recover from panic log including stacktrace to help finding the origin (#5392)

core/dnsserver/config.go

@@ -28,6 +28,9 @@ type Config struct {
 	// Debug controls the panic/recover mechanism that is enabled by default.
 	Debug bool
 
+	// Stacktrace controls including stacktrace as part of log from recover mechanism, it is disabled by default.
+	Stacktrace bool
+
 	// The transport we implement, normally just "dns" over TCP/UDP, but could be
 	// DNS-over-TLS or DNS-over-gRPC.
 	Transport string

core/dnsserver/register.go

@@ -154,6 +154,7 @@ func (h *dnsContext) MakeServers() ([]caddy.Server, error) {
 		c.Plugin = c.firstConfigInBlock.Plugin
 		c.ListenHosts = c.firstConfigInBlock.ListenHosts
 		c.Debug = c.firstConfigInBlock.Debug
+		c.Stacktrace = c.firstConfigInBlock.Stacktrace
 		c.TLSConfig = c.firstConfigInBlock.TLSConfig
 	}
 

core/dnsserver/server.go

@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"net"
 	"runtime"
+	"runtime/debug"
 	"strings"
 	"sync"
 	"time"
@@ -41,6 +42,7 @@ type Server struct {
 	graceTimeout time.Duration      // the maximum duration of a graceful shutdown
 	trace        trace.Trace        // the trace plugin for the server
 	debug        bool               // disable recover()
+	stacktrace   bool               // enable stacktrace in recover error log
 	classChaos   bool               // allow non-INET class queries
 }
 
@@ -67,6 +69,7 @@ func NewServer(addr string, group []*Config) (*Server, error) {
 			s.debug = true
 			log.D.Set()
 		}
+		s.stacktrace = site.Stacktrace
 		// set the config per zone
 		s.zones[site.Zone] = site
 
@@ -213,7 +216,11 @@ func (s *Server) ServeDNS(ctx context.Context, w dns.ResponseWriter, r *dns.Msg)
 			// In case the user doesn't enable error plugin, we still
 			// need to make sure that we stay alive up here
 			if rec := recover(); rec != nil {
-				log.Errorf("Recovered from panic in server: %q %v", s.Addr, rec)
+				if s.stacktrace {
+					log.Errorf("Recovered from panic in server: %q %v\n%s", s.Addr, rec, string(debug.Stack()))
+				} else {
+					log.Errorf("Recovered from panic in server: %q %v", s.Addr, rec)
+				}
 				vars.Panic.Inc()
 				errorAndMetricsFunc(s.Addr, w, r, dns.RcodeServerFailure)
 			}

core/dnsserver/server_test.go

@@ -26,6 +26,7 @@ func testConfig(transport string, p plugin.Handler) *Config {
 		ListenHosts: []string{"127.0.0.1"},
 		Port:        "53",
 		Debug:       false,
+		Stacktrace:  false,
 	}
 
 	c.AddPlugin(func(next plugin.Handler) plugin.Handler { return p })
@@ -76,6 +77,27 @@ func TestDebug(t *testing.T) {
 	}
 }
 
+func TestStacktrace(t *testing.T) {
+	configNoStacktrace, configStacktrace := testConfig("dns", testPlugin{}), testConfig("dns", testPlugin{})
+	configStacktrace.Stacktrace = true
+
+	s1, err := NewServer("127.0.0.1:53", []*Config{configStacktrace, configStacktrace})
+	if err != nil {
+		t.Errorf("Expected no error for NewServer, got %s", err)
+	}
+	if !s1.stacktrace {
+		t.Errorf("Expected stacktrace mode enabled for server s1")
+	}
+
+	s2, err := NewServer("127.0.0.1:53", []*Config{configNoStacktrace})
+	if err != nil {
+		t.Errorf("Expected no error for NewServer, got %s", err)
+	}
+	if s2.stacktrace {
+		t.Errorf("Expected stacktrace disabled for server s2")
+	}
+}
+
 func BenchmarkCoreServeDNS(b *testing.B) {
 	s, err := NewServer("127.0.0.1:53", []*Config{testConfig("dns", testPlugin{})})
 	if err != nil {