Tolerate ErrDumpInterrupted when listing tun addresses (#1835)

connection_manager.go

@@ -105,12 +105,11 @@ func (cm *connectionManager) getInactivityTimeout() time.Duration {
 }
 
 func (cm *connectionManager) In(h *HostInfo) {
-	h.markIn()
+	h.in.Store(true)
 }
 
-// Out records outbound traffic and reports whether the local network changed since this tunnel last sent.
-func (cm *connectionManager) Out(h *HostInfo) bool {
-	return h.markOut(cm.intf.rebindEpoch.Load())
+func (cm *connectionManager) Out(h *HostInfo) {
+	h.out.Store(true)
 }
 
 func (cm *connectionManager) RelayUsed(localIndex uint32) {
@@ -129,7 +128,8 @@ func (cm *connectionManager) RelayUsed(localIndex uint32) {
 // getAndResetTrafficCheck returns if there was any inbound or outbound traffic within the last tick and
 // resets the state for this local index
 func (cm *connectionManager) getAndResetTrafficCheck(h *HostInfo, now time.Time) (bool, bool) {
-	in, out := h.takeTraffic()
+	in := h.in.Swap(false)
+	out := h.out.Swap(false)
 	if in || out {
 		h.lastUsed = now
 	}
@@ -340,7 +340,7 @@ func (cm *connectionManager) makeTrafficDecision(localIndex uint32, now time.Tim
 				"tunnelCheck", m{"state": "alive", "method": "passive"},
 			)
 		}
-		hostinfo.setPendingDeletion(false)
+		hostinfo.pendingDeletion.Store(false)
 
 		if mainHostInfo {
 			decision = tryRehandshake
@@ -363,7 +363,7 @@ func (cm *connectionManager) makeTrafficDecision(localIndex uint32, now time.Tim
 		return decision, hostinfo, primary
 	}
 
-	if hostinfo.isPendingDeletion() {
+	if hostinfo.pendingDeletion.Load() {
 		// We have already sent a test packet and nothing was returned, this hostinfo is dead
 		hostinfo.logger(cm.l).Info("Tunnel status",
 			"tunnelCheck", m{"state": "dead", "method": "active"},
@@ -414,7 +414,7 @@ func (cm *connectionManager) makeTrafficDecision(localIndex uint32, now time.Tim
 		}
 	}
 
-	hostinfo.setPendingDeletion(true)
+	hostinfo.pendingDeletion.Store(true)
 	cm.trafficTimer.Add(hostinfo.localIndexId, cm.pendingDeletionInterval)
 	return decision, hostinfo, nil
 }

connection_manager_test.go

@@ -86,25 +86,25 @@ func Test_NewConnectionManagerTest(t *testing.T) {
 	// We saw traffic out to vpnIp
 	nc.Out(hostinfo)
 	nc.In(hostinfo)
-	assert.False(t, hostinfo.isPendingDeletion())
+	assert.False(t, hostinfo.pendingDeletion.Load())
 	assert.Contains(t, nc.hostMap.Hosts, hostinfo.vpnAddrs[0])
 	assert.Contains(t, nc.hostMap.Indexes, hostinfo.localIndexId)
-	assert.True(t, hostinfo.sentSinceCheck())
-	assert.True(t, (hostinfo.state.Load()&stateIn != 0))
+	assert.True(t, hostinfo.out.Load())
+	assert.True(t, hostinfo.in.Load())
 
 	// Do a traffic check tick, should not be pending deletion but should not have any in/out packets recorded
 	nc.doTrafficCheck(hostinfo.localIndexId, p, nb, out, time.Now())
-	assert.False(t, hostinfo.isPendingDeletion())
-	assert.False(t, hostinfo.sentSinceCheck())
-	assert.False(t, (hostinfo.state.Load()&stateIn != 0))
+	assert.False(t, hostinfo.pendingDeletion.Load())
+	assert.False(t, hostinfo.out.Load())
+	assert.False(t, hostinfo.in.Load())
 
