
- "Enabled",
- true,
- linkAddr1,
- true,
+ name: "EUI64 Invalid MAC",
+ autoGen: true,
+ linkAddr: "\x01\x02\x03",
+ shouldGen: false,
},
{
- "Nil MAC",
- true,
- tcpip.LinkAddress([]byte(nil)),
- false,
+ name: "EUI64 Multicast MAC",
+ autoGen: true,
+ linkAddr: "\x01\x02\x03\x04\x05\x06",
+ shouldGen: false,
},
{
- "Empty MAC",
- true,
- tcpip.LinkAddress(""),
- false,
+ name: "EUI64 Unspecified MAC",
+ autoGen: true,
+ linkAddr: "\x00\x00\x00\x00\x00\x00",
+ shouldGen: false,
},
+
+ // Tests for Opaque IID based addresses.
+ {
+ name: "OIID Enabled",
+ nicName: "nic1",
+ autoGen: true,
+ linkAddr: linkAddr1,
+ iidOpts: stack.OpaqueInterfaceIdentifierOptions{
+ NICNameFromID: nicNameFunc,
+ SecretKey: secretKey[:],
+ },
+ shouldGen: true,
+ expectedAddr: header.LinkLocalAddrWithOpaqueIID("nic1", 0, secretKey[:]),
+ },
+ // These are all cases where we would not have generated a
+ // link-local address if opaque IIDs were disabled.
{
- "Invalid MAC",
- true,
- tcpip.LinkAddress("\x01\x02\x03"),
- false,
+ name: "OIID Empty MAC and empty nicName",
+ autoGen: true,
+ iidOpts: stack.OpaqueInterfaceIdentifierOptions{
+ NICNameFromID: nicNameFunc,
+ SecretKey: secretKey[:1],
+ },
+ shouldGen: true,
+ expectedAddr: header.LinkLocalAddrWithOpaqueIID("", 0, secretKey[:1]),
},
{
- "Multicast MAC",
- true,
- tcpip.LinkAddress("\x01\x02\x03\x04\x05\x06"),
- false,
+ name: "OIID Invalid MAC",
+ nicName: "test",
+ autoGen: true,
+ linkAddr: "\x01\x02\x03",
+ iidOpts: stack.OpaqueInterfaceIdentifierOptions{
+ NICNameFromID: nicNameFunc,
+ SecretKey: secretKey[:2],
+ },
+ shouldGen: true,
+ expectedAddr: header.LinkLocalAddrWithOpaqueIID("test", 0, secretKey[:2]),
},
{
- "Unspecified MAC",
- true,
- tcpip.LinkAddress("\x00\x00\x00\x00\x00\x00"),
- false,
+ name: "OIID Multicast MAC",
+ nicName: "test2",
+ autoGen: true,
+ linkAddr: "\x01\x02\x03\x04\x05\x06",
+ iidOpts: stack.OpaqueInterfaceIdentifierOptions{
+ NICNameFromID: nicNameFunc,
+ SecretKey: secretKey[:3],
+ },
+ shouldGen: true,
+ expectedAddr: header.LinkLocalAddrWithOpaqueIID("test2", 0, secretKey[:3]),
+ },
+ {
+ name: "OIID Unspecified MAC and nil SecretKey",
+ nicName: "test3",
+ autoGen: true,
+ linkAddr: "\x00\x00\x00\x00\x00\x00",
+ iidOpts: stack.OpaqueInterfaceIdentifierOptions{
+ NICNameFromID: nicNameFunc,
+ },
+ shouldGen: true,
+ expectedAddr: header.LinkLocalAddrWithOpaqueIID("test3", 0, nil),
},
}
for _, test := range tests {
t.Run(test.name, func(t *testing.T) {
+ ndpDisp := ndpDispatcher{
+ autoGenAddrC: make(chan ndpAutoGenAddrEvent, 1),
+ }
opts := stack.Options{
- NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ AutoGenIPv6LinkLocal: test.autoGen,
+ NDPDisp: &ndpDisp,
+ OpaqueIIDOpts: test.iidOpts,
}
- 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(0, 1280, test.linkAddr)
+ s := stack.New(opts)
+ nicOpts := stack.NICOptions{Name: test.nicName, Disabled: true}
+ if err := s.CreateNICWithOptions(nicID, e, nicOpts); err != nil {
+ t.Fatalf("CreateNICWithOptions(%d, _, %+v) = %s", nicID, opts, err)
}
- e := channel.New(10, 1280, test.linkAddr)
- s := stack.New(opts)
- if err := s.CreateNIC(1, e); err != nil {
- t.Fatalf("CreateNIC(_) = %s", err)
+ // A new disabled NIC should not have any address, even if auto generation
+ // was enabled.
+ allStackAddrs := s.AllAddresses()
+ allNICAddrs, ok := allStackAddrs[nicID]
+ if !ok {
+ t.Fatalf("entry for %d missing from allStackAddrs = %+v", nicID, allStackAddrs)
+ }
+ if l := len(allNICAddrs); l != 0 {
+ t.Fatalf("got len(allNICAddrs) = %d, want = 0", l)
}
- addr, err := s.GetMainNICAddress(1, header.IPv6ProtocolNumber)
- if err != nil {
- t.Fatalf("stack.GetMainNICAddress(_, _) err = %s", err)
+ // Enabling the NIC should attempt auto-generation of a link-local
+ // address.
+ if err := s.EnableNIC(nicID); err != nil {
+ t.Fatalf("s.EnableNIC(%d): %s", nicID, err)
}
+ var expectedMainAddr tcpip.AddressWithPrefix
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)
+ expectedMainAddr = tcpip.AddressWithPrefix{
+ Address: test.expectedAddr,
+ PrefixLen: header.IPv6LinkLocalPrefix.PrefixLen,
+ }
+
+ // Should have auto-generated an address and resolved immediately (DAD
+ // is disabled).
+ select {
+ case e := <-ndpDisp.autoGenAddrC:
+ if diff := checkAutoGenAddrEvent(e, expectedMainAddr, newAddr); diff != "" {
+ t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected addr auto gen event")
}
} 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)
+ select {
+ case <-ndpDisp.autoGenAddrC:
+ t.Fatal("unexpectedly auto-generated an address")
+ default:
}
}
+
+ gotMainAddr, err := s.GetMainNICAddress(1, header.IPv6ProtocolNumber)
+ if err != nil {
+ t.Fatalf("stack.GetMainNICAddress(_, _) err = %s", err)
+ }
+ if gotMainAddr != expectedMainAddr {
+ t.Fatalf("got stack.GetMainNICAddress(_, _) = %s, want = %s", gotMainAddr, expectedMainAddr)
+ }
+ })
+ }
+}
+
+// TestNoLinkLocalAutoGenForLoopbackNIC tests that IPv6 link-local addresses are
+// not auto-generated for loopback NICs.
+func TestNoLinkLocalAutoGenForLoopbackNIC(t *testing.T) {
+ const nicID = 1
+ const nicName = "nicName"
+
+ tests := []struct {
+ name string
+ opaqueIIDOpts stack.OpaqueInterfaceIdentifierOptions
+ }{
+ {
+ name: "IID From MAC",
+ opaqueIIDOpts: stack.OpaqueInterfaceIdentifierOptions{},
+ },
+ {
+ name: "Opaque IID",
+ opaqueIIDOpts: stack.OpaqueInterfaceIdentifierOptions{
+ NICNameFromID: func(_ tcpip.NICID, nicName string) string {
+ return nicName
+ },
+ },
+ },
+ }
+
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ opts := stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ AutoGenIPv6LinkLocal: true,
+ OpaqueIIDOpts: test.opaqueIIDOpts,
+ }
+
+ e := loopback.New()
+ s := stack.New(opts)
+ nicOpts := stack.NICOptions{Name: nicName}
+ if err := s.CreateNICWithOptions(nicID, e, nicOpts); err != nil {
+ t.Fatalf("CreateNICWithOptions(%d, _, %+v) = %s", nicID, nicOpts, err)
+ }
+
+ addr, err := s.GetMainNICAddress(nicID, header.IPv6ProtocolNumber)
+ if err != nil {
+ t.Fatalf("stack.GetMainNICAddress(%d, _) err = %s", nicID, err)
+ }
+ if want := (tcpip.AddressWithPrefix{}); addr != want {
+ t.Errorf("got stack.GetMainNICAddress(%d, _) = %s, want = %s", nicID, addr, want)
+ }
})
}
}
@@ -1974,6 +2493,8 @@ func TestNICAutoGenAddr(t *testing.T) {
// TestNICAutoGenAddrDoesDAD tests that the successful auto-generation of IPv6
// link-local addresses will only be assigned after the DAD process resolves.
func TestNICAutoGenAddrDoesDAD(t *testing.T) {
+ const nicID = 1
+
ndpDisp := ndpDispatcher{
dadC: make(chan ndpDADEvent),
}
@@ -1985,20 +2506,20 @@ func TestNICAutoGenAddrDoesDAD(t *testing.T) {
NDPDisp: &ndpDisp,
}
- e := channel.New(10, 1280, linkAddr1)
+ e := channel.New(int(ndpConfigs.DupAddrDetectTransmits), 1280, linkAddr1)
s := stack.New(opts)
- if err := s.CreateNIC(1, e); err != nil {
- t.Fatalf("CreateNIC(_) = %s", err)
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
}
// Address should not be considered bound to the
// NIC yet (DAD ongoing).
- addr, err := s.GetMainNICAddress(1, header.IPv6ProtocolNumber)
+ addr, err := s.GetMainNICAddress(nicID, header.IPv6ProtocolNumber)
if err != nil {
- t.Fatalf("got stack.GetMainNICAddress(_, _) = (_, %v), want = (_, nil)", err)
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (_, %v), want = (_, nil)", nicID, header.IPv6ProtocolNumber, err)
}
if want := (tcpip.AddressWithPrefix{}); addr != want {
- t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want)
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (%s, nil), want = (%s, nil)", nicID, header.IPv6ProtocolNumber, addr, want)
}
linkLocalAddr := header.LinkLocalAddr(linkAddr1)
@@ -2012,25 +2533,16 @@ func TestNICAutoGenAddrDoesDAD(t *testing.T) {
// 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")
+ if diff := checkDADEvent(e, nicID, linkLocalAddr, true, nil); diff != "" {
+ t.Errorf("dad event mismatch (-want +got):\n%s", diff)
}
}
- addr, err = s.GetMainNICAddress(1, header.IPv6ProtocolNumber)
+ addr, err = s.GetMainNICAddress(nicID, header.IPv6ProtocolNumber)
if err != nil {
- t.Fatalf("stack.GetMainNICAddress(_, _) err = %s", err)
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (_, %v), want = (_, nil)", nicID, header.IPv6ProtocolNumber, err)
}
if want := (tcpip.AddressWithPrefix{Address: linkLocalAddr, PrefixLen: header.IPv6LinkLocalPrefix.PrefixLen}); addr != want {
- t.Fatalf("got stack.GetMainNICAddress(_, _) = %s, want = %s", addr, want)
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (%s, nil), want = (%s, nil)", nicID, header.IPv6ProtocolNumber, addr, want)
}
}
@@ -2078,7 +2590,7 @@ func TestNewPEBOnPromotionToPermanent(t *testing.T) {
{
subnet, err := tcpip.NewSubnet("\x00", "\x00")
if err != nil {
- t.Fatalf("NewSubnet failed:", err)
+ t.Fatalf("NewSubnet failed: %v", err)
}
s.SetRouteTable([]tcpip.Route{{Destination: subnet, Gateway: "\x00", NIC: 1}})
}
@@ -2092,11 +2604,11 @@ func TestNewPEBOnPromotionToPermanent(t *testing.T) {
// permanentExpired kind.
r, err := s.FindRoute(1, "\x01", "\x02", fakeNetNumber, false)
if err != nil {
- t.Fatal("FindRoute failed:", err)
+ t.Fatalf("FindRoute failed: %v", err)
}
defer r.Release()
if err := s.RemoveAddress(1, "\x01"); err != nil {
- t.Fatalf("RemoveAddress failed:", err)
+ t.Fatalf("RemoveAddress failed: %v", err)
}
//
@@ -2108,7 +2620,7 @@ func TestNewPEBOnPromotionToPermanent(t *testing.T) {
// 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)
+ t.Fatalf("AddAddressWithOptions failed: %v", err)
}
@@ -2116,7 +2628,7 @@ func TestNewPEBOnPromotionToPermanent(t *testing.T) {
// 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)
+ t.Fatalf("AddAddressWithOptions failed: %v", err)
}
var primaryAddrs []tcpip.Address
for _, pa := range s.NICInfo()[1].ProtocolAddresses {
@@ -2149,11 +2661,11 @@ func TestNewPEBOnPromotionToPermanent(t *testing.T) {
// GetMainNICAddress; else, our original address
// should be returned.
if err := s.RemoveAddress(1, "\x03"); err != nil {
- t.Fatalf("RemoveAddress failed:", err)
+ t.Fatalf("RemoveAddress failed: %v", err)
}
addr, err = s.GetMainNICAddress(1, fakeNetNumber)
if err != nil {
- t.Fatal("s.GetMainNICAddress failed:", err)
+ t.Fatalf("s.GetMainNICAddress failed: %v", err)
}
if ps == stack.NeverPrimaryEndpoint {
if want := (tcpip.AddressWithPrefix{}); addr != want {
@@ -2169,3 +2681,858 @@ func TestNewPEBOnPromotionToPermanent(t *testing.T) {
}
}
}
+
+func TestIPv6SourceAddressSelectionScopeAndSameAddress(t *testing.T) {
+ const (
+ linkLocalAddr1 = tcpip.Address("\xfe\x80\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01")
+ linkLocalAddr2 = tcpip.Address("\xfe\x80\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02")
+ linkLocalMulticastAddr = tcpip.Address("\xff\x02\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01")
+ uniqueLocalAddr1 = tcpip.Address("\xfc\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01")
+ uniqueLocalAddr2 = tcpip.Address("\xfd\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02")
+ globalAddr1 = tcpip.Address("\xa0\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01")
+ globalAddr2 = tcpip.Address("\xa0\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02")
+ nicID = 1
+ lifetimeSeconds = 9999
+ )
+
+ prefix1, _, stableGlobalAddr1 := prefixSubnetAddr(0, linkAddr1)
+ prefix2, _, stableGlobalAddr2 := prefixSubnetAddr(1, linkAddr1)
+
+ var tempIIDHistory [header.IIDSize]byte
+ header.InitialTempIID(tempIIDHistory[:], nil, nicID)
+ tempGlobalAddr1 := header.GenerateTempIPv6SLAACAddr(tempIIDHistory[:], stableGlobalAddr1.Address).Address
+ tempGlobalAddr2 := header.GenerateTempIPv6SLAACAddr(tempIIDHistory[:], stableGlobalAddr2.Address).Address
+
+ // Rule 3 is not tested here, and is instead tested by NDP's AutoGenAddr test.
