diff options
author | Sam Balana <sbalana@google.com> | 2020-08-25 11:07:32 -0700 |
---|---|---|
committer | Andrei Vagin <avagin@gmail.com> | 2020-09-09 17:53:10 -0700 |
commit | 232587304de02d5d0634fe8b6118529cfd04bcad (patch) | |
tree | 4fa209d6477112b045ba22f68f122a838c27d26c /pkg/tcpip/network/arp | |
parent | 98e652f6f1d8f3d0bbc4600b1ef2ce471d8e6406 (diff) |
Add option to replace linkAddrCache with neighborCache
This change adds an option to replace the current implementation of ARP through
linkAddrCache, with an implementation of NUD through neighborCache. Switching
to using NUD for both ARP and NDP is beneficial for the reasons described by
RFC 4861 Section 3.1:
"[Using NUD] significantly improves the robustness of packet delivery in the
presence of failing routers, partially failing or partitioned links, or nodes
that change their link-layer addresses. For instance, mobile nodes can move
off-link without losing any connectivity due to stale ARP caches."
"Unlike ARP, Neighbor Unreachability Detection detects half-link failures and
avoids sending traffic to neighbors with which two-way connectivity is
absent."
Along with these changes exposes the API for querying and operating the
neighbor cache. Operations include:
- Create a static entry
- List all entries
- Delete all entries
- Remove an entry by address
This also exposes the API to change the NUD protocol constants on a per-NIC
basis to allow Neighbor Discovery to operate over links with widely varying
performance characteristics. See [RFC 4861 Section 10][1] for the list of
constants.
Finally, an API for subscribing to NUD state changes is exposed through
NUDDispatcher. See [RFC 4861 Appendix C][3] for the list of edges.
Tests:
pkg/tcpip/network/arp:arp_test
+ TestDirectRequest
pkg/tcpip/network/ipv6:ipv6_test
+ TestLinkResolution
+ TestNDPValidation
+ TestNeighorAdvertisementWithTargetLinkLayerOption
+ TestNeighorSolicitationResponse
+ TestNeighorSolicitationWithSourceLinkLayerOption
+ TestRouterAdvertValidation
pkg/tcpip/stack:stack_test
+ TestCacheWaker
+ TestForwardingWithFakeResolver
+ TestForwardingWithFakeResolverManyPackets
+ TestForwardingWithFakeResolverManyResolutions
+ TestForwardingWithFakeResolverPartialTimeout
+ TestForwardingWithFakeResolverTwoPackets
+ TestIPv6SourceAddressSelectionScopeAndSameAddress
[1]: https://tools.ietf.org/html/rfc4861#section-10
[2]: https://tools.ietf.org/html/rfc4861#appendix-C
Fixes #1889
Fixes #1894
Fixes #1895
Fixes #1947
Fixes #1948
Fixes #1949
Fixes #1950
PiperOrigin-RevId: 328365034
Diffstat (limited to 'pkg/tcpip/network/arp')
-rw-r--r-- | pkg/tcpip/network/arp/BUILD | 1 | ||||
-rw-r--r-- | pkg/tcpip/network/arp/arp.go | 47 | ||||
-rw-r--r-- | pkg/tcpip/network/arp/arp_test.go | 331 |
3 files changed, 345 insertions, 34 deletions
diff --git a/pkg/tcpip/network/arp/BUILD b/pkg/tcpip/network/arp/BUILD index eddf7b725..82c073e32 100644 --- a/pkg/tcpip/network/arp/BUILD +++ b/pkg/tcpip/network/arp/BUILD @@ -28,5 +28,6 @@ go_test( "//pkg/tcpip/network/ipv4", "//pkg/tcpip/stack", "//pkg/tcpip/transport/icmp", + "@com_github_google_go_cmp//cmp:go_default_library", ], ) diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index 920872c3f..cbbe5b77f 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -46,6 +46,7 @@ type endpoint