 	// Do another traffic check tick, this host should be pending deletion now
 	nc.Out(hostinfo)
-	assert.True(t, hostinfo.sentSinceCheck())
+	assert.True(t, hostinfo.out.Load())
 	nc.doTrafficCheck(hostinfo.localIndexId, p, nb, out, time.Now())
-	assert.True(t, hostinfo.isPendingDeletion())
-	assert.False(t, hostinfo.sentSinceCheck())
-	assert.False(t, (hostinfo.state.Load()&stateIn != 0))
+	assert.True(t, hostinfo.pendingDeletion.Load())
+	assert.False(t, hostinfo.out.Load())
+	assert.False(t, hostinfo.in.Load())
 	assert.Contains(t, nc.hostMap.Indexes, hostinfo.localIndexId)
 	assert.Contains(t, nc.hostMap.Hosts, hostinfo.vpnAddrs[0])
 
@@ -168,33 +168,33 @@ func Test_NewConnectionManagerTest2(t *testing.T) {
 	// We saw traffic out to vpnIp
 	nc.Out(hostinfo)
 	nc.In(hostinfo)
-	assert.True(t, (hostinfo.state.Load()&stateIn != 0))
-	assert.True(t, hostinfo.sentSinceCheck())
-	assert.False(t, hostinfo.isPendingDeletion())
+	assert.True(t, hostinfo.in.Load())
+	assert.True(t, hostinfo.out.Load())
+	assert.False(t, hostinfo.pendingDeletion.Load())
 	assert.Contains(t, nc.hostMap.Hosts, hostinfo.vpnAddrs[0])
 	assert.Contains(t, nc.hostMap.Indexes, hostinfo.localIndexId)
 
 	// Do a traffic check tick, should not be pending deletion but should not have any in/out packets recorded
 	nc.doTrafficCheck(hostinfo.localIndexId, p, nb, out, time.Now())
-	assert.False(t, hostinfo.isPendingDeletion())
-	assert.False(t, hostinfo.sentSinceCheck())
-	assert.False(t, (hostinfo.state.Load()&stateIn != 0))
+	assert.False(t, hostinfo.pendingDeletion.Load())
+	assert.False(t, hostinfo.out.Load())
+	assert.False(t, hostinfo.in.Load())
 
 	// Do another traffic check tick, this host should be pending deletion now
 	nc.Out(hostinfo)
 	nc.doTrafficCheck(hostinfo.localIndexId, p, nb, out, time.Now())
-	assert.True(t, hostinfo.isPendingDeletion())
-	assert.False(t, hostinfo.sentSinceCheck())
-	assert.False(t, (hostinfo.state.Load()&stateIn != 0))
+	assert.True(t, hostinfo.pendingDeletion.Load())
+	assert.False(t, hostinfo.out.Load())
+	assert.False(t, hostinfo.in.Load())
 	assert.Contains(t, nc.hostMap.Indexes, hostinfo.localIndexId)
 	assert.Contains(t, nc.hostMap.Hosts, hostinfo.vpnAddrs[0])
 
 	// We saw traffic, should no longer be pending deletion
 	nc.In(hostinfo)
 	nc.doTrafficCheck(hostinfo.localIndexId, p, nb, out, time.Now())
-	assert.False(t, hostinfo.isPendingDeletion())
-	assert.False(t, hostinfo.sentSinceCheck())
-	assert.False(t, (hostinfo.state.Load()&stateIn != 0))
+	assert.False(t, hostinfo.pendingDeletion.Load())
+	assert.False(t, hostinfo.out.Load())
+	assert.False(t, hostinfo.in.Load())
 	assert.Contains(t, nc.hostMap.Indexes, hostinfo.localIndexId)
 	assert.Contains(t, nc.hostMap.Hosts, hostinfo.vpnAddrs[0])
 }
@@ -253,31 +253,31 @@ func Test_NewConnectionManager_DisconnectInactive(t *testing.T) {
 	// Do a traffic check tick, in and out should be cleared but should not be pending deletion
 	nc.Out(hostinfo)
 	nc.In(hostinfo)
-	assert.True(t, hostinfo.sentSinceCheck())
-	assert.True(t, (hostinfo.state.Load()&stateIn != 0))
+	assert.True(t, hostinfo.out.Load())
+	assert.True(t, hostinfo.in.Load())
 