+ tests := []struct {
+ name string
+ slaacPrefixForTempAddrBeforeNICAddrAdd tcpip.AddressWithPrefix
+ nicAddrs []tcpip.Address
+ slaacPrefixForTempAddrAfterNICAddrAdd tcpip.AddressWithPrefix
+ connectAddr tcpip.Address
+ expectedLocalAddr tcpip.Address
+ }{
+ // Test Rule 1 of RFC 6724 section 5.
+ {
+ name: "Same Global most preferred (last address)",
+ nicAddrs: []tcpip.Address{linkLocalAddr1, uniqueLocalAddr1, globalAddr1},
+ connectAddr: globalAddr1,
+ expectedLocalAddr: globalAddr1,
+ },
+ {
+ name: "Same Global most preferred (first address)",
+ nicAddrs: []tcpip.Address{globalAddr1, linkLocalAddr1, uniqueLocalAddr1},
+ connectAddr: globalAddr1,
+ expectedLocalAddr: globalAddr1,
+ },
+ {
+ name: "Same Link Local most preferred (last address)",
+ nicAddrs: []tcpip.Address{globalAddr1, uniqueLocalAddr1, linkLocalAddr1},
+ connectAddr: linkLocalAddr1,
+ expectedLocalAddr: linkLocalAddr1,
+ },
+ {
+ name: "Same Link Local most preferred (first address)",
+ nicAddrs: []tcpip.Address{linkLocalAddr1, uniqueLocalAddr1, globalAddr1},
+ connectAddr: linkLocalAddr1,
+ expectedLocalAddr: linkLocalAddr1,
+ },
+ {
+ name: "Same Unique Local most preferred (last address)",
+ nicAddrs: []tcpip.Address{uniqueLocalAddr1, globalAddr1, linkLocalAddr1},
+ connectAddr: uniqueLocalAddr1,
+ expectedLocalAddr: uniqueLocalAddr1,
+ },
+ {
+ name: "Same Unique Local most preferred (first address)",
+ nicAddrs: []tcpip.Address{globalAddr1, linkLocalAddr1, uniqueLocalAddr1},
+ connectAddr: uniqueLocalAddr1,
+ expectedLocalAddr: uniqueLocalAddr1,
+ },
+
+ // Test Rule 2 of RFC 6724 section 5.
+ {
+ name: "Global most preferred (last address)",
+ nicAddrs: []tcpip.Address{linkLocalAddr1, uniqueLocalAddr1, globalAddr1},
+ connectAddr: globalAddr2,
+ expectedLocalAddr: globalAddr1,
+ },
+ {
+ name: "Global most preferred (first address)",
+ nicAddrs: []tcpip.Address{globalAddr1, linkLocalAddr1, uniqueLocalAddr1},
+ connectAddr: globalAddr2,
+ expectedLocalAddr: globalAddr1,
+ },
+ {
+ name: "Link Local most preferred (last address)",
+ nicAddrs: []tcpip.Address{globalAddr1, uniqueLocalAddr1, linkLocalAddr1},
+ connectAddr: linkLocalAddr2,
+ expectedLocalAddr: linkLocalAddr1,
+ },
+ {
+ name: "Link Local most preferred (first address)",
+ nicAddrs: []tcpip.Address{linkLocalAddr1, uniqueLocalAddr1, globalAddr1},
+ connectAddr: linkLocalAddr2,
+ expectedLocalAddr: linkLocalAddr1,
+ },
+ {
+ name: "Link Local most preferred for link local multicast (last address)",
+ nicAddrs: []tcpip.Address{globalAddr1, uniqueLocalAddr1, linkLocalAddr1},
+ connectAddr: linkLocalMulticastAddr,
+ expectedLocalAddr: linkLocalAddr1,
+ },
+ {
+ name: "Link Local most preferred for link local multicast (first address)",
+ nicAddrs: []tcpip.Address{linkLocalAddr1, uniqueLocalAddr1, globalAddr1},
+ connectAddr: linkLocalMulticastAddr,
+ expectedLocalAddr: linkLocalAddr1,
+ },
+ {
+ name: "Unique Local most preferred (last address)",
+ nicAddrs: []tcpip.Address{uniqueLocalAddr1, globalAddr1, linkLocalAddr1},
+ connectAddr: uniqueLocalAddr2,
+ expectedLocalAddr: uniqueLocalAddr1,
+ },
+ {
+ name: "Unique Local most preferred (first address)",
+ nicAddrs: []tcpip.Address{globalAddr1, linkLocalAddr1, uniqueLocalAddr1},
+ connectAddr: uniqueLocalAddr2,
+ expectedLocalAddr: uniqueLocalAddr1,
+ },
+
+ // Test Rule 7 of RFC 6724 section 5.
+ {
+ name: "Temp Global most preferred (last address)",
+ slaacPrefixForTempAddrBeforeNICAddrAdd: prefix1,
+ nicAddrs: []tcpip.Address{linkLocalAddr1, uniqueLocalAddr1, globalAddr1},
+ connectAddr: globalAddr2,
+ expectedLocalAddr: tempGlobalAddr1,
+ },
+ {
+ name: "Temp Global most preferred (first address)",
+ nicAddrs: []tcpip.Address{linkLocalAddr1, uniqueLocalAddr1, globalAddr1},
+ slaacPrefixForTempAddrAfterNICAddrAdd: prefix1,
+ connectAddr: globalAddr2,
+ expectedLocalAddr: tempGlobalAddr1,
+ },
+
+ // Test returning the endpoint that is closest to the front when
+ // candidate addresses are "equal" from the perspective of RFC 6724
+ // section 5.
+ {
+ name: "Unique Local for Global",
+ nicAddrs: []tcpip.Address{linkLocalAddr1, uniqueLocalAddr1, uniqueLocalAddr2},
+ connectAddr: globalAddr2,
+ expectedLocalAddr: uniqueLocalAddr1,
+ },
+ {
+ name: "Link Local for Global",
+ nicAddrs: []tcpip.Address{linkLocalAddr1, linkLocalAddr2},
+ connectAddr: globalAddr2,
+ expectedLocalAddr: linkLocalAddr1,
+ },
+ {
+ name: "Link Local for Unique Local",
+ nicAddrs: []tcpip.Address{linkLocalAddr1, linkLocalAddr2},
+ connectAddr: uniqueLocalAddr2,
+ expectedLocalAddr: linkLocalAddr1,
+ },
+ {
+ name: "Temp Global for Global",
+ slaacPrefixForTempAddrBeforeNICAddrAdd: prefix1,
+ slaacPrefixForTempAddrAfterNICAddrAdd: prefix2,
+ connectAddr: globalAddr1,
+ expectedLocalAddr: tempGlobalAddr2,
+ },
+ }
+
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ HandleRAs: true,
+ AutoGenGlobalAddresses: true,
+ AutoGenTempGlobalAddresses: true,
+ },
+ NDPDisp: &ndpDispatcher{},
+ })
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+ s.SetRouteTable([]tcpip.Route{{
+ Destination: header.IPv6EmptySubnet,
+ Gateway: llAddr3,
+ NIC: nicID,
+ }})
+ s.AddLinkAddress(nicID, llAddr3, linkAddr3)
+
+ if test.slaacPrefixForTempAddrBeforeNICAddrAdd != (tcpip.AddressWithPrefix{}) {
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr3, 0, test.slaacPrefixForTempAddrBeforeNICAddrAdd, true, true, lifetimeSeconds, lifetimeSeconds))
+ }
+
+ for _, a := range test.nicAddrs {
+ if err := s.AddAddress(nicID, ipv6.ProtocolNumber, a); err != nil {
+ t.Errorf("s.AddAddress(%d, %d, %s): %s", nicID, ipv6.ProtocolNumber, a, err)
+ }
+ }
+
+ if test.slaacPrefixForTempAddrAfterNICAddrAdd != (tcpip.AddressWithPrefix{}) {
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr3, 0, test.slaacPrefixForTempAddrAfterNICAddrAdd, true, true, lifetimeSeconds, lifetimeSeconds))
+ }
+
+ if t.Failed() {
+ t.FailNow()
+ }
+
+ if got := addrForNewConnectionTo(t, s, tcpip.FullAddress{Addr: test.connectAddr, NIC: nicID, Port: 1234}); got != test.expectedLocalAddr {
+ t.Errorf("got local address = %s, want = %s", got, test.expectedLocalAddr)
+ }
+ })
+ }
+}
+
+func TestAddRemoveIPv4BroadcastAddressOnNICEnableDisable(t *testing.T) {
+ const nicID = 1
+ broadcastAddr := tcpip.ProtocolAddress{
+ Protocol: header.IPv4ProtocolNumber,
+ AddressWithPrefix: tcpip.AddressWithPrefix{
+ Address: header.IPv4Broadcast,
+ PrefixLen: 32,
+ },
+ }
+
+ e := loopback.New()
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()},
+ })
+ nicOpts := stack.NICOptions{Disabled: true}
+ if err := s.CreateNICWithOptions(nicID, e, nicOpts); err != nil {
+ t.Fatalf("CreateNIC(%d, _, %+v) = %s", nicID, nicOpts, err)
+ }
+
+ {
+ allStackAddrs := s.AllAddresses()
+ if allNICAddrs, ok := allStackAddrs[nicID]; !ok {
+ t.Fatalf("entry for %d missing from allStackAddrs = %+v", nicID, allStackAddrs)
+ } else if containsAddr(allNICAddrs, broadcastAddr) {
+ t.Fatalf("got allNICAddrs = %+v, don't want = %+v", allNICAddrs, broadcastAddr)
+ }
+ }
+
+ // Enabling the NIC should add the IPv4 broadcast address.
+ if err := s.EnableNIC(nicID); err != nil {
+ t.Fatalf("s.EnableNIC(%d): %s", nicID, err)
+ }
+
+ {
+ allStackAddrs := s.AllAddresses()
+ if allNICAddrs, ok := allStackAddrs[nicID]; !ok {
+ t.Fatalf("entry for %d missing from allStackAddrs = %+v", nicID, allStackAddrs)
+ } else if !containsAddr(allNICAddrs, broadcastAddr) {
+ t.Fatalf("got allNICAddrs = %+v, want = %+v", allNICAddrs, broadcastAddr)
+ }
+ }
+
+ // Disabling the NIC should remove the IPv4 broadcast address.
+ if err := s.DisableNIC(nicID); err != nil {
+ t.Fatalf("s.DisableNIC(%d): %s", nicID, err)
+ }
+
+ {
+ allStackAddrs := s.AllAddresses()
+ if allNICAddrs, ok := allStackAddrs[nicID]; !ok {
+ t.Fatalf("entry for %d missing from allStackAddrs = %+v", nicID, allStackAddrs)
+ } else if containsAddr(allNICAddrs, broadcastAddr) {
+ t.Fatalf("got allNICAddrs = %+v, don't want = %+v", allNICAddrs, broadcastAddr)
+ }
+ }
+}
+
+// TestLeaveIPv6SolicitedNodeAddrBeforeAddrRemoval tests that removing an IPv6
+// address after leaving its solicited node multicast address does not result in
+// an error.
+func TestLeaveIPv6SolicitedNodeAddrBeforeAddrRemoval(t *testing.T) {
+ const nicID = 1
+
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ })
+ e := channel.New(10, 1280, linkAddr1)
+ if err := s.CreateNIC(1, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _): %s", nicID, err)
+ }
+
+ if err := s.AddAddress(nicID, ipv6.ProtocolNumber, addr1); err != nil {
+ t.Fatalf("AddAddress(%d, %d, %s): %s", nicID, ipv6.ProtocolNumber, addr1, err)
+ }
+
+ // The NIC should have joined addr1's solicited node multicast address.
+ snmc := header.SolicitedNodeAddr(addr1)
+ in, err := s.IsInGroup(nicID, snmc)
+ if err != nil {
+ t.Fatalf("IsInGroup(%d, %s): %s", nicID, snmc, err)
+ }
+ if !in {
+ t.Fatalf("got IsInGroup(%d, %s) = false, want = true", nicID, snmc)
+ }
+
+ if err := s.LeaveGroup(ipv6.ProtocolNumber, nicID, snmc); err != nil {
+ t.Fatalf("LeaveGroup(%d, %d, %s): %s", ipv6.ProtocolNumber, nicID, snmc, err)
+ }
+ in, err = s.IsInGroup(nicID, snmc)
+ if err != nil {
+ t.Fatalf("IsInGroup(%d, %s): %s", nicID, snmc, err)
+ }
+ if in {
+ t.Fatalf("got IsInGroup(%d, %s) = true, want = false", nicID, snmc)
+ }
+
+ if err := s.RemoveAddress(nicID, addr1); err != nil {
+ t.Fatalf("RemoveAddress(%d, %s) = %s", nicID, addr1, err)
+ }
+}
+
+func TestJoinLeaveMulticastOnNICEnableDisable(t *testing.T) {
+ const nicID = 1
+
+ tests := []struct {
+ name string
+ proto tcpip.NetworkProtocolNumber
+ addr tcpip.Address
+ }{
+ {
+ name: "IPv6 All-Nodes",
+ proto: header.IPv6ProtocolNumber,
+ addr: header.IPv6AllNodesMulticastAddress,
+ },
+ {
+ name: "IPv4 All-Systems",
+ proto: header.IPv4ProtocolNumber,
+ addr: header.IPv4AllSystems,
+ },
+ }
+
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ e := loopback.New()
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()},
+ })
+ nicOpts := stack.NICOptions{Disabled: true}
+ if err := s.CreateNICWithOptions(nicID, e, nicOpts); err != nil {
+ t.Fatalf("CreateNIC(%d, _, %+v) = %s", nicID, nicOpts, err)
+ }
+
+ // Should not be in the multicast group yet because the NIC has not been
+ // enabled yet.
+ if isInGroup, err := s.IsInGroup(nicID, test.addr); err != nil {
+ t.Fatalf("IsInGroup(%d, %s): %s", nicID, test.addr, err)
+ } else if isInGroup {
+ t.Fatalf("got IsInGroup(%d, %s) = true, want = false", nicID, test.addr)
+ }
+
+ // The all-nodes multicast group should be joined when the NIC is enabled.
+ if err := s.EnableNIC(nicID); err != nil {
+ t.Fatalf("s.EnableNIC(%d): %s", nicID, err)
+ }
+
+ if isInGroup, err := s.IsInGroup(nicID, test.addr); err != nil {
+ t.Fatalf("IsInGroup(%d, %s): %s", nicID, test.addr, err)
+ } else if !isInGroup {
+ t.Fatalf("got IsInGroup(%d, %s) = false, want = true", nicID, test.addr)
+ }
+
+ // The multicast group should be left when the NIC is disabled.