struct { nicID tcpip.NICID linkEP stack.LinkEndpoint linkAddrCache stack.LinkAddressCache + nud stack.NUDHandler } // DefaultTTL is unused for ARP. It implements stack.NetworkEndpoint. @@ -78,7 +79,7 @@ func (e *endpoint) WritePacket(*stack.Route, *stack.GSO, stack.NetworkHeaderPara // NetworkProtocolNumber implements stack.NetworkEndpoint.NetworkProtocolNumber. func (e *endpoint) NetworkProtocolNumber() tcpip.NetworkProtocolNumber { - return e.protocol.Number() + return ProtocolNumber } // WritePackets implements stack.NetworkEndpoint.WritePackets. @@ -99,9 +100,25 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { switch h.Op() { case header.ARPRequest: localAddr := tcpip.Address(h.ProtocolAddressTarget()) - if e.linkAddrCache.CheckLocalAddress(e.nicID, header.IPv4ProtocolNumber, localAddr) == 0 { - return // we have no useful answer, ignore the request + + if e.nud == nil { + if e.linkAddrCache.CheckLocalAddress(e.nicID, header.IPv4ProtocolNumber, localAddr) == 0 { + return // we have no useful answer, ignore the request + } + + addr := tcpip.Address(h.ProtocolAddressSender()) + linkAddr := tcpip.LinkAddress(h.HardwareAddressSender()) + e.linkAddrCache.AddLinkAddress(e.nicID, addr, linkAddr) + } else { + if r.Stack().CheckLocalAddress(e.nicID, header.IPv4ProtocolNumber, localAddr) == 0 { + return // we have no useful answer, ignore the request + } + + remoteAddr := tcpip.Address(h.ProtocolAddressSender()) + remoteLinkAddr := tcpip.LinkAddress(h.HardwareAddressSender()) + e.nud.HandleProbe(remoteAddr, localAddr, ProtocolNumber, remoteLinkAddr, e.protocol) } + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ ReserveHeaderBytes: int(e.linkEP.MaxHeaderLength()) + header.ARPSize, }) @@ -113,11 +130,28 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { copy(packet.HardwareAddressTarget(), h.HardwareAddressSender()) copy(packet.ProtocolAddressTarget(), h.ProtocolAddressSender()) _ = e.linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, pkt) - fallthrough // also fill the cache from requests + case header.ARPReply: addr := tcpip.Address(h.ProtocolAddressSender()) linkAddr := tcpip.LinkAddress(h.HardwareAddressSender()) - e.linkAddrCache.AddLinkAddress(e.nicID, addr, linkAddr) + + if e.nud == nil { + e.linkAddrCache.AddLinkAddress(e.nicID, addr, linkAddr) + return + } + + // The solicited, override, and isRouter flags are not available for ARP; + // they are only available for IPv6 Neighbor Advertisements. + e.nud.HandleConfirmation(addr, linkAddr, stack.ReachabilityConfirmationFlags{ + // Solicited and unsolicited (also referred to as gratuitous) ARP Replies + // are handled equivalently to a solicited Neighbor Advertisement. + Solicited: true, + // If a different link address is received than the one cached, the entry + // should always go to Stale. + Override: false, + // ARP does not distinguish between router and non-router hosts. + IsRouter: false, + }) } } @@ -134,12 +168,13 @@ func (*protocol) ParseAddresses(v buffer.View) (src, dst tcpip.Address) { return tcpip.Address(h.ProtocolAddressSender()), ProtocolAddress } -func (p *protocol) NewEndpoint(nicID tcpip.NICID, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, sender stack.LinkEndpoint, st *stack.Stack) stack.NetworkEndpoint { +func (p *protocol) NewEndpoint(nicID