[client] Surface DNS forwarder upstream failures via Extended DNS Errors (#6441)

client/internal/dnsfwd/forwarder.go

@@ -26,6 +26,15 @@ import (
 const errResolveFailed = "failed to resolve query for domain=%s: %v"
 const upstreamTimeout = 15 * time.Second
 
+// EDE info codes the forwarder emits on upstream failures so the querying
+// client can see the reason without inspecting this peer's logs. They live in
+// the RFC 8914 Private Use range (49152-65535); the Go resolver never exposes a
+// real upstream EDE here, so these cannot collide with a genuine code.
+const (
+	edeNetbirdUpstreamTimeout uint16 = 49152
+	edeNetbirdUpstreamFailure uint16 = 49153
+)
+
 type resolver interface {
 	LookupNetIP(ctx context.Context, network, host string) ([]netip.Addr, error)
 }
@@ -220,7 +229,7 @@ func (f *DNSForwarder) handleDNSQuery(logger *log.Entry, w dns.ResponseWriter, q
 
 	result := resutil.LookupIP(ctx, f.resolver, network, qname, question.Qtype)
 	if result.Err != nil {
-		f.handleDNSError(ctx, logger, w, question, resp, qname, result, startTime)
+		f.handleDNSError(ctx, logger, w, question, resp, qname, result, query.IsEdns0() != nil, startTime)
 		return
 	}
 
@@ -333,6 +342,7 @@ func (f *DNSForwarder) handleDNSError(
 	resp *dns.Msg,
 	domain string,
 	result resutil.LookupResult,
+	reqHasEdns bool,
 	startTime time.Time,
 ) {
 	qType := question.Qtype
@@ -374,6 +384,10 @@ func (f *DNSForwarder) handleDNSError(
 		logger.Warnf(errResolveFailed, domain, result.Err)
 	}
 
+	if reqHasEdns {
+		attachEDE(resp, edeCodeFor(dnsErr), edeText(dnsErr))
+	}
+
 	f.writeResponse(logger, w, resp, domain, startTime)
 }
 
@@ -414,3 +428,33 @@ func (f *DNSForwarder) getMatchingEntries(domain string) (route.ResID, []*Forwar
 
 	return selectedResId, matches
 }
+
+// edeCodeFor maps an upstream lookup error to the NetBird EDE info code.
+func edeCodeFor(dnsErr *net.DNSError) uint16 {
+	if dnsErr != nil && dnsErr.IsTimeout {
+		return edeNetbirdUpstreamTimeout
+	}
+	return edeNetbirdUpstreamFailure
+}
+
+// edeText builds the EDE extra-text describing the class of upstream failure.
+// It deliberately omits the upstream server address, which may be an internal
+// resolver and is exposed to any client permitted to use the route; the full
+// detail stays in the forwarder's local log.
+func edeText(dnsErr *net.DNSError) string {
+	if dnsErr != nil && dnsErr.IsTimeout {
+		return "netbird forwarder: upstream timeout"
+	}
+	return "netbird forwarder: upstream failure"
+}
+
+// attachEDE adds an Extended DNS Error (RFC 8914) option to the response,
+// creating the OPT pseudo-record if the response does not already carry one.
+func attachEDE(resp *dns.Msg, code uint16, text string) {
+	opt := resp.IsEdns0()
+	if opt == nil {
+		resp.SetEdns0(dns.DefaultMsgSize, false)
+		opt = resp.IsEdns0()
+	}
+	opt.Option = append(opt.Option, &dns.EDNS0_EDE{InfoCode: code, ExtraText: text})
+}

client/internal/dnsfwd/forwarder_test.go

@@ -16,6 +16,7 @@ import (
 	"github.com/stretchr/testify/require"
 
 	firewall "github.com/netbirdio/netbird/client/firewall/manager"
+	"github.com/netbirdio/netbird/client/internal/dns/resutil"
 	"github.com/netbirdio/netbird/client/internal/dns/test"
 	"github.com/netbirdio/netbird/client/internal/peer"
 	"github.com/netbirdio/netbird/route"
@@ -617,6 +618,85 @@ func TestDNSForwarder_ResponseCodes(t *testing.T) {
 	}
 }
 
+func TestDNSForwarder_UpstreamFailureEDE(t *testing.T) {
+	tests := []struct {
+		name        string
+		lookupErr   error
+		reqEdns     bool
+		wantEDE     bool
+		wantCode    uint16
+		wantTextHas string
+	}{
+		{
+			name:        "timeout with edns0",
+			lookupErr:   &net.DNSError{Err: "i/o timeout", Server: "10.0.0.53:53", IsTimeout: true},
+			reqEdns:     true,
+			wantEDE:     true,
+			wantCode:    edeNetbirdUpstreamTimeout,
+			wantTextHas: "netbird forwarder: upstream timeout",
+		},
+		{
+			name:        "server failure with edns0",
+			lookupErr:   &net.DNSError{Err: "server misbehaving", Server: "10.0.0.53:53"},
+			reqEdns:     true,
+			wantEDE:     true,
+			wantCode:    edeNetbirdUpstreamFailure,
+			wantTextHas: "netbird forwarder: upstream failure",
+		},
+		{
+			name:      "no edns0 in request omits ede",
+			lookupErr: &net.DNSError{Err: "server misbehaving", Server: "10.0.0.53:53"},
+			reqEdns:   false,
+			wantEDE:   false,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			mockResolver := &MockResolver{}
+			forwarder := NewDNSForwarder(netip.MustParseAddrPort("127.0.0.1:0"), 300, nil, &peer.Status{}, nil)
+			forwarder.resolver = mockResolver
+
+			d, err := domain.FromString("example.com")
+			require.NoError(t, err)
+			forwarder.UpdateDomains([]*ForwarderEntry{{Domain: d, ResID: "test-res"}})
+
+			mockResolver.On("LookupNetIP", mock.Anything, "ip4", "example.com.").
+				Return([]netip.Addr(nil), tt.lookupErr).Once()
+
+			query := &dns.Msg{}
+			query.SetQuestion("example.com.", dns.TypeA)
+			if tt.reqEdns {
+				query.SetEdns0(dns.DefaultMsgSize, false)
+			}
+
+			var writtenResp *dns.Msg
+			mockWriter := &test.MockResponseWriter{
+				WriteMsgFunc: func(m *dns.Msg) error {
+					writtenResp = m
+					return nil
+				},
+			}
+
+			forwarder.handleDNSQuery(log.NewEntry(log.StandardLogger()), mockWriter, query, time.Now())
+			mockResolver.AssertExpectations(t)
+
+			require.NotNil(t, writtenResp, "expected a response")
+			assert.Equal(t, dns.RcodeServerFailure, writtenResp.Rcode, "upstream failure must be SERVFAIL")
+
+			ede, ok := resutil.ExtractEDE(writtenResp)
+			if !tt.wantEDE {
+				assert.False(t, ok, "response must not carry EDE")
+				return
+			}
+			require.True(t, ok, "response must carry EDE")
+			assert.Equal(t, tt.wantCode, ede.InfoCode, "EDE info code")
+			assert.Contains(t, ede.ExtraText, tt.wantTextHas, "EDE extra-text")
+			assert.NotContains(t, ede.ExtraText, "10.0.0.53", "must not leak upstream server address")
+		})
+	}
+}
+
 func TestDNSForwarder_TCPTruncation(t *testing.T) {
 	// Test that large UDP responses are truncated with TC bit set
 	mockResolver := &MockResolver{}