net/dns/resolver: skip DNS health warning when doing split DNS (#19959)

When MagicDNS is enabled but no global upstream resolvers are configured,
the forwarder only handles specific suffixes and defers other names to the
system resolver. A query it has no resolver for is expected in that case, so
don't raise the dns-forward-failing warning unless a default "." route makes
Tailscale the default resolver.

Fixes #19931

Signed-off-by: Brendan Creane <bcreane@gmail.com>

net/dns/resolver/forwarder.go

@@ -1199,10 +1199,11 @@ func (f *forwarder) forwardWithDestChan(ctx context.Context, query packet, respo
 	if len(resolvers) == 0 {
 		resolvers = f.resolvers(domain)
 		if len(resolvers) == 0 {
+			// No upstream resolver for this name isn't a forwarder failure:
+			// it's split DNS / a name we weren't asked to handle. Count it
+			// rather than raising dnsForwarderFailing, which is reserved for
+			// resolvers we found but couldn't reach. See tailscale/tailscale#19931.
 			metricDNSFwdErrorNoUpstream.Add(1)
-			if f.acceptDNS {
-				f.health.SetUnhealthy(dnsForwarderFailing, health.Args{health.ArgDNSServers: ""})
-			}
 			f.logf("no upstream resolvers set, returning SERVFAIL")
 
 			res, err := servfailResponse(query)

net/dns/resolver/forwarder_test.go

@@ -1387,6 +1387,52 @@ func TestForwarderHealthOnContextExpiry(t *testing.T) {
 	}
 }
 
+// TestForwarderHealthNoUpstreamResolvers verifies that a query with no upstream
+// resolver never raises dnsForwarderFailing, regardless of acceptDNS; that
+// warning is reserved for resolvers we found but couldn't reach (see
+// TestForwarderHealthOnContextExpiry). See tailscale/tailscale#19931.
+func TestForwarderHealthNoUpstreamResolvers(t *testing.T) {
+	const domain = "no-resolver.example.com."
+
+	for _, acceptDNS := range []bool{true, false} {
+		t.Run(fmt.Sprintf("acceptDNS=%v", acceptDNS), func(t *testing.T) {
+			request := makeTestRequest(t, domain, dns.TypeA, 0)
+			logf := tstest.WhileTestRunningLogger(t)
+			bus := eventbustest.NewBus(t)
+			netMon, err := netmon.New(bus, logf)
+			if err != nil {
+				t.Fatal(err)
+			}
+
+			var dialer tsdial.Dialer
+			dialer.SetNetMon(netMon)
+			dialer.SetBus(bus)
+
+			ht := health.NewTracker(bus)
+			fwd := newForwarder(logf, netMon, nil, &dialer, ht, nil)
+			fwd.acceptDNS = acceptDNS
+			// No routes are configured, so the forwarder has no upstream
+			// resolver for the query and returns SERVFAIL.
+
+			rpkt := packet{
+				bs:     request,
+				family: "udp",
+				addr:   netip.MustParseAddrPort("127.0.0.1:12345"),
+			}
+			// Buffered so the SERVFAIL response can be sent without a reader.
+			responseChan := make(chan packet, 1)
+
+			if err := fwd.forwardWithDestChan(context.Background(), rpkt, responseChan); err != nil {
+				t.Fatalf("forwardWithDestChan: %v", err)
+			}
+
+			if ht.IsUnhealthy(dnsForwarderFailing) {
+				t.Error("dnsForwarderFailing was raised for a query with no upstream resolver; want healthy")
+			}
+		})
+	}
+}
+
 func TestResolversCustomScheme(t *testing.T) {
 	t.Parallel()
 	tests := []struct {