tcpip.NICID, linkAddrCache stack.LinkAddressCache, nud stack.NUDHandler, dispatcher stack.TransportDispatcher, sender stack.LinkEndpoint, st *stack.Stack) stack.NetworkEndpoint { return &endpoint{ protocol: p, nicID: nicID, linkEP: sender, linkAddrCache: linkAddrCache, + nud: nud, } } diff --git a/pkg/tcpip/network/arp/arp_test.go b/pkg/tcpip/network/arp/arp_test.go index c2c3e6891..9c9a859e3 100644 --- a/pkg/tcpip/network/arp/arp_test.go +++ b/pkg/tcpip/network/arp/arp_test.go @@ -16,10 +16,12 @@ package arp_test import ( "context" + "fmt" "strconv" "testing" "time" + "github.com/google/go-cmp/cmp" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -32,57 +34,192 @@ import ( ) const ( - stackLinkAddr1 = tcpip.LinkAddress("\x0a\x0a\x0b\x0b\x0c\x0c") - stackLinkAddr2 = tcpip.LinkAddress("\x0b\x0b\x0c\x0c\x0d\x0d") - stackAddr1 = tcpip.Address("\x0a\x00\x00\x01") - stackAddr2 = tcpip.Address("\x0a\x00\x00\x02") - stackAddrBad = tcpip.Address("\x0a\x00\x00\x03") + nicID = 1 + + stackAddr = tcpip.Address("\x0a\x00\x00\x01") + stackLinkAddr = tcpip.LinkAddress("\x0a\x0a\x0b\x0b\x0c\x0c") + + remoteAddr = tcpip.Address("\x0a\x00\x00\x02") + remoteLinkAddr = tcpip.LinkAddress("\x01\x02\x03\x04\x05\x06") + + unknownAddr = tcpip.Address("\x0a\x00\x00\x03") defaultChannelSize = 1 defaultMTU = 65536 + + // eventChanSize defines the size of event channels used by the neighbor + // cache's event dispatcher. The size chosen here needs to be sufficient to + // queue all the events received during tests before consumption. + // If eventChanSize is too small, the tests may deadlock. + eventChanSize = 32 +) + +type eventType uint8 + +const ( + entryAdded eventType = iota + entryChanged + entryRemoved ) +func (t eventType) String() string { + switch t { + case entryAdded: + return "add" + case entryChanged: + return "change" + case entryRemoved: + return "remove" + default: + return fmt.Sprintf("unknown (%d)", t) + } +} + +type eventInfo struct { + eventType eventType + nicID tcpip.NICID + addr tcpip.Address + linkAddr tcpip.LinkAddress + state stack.NeighborState +} + +func (e eventInfo) 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) +} + +// arpDispatcher implements NUDDispatcher to validate the dispatching of +// events upon certain NUD state machine events. +type arpDispatcher struct { + // C is where events are queued + C chan eventInfo +} + +var _ stack.NUDDispatcher = (*arpDispatcher)(nil) + +func (d *arpDispatcher) OnNeighborAdded(nicID tcpip.NICID, addr tcpip.Address, linkAddr tcpip.LinkAddress, state stack.NeighborState, updatedAt time.Time) { + e := eventInfo{ + eventType: entryAdded, + nicID: nicID, + addr: addr, + linkAddr: linkAddr, + state: state, + } + d.C <- e +} + +func (d *arpDispatcher) OnNeighborChanged(nicID tcpip.NICID, addr tcpip.Address, linkAddr tcpip.LinkAddress, state stack.NeighborState, updatedAt time.Time) { + e := eventInfo{ + eventType: entryChanged, + nicID: nicID, + addr: addr, + linkAddr: linkAddr, + state: state, + } + d.C <- e +} + +func (d *arpDispatcher) OnNeighborRemoved(nicID tcpip.NICID, addr tcpip.Address, linkAddr tcpip.LinkAddress, state stack.NeighborState, updatedAt time.Time) { + e := eventInfo{ + eventType: entryRemoved, + nicID: nicID, + addr: addr, + linkAddr: linkAddr, + state: