v1.5.1 增强程序稳定性

EdgeHttpDNS/internal/nodes/httpdns_node.go

@@ -2,6 +2,7 @@ package nodes
 
 import (
 	"errors"
+	"fmt"
 	"log"
 	"net"
 	"os"
@@ -46,7 +47,14 @@ func (n *HTTPDNSNode) Run() {
 		return
 	}
 
-	go n.start()
+	go func() {
+		defer func() {
+			if r := recover(); r != nil {
+				log.Println(fmt.Sprintf("[HTTPDNS_NODE]goroutine panic: %v", r))
+			}
+		}()
+		n.start()
+	}()
 	select {}
 }
 
@@ -108,6 +116,12 @@ func (n *HTTPDNSNode) listenSock() error {
 	}
 
 	go func() {
+		defer func() {
+			if r := recover(); r != nil {
+				log.Println(fmt.Sprintf("[HTTPDNS_NODE]sock goroutine panic: %v", r))
+			}
+		}()
+
 		n.sock.OnCommand(func(cmd *gosock.Command) {
 			switch cmd.Code {
 			case "pid":
@@ -152,12 +166,47 @@ func (n *HTTPDNSNode) start() {
 	taskManager := NewTaskManager(n.quitCh, snapshotManager)
 	resolveServer := NewResolveServer(n.quitCh, snapshotManager)
 
-	go snapshotManager.Start()
-	go statusManager.Start()
-	go taskManager.Start()
-	go resolveServer.Start()
+	go func() {
+		defer func() {
+			if r := recover(); r != nil {
+				log.Println(fmt.Sprintf("[HTTPDNS_NODE]snapshot goroutine panic: %v", r))
+			}
+		}()
+		snapshotManager.Start()
+	}()
+	go func() {
+		defer func() {
+			if r := recover(); r != nil {
+				log.Println(fmt.Sprintf("[HTTPDNS_NODE]status goroutine panic: %v", r))
+			}
+		}()
+		statusManager.Start()
+	}()
+	go func() {
+		defer func() {
+			if r := recover(); r != nil {
+				log.Println(fmt.Sprintf("[HTTPDNS_NODE]task goroutine panic: %v", r))
+			}
+		}()
+		taskManager.Start()
+	}()
+	go func() {
+		defer func() {
+			if r := recover(); r != nil {
+				log.Println(fmt.Sprintf("[HTTPDNS_NODE]resolve goroutine panic: %v", r))
+			}
+		}()
+		resolveServer.Start()
+	}()
 
-	go NewUpgradeManager().Loop()
+	go func() {
+		defer func() {
+			if r := recover(); r != nil {
+				log.Println(fmt.Sprintf("[HTTPDNS_NODE]upgrade goroutine panic: %v", r))
+			}
+		}()
+		NewUpgradeManager().Loop()
+	}()
 }
 
 func (n *HTTPDNSNode) stop() {

EdgeHttpDNS/internal/nodes/resolve_server.go

@@ -166,7 +166,14 @@ func NewResolveServer(quitCh <-chan struct{}, snapshotManager *SnapshotManager)
 }
 
 func (s *ResolveServer) Start() {
-	go s.startAccessLogFlusher()
+	go func() {
+		defer func() {
+			if r := recover(); r != nil {
+				log.Println(fmt.Sprintf("[HTTPDNS_NODE][resolve]access log flusher panic: %v", r))
+			}
+		}()
+		s.startAccessLogFlusher()
+	}()
 
 	// 1. Load initial certificate from file (fallback)
 	if len(s.certFile) > 0 && len(s.keyFile) > 0 {
@@ -323,6 +330,12 @@ func (s *ResolveServer) startListener(addr string) error {
 	}
 
 	go func() {
+		defer func() {
+			if r := recover(); r != nil {
+				log.Println(fmt.Sprintf("[HTTPDNS_NODE][resolve]serve goroutine panic: %v", r))
+			}
+		}()
+
 		if err := srv.Serve(tlsLn); err != nil && !errors.Is(err, http.ErrServerClosed) {
 			log.Println("[HTTPDNS_NODE][resolve]serve failed on", addr, ":", err.Error())
 		}
@@ -498,11 +511,18 @@ func (s *ResolveServer) handleResolve(writer http.ResponseWriter, request *http.
 	clientIP := detectClientIP(request, query.Get("cip"))
 	clientProfile := buildClientRouteProfile(clientIP)
 
