fix: rollback reloading and refresh fixed outbound alive state

cmd/run.go

@@ -186,6 +186,7 @@ loop:
 						"err": err,
 					}).Fatalln("[Reload] Failed to roll back configuration")
 				}
+				newConf = conf
 				log.Warnln("[Reload] Last reload failed; rolled back configuration")
 			} else {
 				log.Warnln("[Reload] Stopped old control plane")

component/outbound/dialer_group.go

@@ -7,10 +7,10 @@ package outbound
 
 import (
 	"fmt"
-	"github.com/mzz2017/softwind/netproxy"
-	"github.com/sirupsen/logrus"
 	"github.com/daeuniverse/dae/common/consts"
 	"github.com/daeuniverse/dae/component/outbound/dialer"
+	"github.com/mzz2017/softwind/netproxy"
+	"github.com/sirupsen/logrus"
 	"time"
 )
 
@@ -38,84 +38,96 @@ func NewDialerGroup(option *dialer.GlobalOption, name string, dialers []*dialer.
 	var aliveDnsUdp4DialerSet *dialer.AliveDialerSet
 	var aliveDnsUdp6DialerSet *dialer.AliveDialerSet
 
+	var needAliveState bool
+
 	switch p.Policy {
 	case consts.DialerSelectionPolicy_Random,
 		consts.DialerSelectionPolicy_MinLastLatency,
 		consts.DialerSelectionPolicy_MinAverage10Latencies,
 		consts.DialerSelectionPolicy_MinMovingAverageLatencies:
-
 		// Need to know the alive state or latency.
-		networkType := &dialer.NetworkType{
+		needAliveState = true
-			L4Proto:   consts.L4ProtoStr_TCP,
+
-			IpVersion: consts.IpVersionStr_4,
+	case consts.DialerSelectionPolicy_Fixed:
-			IsDns:     false,
+		// No need to know if the dialer is alive.
-		}
+		needAliveState = false
+
+	default:
+		log.Panicf("Unexpected dialer selection policy: %v", p.Policy)
+	}
+
+	networkType := &dialer.NetworkType{
+		L4Proto:   consts.L4ProtoStr_TCP,
+		IpVersion: consts.IpVersionStr_4,
+		IsDns:     false,
+	}
+	if needAliveState {
 		aliveTcp4DialerSet = dialer.NewAliveDialerSet(
 			log, name, networkType, option.CheckTolerance, p.Policy, dialers,
 			func(networkType *dialer.NetworkType) func(alive bool) {
 				// Use the trick to copy a pointer of *dialer.NetworkType.
 				return func(alive bool) { aliveChangeCallback(alive, networkType, false) }
 			}(networkType), true)
-		aliveChangeCallback(true, networkType, true)
+	}
+	aliveChangeCallback(true, networkType, true)
 
-		networkType = &dialer.NetworkType{
+	networkType = &dialer.NetworkType{
-			L4Proto:   consts.L4ProtoStr_TCP,
+		L4Proto:   consts.L4ProtoStr_TCP,
-			IpVersion: consts.IpVersionStr_6,
+		IpVersion: consts.IpVersionStr_6,
-			IsDns:     false,
+		IsDns:     false,
-		}
+	}
+	if needAliveState {
 		aliveTcp6DialerSet = dialer.NewAliveDialerSet(
 			log, name, networkType, option.CheckTolerance, p.Policy, dialers,
 			func(networkType *dialer.NetworkType) func(alive bool) {
 				// Use the trick to copy a pointer of *dialer.NetworkType.
 				return func(alive bool) { aliveChangeCallback(alive, networkType, false) }
 			}(networkType), true)
-		aliveChangeCallback(true, networkType, true)
+	}
+	aliveChangeCallback(true, networkType, true)
 
-		networkType = &dialer.NetworkType{
+	networkType = &dialer.NetworkType{
-			L4Proto:   consts.L4ProtoStr_UDP,
+		L4Proto:   consts.L4ProtoStr_UDP,
-			IpVersion: consts.IpVersionStr_4,
+		IpVersion: consts.IpVersionStr_4,
-			IsDns:     true,
+		IsDns:     true,
-		}
+	}
+	if needAliveState {
 		aliveDnsUdp4DialerSet = dialer.NewAliveDialerSet(
 			log, name, networkType, option.CheckTolerance, p.Policy, dialers,
 			func(networkType *dialer.NetworkType) func(alive bool) {
 				// Use the trick to copy a pointer of *dialer.NetworkType.
 				return func(alive bool) { aliveChangeCallback(alive, networkType, false) }
 			}(networkType), true)
-		aliveChangeCallback(true, networkType, true)
+	}
+	aliveChangeCallback(true, networkType, true)
 