+ if err := s.DisableNIC(nicID); err != nil {
+ t.Fatalf("s.DisableNIC(%d): %s", nicID, err)
+ }
+
+ if isInGroup, err := s.IsInGroup(nicID, test.addr); err != nil {
+ t.Fatalf("IsInGroup(%d, %s): %s", nicID, test.addr, err)
+ } else if isInGroup {
+ t.Fatalf("got IsInGroup(%d, %s) = true, want = false", nicID, test.addr)
+ }
+
+ // The all-nodes multicast group should be joined when the NIC is enabled.
+ if err := s.EnableNIC(nicID); err != nil {
+ t.Fatalf("s.EnableNIC(%d): %s", nicID, err)
+ }
+
+ if isInGroup, err := s.IsInGroup(nicID, test.addr); err != nil {
+ t.Fatalf("IsInGroup(%d, %s): %s", nicID, test.addr, err)
+ } else if !isInGroup {
+ t.Fatalf("got IsInGroup(%d, %s) = false, want = true", nicID, test.addr)
+ }
+
+ // Leaving the group before disabling the NIC should not cause an error.
+ if err := s.LeaveGroup(test.proto, nicID, test.addr); err != nil {
+ t.Fatalf("s.LeaveGroup(%d, %d, %s): %s", test.proto, nicID, test.addr, err)
+ }
+
+ if err := s.DisableNIC(nicID); err != nil {
+ t.Fatalf("s.DisableNIC(%d): %s", nicID, err)
+ }
+
+ if isInGroup, err := s.IsInGroup(nicID, test.addr); err != nil {
+ t.Fatalf("IsInGroup(%d, %s): %s", nicID, test.addr, err)
+ } else if isInGroup {
+ t.Fatalf("got IsInGroup(%d, %s) = true, want = false", nicID, test.addr)
+ }
+ })
+ }
+}
+
+// TestDoDADWhenNICEnabled tests that IPv6 endpoints that were added while a NIC
+// was disabled have DAD performed on them when the NIC is enabled.
+func TestDoDADWhenNICEnabled(t *testing.T) {
+ const dadTransmits = 1
+ const retransmitTimer = time.Second
+ const nicID = 1
+
+ ndpDisp := ndpDispatcher{
+ dadC: make(chan ndpDADEvent),
+ }
+ opts := stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ DupAddrDetectTransmits: dadTransmits,
+ RetransmitTimer: retransmitTimer,
+ },
+ NDPDisp: &ndpDisp,
+ }
+
+ e := channel.New(dadTransmits, 1280, linkAddr1)
+ s := stack.New(opts)
+ nicOpts := stack.NICOptions{Disabled: true}
+ if err := s.CreateNICWithOptions(nicID, e, nicOpts); err != nil {
+ t.Fatalf("CreateNIC(%d, _, %+v) = %s", nicID, nicOpts, err)
+ }
+
+ addr := tcpip.ProtocolAddress{
+ Protocol: header.IPv6ProtocolNumber,
+ AddressWithPrefix: tcpip.AddressWithPrefix{
+ Address: llAddr1,
+ PrefixLen: 128,
+ },
+ }
+ if err := s.AddProtocolAddress(nicID, addr); err != nil {
+ t.Fatalf("AddProtocolAddress(%d, %+v): %s", nicID, addr, err)
+ }
+
+ // Address should be in the list of all addresses.
+ if addrs := s.AllAddresses()[nicID]; !containsV6Addr(addrs, addr.AddressWithPrefix) {
+ t.Fatalf("got s.AllAddresses()[%d] = %+v, want = %+v", nicID, addrs, addr)
+ }
+
+ // Address should be tentative so it should not be a main address.
+ got, err := s.GetMainNICAddress(nicID, header.IPv6ProtocolNumber)
+ if err != nil {
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (_, %v), want = (_, nil)", nicID, header.IPv6ProtocolNumber, err)
+ }
+ if want := (tcpip.AddressWithPrefix{}); got != want {
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (%s, nil), want = (%s, nil)", nicID, header.IPv6ProtocolNumber, got, want)
+ }
+
+ // Enabling the NIC should start DAD for the address.
+ if err := s.EnableNIC(nicID); err != nil {
+ t.Fatalf("s.EnableNIC(%d): %s", nicID, err)
+ }
+ if addrs := s.AllAddresses()[nicID]; !containsV6Addr(addrs, addr.AddressWithPrefix) {
+ t.Fatalf("got s.AllAddresses()[%d] = %+v, want = %+v", nicID, addrs, addr)
+ }
+
+ // Address should not be considered bound to the NIC yet (DAD ongoing).
+ got, err = s.GetMainNICAddress(nicID, header.IPv6ProtocolNumber)
+ if err != nil {
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (_, %v), want = (_, nil)", nicID, header.IPv6ProtocolNumber, err)
+ }
+ if want := (tcpip.AddressWithPrefix{}); got != want {
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (%s, nil), want = (%s, nil)", nicID, header.IPv6ProtocolNumber, got, want)
+ }
+
+ // Wait for DAD to resolve.
+ select {
+ case <-time.After(dadTransmits*retransmitTimer + defaultAsyncPositiveEventTimeout):
+ t.Fatal("timed out waiting for DAD resolution")
+ case e := <-ndpDisp.dadC:
+ if diff := checkDADEvent(e, nicID, addr.AddressWithPrefix.Address, true, nil); diff != "" {
+ t.Errorf("dad event mismatch (-want +got):\n%s", diff)
+ }
+ }
+ if addrs := s.AllAddresses()[nicID]; !containsV6Addr(addrs, addr.AddressWithPrefix) {
+ t.Fatalf("got s.AllAddresses()[%d] = %+v, want = %+v", nicID, addrs, addr)
+ }
+ got, err = s.GetMainNICAddress(nicID, header.IPv6ProtocolNumber)
+ if err != nil {
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (_, %v), want = (_, nil)", nicID, header.IPv6ProtocolNumber, err)
+ }
+ if got != addr.AddressWithPrefix {
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = %s, want = %s", nicID, header.IPv6ProtocolNumber, got, addr.AddressWithPrefix)
+ }
+
+ // Enabling the NIC again should be a no-op.
+ if err := s.EnableNIC(nicID); err != nil {
+ t.Fatalf("s.EnableNIC(%d): %s", nicID, err)
+ }
+ if addrs := s.AllAddresses()[nicID]; !containsV6Addr(addrs, addr.AddressWithPrefix) {
+ t.Fatalf("got s.AllAddresses()[%d] = %+v, want = %+v", nicID, addrs, addr)
+ }
+ got, err = s.GetMainNICAddress(nicID, header.IPv6ProtocolNumber)
+ if err != nil {
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (_, %v), want = (_, nil)", nicID, header.IPv6ProtocolNumber, err)
+ }
+ if got != addr.AddressWithPrefix {
+ t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (%s, nil), want = (%s, nil)", nicID, header.IPv6ProtocolNumber, got, addr.AddressWithPrefix)
+ }
+}
+
+func TestStackReceiveBufferSizeOption(t *testing.T) {
+ const sMin = stack.MinBufferSize
+ testCases := []struct {
+ name string
+ rs stack.ReceiveBufferSizeOption
+ err *tcpip.Error
+ }{
+ // Invalid configurations.
+ {"min_below_zero", stack.ReceiveBufferSizeOption{Min: -1, Default: sMin, Max: sMin}, tcpip.ErrInvalidOptionValue},
+ {"min_zero", stack.ReceiveBufferSizeOption{Min: 0, Default: sMin, Max: sMin}, tcpip.ErrInvalidOptionValue},
+ {"default_below_min", stack.ReceiveBufferSizeOption{Min: sMin, Default: sMin - 1, Max: sMin - 1}, tcpip.ErrInvalidOptionValue},
+ {"default_above_max", stack.ReceiveBufferSizeOption{Min: sMin, Default: sMin + 1, Max: sMin}, tcpip.ErrInvalidOptionValue},
+ {"max_below_min", stack.ReceiveBufferSizeOption{Min: sMin, Default: sMin + 1, Max: sMin - 1}, tcpip.ErrInvalidOptionValue},
+
+ // Valid Configurations
+ {"in_ascending_order", stack.ReceiveBufferSizeOption{Min: sMin, Default: sMin + 1, Max: sMin + 2}, nil},
+ {"all_equal", stack.ReceiveBufferSizeOption{Min: sMin, Default: sMin, Max: sMin}, nil},
+ {"min_default_equal", stack.ReceiveBufferSizeOption{Min: sMin, Default: sMin, Max: sMin + 1}, nil},
+ {"default_max_equal", stack.ReceiveBufferSizeOption{Min: sMin, Default: sMin + 1, Max: sMin + 1}, nil},
+ }
+ for _, tc := range testCases {
+ t.Run(tc.name, func(t *testing.T) {
+ s := stack.New(stack.Options{})
+ defer s.Close()
+ if err := s.SetOption(tc.rs); err != tc.err {
+ t.Fatalf("s.SetOption(%#v) = %v, want: %v", tc.rs, err, tc.err)
+ }
+ var rs stack.ReceiveBufferSizeOption
+ if tc.err == nil {
+ if err := s.Option(&rs); err != nil {
+ t.Fatalf("s.Option(%#v) = %v, want: nil", rs, err)
+ }
+ if got, want := rs, tc.rs; got != want {
+ t.Fatalf("s.Option(..) returned unexpected value got: %#v, want: %#v", got, want)
+ }
+ }
+ })
+ }
+}
+
+func TestStackSendBufferSizeOption(t *testing.T) {
+ const sMin = stack.MinBufferSize
+ testCases := []struct {
+ name string
+ ss stack.SendBufferSizeOption
+ err *tcpip.Error
+ }{
+ // Invalid configurations.
+ {"min_below_zero", stack.SendBufferSizeOption{Min: -1, Default: sMin, Max: sMin}, tcpip.ErrInvalidOptionValue},
+ {"min_zero", stack.SendBufferSizeOption{Min: 0, Default: sMin, Max: sMin}, tcpip.ErrInvalidOptionValue},
+ {"default_below_min", stack.SendBufferSizeOption{Min: 0, Default: sMin - 1, Max: sMin - 1}, tcpip.ErrInvalidOptionValue},
+ {"default_above_max", stack.SendBufferSizeOption{Min: 0, Default: sMin + 1, Max: sMin}, tcpip.ErrInvalidOptionValue},
+ {"max_below_min", stack.SendBufferSizeOption{Min: sMin, Default: sMin + 1, Max: sMin - 1}, tcpip.ErrInvalidOptionValue},
+
+ // Valid Configurations
+ {"in_ascending_order", stack.SendBufferSizeOption{Min: sMin, Default: sMin + 1, Max: sMin + 2}, nil},
+ {"all_equal", stack.SendBufferSizeOption{Min: sMin, Default: sMin, Max: sMin}, nil},
+ {"min_default_equal", stack.SendBufferSizeOption{Min: sMin, Default: sMin, Max: sMin + 1}, nil},
+ {"default_max_equal", stack.SendBufferSizeOption{Min: sMin, Default: sMin + 1, Max: sMin + 1}, nil},
+ }
+ for _, tc := range testCases {
+ t.Run(tc.name, func(t *testing.T) {
+ s := stack.New(stack.Options{})
+ defer s.Close()
+ if err := s.SetOption(tc.ss); err != tc.err {
+ t.Fatalf("s.SetOption(%+v) = %v, want: %v", tc.ss, err, tc.err)
+ }
+ var ss stack.SendBufferSizeOption
+ if tc.err == nil {
+ if err := s.Option(&ss); err != nil {
+ t.Fatalf("s.Option(%+v) = %v, want: nil", ss, err)
+ }
+ if got, want := ss, tc.ss; got != want {
+ t.Fatalf("s.Option(..) returned unexpected value got: %#v, want: %#v", got, want)
+ }
+ }
+ })
+ }
+}
+
+func TestOutgoingSubnetBroadcast(t *testing.T) {
+ const (
+ unspecifiedNICID = 0
+ nicID1 = 1
+ )
+
+ defaultAddr := tcpip.AddressWithPrefix{
+ Address: header.IPv4Any,
+ PrefixLen: 0,
+ }
+ defaultSubnet := defaultAddr.Subnet()
+ ipv4Addr := tcpip.AddressWithPrefix{
+ Address: "\xc0\xa8\x01\x3a",
+ PrefixLen: 24,
+ }
+ ipv4Subnet := ipv4Addr.Subnet()
+ ipv4SubnetBcast := ipv4Subnet.Broadcast()
+ ipv4Gateway := tcpip.Address("\xc0\xa8\x01\x01")
+ ipv4AddrPrefix31 := tcpip.AddressWithPrefix{
+ Address: "\xc0\xa8\x01\x3a",
+ PrefixLen: 31,
+ }
+ ipv4Subnet31 := ipv4AddrPrefix31.Subnet()
+ ipv4Subnet31Bcast := ipv4Subnet31.Broadcast()
+ ipv4AddrPrefix32 := tcpip.AddressWithPrefix{
+ Address: "\xc0\xa8\x01\x3a",
+ PrefixLen: 32,
+ }
+ ipv4Subnet32 := ipv4AddrPrefix32.Subnet()
+ ipv4Subnet32Bcast := ipv4Subnet32.Broadcast()
+ ipv6Addr := tcpip.AddressWithPrefix{
+ Address: "\x20\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01",
+ PrefixLen: 64,
+ }
+ ipv6Subnet := ipv6Addr.Subnet()
+ ipv6SubnetBcast := ipv6Subnet.Broadcast()
+ remNetAddr := tcpip.AddressWithPrefix{
+ Address: "\x64\x0a\x7b\x18",
+ PrefixLen: 24,
+ }
+ remNetSubnet := remNetAddr.Subnet()
+ remNetSubnetBcast := remNetSubnet.Broadcast()
+
+ tests := []struct {
+ name string
+ nicAddr tcpip.ProtocolAddress
+ routes []tcpip.Route
+ remoteAddr tcpip.Address
+ expectedRoute stack.Route
+ }{
+ // Broadcast to a locally attached subnet populates the broadcast MAC.
+ {
+ name: "IPv4 Broadcast to local subnet",
+ nicAddr: tcpip.ProtocolAddress{
+ Protocol: header.IPv4ProtocolNumber,
+ AddressWithPrefix: ipv4Addr,
+ },
+ routes: []tcpip.Route{
+ {
+ Destination: ipv4Subnet,
+ NIC: nicID1,
+ },
+ },
+ remoteAddr: ipv4SubnetBcast,
+ expectedRoute: stack.Route{
+ LocalAddress: ipv4Addr.Address,
+ RemoteAddress: ipv4SubnetBcast,
+ RemoteLinkAddress: header.EthernetBroadcastAddress,
+ NetProto: header.IPv4ProtocolNumber,
+ Loop: stack.PacketOut,
+ },
+ },
+ // Broadcast to a locally attached /31 subnet does not populate the
+ // broadcast MAC.
