feat: implement referral resolution for nameserver names without glue records (HAN-381) (#7)

Co-authored-by: Hansen IT Solutions <gary@hansenits.com>
Co-committed-by: Hansen IT Solutions <gary@hansenits.com>

internal/traverse/traverser.go

@@ -36,6 +36,9 @@ type TraversalResult struct {
 type Traverser struct {
 	config   *TraverserConfig
 	exchange dns.ExchangeFunc
+	visited  map[string]bool
+	depth    int
+	mu       sync.Mutex
 }
 
 func NewTraverser(cfg *TraverserConfig) *Traverser {
@@ -45,6 +48,8 @@ func NewTraverser(cfg *TraverserConfig) *Traverser {
 	return &Traverser{
 		config:   cfg,
 		exchange: nil,
+		visited:  make(map[string]bool),
+		depth:    0,
 	}
 }
 
@@ -155,13 +160,32 @@ func (t *Traverser) discoverRoots(ctx context.Context) ([]net.IP, error) {
 
 func (t *Traverser) processReferral(ctx context.Context, ref *Referral, cache *InfoCache) *Response {
 	if !ref.HasAddresses() {
+		t.mu.Lock()
+		visitedCopy := make(map[string]bool)
+		for k, v := range t.visited {
+			visitedCopy[k] = v
+		}
+		t.mu.Unlock()
+
 		ref.Addresses = t.resolveGlueViaSystem(ctx, ref.Name, cache)
 		if len(ref.Addresses) > 0 {
 			ref.State = StateResolved
 		} else {
-			return &Response{
-				Referral: ref,
-				Type:     RespError,
+			addrs, err := t.ResolveNS(ctx, ref.Name, cache, visitedCopy, t.depth)
+			if err != nil {
+				return &Response{
+					Referral: ref,
+					Type:     RespError,
+				}
+			}
+			if len(addrs) > 0 {
+				ref.Addresses = addrs
+				ref.State = StateResolved
+			} else {
+				return &Response{
+					Referral: ref,
+					Type:     RespError,
+				}
 			}
 		}
 	}
@@ -179,6 +203,131 @@ func (t *Traverser) processReferral(ctx context.Context, ref *Referral, cache *I
 	}
 }
 
+func (t *Traverser) ResolveNS(ctx context.Context, nsName string, cache *InfoCache, visited map[string]bool, depth int) ([]net.IP, error) {
+	if cache != nil {
+		if addrs := cache.LookupGlue(nsName); len(addrs) > 0 {
+			return addrs, nil
+		}
+	}
+
+	if visited != nil {
+		if visited[nsName] {
+			return nil, &CircularReferralError{
+				Name:  nsName,
+				Chain: getVisitedNames(visited),
+			}
+		}
+		visited[nsName] = true
+	}
+
+	if depth > DefaultMaxDepth {
+		return nil, &UnresolvableNameserverError{
+			Name:   nsName,
+			Reason: "max depth exceeded",
+		}
+	}
+
+	roots, err := t.discoverRoots(ctx)
+	if err != nil {
+		return nil, fmt.Errorf("root discovery: %w", err)
+	}
+
+	ref := NewReferral(nsName, dns.TypeA, ".", 0, 1.0, nil)
+	ref.Addresses = roots
+	ref.State = StateResolved
+
+	traversalCache := NewInfoCache(nil)
+	if visited != nil {
+		for name := range visited {
+			traversalCache.StoreGlue(name, []net.IP{})
+		}
+	}
+
+	var addrs []net.IP
+	var lastErr error
+
+	stack := NewStack(DefaultMaxDepth)
+	stack.Push(ref)
+
+	for {
+		select {
+		case <-ctx.Done():
+			return nil, fmt.Errorf("resolution cancelled: %w", ctx.Err())
+		default:
+		}
+
+		current := stack.Pop()
+		if current == nil {
+			break
+		}
+
+		cacheForStep := traversalCache
+		if current.Parent != nil {
+			cacheForStep = traversalCache.Child()
+		}
+
+		resp := t.processReferral(ctx, current, cacheForStep)
+
+		if resp.Type == RespAnswer && len(resp.Decoded.Answers) > 0 {
+			for _, rr := range resp.Decoded.Answers {
+				if a, ok := rr.(*miekgdns.A); ok {
+					addrs = append(addrs, a.A)
+				}
+				if aaaa, ok := rr.(*miekgdns.AAAA); ok {
+					addrs = append(addrs, aaaa.AAAA)
+				}
+			}
+			if len(addrs) > 0 {
+				if cache != nil {
+					cache.StoreGlue(nsName, addrs)
+				}
+				return addrs, nil
+			}
+		}
+
+		if resp.Type == RespNXDOMAIN {
+			lastErr = &UnresolvableNameserverError{
+				Name:   nsName,
+				Reason: "NXDOMAIN",
+			}
+			break
+		}
+
+		if resp.Type == RespSERVFAIL || resp.Type == RespError {
+			lastErr = fmt.Errorf("server error resolving %s: %s", nsName, resp.Type)
+			continue
+		}
+
+		if resp.Type == RespReferral {
+			children := resp.ChildReferrals()
+			for _, child := range children {
+				if visited != nil && visited[child.Name] {
+					continue
+				}
+				if !stack.Push(child) {
+					lastErr = &UnresolvableNameserverError{
+						Name:   nsName,
+						Reason: "max depth exceeded during resolution",
+					}
+				}
+			}
+		}
+	}
+
+	if len(addrs) > 0 {
+		return addrs, nil
+	}
+
+	if lastErr != nil {
+		return nil, lastErr
+	}
+
+	return nil, &UnresolvableNameserverError{
+		Name:   nsName,
+		Reason: "resolution exhausted without answer",
+	}
+}
+
 func (t *Traverser) queryServer(ctx context.Context, ref *Referral, server net.IP, cache *InfoCache) *Response {
 	var msg *miekgdns.Msg
 	var err error