-		networkType = &dialer.NetworkType{
+	networkType = &dialer.NetworkType{
-			L4Proto:   consts.L4ProtoStr_UDP,
+		L4Proto:   consts.L4ProtoStr_UDP,
-			IpVersion: consts.IpVersionStr_6,
+		IpVersion: consts.IpVersionStr_6,
-			IsDns:     true,
+		IsDns:     true,
-		}
+	}
+	if needAliveState {
 		aliveDnsUdp6DialerSet = dialer.NewAliveDialerSet(
 			log, name, networkType, option.CheckTolerance, p.Policy, dialers,
 			func(networkType *dialer.NetworkType) func(alive bool) {
 				// Use the trick to copy a pointer of *dialer.NetworkType.
 				return func(alive bool) { aliveChangeCallback(alive, networkType, false) }
 			}(networkType), true)
-		aliveChangeCallback(true, networkType, true)
+	}
+	aliveChangeCallback(true, networkType, true)
 
-		if option.CheckDnsTcp {
+	if option.CheckDnsTcp && needAliveState {
-			aliveDnsTcp4DialerSet = dialer.NewAliveDialerSet(log, name, &dialer.NetworkType{
+		aliveDnsTcp4DialerSet = dialer.NewAliveDialerSet(log, name, &dialer.NetworkType{
-				L4Proto:   consts.L4ProtoStr_TCP,
+			L4Proto:   consts.L4ProtoStr_TCP,
-				IpVersion: consts.IpVersionStr_4,
+			IpVersion: consts.IpVersionStr_4,
-				IsDns:     true,
+			IsDns:     true,
-			}, option.CheckTolerance, p.Policy, dialers, func(alive bool) {}, true)
+		}, option.CheckTolerance, p.Policy, dialers, func(alive bool) {}, true)
 
-			aliveDnsTcp6DialerSet = dialer.NewAliveDialerSet(log, name, &dialer.NetworkType{
+		aliveDnsTcp6DialerSet = dialer.NewAliveDialerSet(log, name, &dialer.NetworkType{
-				L4Proto:   consts.L4ProtoStr_TCP,
+			L4Proto:   consts.L4ProtoStr_TCP,
-				IpVersion: consts.IpVersionStr_6,
+			IpVersion: consts.IpVersionStr_6,
-				IsDns:     true,
+			IsDns:     true,
-			}, option.CheckTolerance, p.Policy, dialers, func(alive bool) {}, true)
+		}, option.CheckTolerance, p.Policy, dialers, func(alive bool) {}, true)
-		}
-
-	case consts.DialerSelectionPolicy_Fixed:
-		// No need to know if the dialer is alive.
-
-	default:
-		log.Panicf("Unexpected dialer selection policy: %v", p.Policy)
 	}
 
 	for _, d := range dialers {

control/connectivity.go

@@ -7,8 +7,8 @@ package control
 
 import (
 	"github.com/cilium/ebpf"
-	"github.com/sirupsen/logrus"
 	"github.com/daeuniverse/dae/component/outbound/dialer"
+	"github.com/sirupsen/logrus"
 	"golang.org/x/sys/unix"
 	"strconv"
 )
@@ -25,11 +25,17 @@ func FormatL4Proto(l4proto uint8) string {
 
 func (c *controlPlaneCore) OutboundAliveChangeCallback(outbound uint8) func(alive bool, networkType *dialer.NetworkType, isInit bool) {
 	return func(alive bool, networkType *dialer.NetworkType, isInit bool) {
-		if !isInit {
+		select {
+		case <-c.closed.Done():
+			return
+		default:
+		}
+		if !isInit || c.log.IsLevelEnabled(logrus.TraceLevel) {
 			c.log.WithFields(logrus.Fields{
-				"alive":    alive,
+				"alive":      alive,
-				"network":  networkType.StringWithoutDns(),
+				"network":    networkType.StringWithoutDns(),
-				"outbound": c.outboundId2Name[outbound],
+				"outboundId": outbound,
+				"outbound":   c.outboundId2Name[outbound],
 			}).Warnf("Outbound alive state changed, notify the kernel program.")
 		}
 