+ {
+ name: "IPv4 Broadcast to local /31 subnet",
+ nicAddr: tcpip.ProtocolAddress{
+ Protocol: header.IPv4ProtocolNumber,
+ AddressWithPrefix: ipv4AddrPrefix31,
+ },
+ routes: []tcpip.Route{
+ {
+ Destination: ipv4Subnet31,
+ NIC: nicID1,
+ },
+ },
+ remoteAddr: ipv4Subnet31Bcast,
+ expectedRoute: stack.Route{
+ LocalAddress: ipv4AddrPrefix31.Address,
+ RemoteAddress: ipv4Subnet31Bcast,
+ NetProto: header.IPv4ProtocolNumber,
+ Loop: stack.PacketOut,
+ },
+ },
+ // Broadcast to a locally attached /32 subnet does not populate the
+ // broadcast MAC.
+ {
+ name: "IPv4 Broadcast to local /32 subnet",
+ nicAddr: tcpip.ProtocolAddress{
+ Protocol: header.IPv4ProtocolNumber,
+ AddressWithPrefix: ipv4AddrPrefix32,
+ },
+ routes: []tcpip.Route{
+ {
+ Destination: ipv4Subnet32,
+ NIC: nicID1,
+ },
+ },
+ remoteAddr: ipv4Subnet32Bcast,
+ expectedRoute: stack.Route{
+ LocalAddress: ipv4AddrPrefix32.Address,
+ RemoteAddress: ipv4Subnet32Bcast,
+ NetProto: header.IPv4ProtocolNumber,
+ Loop: stack.PacketOut,
+ },
+ },
+ // IPv6 has no notion of a broadcast.
+ {
+ name: "IPv6 'Broadcast' to local subnet",
+ nicAddr: tcpip.ProtocolAddress{
+ Protocol: header.IPv6ProtocolNumber,
+ AddressWithPrefix: ipv6Addr,
+ },
+ routes: []tcpip.Route{
+ {
+ Destination: ipv6Subnet,
+ NIC: nicID1,
+ },
+ },
+ remoteAddr: ipv6SubnetBcast,
+ expectedRoute: stack.Route{
+ LocalAddress: ipv6Addr.Address,
+ RemoteAddress: ipv6SubnetBcast,
+ NetProto: header.IPv6ProtocolNumber,
+ Loop: stack.PacketOut,
+ },
+ },
+ // Broadcast to a remote subnet in the route table is send to the next-hop
+ // gateway.
+ {
+ name: "IPv4 Broadcast to remote subnet",
+ nicAddr: tcpip.ProtocolAddress{
+ Protocol: header.IPv4ProtocolNumber,
+ AddressWithPrefix: ipv4Addr,
+ },
+ routes: []tcpip.Route{
+ {
+ Destination: remNetSubnet,
+ Gateway: ipv4Gateway,
+ NIC: nicID1,
+ },
+ },
+ remoteAddr: remNetSubnetBcast,
+ expectedRoute: stack.Route{
+ LocalAddress: ipv4Addr.Address,
+ RemoteAddress: remNetSubnetBcast,
+ NextHop: ipv4Gateway,
+ NetProto: header.IPv4ProtocolNumber,
+ Loop: stack.PacketOut,
+ },
+ },
+ // Broadcast to an unknown subnet follows the default route. Note that this
+ // is essentially just routing an unknown destination IP, because w/o any
+ // subnet prefix information a subnet broadcast address is just a normal IP.
+ {
+ name: "IPv4 Broadcast to unknown subnet",
+ nicAddr: tcpip.ProtocolAddress{
+ Protocol: header.IPv4ProtocolNumber,
+ AddressWithPrefix: ipv4Addr,
+ },
+ routes: []tcpip.Route{
+ {
+ Destination: defaultSubnet,
+ Gateway: ipv4Gateway,
+ NIC: nicID1,
+ },
+ },
+ remoteAddr: remNetSubnetBcast,
+ expectedRoute: stack.Route{
+ LocalAddress: ipv4Addr.Address,
+ RemoteAddress: remNetSubnetBcast,
+ NextHop: ipv4Gateway,
+ NetProto: header.IPv4ProtocolNumber,
+ Loop: stack.PacketOut,
+ },
+ },
+ }
+
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()},
+ })
+ ep := channel.New(0, defaultMTU, "")
+ if err := s.CreateNIC(nicID1, ep); err != nil {
+ t.Fatalf("CreateNIC(%d, _): %s", nicID1, err)
+ }
+ if err := s.AddProtocolAddress(nicID1, test.nicAddr); err != nil {
+ t.Fatalf("AddProtocolAddress(%d, %+v): %s", nicID1, test.nicAddr, err)
+ }
+
+ s.SetRouteTable(test.routes)
+
+ var netProto tcpip.NetworkProtocolNumber
+ switch l := len(test.remoteAddr); l {
+ case header.IPv4AddressSize:
+ netProto = header.IPv4ProtocolNumber
+ case header.IPv6AddressSize:
+ netProto = header.IPv6ProtocolNumber
+ default:
+ t.Fatalf("got unexpected address length = %d bytes", l)
+ }
+
+ if r, err := s.FindRoute(unspecifiedNICID, "" /* localAddr */, test.remoteAddr, netProto, false /* multicastLoop */); err != nil {
+ t.Fatalf("FindRoute(%d, '', %s, %d): %s", unspecifiedNICID, test.remoteAddr, netProto, err)
+ } else if diff := cmp.Diff(r, test.expectedRoute, cmpopts.IgnoreUnexported(r)); diff != "" {
+ t.Errorf("route mismatch (-want +got):\n%s", diff)
+ }
+ })
+ }
+}
+
+func TestResolveWith(t *testing.T) {
+ const (
+ unspecifiedNICID = 0
+ nicID = 1
+ )
+
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), arp.NewProtocol()},
+ })
+ ep := channel.New(0, defaultMTU, "")
+ ep.LinkEPCapabilities |= stack.CapabilityResolutionRequired
+ if err := s.CreateNIC(nicID, ep); err != nil {
+ t.Fatalf("CreateNIC(%d, _): %s", nicID, err)
+ }
+ addr := tcpip.ProtocolAddress{
+ Protocol: header.IPv4ProtocolNumber,
+ AddressWithPrefix: tcpip.AddressWithPrefix{
+ Address: tcpip.Address(net.ParseIP("192.168.1.58").To4()),
+ PrefixLen: 24,
+ },
+ }
+ if err := s.AddProtocolAddress(nicID, addr); err != nil {
+ t.Fatalf("AddProtocolAddress(%d, %+v): %s", nicID, addr, err)
+ }
+
+ s.SetRouteTable([]tcpip.Route{{Destination: header.IPv4EmptySubnet, NIC: nicID}})
+
+ remoteAddr := tcpip.Address(net.ParseIP("192.168.1.59").To4())
+ r, err := s.FindRoute(unspecifiedNICID, "" /* localAddr */, remoteAddr, header.IPv4ProtocolNumber, false /* multicastLoop */)
+ if err != nil {
+ t.Fatalf("FindRoute(%d, '', %s, %d): %s", unspecifiedNICID, remoteAddr, header.IPv4ProtocolNumber, err)
+ }
+ defer r.Release()
+
+ // Should initially require resolution.
+ if !r.IsResolutionRequired() {
+ t.Fatal("got r.IsResolutionRequired() = false, want = true")
+ }
+
+ // Manually resolving the route should no longer require resolution.
+ r.ResolveWith("\x01")
+ if r.IsResolutionRequired() {
+ t.Fatal("got r.IsResolutionRequired() = true, want = false")
+ }
+}
diff --git a/pkg/tcpip/stack/transport_demuxer.go b/pkg/tcpip/stack/transport_demuxer.go
index 97a1aec4b..b902c6ca9 100644
--- a/pkg/tcpip/stack/transport_demuxer.go
+++ b/pkg/tcpip/stack/transport_demuxer.go
@@ -17,12 +17,12 @@ package stack
import (
"fmt"
"math/rand"
- "sync"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
- "gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/tcpip/hash/jenkins"
"gvisor.dev/gvisor/pkg/tcpip/header"
+ "gvisor.dev/gvisor/pkg/tcpip/ports"
)
type protocolIDs struct {
@@ -35,28 +35,130 @@ type protocolIDs struct {
type transportEndpoints struct {
// mu protects all fields of the transportEndpoints.
mu sync.RWMutex
- endpoints map[TransportEndpointID]*endpointsByNic
+ endpoints map[TransportEndpointID]*endpointsByNIC
// rawEndpoints contains endpoints for raw sockets, which receive all
// traffic of a given protocol regardless of port.
rawEndpoints []RawTransportEndpoint
}
-type endpointsByNic struct {
+// unregisterEndpoint unregisters the endpoint with the given id such that it
+// won't receive any more packets.
+func (eps *transportEndpoints) unregisterEndpoint(id TransportEndpointID, ep TransportEndpoint, flags ports.Flags, bindToDevice tcpip.NICID) {
+ eps.mu.Lock()
+ defer eps.mu.Unlock()
+ epsByNIC, ok := eps.endpoints[id]
+ if !ok {
+ return
+ }
+ if !epsByNIC.unregisterEndpoint(bindToDevice, ep, flags) {
+ return
+ }
+ delete(eps.endpoints, id)
+}
+
+func (eps *transportEndpoints) transportEndpoints() []TransportEndpoint {
+ eps.mu.RLock()
+ defer eps.mu.RUnlock()
+ es := make([]TransportEndpoint, 0, len(eps.endpoints))
+ for _, e := range eps.endpoints {
+ es = append(es, e.transportEndpoints()...)
+ }
+ return es
+}
+
+// iterEndpointsLocked yields all endpointsByNIC in eps that match id, in
+// descending order of match quality. If a call to yield returns false,
+// iterEndpointsLocked stops iteration and returns immediately.
+//
+// Preconditions: eps.mu must be locked.
+func (eps *transportEndpoints) iterEndpointsLocked(id TransportEndpointID, yield func(*endpointsByNIC) bool) {
+ // Try to find a match with the id as provided.
+ if ep, ok := eps.endpoints[id]; ok {
+ if !yield(ep) {
+ return
+ }
+ }
+
+ // Try to find a match with the id minus the local address.
+ nid := id
+
+ nid.LocalAddress = ""
+ if ep, ok := eps.endpoints[nid]; ok {
+ if !yield(ep) {
+ return
+ }
+ }
+
+ // Try to find a match with the id minus the remote part.
+ nid.LocalAddress = id.LocalAddress
+ nid.RemoteAddress = ""
+ nid.RemotePort = 0
+ if ep, ok := eps.endpoints[nid]; ok {
+ if !yield(ep) {
+ return
+ }
+ }
+
+ // Try to find a match with only the local port.
+ nid.LocalAddress = ""
+ if ep, ok := eps.endpoints[nid]; ok {
+ if !yield(ep) {
+ return
+ }
+ }
+}
+
+// findAllEndpointsLocked returns all endpointsByNIC in eps that match id, in
+// descending order of match quality.
+//
+// Preconditions: eps.mu must be locked.
+func (eps *transportEndpoints) findAllEndpointsLocked(id TransportEndpointID) []*endpointsByNIC {
+ var matchedEPs []*endpointsByNIC
+ eps.iterEndpointsLocked(id, func(ep *endpointsByNIC) bool {
+ matchedEPs = append(matchedEPs, ep)
+ return true
+ })
+ return matchedEPs
+}
+
+// findEndpointLocked returns the endpoint that most closely matches the given id.
+//
+// Preconditions: eps.mu must be locked.
+func (eps *transportEndpoints) findEndpointLocked(id TransportEndpointID) *endpointsByNIC {
+ var matchedEP *endpointsByNIC
+ eps.iterEndpointsLocked(id, func(ep *endpointsByNIC) bool {
+ matchedEP = ep
+ return false
+ })
+ return matchedEP
+}
+
+type endpointsByNIC struct {
mu sync.RWMutex
endpoints map[tcpip.NICID]*multiPortEndpoint
// seed is a random secret for a jenkins hash.
seed uint32
}
+func (epsByNIC *endpointsByNIC) transportEndpoints() []TransportEndpoint {
+ epsByNIC.mu.RLock()
+ defer epsByNIC.mu.RUnlock()
+ var eps []TransportEndpoint
+ for _, ep := range epsByNIC.endpoints {
+ eps = append(eps, ep.transportEndpoints()...)
+ }
+ return eps
+}
+
// HandlePacket is called by the stack when new packets arrive to this transport
// endpoint.
-func (epsByNic *endpointsByNic) handlePacket(r *Route, id TransportEndpointID, vv buffer.VectorisedView) {
- epsByNic.mu.RLock()
+func (epsByNIC *endpointsByNIC) handlePacket(r *Route, id TransportEndpointID, pkt *PacketBuffer) {
+ epsByNIC.mu.RLock()
- mpep, ok := epsByNic.endpoints[r.ref.nic.ID()]
+ mpep, ok := epsByNIC.endpoints[r.ref.nic.ID()]
if !ok {
- if mpep, ok = epsByNic.endpoints[0]; !ok {
- epsByNic.mu.RUnlock() // Don't use defer for performance reasons.
+ if mpep, ok = epsByNIC.endpoints[0]; !ok {
+ epsByNIC.mu.RUnlock() // Don't use defer for performance reasons.
return
}
}
@@ -64,24 +166,30 @@ func (epsByNic *endpointsByNic) handlePacket(r *Route, id TransportEndpointID, v
// If this is a broadcast or multicast datagram, deliver the datagram to all
// endpoints bound to the right device.
if isMulticastOrBroadcast(id.LocalAddress) {
- mpep.handlePacketAll(r, id, vv)
- epsByNic.mu.RUnlock() // Don't use defer for performance reasons.
+ mpep.handlePacketAll(r, id, pkt)
+ epsByNIC.mu.RUnlock() // Don't use defer for performance reasons.
return
}
-
// multiPortEndpoints are guaranteed to have at least one element.
- selectEndpoint(id, mpep, epsByNic.seed).HandlePacket(r, id, vv)
- epsByNic.mu.RUnlock() // Don't use defer for performance reasons.
+ transEP := selectEndpoint(id, mpep, epsByNIC.seed)
+ if queuedProtocol, mustQueue := mpep.demux.queuedProtocols[protocolIDs{mpep.netProto, mpep.transProto}]; mustQueue {
+ queuedProtocol.QueuePacket(r, transEP, id, pkt)
+ epsByNIC.mu.RUnlock()
+ return
+ }
+
+ transEP.HandlePacket(r, id, pkt)
+ epsByNIC.mu.RUnlock() // Don't use defer for performance reasons.
}
// HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket.