 	now := time.Now()
 	decision, _, _ := nc.makeTrafficDecision(hostinfo.localIndexId, now)
 	assert.Equal(t, tryRehandshake, decision)
 	assert.Equal(t, now, hostinfo.lastUsed)
-	assert.False(t, hostinfo.isPendingDeletion())
-	assert.False(t, hostinfo.sentSinceCheck())
-	assert.False(t, (hostinfo.state.Load()&stateIn != 0))
+	assert.False(t, hostinfo.pendingDeletion.Load())
+	assert.False(t, hostinfo.out.Load())
+	assert.False(t, hostinfo.in.Load())
 
 	decision, _, _ = nc.makeTrafficDecision(hostinfo.localIndexId, now.Add(time.Second*5))
 	assert.Equal(t, doNothing, decision)
 	assert.Equal(t, now, hostinfo.lastUsed)
-	assert.False(t, hostinfo.isPendingDeletion())
-	assert.False(t, hostinfo.sentSinceCheck())
-	assert.False(t, (hostinfo.state.Load()&stateIn != 0))
+	assert.False(t, hostinfo.pendingDeletion.Load())
+	assert.False(t, hostinfo.out.Load())
+	assert.False(t, hostinfo.in.Load())
 
 	// Do another traffic check tick, should still not be pending deletion
 	decision, _, _ = nc.makeTrafficDecision(hostinfo.localIndexId, now.Add(time.Second*10))
 	assert.Equal(t, doNothing, decision)
 	assert.Equal(t, now, hostinfo.lastUsed)
-	assert.False(t, hostinfo.isPendingDeletion())
-	assert.False(t, hostinfo.sentSinceCheck())
-	assert.False(t, (hostinfo.state.Load()&stateIn != 0))
+	assert.False(t, hostinfo.pendingDeletion.Load())
+	assert.False(t, hostinfo.out.Load())
+	assert.False(t, hostinfo.in.Load())
 	assert.Contains(t, nc.hostMap.Indexes, hostinfo.localIndexId)
 	assert.Contains(t, nc.hostMap.Hosts, hostinfo.vpnAddrs[0])
 
@@ -285,9 +285,9 @@ func Test_NewConnectionManager_DisconnectInactive(t *testing.T) {
 	decision, _, _ = nc.makeTrafficDecision(hostinfo.localIndexId, now.Add(time.Minute*10))
 	assert.Equal(t, closeTunnel, decision)
 	assert.Equal(t, now, hostinfo.lastUsed)
-	assert.False(t, hostinfo.isPendingDeletion())
-	assert.False(t, hostinfo.sentSinceCheck())
-	assert.False(t, (hostinfo.state.Load()&stateIn != 0))
+	assert.False(t, hostinfo.pendingDeletion.Load())
+	assert.False(t, hostinfo.out.Load())
+	assert.False(t, hostinfo.in.Load())
 	assert.Contains(t, nc.hostMap.Indexes, hostinfo.localIndexId)
 	assert.Contains(t, nc.hostMap.Hosts, hostinfo.vpnAddrs[0])
 }

control.go

@@ -212,7 +212,7 @@ func (c *Control) RebindUDPServer() {
 	c.f.lightHouse.SendUpdate()
 
 	// Let the main interface know that we rebound so that underlying tunnels know to trigger punches from their remotes
-	c.f.rebindEpoch.Add(1)
+	c.f.rebindCount++
 }
 
 // ListHostmapHosts returns details about the actual or pending (handshaking) hostmap by vpn ip

hostmap.go

@@ -238,30 +238,18 @@ const (
 )
 
 type HostInfo struct {
-	// The first cache line is everything the packet paths touch. Grouping them here means a send or receive
-	// pulls in one line instead of two, which is what the layout looked like when state lived at the end.
-
 	remote          atomic.Pointer[netip.AddrPort]
+	remotes         *RemoteList
+	promoteCounter  atomic.Uint32
 	ConnectionState *ConnectionState
-
-	// state holds everything the hot paths need to touch per packet, in one word: whether we have seen traffic
-	// each way since the connection manager last looked, whether it has given up on us, and the
-	// Interface.rebindEpoch this tunnel last sent under. Keeping the epoch here means it survives the traffic
-	// bits being cleared, so a tunnel that has not sent since a rebind still notices when it does.
-	state atomic.Uint32
-
-	promoteCounter atomic.Uint32
-	remoteIndexId  uint32
-	localIndexId   uint32
-	remotes        *RemoteList
+	remoteIndexId   uint32
+	localIndexId    uint32
 