state, + } + d.C <- e +} + +func (d *arpDispatcher) waitForEvent(ctx context.Context, want eventInfo) error { + select { + case got := <-d.C: + if diff := cmp.Diff(got, want, cmp.AllowUnexported(got)); diff != "" { + return fmt.Errorf("got invalid event (-got +want):\n%s", diff) + } + case <-ctx.Done(): + return fmt.Errorf("%s for %s", ctx.Err(), want) + } + return nil +} + +func (d *arpDispatcher) waitForEventWithTimeout(want eventInfo, timeout time.Duration) error { + ctx, cancel := context.WithTimeout(context.Background(), timeout) + defer cancel() + return d.waitForEvent(ctx, want) +} + +func (d *arpDispatcher) nextEvent() (eventInfo, bool) { + select { + case event := <-d.C: + return event, true + default: + return eventInfo{}, false + } +} + type testContext struct { - t *testing.T - linkEP *channel.Endpoint - s *stack.Stack + s *stack.Stack + linkEP *channel.Endpoint + nudDisp *arpDispatcher } -func newTestContext(t *testing.T) *testContext { +func newTestContext(t *testing.T, useNeighborCache bool) *testContext { + c := stack.DefaultNUDConfigurations() + // Transition from Reachable to Stale almost immediately to test if receiving + // probes refreshes positive reachability. + c.BaseReachableTime = time.Microsecond + + d := arpDispatcher{ + // Create an event channel large enough so the neighbor cache doesn't block + // while dispatching events. Blocking could interfere with the timing of + // NUD transitions. + C: make(chan eventInfo, eventChanSize), + } + s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), arp.NewProtocol()}, TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol4()}, + NUDConfigs: c, + NUDDisp: &d, + UseNeighborCache: useNeighborCache, }) - ep := channel.New(defaultChannelSize, defaultMTU, stackLinkAddr1) + ep := channel.New(defaultChannelSize, defaultMTU, stackLinkAddr) + ep.LinkEPCapabilities |= stack.CapabilityResolutionRequired + wep := stack.LinkEndpoint(ep) if testing.Verbose() { wep = sniffer.New(ep) } - if err := s.CreateNIC(1, wep); err != nil { + if err := s.CreateNIC(nicID, wep); err != nil { t.Fatalf("CreateNIC failed: %v", err) } - if err := s.AddAddress(1, ipv4.ProtocolNumber, stackAddr1); err != nil { + if err := s.AddAddress(nicID, ipv4.ProtocolNumber, stackAddr); err != nil { t.Fatalf("AddAddress for ipv4 failed: %v", err) } - if err := s.AddAddress(1, ipv4.ProtocolNumber, stackAddr2); err != nil { - t.Fatalf("AddAddress for ipv4 failed: %v", err) + if !useNeighborCache { + // The remote address needs to be assigned to the NIC so we can receive and + // verify outgoing ARP packets. The neighbor cache isn't concerned with + // this; the tests that use linkAddrCache expect the ARP responses to be + // received by the same NIC. + if err := s.AddAddress(nicID, ipv4.ProtocolNumber, remoteAddr); err != nil { + t.Fatalf("AddAddress for ipv4 failed: %v", err) + } } - if err := s.AddAddress(1, arp.ProtocolNumber, arp.ProtocolAddress); err != nil { + if err := s.AddAddress(nicID, arp.ProtocolNumber, arp.ProtocolAddress); err != nil { t.Fatalf("AddAddress for arp failed: %v", err) } s.SetRouteTable([]tcpip.Route{{ Destination: header.IPv4EmptySubnet, - NIC: 1, + NIC: nicID, }}) return &testContext{ - t: t, - s: s, - linkEP: ep, + s: s, + linkEP: ep, + nudDisp: &d, } } @@ -91,7 +228,7 @@ func (c *testContext) cleanup() { } func TestDirectRequest(t *testing.T) { - c := newTestContext(t) + c := newTestContext(t, false /* useNeighborCache */) defer c.cleanup() const senderMAC = "\x01\x02\x03\x04\x05\x06" @@ -111,7 +248,7 @@ func TestDirectRequest(t *testing.T) { })) } - for i, address := range []tcpip.Address{stackAddr1, stackAddr2} { + for i, address := range []tcpip.Address{stackAddr, remoteAddr} { t.Run(strconv.Itoa(i), func(t *testing.T) { inject(address) pi, _ := c.linkEP.ReadContext(context.Background()) @@ -122,7 +259,7 @@ func TestDirectRequest(t *testing.T) { if !rep.IsValid() { t.Fatalf("invalid ARP response: len = %d; response = %x", len(rep), rep) } - if got, want := tcpip.LinkAddress(rep.HardwareAddressSender()), stackLinkAddr1; got != want { + if got, want := tcpip.LinkAddress(rep.HardwareAddressSender()), stackLinkAddr; got != want { t.Errorf("got HardwareAddressSender = %s, want = %s", got, want) } if got, want := tcpip.Address(rep.ProtocolAddressSender()), tcpip.Address(h.ProtocolAddressTarget()); got != want { @@ -137,7 +274,7 @@ func TestDirectRequest(t *testing.T) { }) } - inject(stackAddrBad) + inject(unknownAddr) // Sleep tests are gross, but this will only potentially flake // if there's a bug. If there is no bug this will reliably // succeed. @@ -148,6 +285,144 @@ func TestDirectRequest(t *testing.T) { } } +func TestDirectRequestWithNeighborCache(t *testing.T) { + c := newTestContext(t, true /* useNeighborCache */) + defer c.cleanup() + + tests := []struct { + name string + senderAddr tcpip.Address + senderLinkAddr tcpip.LinkAddress + targetAddr tcpip.Address + isValid bool + }{ + { + name: "Loopback", + senderAddr: stackAddr, + senderLinkAddr: stackLinkAddr, + targetAddr: stackAddr, + isValid: true, + }, + { + name: "Remote", + senderAddr: remoteAddr, + senderLinkAddr: remoteLinkAddr, + targetAddr: stackAddr, + isValid: true, + }, + { + name: "RemoteInvalidTarget", + senderAddr: remoteAddr, + senderLinkAddr: remoteLinkAddr, + targetAddr: unknownAddr, + isValid: false, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + // Inject an incoming ARP request. + v := make(buffer.View, header.ARPSize) + h := header.ARP(v) + h.SetIPv4OverEthernet() + h.SetOp(header.ARPRequest) + copy(h.HardwareAddressSender(), test.senderLinkAddr) + copy(h.ProtocolAddressSender(), test.senderAddr) + copy(h.ProtocolAddressTarget(), test.targetAddr) + c.linkEP.InjectInbound(arp.ProtocolNumber, &stack.PacketBuffer{ + Data: v.ToVectorisedView(), + }) + + if !test.isValid { + // No packets should be sent after receiving an invalid ARP request. + // There is no need to perform a blocking read here, since packets are + // sent in the same function that handles ARP requests. + if pkt, ok := c.linkEP.Read(); ok { + t.Errorf("unexpected packet sent with network protocol number %d", pkt.Proto) + } + return + } + + // Verify an ARP response was sent. + pi, ok := c.linkEP.Read() + if !ok { + t.Fatal("expected ARP response to be sent, got none") + } + + if pi.Proto != arp.ProtocolNumber { + t.Fatalf("expected ARP response, got network protocol number %d", pi.Proto) + } + rep := header.ARP(pi.Pkt.NetworkHeader().View()) + if !rep.IsValid() { + t.Fatalf("invalid ARP response: len = %d; response = %x", len(rep), rep) + } + if got, want := tcpip.LinkAddress(rep.HardwareAddressSender()), stackLinkAddr; got != want { + t.Errorf("got HardwareAddressSender() = %s, want = %s", got, want) + } + if got, want := tcpip.Address(rep.ProtocolAddressSender()), tcpip.Address(h.ProtocolAddressTarget()); got != want { + t.Errorf("got ProtocolAddressSender() = %s, want = %s", got, want) + } + if got, want := tcpip.LinkAddress(rep.HardwareAddressTarget()), tcpip.LinkAddress(h.HardwareAddressSender()); got != want { + t.Errorf("got HardwareAddressTarget() = %s, want = %s", got, want) + } + if got, want := tcpip.Address(rep.ProtocolAddressTarget()), tcpip.Address(h.ProtocolAddressSender()); got != want { + t.Errorf("got ProtocolAddressTarget() = %s, want = %s", got, want) + } + + // Verify the sender was saved in the neighbor cache. + wantEvent := eventInfo{ + eventType: entryAdded, + nicID: nicID, + addr: test.senderAddr, + linkAddr: tcpip.LinkAddress(test.senderLinkAddr), + state: stack.Stale, + } + if err := c.nudDisp.waitForEventWithTimeout(wantEvent, time.Second); err != nil { + t.Fatal(err) + } + + neighbors, err := c.s.Neighbors(nicID) + if err != nil { + t.Fatalf("c.s.Neighbors(%d): %s", nicID, err) + } + + neighborByAddr := make(map[tcpip.Address]stack.NeighborEntry) + for _, n := range neighbors { + if existing, ok := neighborByAddr[n.Addr]; ok { + if diff := cmp.Diff(existing, n); diff != "" { + t.Fatalf("duplicate neighbor entry found (-existing +got):\n%s", diff) + } + t.Fatalf("exact neighbor entry duplicate found for addr=%s", n.Addr) + } + neighborByAddr[n.Addr] = n + } + + neigh, ok := neighborByAddr[test.senderAddr] + if !ok { + t.Fatalf("expected neighbor entry with Addr = %s", test.senderAddr) + } + if got, want := neigh.LinkAddr, test.senderLinkAddr; got != want { + t.Errorf("got neighbor LinkAddr = %s, want = %s", got, want) + } + if got, want := neigh.LocalAddr, stackAddr; got != want { + t.Errorf("got neighbor LocalAddr = %s, want = %s", got, want) + } + if got, want := neigh.State, stack.Stale; got != want { + t.Errorf("got neighbor State = %s, want = %s", got, want) + } + + // No more events should be dispatched + for { + event, ok := c.nudDisp.nextEvent() + if !ok { + break + } + t.Errorf("unexpected %s", event) + } + }) + } +} + func TestLinkAddressRequest(t *testing.T) { tests := []struct { name string @@ -156,8 +431,8 @@ func TestLinkAddressRequest(t *testing.T) { }{ { name: "Unicast", - remoteLinkAddr: stackLinkAddr2, - expectLinkAddr: stackLinkAddr2, + remoteLinkAddr: remoteLinkAddr, + expectLinkAddr: remoteLinkAddr, }, { name: "Multicast", @@ -173,9 +448,9 @@ func TestLinkAddressRequest(t *testing.T) { t.Fatal("expected ARP protocol to implement stack.LinkAddressResolver") } - linkEP := channel.New(defaultChannelSize, defaultMTU, stackLinkAddr1) - if err := linkRes.LinkAddressRequest(stackAddr1, stackAddr2, test.remoteLinkAddr, linkEP); err != nil { - t.Errorf("got p.LinkAddressRequest(%s, %s, %s, _) = %s", stackAddr1, stackAddr2, test.remoteLinkAddr, err) + linkEP := channel.New(defaultChannelSize, defaultMTU, stackLinkAddr) + if err := linkRes.LinkAddressRequest(stackAddr, remoteAddr, test.remoteLinkAddr, linkEP); err != nil { + t.Errorf("got p.LinkAddressRequest(%s, %s, %s, _) = %s", stackAddr, remoteAddr, test.remoteLinkAddr, err) } pkt, ok := linkEP.Read() |