-func (epsByNic *endpointsByNic) handleControlPacket(n *NIC, id TransportEndpointID, typ ControlType, extra uint32, vv buffer.VectorisedView) {
- epsByNic.mu.RLock()
- defer epsByNic.mu.RUnlock()
+func (epsByNIC *endpointsByNIC) handleControlPacket(n *NIC, id TransportEndpointID, typ ControlType, extra uint32, pkt *PacketBuffer) {
+ epsByNIC.mu.RLock()
+ defer epsByNIC.mu.RUnlock()
- mpep, ok := epsByNic.endpoints[n.ID()]
+ mpep, ok := epsByNIC.endpoints[n.ID()]
if !ok {
- mpep, ok = epsByNic.endpoints[0]
+ mpep, ok = epsByNIC.endpoints[0]
}
if !ok {
return
@@ -91,55 +199,52 @@ func (epsByNic *endpointsByNic) handleControlPacket(n *NIC, id TransportEndpoint
// broadcast like we are doing with handlePacket above?
// multiPortEndpoints are guaranteed to have at least one element.
- selectEndpoint(id, mpep, epsByNic.seed).HandleControlPacket(id, typ, extra, vv)
+ selectEndpoint(id, mpep, epsByNIC.seed).HandleControlPacket(id, typ, extra, pkt)
}
// registerEndpoint returns true if it succeeds. It fails and returns
// false if ep already has an element with the same key.
-func (epsByNic *endpointsByNic) registerEndpoint(t TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error {
- epsByNic.mu.Lock()
- defer epsByNic.mu.Unlock()
+func (epsByNIC *endpointsByNIC) registerEndpoint(d *transportDemuxer, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, t TransportEndpoint, flags ports.Flags, bindToDevice tcpip.NICID) *tcpip.Error {
+ epsByNIC.mu.Lock()
+ defer epsByNIC.mu.Unlock()
- if multiPortEp, ok := epsByNic.endpoints[bindToDevice]; ok {
- // There was already a bind.
- return multiPortEp.singleRegisterEndpoint(t, reusePort)
+ multiPortEp, ok := epsByNIC.endpoints[bindToDevice]
+ if !ok {
+ multiPortEp = &multiPortEndpoint{
+ demux: d,
+ netProto: netProto,
+ transProto: transProto,
+ }
+ epsByNIC.endpoints[bindToDevice] = multiPortEp
}
- // This is a new binding.
- multiPortEp := &multiPortEndpoint{}
- multiPortEp.endpointsMap = make(map[TransportEndpoint]int)
- multiPortEp.reuse = reusePort
- epsByNic.endpoints[bindToDevice] = multiPortEp
- return multiPortEp.singleRegisterEndpoint(t, reusePort)
+ return multiPortEp.singleRegisterEndpoint(t, flags)
}
-// unregisterEndpoint returns true if endpointsByNic has to be unregistered.
-func (epsByNic *endpointsByNic) unregisterEndpoint(bindToDevice tcpip.NICID, t TransportEndpoint) bool {
- epsByNic.mu.Lock()
- defer epsByNic.mu.Unlock()
- multiPortEp, ok := epsByNic.endpoints[bindToDevice]
+func (epsByNIC *endpointsByNIC) checkEndpoint(d *transportDemuxer, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, flags ports.Flags, bindToDevice tcpip.NICID) *tcpip.Error {
+ epsByNIC.mu.RLock()
+ defer epsByNIC.mu.RUnlock()
+
+ multiPortEp, ok := epsByNIC.endpoints[bindToDevice]
if !ok {
- return false
- }
- if multiPortEp.unregisterEndpoint(t) {
- delete(epsByNic.endpoints, bindToDevice)
+ return nil
}
- return len(epsByNic.endpoints) == 0
+
+ return multiPortEp.singleCheckEndpoint(flags)
}
-// unregisterEndpoint unregisters the endpoint with the given id such that it
-// won't receive any more packets.
-func (eps *transportEndpoints) unregisterEndpoint(id TransportEndpointID, ep TransportEndpoint, bindToDevice tcpip.NICID) {
- eps.mu.Lock()
- defer eps.mu.Unlock()
- epsByNic, ok := eps.endpoints[id]
+// unregisterEndpoint returns true if endpointsByNIC has to be unregistered.
+func (epsByNIC *endpointsByNIC) unregisterEndpoint(bindToDevice tcpip.NICID, t TransportEndpoint, flags ports.Flags) bool {
+ epsByNIC.mu.Lock()
+ defer epsByNIC.mu.Unlock()
+ multiPortEp, ok := epsByNIC.endpoints[bindToDevice]
if !ok {
- return
+ return false
}
- if !epsByNic.unregisterEndpoint(bindToDevice, ep) {
- return
+ if multiPortEp.unregisterEndpoint(t, flags) {
+ delete(epsByNIC.endpoints, bindToDevice)
}
- delete(eps.endpoints, id)
+ return len(epsByNIC.endpoints) == 0
}
// transportDemuxer demultiplexes packets targeted at a transport endpoint
@@ -149,17 +254,33 @@ func (eps *transportEndpoints) unregisterEndpoint(id TransportEndpointID, ep Tra
// newTransportDemuxer.
type transportDemuxer struct {
// protocol is immutable.
- protocol map[protocolIDs]*transportEndpoints
+ protocol map[protocolIDs]*transportEndpoints
+ queuedProtocols map[protocolIDs]queuedTransportProtocol
+}
+
+// queuedTransportProtocol if supported by a protocol implementation will cause
+// the dispatcher to delivery packets to the QueuePacket method instead of
+// calling HandlePacket directly on the endpoint.
+type queuedTransportProtocol interface {
+ QueuePacket(r *Route, ep TransportEndpoint, id TransportEndpointID, pkt *PacketBuffer)
}
func newTransportDemuxer(stack *Stack) *transportDemuxer {
- d := &transportDemuxer{protocol: make(map[protocolIDs]*transportEndpoints)}
+ d := &transportDemuxer{
+ protocol: make(map[protocolIDs]*transportEndpoints),
+ queuedProtocols: make(map[protocolIDs]queuedTransportProtocol),
+ }
// Add each network and transport pair to the demuxer.
for netProto := range stack.networkProtocols {
for proto := range stack.transportProtocols {
- d.protocol[protocolIDs{netProto, proto}] = &transportEndpoints{
- endpoints: make(map[TransportEndpointID]*endpointsByNic),
+ protoIDs := protocolIDs{netProto, proto}
+ d.protocol[protoIDs] = &transportEndpoints{
+ endpoints: make(map[TransportEndpointID]*endpointsByNIC),
+ }
+ qTransProto, isQueued := (stack.transportProtocols[proto].proto).(queuedTransportProtocol)
+ if isQueued {
+ d.queuedProtocols[protoIDs] = qTransProto
}
}
}
@@ -169,10 +290,21 @@ func newTransportDemuxer(stack *Stack) *transportDemuxer {
// registerEndpoint registers the given endpoint with the dispatcher such that
// packets that match the endpoint ID are delivered to it.
-func (d *transportDemuxer) registerEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error {
+func (d *transportDemuxer) registerEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, flags ports.Flags, bindToDevice tcpip.NICID) *tcpip.Error {
for i, n := range netProtos {
- if err := d.singleRegisterEndpoint(n, protocol, id, ep, reusePort, bindToDevice); err != nil {
- d.unregisterEndpoint(netProtos[:i], protocol, id, ep, bindToDevice)
+ if err := d.singleRegisterEndpoint(n, protocol, id, ep, flags, bindToDevice); err != nil {
+ d.unregisterEndpoint(netProtos[:i], protocol, id, ep, flags, bindToDevice)
+ return err
+ }
+ }
+
+ return nil
+}
+
+// checkEndpoint checks if an endpoint can be registered with the dispatcher.
+func (d *transportDemuxer) checkEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, flags ports.Flags, bindToDevice tcpip.NICID) *tcpip.Error {
+ for _, n := range netProtos {
+ if err := d.singleCheckEndpoint(n, protocol, id, flags, bindToDevice); err != nil {
return err
}
}
@@ -183,12 +315,29 @@ func (d *transportDemuxer) registerEndpoint(netProtos []tcpip.NetworkProtocolNum
// multiPortEndpoint is a container for TransportEndpoints which are bound to
// the same pair of address and port. endpointsArr always has at least one
// element.
+//
+// FIXME(gvisor.dev/issue/873): Restore this properly. Currently, we just save
+// this to ensure that the underlying endpoints get saved/restored, but not not
+// use the restored copy.
+//
+// +stateify savable
type multiPortEndpoint struct {
- mu sync.RWMutex
- endpointsArr []TransportEndpoint
- endpointsMap map[TransportEndpoint]int
- // reuse indicates if more than one endpoint is allowed.
- reuse bool
+ mu sync.RWMutex `state:"nosave"`
+ demux *transportDemuxer
+ netProto tcpip.NetworkProtocolNumber
+ transProto tcpip.TransportProtocolNumber
+
+ // endpoints stores the transport endpoints in the order in which they
+ // were bound. This is required for UDP SO_REUSEADDR.
+ endpoints []TransportEndpoint
+ flags ports.FlagCounter
+}
+
+func (ep *multiPortEndpoint) transportEndpoints() []TransportEndpoint {
+ ep.mu.RLock()
+ eps := append([]TransportEndpoint(nil), ep.endpoints...)
+ ep.mu.RUnlock()
+ return eps
}
// reciprocalScale scales a value into range [0, n).
@@ -203,8 +352,12 @@ func reciprocalScale(val, n uint32) uint32 {
// ports then uses it to select a socket. In this case, all packets from one
// address will be sent to same endpoint.
func selectEndpoint(id TransportEndpointID, mpep *multiPortEndpoint, seed uint32) TransportEndpoint {
- if len(mpep.endpointsArr) == 1 {
- return mpep.endpointsArr[0]
+ if len(mpep.endpoints) == 1 {
+ return mpep.endpoints[0]
+ }
+
+ if mpep.flags.IntersectionRefs().ToFlags().Effective().MostRecent {
+ return mpep.endpoints[len(mpep.endpoints)-1]
}
payload := []byte{
@@ -220,72 +373,89 @@ func selectEndpoint(id TransportEndpointID, mpep *multiPortEndpoint, seed uint32
h.Write([]byte(id.RemoteAddress))
hash := h.Sum32()
- idx := reciprocalScale(hash, uint32(len(mpep.endpointsArr)))
- return mpep.endpointsArr[idx]
+ idx := reciprocalScale(hash, uint32(len(mpep.endpoints)))
+ return mpep.endpoints[idx]
}
-func (ep *multiPortEndpoint) handlePacketAll(r *Route, id TransportEndpointID, vv buffer.VectorisedView) {
+func (ep *multiPortEndpoint) handlePacketAll(r *Route, id TransportEndpointID, pkt *PacketBuffer) {
ep.mu.RLock()
- for i, endpoint := range ep.endpointsArr {
- // HandlePacket modifies vv, so each endpoint needs its own copy except for
- // the final one.
- if i == len(ep.endpointsArr)-1 {
- endpoint.HandlePacket(r, id, vv)
- break
+ queuedProtocol, mustQueue := ep.demux.queuedProtocols[protocolIDs{ep.netProto, ep.transProto}]
+ // HandlePacket takes ownership of pkt, so each endpoint needs
+ // its own copy except for the final one.
+ for _, endpoint := range ep.endpoints[:len(ep.endpoints)-1] {
+ if mustQueue {
+ queuedProtocol.QueuePacket(r, endpoint, id, pkt.Clone())
+ } else {
+ endpoint.HandlePacket(r, id, pkt.Clone())
}
- vvCopy := buffer.NewView(vv.Size())
- copy(vvCopy, vv.ToView())
- endpoint.HandlePacket(r, id, vvCopy.ToVectorisedView())
+ }
+ if endpoint := ep.endpoints[len(ep.endpoints)-1]; mustQueue {
+ queuedProtocol.QueuePacket(r, endpoint, id, pkt)
+ } else {
+ endpoint.HandlePacket(r, id, pkt)
}
ep.mu.RUnlock() // Don't use defer for performance reasons.
}
// singleRegisterEndpoint tries to add an endpoint to the multiPortEndpoint
// list. The list might be empty already.
-func (ep *multiPortEndpoint) singleRegisterEndpoint(t TransportEndpoint, reusePort bool) *tcpip.Error {
+func (ep *multiPortEndpoint) singleRegisterEndpoint(t TransportEndpoint, flags ports.Flags) *tcpip.Error {
ep.mu.Lock()
defer ep.mu.Unlock()
- if len(ep.endpointsArr) > 0 {
+ bits := flags.Bits() & ports.MultiBindFlagMask
+
+ if len(ep.endpoints) != 0 {
+ // If it was previously bound, we need to check if we can bind again.
+ if ep.flags.TotalRefs() > 0 && bits&ep.flags.IntersectionRefs() == 0 {
+ return tcpip.ErrPortInUse
+ }
+ }
+
+ ep.endpoints = append(ep.endpoints, t)
+ ep.flags.AddRef(bits)
+
+ return nil
+}
+
+func (ep *multiPortEndpoint) singleCheckEndpoint(flags ports.Flags) *tcpip.Error {
+ ep.mu.RLock()
+ defer ep.mu.RUnlock()
+
+ bits := flags.Bits() & ports.MultiBindFlagMask
+
+ if len(ep.endpoints) != 0 {
// If it was previously bound, we need to check if we can bind again.
- if !ep.reuse || !reusePort {
+ if ep.flags.TotalRefs() > 0 && bits&ep.flags.IntersectionRefs() == 0 {
return tcpip.ErrPortInUse
}
}
- // A new endpoint is added into endpointsArr and its index there is saved in
- // endpointsMap. This will allow us to remove endpoint from the array fast.
- ep.endpointsMap[t] = len(ep.endpointsArr)
- ep.endpointsArr = append(ep.endpointsArr, t)
return nil
}
// unregisterEndpoint returns true if multiPortEndpoint has to be unregistered.
-func (ep *multiPortEndpoint) unregisterEndpoint(t TransportEndpoint) bool {
+func (ep *multiPortEndpoint) unregisterEndpoint(t TransportEndpoint, flags ports.Flags) bool {
ep.mu.Lock()
defer ep.mu.Unlock()
- idx, ok := ep.endpointsMap[t]
- if !ok {
- return false
- }
- delete(ep.endpointsMap, t)
- l := len(ep.endpointsArr)
- if l > 1 {
- // The last endpoint in endpointsArr is moved instead of the deleted one.
- lastEp := ep.endpointsArr[l-1]
- ep.endpointsArr[idx] = lastEp
- ep.endpointsMap[lastEp] = idx
- ep.endpointsArr = ep.endpointsArr[0 : l-1]
- return false
+ for i, endpoint := range ep.endpoints {
+ if endpoint == t {
+ copy(ep.endpoints[i:], ep.endpoints[i+1:])
+ ep.endpoints[len(ep.endpoints)-1] = nil
+ ep.endpoints = ep.endpoints[:len(ep.endpoints)-1]
+
+ ep.flags.DropRef(flags.Bits() & ports.MultiBindFlagMask)
+ break
+ }
}
- return true
+ return len(ep.endpoints) == 0
}
-func (d *transportDemuxer) singleRegisterEndpoint(netProto tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error {
+func (d *transportDemuxer) singleRegisterEndpoint(netProto tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, flags ports.Flags, bindToDevice tcpip.NICID) *tcpip.Error {
if id.RemotePort != 0 {
- // TODO(eyalsoha): Why?