 	// vpnAddrs is a list of vpn addresses assigned to this host that are within our own vpn networks
 	// The host may have other vpn addresses that are outside our
 	// vpn networks but were removed because they are not usable
 	vpnAddrs []netip.Addr
 
-	// Everything below is off the packet path: handshakes, relays, roaming and the connection manager.
-
 	// networks is a combination of specific vpn addresses (not prefixes!) and full unsafe networks assigned to this host.
 	networks   *bart.Table[NetworkType]
 	relayState RelayState
@@ -274,6 +262,11 @@ type HostInfo struct {
 	// This is used to limit lighthouse re-queries in chatty clients
 	nextLHQuery atomic.Int64
 
+	// lastRebindCount is the other side of Interface.rebindCount, if these values don't match then we need to ask LH
+	// for a punch from the remote end of this tunnel. The goal being to prime their conntrack for our traffic just like
+	// with a handshake
+	lastRebindCount int8
+
 	// lastHandshakeTime records the time the remote side told us about at the stage when the handshake was completed locally
 	// Stage 1 packet will contain it if I am a responder, stage 2 packet if I am an initiator
 	// This is used to avoid an attack where a handshake packet is replayed after some time
@@ -282,6 +275,9 @@ type HostInfo struct {
 	lastRoam       time.Time
 	lastRoamRemote netip.AddrPort
 
+	//TODO: in, out, and others might benefit from being an atomic.Int32. We could collapse connectionManager pendingDeletion, relayUsed, and in/out into this 1 thing
+	in, out, pendingDeletion atomic.Bool
+
 	// lastUsed tracks the last time ConnectionManager checked the tunnel and it was in use.
 	// This value will be behind against actual tunnel utilization in the hot path.
 	// This should only be used by the ConnectionManagers ticker routine.
@@ -662,7 +658,7 @@ func (hm *HostMap) unlockedAddHostInfo(hostinfo *HostInfo, f *Interface) {
 	hm.Indexes[hostinfo.localIndexId] = hostinfo
 	hm.RemoteIndexes[hostinfo.remoteIndexId] = hostinfo
 
-	hostinfo.markOut(f.rebindEpoch.Load())
+	hostinfo.out.Store(true)
 	if f.connectionManager != nil { // f.connectionManager is only nil in some unit tests
 		f.connectionManager.trafficTimer.Add(hostinfo.localIndexId, f.connectionManager.checkInterval)
 	}
@@ -763,64 +759,6 @@ func (i *HostInfo) TryPromoteBest(preferredRanges []netip.Prefix, ifce *Interfac
 	}
 }
 
-// Bits within HostInfo.state. Everything above stateEpochShift is the rebind epoch.
-const (
-	stateIn uint32 = 1 << iota
-	stateOut
-	statePendingDeletion
-
-	stateFlags      = stateIn | stateOut | statePendingDeletion
-	stateEpochShift = 3
-)
-
-// markIn records inbound traffic. Reading first keeps the cache line shared on the common path, where the bit
-// is already set.
-func (i *HostInfo) markIn() {
-	if i.state.Load()&stateIn == 0 {
-		i.state.Or(stateIn)
-	}
-}
-
-// markOut records that we sent on this tunnel under the given rebind epoch. It reports whether the epoch moved
-// since our last send, which means the local network changed and we want the far side to punch at us again.
-// The common path is a single load that matches and returns.
-func (i *HostInfo) markOut(epoch uint32) bool {
-	e := epoch << stateEpochShift
-	for {
-		old := i.state.Load()
-		if old&stateOut != 0 && old&^stateFlags == e {
-			return false
-		}
-
-		if i.state.CompareAndSwap(old, old&stateFlags|stateOut|e) {
-			return old&^stateFlags != e
-		}
-	}
-}
-
-// sentSinceCheck reports whether anything has been sent since the connection manager last looked.
-func (i *HostInfo) sentSinceCheck() bool {
-	return i.state.Load()&stateOut != 0
-}
-
-// takeTraffic clears both traffic bits, leaving the epoch alone, and reports what they were.
-func (i *HostInfo) takeTraffic() (in bool, out bool) {
-	old := i.state.And(^(stateIn | stateOut))
-	return old&stateIn != 0, old&stateOut != 0
-}
-
-func (i *HostInfo) setPendingDeletion(v bool) {
-	if v {
-		i.state.Or(statePendingDeletion)
-	} else {
-		i.state.And(^statePendingDeletion)
-	}
-}
-
-func (i *HostInfo) isPendingDeletion() bool {
-	return i.state.Load()&statePendingDeletion != 0
-}
-
 func (i *HostInfo) GetCert() *cert.CachedCertificate {
 	if i.ConnectionState != nil {
 		return i.ConnectionState.peerCert

