net/dns/resolver: count errors when racing DNS queries, fail earlier

If all N queries failed, we waited until context timeout (in 5
seconds) to return.

This makes (*forwarder).forward fail fast when the network's
unavailable.

Change-Id: Ibbb3efea7ed34acd3f3b29b5fee00ba8c7492569
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>

net/dns/resolver/forwarder.go

@@ -615,6 +615,10 @@ func (f *forwarder) forwardWithDestChan(ctx context.Context, query packet, respo
 		return err
 	}
 
+	// Guarantee that the ctx we use below is done when this function returns.
+	ctx, cancel := context.WithCancel(ctx)
+	defer cancel()
+
 	// Drop DNS service discovery spam, primarily for battery life
 	// on mobile.  Things like Spotify on iOS generate this traffic,
 	// when browsing for LAN devices.  But even when filtering this
@@ -655,12 +659,8 @@ func (f *forwarder) forwardWithDestChan(ctx context.Context, query packet, respo
 	}
 	defer fq.closeOnCtxDone.Close()
 
-	resc := make(chan []byte, 1)
+	resc := make(chan []byte, 1) // it's fine buffered or not
-	var (
+	errc := make(chan error, 1)  // it's fine buffered or not too
-		mu       sync.Mutex
-		firstErr error
-	)
-
 	for i := range resolvers {
 		go func(rr *resolverAndDelay) {
 			if rr.startDelay > 0 {
@@ -674,39 +674,48 @@ func (f *forwarder) forwardWithDestChan(ctx context.Context, query packet, respo
 			}
 			resb, err := f.send(ctx, fq, *rr)
 			if err != nil {
-				mu.Lock()
+				select {
-				defer mu.Unlock()
+				case errc <- err:
-				if firstErr == nil {
+				case <-ctx.Done():
-					firstErr = err
 				}
 				return
 			}
 			select {
 			case resc <- resb:
-			default:
+			case <-ctx.Done():
 			}
 		}(&resolvers[i])
 	}
 
-	select {
+	var firstErr error
-	case v := <-resc:
+	var numErr int
+	for {
 		select {
+		case v := <-resc:
+			select {
+			case <-ctx.Done():
+				metricDNSFwdErrorContext.Add(1)
+				return ctx.Err()
+			case responseChan <- packet{v, query.addr}:
+				metricDNSFwdSuccess.Add(1)
+				return nil
+			}
+		case err := <-errc:
+			if firstErr == nil {
+				firstErr = err
+			}
+			numErr++
+			if numErr == len(resolvers) {
+				return firstErr
+			}
 		case <-ctx.Done():
 			metricDNSFwdErrorContext.Add(1)
+			if firstErr != nil {
+				metricDNSFwdErrorContextGotError.Add(1)
+				return firstErr
+			}
 			return ctx.Err()
-		case responseChan <- packet{v, query.addr}:
-			metricDNSFwdSuccess.Add(1)
-			return nil
 		}
-	case <-ctx.Done():
-		mu.Lock()
-		defer mu.Unlock()
-		metricDNSFwdErrorContext.Add(1)
-		if firstErr != nil {
-			metricDNSFwdErrorContextGotError.Add(1)
-			return firstErr
-		}
-		return ctx.Err()
 	}
 }