- reusePort = false
+ // SO_REUSEPORT only applies to bound/listening endpoints.
+ flags.LoadBalanced = false
}
eps, ok := d.protocol[protocolIDs{netProto, protocol}]
@@ -296,82 +466,109 @@ func (d *transportDemuxer) singleRegisterEndpoint(netProto tcpip.NetworkProtocol
eps.mu.Lock()
defer eps.mu.Unlock()
- if epsByNic, ok := eps.endpoints[id]; ok {
- // There was already a binding.
- return epsByNic.registerEndpoint(ep, reusePort, bindToDevice)
+ epsByNIC, ok := eps.endpoints[id]
+ if !ok {
+ epsByNIC = &endpointsByNIC{
+ endpoints: make(map[tcpip.NICID]*multiPortEndpoint),
+ seed: rand.Uint32(),
+ }
+ eps.endpoints[id] = epsByNIC
+ }
+
+ return epsByNIC.registerEndpoint(d, netProto, protocol, ep, flags, bindToDevice)
+}
+
+func (d *transportDemuxer) singleCheckEndpoint(netProto tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, flags ports.Flags, bindToDevice tcpip.NICID) *tcpip.Error {
+ if id.RemotePort != 0 {
+ // SO_REUSEPORT only applies to bound/listening endpoints.
+ flags.LoadBalanced = false
+ }
+
+ eps, ok := d.protocol[protocolIDs{netProto, protocol}]
+ if !ok {
+ return tcpip.ErrUnknownProtocol
}
- // This is a new binding.
- epsByNic := &endpointsByNic{
- endpoints: make(map[tcpip.NICID]*multiPortEndpoint),
- seed: rand.Uint32(),
+ eps.mu.RLock()
+ defer eps.mu.RUnlock()
+
+ epsByNIC, ok := eps.endpoints[id]
+ if !ok {
+ return nil
}
- eps.endpoints[id] = epsByNic
- return epsByNic.registerEndpoint(ep, reusePort, bindToDevice)
+ return epsByNIC.checkEndpoint(d, netProto, protocol, flags, bindToDevice)
}
// unregisterEndpoint unregisters the endpoint with the given id such that it
// won't receive any more packets.
-func (d *transportDemuxer) unregisterEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, bindToDevice tcpip.NICID) {
+func (d *transportDemuxer) unregisterEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, flags ports.Flags, bindToDevice tcpip.NICID) {
+ if id.RemotePort != 0 {
+ // SO_REUSEPORT only applies to bound/listening endpoints.
+ flags.LoadBalanced = false
+ }
+
for _, n := range netProtos {
if eps, ok := d.protocol[protocolIDs{n, protocol}]; ok {
- eps.unregisterEndpoint(id, ep, bindToDevice)
+ eps.unregisterEndpoint(id, ep, flags, bindToDevice)
}
}
}
-var loopbackSubnet = func() tcpip.Subnet {
- sn, err := tcpip.NewSubnet("\x7f\x00\x00\x00", "\xff\x00\x00\x00")
- if err != nil {
- panic(err)
- }
- return sn
-}()
-
// deliverPacket attempts to find one or more matching transport endpoints, and
-// then, if matches are found, delivers the packet to them. Returns true if it
-// found one or more endpoints, false otherwise.
-func (d *transportDemuxer) deliverPacket(r *Route, protocol tcpip.TransportProtocolNumber, netHeader buffer.View, vv buffer.VectorisedView, id TransportEndpointID) bool {
+// then, if matches are found, delivers the packet to them. Returns true if
+// the packet no longer needs to be handled.
+func (d *transportDemuxer) deliverPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt *PacketBuffer, id TransportEndpointID) bool {
eps, ok := d.protocol[protocolIDs{r.NetProto, protocol}]
if !ok {
return false
}
- eps.mu.RLock()
-
- // Determine which transport endpoint or endpoints to deliver this packet to.
- // If the packet is a broadcast or multicast, then find all matching
+ // If the packet is a UDP broadcast or multicast, then find all matching
// transport endpoints.
- var destEps []*endpointsByNic
if protocol == header.UDPProtocolNumber && isMulticastOrBroadcast(id.LocalAddress) {
- destEps = d.findAllEndpointsLocked(eps, vv, id)
- } else if ep := d.findEndpointLocked(eps, vv, id); ep != nil {
- destEps = append(destEps, ep)
+ eps.mu.RLock()
+ destEPs := eps.findAllEndpointsLocked(id)
+ eps.mu.RUnlock()
+ // Fail if we didn't find at least one matching transport endpoint.
+ if len(destEPs) == 0 {
+ r.Stats().UDP.UnknownPortErrors.Increment()
+ return false
+ }
+ // handlePacket takes ownership of pkt, so each endpoint needs its own
+ // copy except for the final one.
+ for _, ep := range destEPs[:len(destEPs)-1] {
+ ep.handlePacket(r, id, pkt.Clone())
+ }
+ destEPs[len(destEPs)-1].handlePacket(r, id, pkt)
+ return true
}
- eps.mu.RUnlock()
+ // If the packet is a TCP packet with a non-unicast source or destination
+ // address, then do nothing further and instruct the caller to do the same.
+ if protocol == header.TCPProtocolNumber && (!isUnicast(r.LocalAddress) || !isUnicast(r.RemoteAddress)) {
+ // TCP can only be used to communicate between a single source and a
+ // single destination; the addresses must be unicast.
+ r.Stats().TCP.InvalidSegmentsReceived.Increment()
+ return true
+ }
- // Fail if we didn't find at least one matching transport endpoint.
- if len(destEps) == 0 {
- // UDP packet could not be delivered to an unknown destination port.
+ eps.mu.RLock()
+ ep := eps.findEndpointLocked(id)
+ eps.mu.RUnlock()
+ if ep == nil {
if protocol == header.UDPProtocolNumber {
r.Stats().UDP.UnknownPortErrors.Increment()
}
return false
}
-
- // Deliver the packet.
- for _, ep := range destEps {
- ep.handlePacket(r, id, vv)
- }
-
+ ep.handlePacket(r, id, pkt)
return true
}
// deliverRawPacket attempts to deliver the given packet and returns whether it
// was delivered successfully.
-func (d *transportDemuxer) deliverRawPacket(r *Route, protocol tcpip.TransportProtocolNumber, netHeader buffer.View, vv buffer.VectorisedView) bool {
+func (d *transportDemuxer) deliverRawPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt *PacketBuffer) bool {
eps, ok := d.protocol[protocolIDs{r.NetProto, protocol}]
if !ok {
return false
@@ -385,7 +582,7 @@ func (d *transportDemuxer) deliverRawPacket(r *Route, protocol tcpip.TransportPr
for _, rawEP := range eps.rawEndpoints {
// Each endpoint gets its own copy of the packet for the sake
// of save/restore.
- rawEP.HandlePacket(r, buffer.NewViewFromBytes(netHeader), vv.ToView().ToVectorisedView())
+ rawEP.HandlePacket(r, pkt)
foundRaw = true
}
eps.mu.RUnlock()
@@ -395,67 +592,51 @@ func (d *transportDemuxer) deliverRawPacket(r *Route, protocol tcpip.TransportPr
// deliverControlPacket attempts to deliver the given control packet. Returns
// true if it found an endpoint, false otherwise.
-func (d *transportDemuxer) deliverControlPacket(n *NIC, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, vv buffer.VectorisedView, id TransportEndpointID) bool {
+func (d *transportDemuxer) deliverControlPacket(n *NIC, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, pkt *PacketBuffer, id TransportEndpointID) bool {
eps, ok := d.protocol[protocolIDs{net, trans}]
if !ok {
return false
}
- // Try to find the endpoint.
eps.mu.RLock()
- ep := d.findEndpointLocked(eps, vv, id)
+ ep := eps.findEndpointLocked(id)
eps.mu.RUnlock()
-
- // Fail if we didn't find one.
if ep == nil {
return false
}
- // Deliver the packet.
- ep.handleControlPacket(n, id, typ, extra, vv)
-
+ ep.handleControlPacket(n, id, typ, extra, pkt)
return true
}
-func (d *transportDemuxer) findAllEndpointsLocked(eps *transportEndpoints, vv buffer.VectorisedView, id TransportEndpointID) []*endpointsByNic {
- var matchedEPs []*endpointsByNic
- // Try to find a match with the id as provided.
- if ep, ok := eps.endpoints[id]; ok {
- matchedEPs = append(matchedEPs, ep)
+// findTransportEndpoint find a single endpoint that most closely matches the provided id.
+func (d *transportDemuxer) findTransportEndpoint(netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, id TransportEndpointID, r *Route) TransportEndpoint {
+ eps, ok := d.protocol[protocolIDs{netProto, transProto}]
+ if !ok {
+ return nil
}
- // Try to find a match with the id minus the local address.
- nid := id
-
- nid.LocalAddress = ""
- if ep, ok := eps.endpoints[nid]; ok {
- matchedEPs = append(matchedEPs, ep)
+ eps.mu.RLock()
+ epsByNIC := eps.findEndpointLocked(id)
+ if epsByNIC == nil {
+ eps.mu.RUnlock()
+ return nil
}
- // Try to find a match with the id minus the remote part.
- nid.LocalAddress = id.LocalAddress
- nid.RemoteAddress = ""
- nid.RemotePort = 0
- if ep, ok := eps.endpoints[nid]; ok {
- matchedEPs = append(matchedEPs, ep)
- }
+ epsByNIC.mu.RLock()
+ eps.mu.RUnlock()
- // Try to find a match with only the local port.
- nid.LocalAddress = ""
- if ep, ok := eps.endpoints[nid]; ok {
- matchedEPs = append(matchedEPs, ep)
+ mpep, ok := epsByNIC.endpoints[r.ref.nic.ID()]
+ if !ok {
+ if mpep, ok = epsByNIC.endpoints[0]; !ok {
+ epsByNIC.mu.RUnlock() // Don't use defer for performance reasons.
+ return nil
+ }
}
- return matchedEPs
-}
-
-// findEndpointLocked returns the endpoint that most closely matches the given
-// id.
-func (d *transportDemuxer) findEndpointLocked(eps *transportEndpoints, vv buffer.VectorisedView, id TransportEndpointID) *endpointsByNic {
- if matchedEPs := d.findAllEndpointsLocked(eps, vv, id); len(matchedEPs) > 0 {
- return matchedEPs[0]
- }
- return nil
+ ep := selectEndpoint(id, mpep, epsByNIC.seed)
+ epsByNIC.mu.RUnlock()
+ return ep
}
// registerRawEndpoint registers the given endpoint with the dispatcher such
@@ -469,8 +650,8 @@ func (d *transportDemuxer) registerRawEndpoint(netProto tcpip.NetworkProtocolNum
}
eps.mu.Lock()
- defer eps.mu.Unlock()
eps.rawEndpoints = append(eps.rawEndpoints, ep)
+ eps.mu.Unlock()
return nil
}
@@ -484,15 +665,22 @@ func (d *transportDemuxer) unregisterRawEndpoint(netProto tcpip.NetworkProtocolN
}
eps.mu.Lock()
- defer eps.mu.Unlock()
for i, rawEP := range eps.rawEndpoints {
if rawEP == ep {
- eps.rawEndpoints = append(eps.rawEndpoints[:i], eps.rawEndpoints[i+1:]...)
- return
+ lastIdx := len(eps.rawEndpoints) - 1
+ eps.rawEndpoints[i] = eps.rawEndpoints[lastIdx]
+ eps.rawEndpoints[lastIdx] = nil
+ eps.rawEndpoints = eps.rawEndpoints[:lastIdx]
+ break
}
}
+ eps.mu.Unlock()
}
func isMulticastOrBroadcast(addr tcpip.Address) bool {
return addr == header.IPv4Broadcast || header.IsV4MulticastAddress(addr) || header.IsV6MulticastAddress(addr)
}
+
+func isUnicast(addr tcpip.Address) bool {
+ return addr != header.IPv4Any && addr != header.IPv6Any && !isMulticastOrBroadcast(addr)
+}
diff --git a/pkg/tcpip/stack/transport_demuxer_test.go b/pkg/tcpip/stack/transport_demuxer_test.go
index 210233dc0..1339edc2d 100644
--- a/pkg/tcpip/stack/transport_demuxer_test.go
+++ b/pkg/tcpip/stack/transport_demuxer_test.go
@@ -25,96 +25,65 @@ import (
"gvisor.dev/gvisor/pkg/tcpip/link/channel"
"gvisor.dev/gvisor/pkg/tcpip/network/ipv4"
"gvisor.dev/gvisor/pkg/tcpip/network/ipv6"
+ "gvisor.dev/gvisor/pkg/tcpip/ports"
"gvisor.dev/gvisor/pkg/tcpip/stack"
"gvisor.dev/gvisor/pkg/tcpip/transport/udp"
"gvisor.dev/gvisor/pkg/waiter"
)
const (
- stackV6Addr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01"
- testV6Addr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02"
+ testSrcAddrV6 = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01"
+ testDstAddrV6 = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02"
- stackAddr = "\x0a\x00\x00\x01"
- stackPort = 1234
- testPort = 4096
+ testSrcAddrV4 = "\x0a\x00\x00\x01"
+ testDstAddrV4 = "\x0a\x00\x00\x02"
+
+ testDstPort = 1234
+ testSrcPort = 4096
)
type testContext struct {
- t *testing.T
- linkEPs map[string]*channel.Endpoint
+ linkEps map[tcpip.NICID]*channel.Endpoint
s *stack.Stack
-
- ep tcpip.Endpoint
- wq waiter.Queue
+ wq waiter.Queue
}
-func (c *testContext) cleanup() {
- if c.ep != nil {
- c.ep.Close()
- }
-}
-
-func (c *testContext) createV6Endpoint(v6only bool) {
- var err *tcpip.Error
- c.ep, err = c.s.NewEndpoint(udp.ProtocolNumber, ipv6.ProtocolNumber, &c.wq)
- if err != nil {
- c.t.Fatalf("NewEndpoint failed: %v", err)
- }
-
- var v tcpip.V6OnlyOption
- if v6only {
- v = 1
- }
- if err := c.ep.SetSockOpt(v); err != nil {
- c.t.Fatalf("SetSockOpt failed: %v", err)
- }
-}
-
-// newDualTestContextMultiNic creates the testing context and also linkEpNames
-// named NICs.
-func newDualTestContextMultiNic(t *testing.T, mtu uint32, linkEpNames []string) *testContext {
+// newDualTestContextMultiNIC creates the testing context and also linkEpIDs NICs.