inside.go

@@ -365,12 +365,17 @@ func (f *Interface) sendNoMetrics(t header.MessageType, st header.MessageSubType
 
 	//l.WithField("trace", string(debug.Stack())).Error("out Header ", &Header{Version, t, st, 0, hostinfo.remoteIndexId, c}, p)
 	out = header.Encode(out, header.Version, t, st, hostinfo.remoteIndexId, c)
-	// We rebound since this tunnel last sent, so the local network moved. Ask the lighthouse to have the far side
-	// punch at where we are now, which primes their conntrack the same way a handshake would.
-	if f.connectionManager.Out(hostinfo) && t != header.CloseTunnel {
+	f.connectionManager.Out(hostinfo)
+
+	// Query our LH if we haven't since the last time we've been rebound, this will cause the remote to punch against
+	// all our addrs and enable a faster roaming.
+	if t != header.CloseTunnel && hostinfo.lastRebindCount != f.rebindCount {
+		//NOTE: there is an update hole if a tunnel isn't used and exactly 256 rebinds occur before the tunnel is
+		// finally used again. This tunnel would eventually be torn down and recreated if this action didn't help.
 		f.lightHouse.QueryServer(hostinfo.vpnAddrs[0])
+		hostinfo.lastRebindCount = f.rebindCount
 		if f.l.Enabled(context.Background(), slog.LevelDebug) {
-			f.l.Debug("Lighthouse update triggered for punch due to rebind epoch",
+			f.l.Debug("Lighthouse update triggered for punch due to rebind counter",
 				"vpnAddrs", hostinfo.vpnAddrs,
 			)
 		}

interface.go

@@ -82,10 +82,8 @@ type Interface struct {
 	sendRecvErrorConfig   recvErrorConfig
 	acceptRecvErrorConfig recvErrorConfig
 
-	// rebindEpoch bumps every time the udp listener is rebound, which means the local network moved. Tunnels
-	// compare it against their own copy to decide they need a punch from the far side. Read on every send, only
-	// written on a rebind, so the cache line stays shared across the routines.
-	rebindEpoch atomic.Uint32
+	// rebindCount is used to decide if an active tunnel should trigger a punch notification through a lighthouse
+	rebindCount int8
 	version     string
 
 	conntrackCacheTimeout time.Duration

overlay/tun_linux.go

@@ -5,6 +5,7 @@ package overlay
 
 import (
 	"encoding/binary"
+	"errors"
 	"fmt"
 	"io"
 	"log/slog"
@@ -483,7 +484,16 @@ func (t *tun) addIPs(link netlink.Link) error {
 	//iterate over remainder, remove whoever shouldn't be there
 	al, err := netlink.AddrList(link, netlink.FAMILY_ALL)
 	if err != nil {
-		return fmt.Errorf("failed to get tun address list: %s", err)
+		//RTM_GETADDR dumps the whole system, so any concurrent address change
+		//interrupts it - including the kernel's async tentative->preferred
+		//flip of an IPv6 address the AddrReplace calls above just added,
+		//which makes this a race against our own setup. Partial results are
+		//still returned; the worst case is a stale address surviving until
+		//the next config reload, which beats failing startup over it.
+		if !errors.Is(err, netlink.ErrDumpInterrupted) {
+			return fmt.Errorf("failed to get tun address list: %s", err)
+		}
+		t.l.Warn("tun address list dump was interrupted, stale addresses may remain")
 	}
 
 	for i := range al {