@@ -37,10 +43,16 @@ func (c *controlPlaneCore) OutboundAliveChangeCallback(outbound uint8) func(aliv
 		if alive {
 			value = 1
 		}
-		_ = c.bpf.OutboundConnectivityMap.Update(bpfOutboundConnectivityQuery{
+		if err := c.bpf.OutboundConnectivityMap.Update(bpfOutboundConnectivityQuery{
 			Outbound:  outbound,
 			L4proto:   networkType.L4Proto.ToL4Proto(),
 			Ipversion: networkType.IpVersion.ToIpVersion(),
-		}, value, ebpf.UpdateAny)
+		}, value, ebpf.UpdateAny); err != nil {
+			c.log.WithFields(logrus.Fields{
+				"alive":    alive,
+				"network":  networkType.StringWithoutDns(),
+				"outbound": c.outboundId2Name[outbound],
+			}).Warnf("Failed to notify the kernel program: %v", err)
+		}
 	}
 }

control/control_plane_core.go

@@ -6,15 +6,16 @@
 package control
 
 import (
+	"context"
 	"fmt"
 	"github.com/cilium/ebpf"
 	ciliumLink "github.com/cilium/ebpf/link"
-	"github.com/mohae/deepcopy"
-	"github.com/safchain/ethtool"
-	"github.com/sirupsen/logrus"
 	"github.com/daeuniverse/dae/common"
 	"github.com/daeuniverse/dae/common/consts"
 	internal "github.com/daeuniverse/dae/pkg/ebpf_internal"
+	"github.com/mohae/deepcopy"
+	"github.com/safchain/ethtool"
+	"github.com/sirupsen/logrus"
 	"github.com/vishvananda/netlink"
 	"golang.org/x/net/dns/dnsmessage"
 	"golang.org/x/sys/unix"
@@ -38,6 +39,9 @@ type controlPlaneCore struct {
 	flip       int
 	isReload   bool
 	bpfEjected bool
+
+	closed context.Context
+	close  context.CancelFunc
 }
 
 func newControlPlaneCore(log *logrus.Logger,
@@ -53,6 +57,7 @@ func newControlPlaneCore(log *logrus.Logger,
 	if !isReload {
 		deferFuncs = append(deferFuncs, bpf.Close)
 	}
+	closed, toClose := context.WithCancel(context.Background())
 	return &controlPlaneCore{
 		log:             log,
 		deferFuncs:      deferFuncs,
@@ -61,6 +66,9 @@ func newControlPlaneCore(log *logrus.Logger,
 		kernelVersion:   kernelVersion,
 		flip:            coreFlip,
 		isReload:        isReload,
+		bpfEjected:      false,
+		closed:          closed,
+		close:           toClose,
 	}
 }
 
@@ -68,6 +76,11 @@ func (c *controlPlaneCore) Flip() {
 	coreFlip = coreFlip&1 ^ 1
 }
 func (c *controlPlaneCore) Close() (err error) {
+	select {
+	case <-c.closed.Done():
+		return nil
+	default:
+	}
 	// Invoke defer funcs in reverse order.
 	for i := len(c.deferFuncs) - 1; i >= 0; i-- {
 		if e := c.deferFuncs[i](); e != nil {
@@ -79,6 +92,7 @@ func (c *controlPlaneCore) Close() (err error) {
 			}
 		}
 	}
+	c.close()
 	return err
 }
 

control/dns_control.go

@@ -289,6 +289,12 @@ type dialArgument struct {
 }
 
 func (c *DnsController) Handle_(dnsMessage *dnsmessage.Message, req *udpRequest) (err error) {
+	if c.log.IsLevelEnabled(logrus.TraceLevel) && len(dnsMessage.Questions) > 0 {
+		q := dnsMessage.Questions[0]
+		c.log.Tracef("Received UDP(DNS) %v <-> %v: %v %v",
+			RefineSourceToShow(req.realSrc, req.realDst.Addr(), req.lanWanFlag), req.realDst.String(), strings.ToLower(q.Name.String()), q.Type,
+		)
+	}
 	if resp := c.LookupDnsRespCache_(dnsMessage); resp != nil {
 		// Send cache to client directly.
 		if err = sendPkt(resp, req.realDst, req.realSrc, req.src, req.lConn, req.lanWanFlag); err != nil {
@@ -296,7 +302,7 @@ func (c *DnsController) Handle_(dnsMessage *dnsmessage.Message, req *udpRequest)
 		}
 		if c.log.IsLevelEnabled(logrus.DebugLevel) && len(dnsMessage.Questions) > 0 {
 			q := dnsMessage.Questions[0]
-			c.log.Tracef("UDP(DNS) %v <-> Cache: %v %v",
+			c.log.Debugf("UDP(DNS) %v <-> Cache: %v %v",
 				RefineSourceToShow(req.realSrc, req.realDst.Addr(), req.lanWanFlag), strings.ToLower(q.Name.String()), q.Type,
 			)
 		}