+func newDualTestContextMultiNIC(t *testing.T, mtu uint32, linkEpIDs []tcpip.NICID) *testContext {
s := stack.New(stack.Options{
NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()},
- TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}})
- linkEPs := make(map[string]*channel.Endpoint)
- for i, linkEpName := range linkEpNames {
- channelEP := channel.New(256, mtu, "")
- nicid := tcpip.NICID(i + 1)
- if err := s.CreateNamedNIC(nicid, linkEpName, channelEP); err != nil {
- t.Fatalf("CreateNIC failed: %v", err)
+ TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()},
+ })
+ linkEps := make(map[tcpip.NICID]*channel.Endpoint)
+ for _, linkEpID := range linkEpIDs {
+ channelEp := channel.New(256, mtu, "")
+ if err := s.CreateNIC(linkEpID, channelEp); err != nil {
+ t.Fatalf("CreateNIC failed: %s", err)
}
- linkEPs[linkEpName] = channelEP
+ linkEps[linkEpID] = channelEp
- if err := s.AddAddress(nicid, ipv4.ProtocolNumber, stackAddr); err != nil {
- t.Fatalf("AddAddress IPv4 failed: %v", err)
+ if err := s.AddAddress(linkEpID, ipv4.ProtocolNumber, testDstAddrV4); err != nil {
+ t.Fatalf("AddAddress IPv4 failed: %s", err)
}
- if err := s.AddAddress(nicid, ipv6.ProtocolNumber, stackV6Addr); err != nil {
- t.Fatalf("AddAddress IPv6 failed: %v", err)
+ if err := s.AddAddress(linkEpID, ipv6.ProtocolNumber, testDstAddrV6); err != nil {
+ t.Fatalf("AddAddress IPv6 failed: %s", err)
}
}
s.SetRouteTable([]tcpip.Route{
- {
- Destination: header.IPv4EmptySubnet,
- NIC: 1,
- },
- {
- Destination: header.IPv6EmptySubnet,
- NIC: 1,
- },
+ {Destination: header.IPv4EmptySubnet, NIC: 1},
+ {Destination: header.IPv6EmptySubnet, NIC: 1},
})
return &testContext{
- t: t,
s: s,
- linkEPs: linkEPs,
+ linkEps: linkEps,
}
}
type headers struct {
- srcPort uint16
- dstPort uint16
+ srcPort, dstPort uint16
}
func newPayload() []byte {
@@ -125,7 +94,47 @@ func newPayload() []byte {
return b
}
-func (c *testContext) sendV6Packet(payload []byte, h *headers, linkEpName string) {
+func (c *testContext) sendV4Packet(payload []byte, h *headers, linkEpID tcpip.NICID) {
+ buf := buffer.NewView(header.UDPMinimumSize + header.IPv4MinimumSize + len(payload))
+ payloadStart := len(buf) - len(payload)
+ copy(buf[payloadStart:], payload)
+
+ // Initialize the IP header.
+ ip := header.IPv4(buf)
+ ip.Encode(&header.IPv4Fields{
+ IHL: header.IPv4MinimumSize,
+ TOS: 0x80,
+ TotalLength: uint16(len(buf)),
+ TTL: 65,
+ Protocol: uint8(udp.ProtocolNumber),
+ SrcAddr: testSrcAddrV4,
+ DstAddr: testDstAddrV4,
+ })
+ ip.SetChecksum(^ip.CalculateChecksum())
+
+ // Initialize the UDP header.
+ u := header.UDP(buf[header.IPv4MinimumSize:])
+ u.Encode(&header.UDPFields{
+ SrcPort: h.srcPort,
+ DstPort: h.dstPort,
+ Length: uint16(header.UDPMinimumSize + len(payload)),
+ })
+
+ // Calculate the UDP pseudo-header checksum.
+ xsum := header.PseudoHeaderChecksum(udp.ProtocolNumber, testSrcAddrV4, testDstAddrV4, uint16(len(u)))
+
+ // Calculate the UDP checksum and set it.
+ xsum = header.Checksum(payload, xsum)
+ u.SetChecksum(^u.CalculateChecksum(xsum))
+
+ // Inject packet.
+ pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ })
+ c.linkEps[linkEpID].InjectInbound(ipv4.ProtocolNumber, pkt)
+}
+
+func (c *testContext) sendV6Packet(payload []byte, h *headers, linkEpID tcpip.NICID) {
// Allocate a buffer for data and headers.
buf := buffer.NewView(header.UDPMinimumSize + header.IPv6MinimumSize + len(payload))
copy(buf[len(buf)-len(payload):], payload)
@@ -136,8 +145,8 @@ func (c *testContext) sendV6Packet(payload []byte, h *headers, linkEpName string
PayloadLength: uint16(header.UDPMinimumSize + len(payload)),
NextHeader: uint8(udp.ProtocolNumber),
HopLimit: 65,
- SrcAddr: testV6Addr,
- DstAddr: stackV6Addr,
+ SrcAddr: testSrcAddrV6,
+ DstAddr: testDstAddrV6,
})
// Initialize the UDP header.
@@ -149,14 +158,17 @@ func (c *testContext) sendV6Packet(payload []byte, h *headers, linkEpName string
})
// Calculate the UDP pseudo-header checksum.
- xsum := header.PseudoHeaderChecksum(udp.ProtocolNumber, testV6Addr, stackV6Addr, uint16(len(u)))
+ xsum := header.PseudoHeaderChecksum(udp.ProtocolNumber, testSrcAddrV6, testDstAddrV6, uint16(len(u)))
// Calculate the UDP checksum and set it.
xsum = header.Checksum(payload, xsum)
u.SetChecksum(^u.CalculateChecksum(xsum))
// Inject packet.
- c.linkEPs[linkEpName].Inject(ipv6.ProtocolNumber, buf.ToVectorisedView())
+ pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ })
+ c.linkEps[linkEpID].InjectInbound(ipv6.ProtocolNumber, pkt)
}
func TestTransportDemuxerRegister(t *testing.T) {
@@ -171,95 +183,105 @@ func TestTransportDemuxerRegister(t *testing.T) {
t.Run(test.name, func(t *testing.T) {
s := stack.New(stack.Options{
NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()},
- TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}})
- if got, want := s.RegisterTransportEndpoint(0, []tcpip.NetworkProtocolNumber{test.proto}, udp.ProtocolNumber, stack.TransportEndpointID{}, nil, false, 0), test.want; got != want {
- t.Fatalf("s.RegisterTransportEndpoint(...) = %v, want %v", got, want)
+ TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()},
+ })
+ var wq waiter.Queue
+ ep, err := s.NewEndpoint(udp.ProtocolNumber, ipv4.ProtocolNumber, &wq)
+ if err != nil {
+ t.Fatal(err)
+ }
+ tEP, ok := ep.(stack.TransportEndpoint)
+ if !ok {
+ t.Fatalf("%T does not implement stack.TransportEndpoint", ep)
+ }
+ if got, want := s.RegisterTransportEndpoint(0, []tcpip.NetworkProtocolNumber{test.proto}, udp.ProtocolNumber, stack.TransportEndpointID{}, tEP, ports.Flags{}, 0), test.want; got != want {
+ t.Fatalf("s.RegisterTransportEndpoint(...) = %s, want %s", got, want)
}
})
}
}
-// TestReuseBindToDevice injects varied packets on input devices and checks that
+// TestBindToDeviceDistribution injects varied packets on input devices and checks that
// the distribution of packets received matches expectations.
-func TestDistribution(t *testing.T) {
+func TestBindToDeviceDistribution(t *testing.T) {
type endpointSockopts struct {
- reuse int
- bindToDevice string
+ reuse bool
+ bindToDevice tcpip.NICID
}
for _, test := range []struct {
name string
// endpoints will received the inject packets.
endpoints []endpointSockopts
- // wantedDistribution is the wanted ratio of packets received on each
+ // wantDistributions is the want ratio of packets received on each
// endpoint for each NIC on which packets are injected.
- wantedDistributions map[string][]float64
+ wantDistributions map[tcpip.NICID][]float64
}{
{
"BindPortReuse",
// 5 endpoints that all have reuse set.
[]endpointSockopts{
- endpointSockopts{1, ""},
- endpointSockopts{1, ""},
- endpointSockopts{1, ""},
- endpointSockopts{1, ""},
- endpointSockopts{1, ""},
+ {reuse: true, bindToDevice: 0},
+ {reuse: true, bindToDevice: 0},
+ {reuse: true, bindToDevice: 0},
+ {reuse: true, bindToDevice: 0},
+ {reuse: true, bindToDevice: 0},
},
- map[string][]float64{
+ map[tcpip.NICID][]float64{
// Injected packets on dev0 get distributed evenly.
- "dev0": []float64{0.2, 0.2, 0.2, 0.2, 0.2},
+ 1: {0.2, 0.2, 0.2, 0.2, 0.2},
},
},
{
"BindToDevice",
// 3 endpoints with various bindings.
[]endpointSockopts{
- endpointSockopts{0, "dev0"},
- endpointSockopts{0, "dev1"},
- endpointSockopts{0, "dev2"},
+ {reuse: false, bindToDevice: 1},
+ {reuse: false, bindToDevice: 2},
+ {reuse: false, bindToDevice: 3},
},
- map[string][]float64{
+ map[tcpip.NICID][]float64{
// Injected packets on dev0 go only to the endpoint bound to dev0.
- "dev0": []float64{1, 0, 0},
+ 1: {1, 0, 0},
// Injected packets on dev1 go only to the endpoint bound to dev1.
- "dev1": []float64{0, 1, 0},
+ 2: {0, 1, 0},
// Injected packets on dev2 go only to the endpoint bound to dev2.
- "dev2": []float64{0, 0, 1},
+ 3: {0, 0, 1},
},
},
{
"ReuseAndBindToDevice",
// 6 endpoints with various bindings.
[]endpointSockopts{
- endpointSockopts{1, "dev0"},
- endpointSockopts{1, "dev0"},
- endpointSockopts{1, "dev1"},
- endpointSockopts{1, "dev1"},
- endpointSockopts{1, "dev1"},
- endpointSockopts{1, ""},
+ {reuse: true, bindToDevice: 1},
+ {reuse: true, bindToDevice: 1},
+ {reuse: true, bindToDevice: 2},
+ {reuse: true, bindToDevice: 2},
+ {reuse: true, bindToDevice: 2},
+ {reuse: true, bindToDevice: 0},
},
- map[string][]float64{
+ map[tcpip.NICID][]float64{
// Injected packets on dev0 get distributed among endpoints bound to
// dev0.
- "dev0": []float64{0.5, 0.5, 0, 0, 0, 0},
+ 1: {0.5, 0.5, 0, 0, 0, 0},
// Injected packets on dev1 get distributed among endpoints bound to
// dev1 or unbound.
- "dev1": []float64{0, 0, 1. / 3, 1. / 3, 1. / 3, 0},
+ 2: {0, 0, 1. / 3, 1. / 3, 1. / 3, 0},
// Injected packets on dev999 go only to the unbound.
- "dev999": []float64{0, 0, 0, 0, 0, 1},
+ 1000: {0, 0, 0, 0, 0, 1},
},
},
} {
- t.Run(test.name, func(t *testing.T) {
- for device, wantedDistribution := range test.wantedDistributions {
- t.Run(device, func(t *testing.T) {
- var devices []string
- for d := range test.wantedDistributions {
+ for protoName, netProtoNum := range map[string]tcpip.NetworkProtocolNumber{
+ "IPv4": ipv4.ProtocolNumber,
+ "IPv6": ipv6.ProtocolNumber,
+ } {
+ for device, wantDistribution := range test.wantDistributions {
+ t.Run(test.name+protoName+string(device), func(t *testing.T) {
+ var devices []tcpip.NICID
+ for d := range test.wantDistributions {
devices = append(devices, d)
}
- c := newDualTestContextMultiNic(t, defaultMTU, devices)
- defer c.cleanup()
-
- c.createV6Endpoint(false)
+ c := newDualTestContextMultiNIC(t, defaultMTU, devices)
eps := make(map[tcpip.Endpoint]int)
@@ -273,9 +295,9 @@ func TestDistribution(t *testing.T) {
defer close(ch)
var err *tcpip.Error
- ep, err := c.s.NewEndpoint(udp.ProtocolNumber, ipv6.ProtocolNumber, &wq)
+ ep, err := c.s.NewEndpoint(udp.ProtocolNumber, netProtoNum, &wq)
if err != nil {
- c.t.Fatalf("NewEndpoint failed: %v", err)
+ t.Fatalf("NewEndpoint failed: %s", err)
}
eps[ep] = i
@@ -286,22 +308,31 @@ func TestDistribution(t *testing.T) {
}(ep)
defer ep.Close()
- reusePortOption := tcpip.ReusePortOption(endpoint.reuse)
- if err := ep.SetSockOpt(reusePortOption); err != nil {
- c.t.Fatalf("SetSockOpt(%#v) on endpoint %d failed: %v", reusePortOption, i, err)
+ if err := ep.SetSockOptBool(tcpip.ReusePortOption, endpoint.reuse); err != nil {
+ t.Fatalf("SetSockOptBool(ReusePortOption, %t) on endpoint %d failed: %s", endpoint.reuse, i, err)
}
bindToDeviceOption := tcpip.BindToDeviceOption(endpoint.bindToDevice)
if err := ep.SetSockOpt(bindToDeviceOption); err != nil {
- c.t.Fatalf("SetSockOpt(%#v) on endpoint %d failed: %v", bindToDeviceOption, i, err)
+ t.Fatalf("SetSockOpt(%#v) on endpoint %d failed: %s", bindToDeviceOption, i, err)
+ }
+
+ var dstAddr tcpip.Address
+ switch netProtoNum {
+ case ipv4.ProtocolNumber:
+ dstAddr = testDstAddrV4
+ case ipv6.ProtocolNumber:
+ dstAddr = testDstAddrV6
+ default:
+ t.Fatalf("unexpected protocol number: %d", netProtoNum)
}
- if err := ep.Bind(tcpip.FullAddress{Addr: stackV6Addr, Port: stackPort}); err != nil {
- t.Fatalf("ep.Bind(...) on endpoint %d failed: %v", i, err)
+ if err := ep.Bind(tcpip.FullAddress{Addr: dstAddr, Port: testDstPort}); err != nil {
+ t.Fatalf("ep.Bind(...) on endpoint %d failed: %s", i, err)
}
}
npackets := 100000
nports := 10000
- if got, want := len(test.endpoints), len(wantedDistribution); got != want {
+ if got, want := len(test.endpoints), len(wantDistribution); got != want {
t.Fatalf("got len(test.endpoints) = %d, want %d", got, want)
}
ports := make(map[uint16]tcpip.Endpoint)
@@ -310,17 +341,22 @@ func TestDistribution(t *testing.T) {
// Send a packet.
port := uint16(i % nports)
payload := newPayload()
- c.sendV6Packet(payload,
- &headers{
- srcPort: testPort + port,
- dstPort: stackPort},
- device)
+ hdrs := &headers{
+ srcPort: testSrcPort + port,
+ dstPort: testDstPort,
+ }
+ switch netProtoNum {
+ case ipv4.ProtocolNumber:
+ c.sendV4Packet(payload, hdrs, device)
+ case ipv6.ProtocolNumber:
+ c.sendV6Packet(payload, hdrs, device)
+ default:
+ t.Fatalf("unexpected protocol number: %d", netProtoNum)
+ }
- var addr tcpip.FullAddress
ep := <-pollChannel
- _, _, err := ep.Read(&addr)
- if err != nil {
- c.t.Fatalf("Read on endpoint %d failed: %v", eps[ep], err)
+ if _, _, err := ep.Read(nil); err != nil {
+ t.Fatalf("Read on endpoint %d failed: %s", eps[ep], err)
}
stats[ep]++
if i < nports {
@@ -336,17 +372,17 @@ func TestDistribution(t *testing.T) {
// Check that a packet distribution is as expected.
for ep, i := range eps {
- wantedRatio := wantedDistribution[i]
- wantedRecv := wantedRatio * float64(npackets)
+ wantRatio := wantDistribution[i]
+ wantRecv := wantRatio * float64(npackets)
actualRecv := stats[ep]
actualRatio := float64(stats[ep]) / float64(npackets)
// The deviation is less than 10%.
- if math.Abs(actualRatio-wantedRatio) > 0.05 {
- t.Errorf("wanted about %.0f%% (%.0f of %d) packets to arrive on endpoint %d, got %.0f%% (%d of %d)", wantedRatio*100, wantedRecv, npackets, i, actualRatio*100, actualRecv, npackets)
+ if math.Abs(actualRatio-wantRatio) > 0.05 {
+ t.Errorf("want about %.0f%% (%.0f of %d) packets to arrive on endpoint %d, got %.0f%% (%d of %d)", wantRatio*100, wantRecv, npackets, i, actualRatio*100, actualRecv, npackets)
}
}
})
}
- })
+ }
}
}
diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go
index 6d3daed24..fa4b14ba6 100644
--- a/pkg/tcpip/stack/transport_test.go
+++ b/pkg/tcpip/stack/transport_test.go
@@ -19,9 +19,9 @@ import (
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
- "gvisor.dev/gvisor/pkg/tcpip/iptables"
"gvisor.dev/gvisor/pkg/tcpip/link/channel"
"gvisor.dev/gvisor/pkg/tcpip/link/loopback"
+ "gvisor.dev/gvisor/pkg/tcpip/ports"
"gvisor.dev/gvisor/pkg/tcpip/stack"
"gvisor.dev/gvisor/pkg/waiter"
)
@@ -43,6 +43,7 @@ type fakeTransportEndpoint struct {
proto *fakeTransportProtocol
peerAddr tcpip.Address
route stack.Route
+ uniqueID uint64
// acceptQueue is non-nil iff bound.
acceptQueue []fakeTransportEndpoint
@@ -56,8 +57,14 @@ func (f *fakeTransportEndpoint) Stats() tcpip.EndpointStats {
return nil
}
-func newFakeTransportEndpoint(s *stack.Stack, proto *fakeTransportProtocol, netProto tcpip.NetworkProtocolNumber) tcpip.Endpoint {
- return &fakeTransportEndpoint{stack: s, TransportEndpointInfo: stack.TransportEndpointInfo{NetProto: netProto}, proto: proto}
+func (f *fakeTransportEndpoint) SetOwner(owner tcpip.PacketOwner) {}
+
+func newFakeTransportEndpoint(s *stack.Stack, proto *fakeTransportProtocol, netProto tcpip.NetworkProtocolNumber, uniqueID uint64) tcpip.Endpoint {
+ return &fakeTransportEndpoint{stack: s, TransportEndpointInfo: stack.TransportEndpointInfo{NetProto: netProto}, proto: proto, uniqueID: uniqueID}
+}
+
+func (f *fakeTransportEndpoint) Abort() {
+ f.Close()
}
func (f *fakeTransportEndpoint) Close() {
@@ -77,12 +84,16 @@ func (f *fakeTransportEndpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions
return 0, nil, tcpip.ErrNoRoute
}
- hdr := buffer.NewPrependable(int(f.route.MaxHeaderLength()))
v, err := p.FullPayload()
if err != nil {
return 0, nil, err
}
- if err := f.route.WritePacket(nil /* gso */, hdr, buffer.View(v).ToVectorisedView(), stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}); err != nil {
+ pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{
+ ReserveHeaderBytes: int(f.route.MaxHeaderLength()) + fakeTransHeaderLen,
+ Data: buffer.View(v).ToVectorisedView(),
+ })
+ _ = pkt.TransportHeader().Push(fakeTransHeaderLen)
+ if err := f.route.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}, pkt); err != nil {
return 0, nil, err
}
@@ -98,13 +109,23 @@ func (*fakeTransportEndpoint) SetSockOpt(interface{}) *tcpip.Error {
return tcpip.ErrInvalidEndpointState
}
+// SetSockOptBool sets a socket option. Currently not supported.
+func (*fakeTransportEndpoint) SetSockOptBool(tcpip.SockOptBool, bool) *tcpip.Error {
+ return tcpip.ErrInvalidEndpointState
+}
+
// SetSockOptInt sets a socket option. Currently not supported.
-func (*fakeTransportEndpoint) SetSockOptInt(tcpip.SockOpt, int) *tcpip.Error {
+func (*fakeTransportEndpoint) SetSockOptInt(tcpip.SockOptInt, int) *tcpip.Error {
return tcpip.ErrInvalidEndpointState
}
+// GetSockOptBool implements tcpip.Endpoint.GetSockOptBool.
+func (*fakeTransportEndpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) {
+ return false, tcpip.ErrUnknownProtocolOption
+}
+
// GetSockOptInt implements tcpip.Endpoint.GetSockOptInt.
-func (*fakeTransportEndpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) {
+func (*fakeTransportEndpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) {
return -1, tcpip.ErrUnknownProtocolOption
}
@@ -134,7 +155,7 @@ func (f *fakeTransportEndpoint) Connect(addr tcpip.FullAddress) *tcpip.Error {
// Try to register so that we can start receiving packets.
f.ID.RemoteAddress = addr.Addr
- err = f.stack.RegisterTransportEndpoint(0, []tcpip.NetworkProtocolNumber{fakeNetNumber}, fakeTransNumber, f.ID, f, false /* reuse */, 0 /* bindToDevice */)
+ err = f.stack.RegisterTransportEndpoint(0, []tcpip.NetworkProtocolNumber{fakeNetNumber}, fakeTransNumber, f.ID, f, ports.Flags{}, 0 /* bindToDevice */)
if err != nil {
return err
}
@@ -144,6 +165,10 @@ func (f *fakeTransportEndpoint) Connect(addr tcpip.FullAddress) *tcpip.Error {
return nil
}
+func (f *fakeTransportEndpoint) UniqueID() uint64 {
+ return f.uniqueID
+}
+
func (f *fakeTransportEndpoint) ConnectEndpoint(e tcpip.Endpoint) *tcpip.Error {
return nil
}
@@ -175,8 +200,8 @@ func (f *fakeTransportEndpoint) Bind(a tcpip.FullAddress) *tcpip.Error {
fakeTransNumber,
stack.TransportEndpointID{LocalAddress: a.Addr},
f,
- false, /* reuse */
- 0, /* bindtoDevice */
+ ports.Flags{},
+ 0, /* bindtoDevice */
); err != nil {
return err
}
@@ -192,7 +217,7 @@ func (*fakeTransportEndpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Erro
return tcpip.FullAddress{}, nil
}
-func (f *fakeTransportEndpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, _ buffer.VectorisedView) {
+func (f *fakeTransportEndpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, _ *stack.PacketBuffer) {
// Increment the number of received packets.
f.proto.packetCount++
if f.acceptQueue != nil {
@@ -209,7 +234,7 @@ func (f *fakeTransportEndpoint) HandlePacket(r *stack.Route, id stack.TransportE
}
}
-func (f *fakeTransportEndpoint) HandleControlPacket(stack.TransportEndpointID, stack.ControlType, uint32, buffer.VectorisedView) {
+func (f *fakeTransportEndpoint) HandleControlPacket(stack.TransportEndpointID, stack.ControlType, uint32, *stack.PacketBuffer) {
// Increment the number of received control packets.
f.proto.controlCount++
}
@@ -218,15 +243,15 @@ func (f *fakeTransportEndpoint) State() uint32 {
return 0
}
-func (f *fakeTransportEndpoint) ModerateRecvBuf(copied int) {
-}
+func (f *fakeTransportEndpoint) ModerateRecvBuf(copied int) {}
-func (f *fakeTransportEndpoint) IPTables() (iptables.IPTables, error) {
- return iptables.IPTables{}, nil
+func (f *fakeTransportEndpoint) IPTables() (stack.IPTables, error) {
+ return stack.IPTables{}, nil
}
-func (f *fakeTransportEndpoint) Resume(*stack.Stack) {
-}
+func (f *fakeTransportEndpoint) Resume(*stack.Stack) {}
+
+func (f *fakeTransportEndpoint) Wait() {}
type fakeTransportGoodOption bool
@@ -251,10 +276,10 @@ func (*fakeTransportProtocol) Number() tcpip.TransportProtocolNumber {
}
func (f *fakeTransportProtocol) NewEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, _ *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) {
- return newFakeTransportEndpoint(stack, f, netProto), nil
+ return newFakeTransportEndpoint(stack, f, netProto, stack.UniqueID()), nil
}
-func (f *fakeTransportProtocol) NewRawEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, _ *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) {
+func (*fakeTransportProtocol) NewRawEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, _ *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) {
return nil, tcpip.ErrUnknownProtocol
}
@@ -266,7 +291,7 @@ func (*fakeTransportProtocol) ParsePorts(buffer.View) (src, dst uint16, err *tcp
return 0, 0, nil
}
-func (*fakeTransportProtocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, buffer.View, buffer.VectorisedView) bool {
+func (*fakeTransportProtocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, *stack.PacketBuffer) bool {
return true
}
@@ -292,6 +317,21 @@ func (f *fakeTransportProtocol) Option(option interface{}) *tcpip.Error {
}
}
+// Abort implements TransportProtocol.Abort.
+func (*fakeTransportProtocol) Abort() {}
+
+// Close implements tcpip.Endpoint.Close.
+func (*fakeTransportProtocol) Close() {}
+
+// Wait implements TransportProtocol.Wait.
+func (*fakeTransportProtocol) Wait() {}
+
+// Parse implements TransportProtocol.Parse.
+func (*fakeTransportProtocol) Parse(pkt *stack.PacketBuffer) bool {
+ _, ok := pkt.TransportHeader().Consume(fakeTransHeaderLen)
+ return ok
+}
+
func fakeTransFactory() stack.TransportProtocol {
return &fakeTransportProtocol{}
}
@@ -337,7 +377,9 @@ func TestTransportReceive(t *testing.T) {
// Make sure packet with wrong protocol is not delivered.
buf[0] = 1
buf[2] = 0
- linkEP.Inject(fakeNetNumber, buf.ToVectorisedView())
+ linkEP.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ }))
if fakeTrans.packetCount != 0 {
t.Errorf("packetCount = %d, want %d", fakeTrans.packetCount, 0)
}
@@ -346,7 +388,9 @@ func TestTransportReceive(t *testing.T) {
buf[0] = 1
buf[1] = 3
buf[2] = byte(fakeTransNumber)
- linkEP.Inject(fakeNetNumber, buf.ToVectorisedView())
+ linkEP.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ }))
if fakeTrans.packetCount != 0 {
t.Errorf("packetCount = %d, want %d", fakeTrans.packetCount, 0)
}
@@ -355,7 +399,9 @@ func TestTransportReceive(t *testing.T) {
buf[0] = 1
buf[1] = 2
buf[2] = byte(fakeTransNumber)
- linkEP.Inject(fakeNetNumber, buf.ToVectorisedView())
+ linkEP.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ }))
if fakeTrans.packetCount != 1 {
t.Errorf("packetCount = %d, want %d", fakeTrans.packetCount, 1)
}
@@ -408,7 +454,9 @@ func TestTransportControlReceive(t *testing.T) {
buf[fakeNetHeaderLen+0] = 0
buf[fakeNetHeaderLen+1] = 1
buf[fakeNetHeaderLen+2] = 0
- linkEP.Inject(fakeNetNumber, buf.ToVectorisedView())
+ linkEP.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ }))
if fakeTrans.controlCount != 0 {
t.Errorf("controlCount = %d, want %d", fakeTrans.controlCount, 0)
}
@@ -417,7 +465,9 @@ func TestTransportControlReceive(t *testing.T) {
buf[fakeNetHeaderLen+0] = 3
buf[fakeNetHeaderLen+1] = 1
buf[fakeNetHeaderLen+2] = byte(fakeTransNumber)
- linkEP.Inject(fakeNetNumber, buf.ToVectorisedView())
+ linkEP.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ }))
if fakeTrans.controlCount != 0 {
t.Errorf("controlCount = %d, want %d", fakeTrans.controlCount, 0)
}
@@ -426,7 +476,9 @@ func TestTransportControlReceive(t *testing.T) {
buf[fakeNetHeaderLen+0] = 2
buf[fakeNetHeaderLen+1] = 1
buf[fakeNetHeaderLen+2] = byte(fakeTransNumber)
- linkEP.Inject(fakeNetNumber, buf.ToVectorisedView())
+ linkEP.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{
+ Data: buf.ToVectorisedView(),
+ }))
if fakeTrans.controlCount != 1 {
t.Errorf("controlCount = %d, want %d", fakeTrans.controlCount, 1)
}
@@ -579,7 +631,9 @@ func TestTransportForwarding(t *testing.T) {
req[0] = 1
req[1] = 3
req[2] = byte(fakeTransNumber)
- ep2.Inject(fakeNetNumber, req.ToVectorisedView())
+ ep2.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{
+ Data: req.ToVectorisedView(),
+ }))
aep, _, err := ep.Accept()
if err != nil || aep == nil {
@@ -591,17 +645,16 @@ func TestTransportForwarding(t *testing.T) {
t.Fatalf("Write failed: %v", err)
}
- var p channel.PacketInfo
- select {
- case p = <-ep2.C:
- default:
+ p, ok := ep2.Read()
+ if !ok {
t.Fatal("Response packet not forwarded")
}
- if dst := p.Header[0]; dst != 3 {
+ nh := stack.PayloadSince(p.Pkt.NetworkHeader())
+ if dst := nh[0]; dst != 3 {
t.Errorf("Response packet has incorrect destination addresss: got = %d, want = 3", dst)
}
- if src := p.Header[1]; src != 1 {
+ if src := nh[1]; src != 1 {
t.Errorf("Response packet has incorrect source addresss: got = %d, want = 3", src)
}
}