+	// ECS 只使用可信来源的 IP（cip 参数或直连 RemoteAddr），
+	// 不信任 X-Forwarded-For 等可被客户端伪造的转发头
+	ecsIP := normalizeIPCandidate(query.Get("cip"))
+	if len(ecsIP) == 0 {
+		ecsIP = normalizeIPCandidate(request.RemoteAddr)
+	}
+
 	clusterTTL := pickDefaultTTL(snapshot, loadedApp.App)
 	rule, records, ttl := pickRuleRecords(loadedDomain.Rules, qtype, clientProfile, clusterTTL)
 	if len(records) == 0 {
 		// Fallback：回源上游 DNS 查询真实记录
-		fallbackRecords, fallbackTTL, fallbackErr := fallbackResolve(domain, qtype)
+		fallbackRecords, fallbackTTL, fallbackErr := fallbackResolve(domain, qtype, ecsIP, isMainlandChinaCountry(clientProfile.Country))
 		if fallbackErr != nil || len(fallbackRecords) == 0 {
 			errMsg := "未找到解析记录"
 			if fallbackErr != nil {
@@ -929,8 +949,49 @@ func pickRuleRecords(rules []*pb.HTTPDNSCustomRule, qtype string, profile *clien
 	return bestRule, bestRecords, bestTTL
 }
 
+// addECSOption 向 DNS 请求中设置 EDNS Client Subnet (ECS)。
+// 如果请求已携带 ECS 则覆盖（避免双 ECS 导致上游 malformed request）。
+func addECSOption(m *dns.Msg, clientIP string) {
+	if len(clientIP) == 0 {
+		return
+	}
+	ip := net.ParseIP(clientIP)
+	if ip == nil {
+		return
+	}
+
+	ecs := &dns.EDNS0_SUBNET{
+		Code: dns.EDNS0SUBNET,
+	}
+	if ip.To4() != nil {
+		ecs.Family = 1 // IPv4
+		ecs.SourceNetmask = 24
+		ecs.Address = ip.To4()
+	} else {
+		ecs.Family = 2 // IPv6
+		ecs.SourceNetmask = 56
+		ecs.Address = ip
+	}
+
+	opt := m.IsEdns0()
+	if opt == nil {
+		m.SetEdns0(4096, false)
+		opt = m.IsEdns0()
+	}
+	if opt != nil {
+		// 删除已有的 ECS option，避免出现双 EDNS0_SUBNET
+		var filtered []dns.EDNS0
+		for _, o := range opt.Option {
+			if o.Option() != dns.EDNS0SUBNET {
+				filtered = append(filtered, o)
+			}
+		}
+		opt.Option = append(filtered, ecs)
+	}
+}
+
 // fallbackResolve 当无自定义规则命中时，回源上游 DNS 查询真实记录（对齐 EdgeDNS 做法）
-func fallbackResolve(domain string, qtype string) ([]*resolveRecord, int32, error) {
+func fallbackResolve(domain string, qtype string, clientIP string, isChina bool) ([]*resolveRecord, int32, error) {
 	var dnsType uint16
 	switch qtype {
 	case "A":
@@ -945,16 +1006,18 @@ func fallbackResolve(domain string, qtype string) ([]*resolveRecord, int32, erro
 	m.SetQuestion(dns.Fqdn(domain), dnsType)
 	m.RecursionDesired = true
 
-	// 优先使用本机 /etc/resolv.conf 中的 DNS 服务器（对齐 EdgeDNS）
-	var upstream = "223.5.5.5:53"
-	resolveConfig, confErr := dns.ClientConfigFromFile("/etc/resolv.conf")
-	if confErr == nil && len(resolveConfig.Servers) > 0 {
-		port := resolveConfig.Port
-		if len(port) == 0 {
-			port = "53"
-		}
-		server := resolveConfig.Servers[rands.Int(0, len(resolveConfig.Servers)-1)]
-		upstream = server + ":" + port
+	// 携带客户端真实 IP（ECS）向上游查询，确保 GeoIP 结果准确
+	addECSOption(m, clientIP)
+
+	// 根据客户端 IP 地理位置选择上游 DNS
+	// 中国大陆客户端 → 阿里 DNS（出口在中国，ECS + 出口双重匹配）
+	// 海外客户端 → Google DNS（出口在海外）
+	// 这样递归 DNS 出口和 ECS 子网指向同一区域，CDN 调度更准确
+	var upstream string
+	if isChina {
+		upstream = "223.5.5.5:53"
+	} else {
+		upstream = "8.8.8.8:53"
 	}
 
 	r, _, err := sharedRecursionDNSClient.Exchange(m, upstream)

EdgeHttpDNS/internal/nodes/upgrade_manager.go

@@ -91,6 +91,12 @@ func (this *UpgradeManager) Start() {
 	log.Println("[UPGRADE_MANAGER]upgrade successfully")
 
 	go func() {
+		defer func() {
+			if r := recover(); r != nil {
+				log.Println("[UPGRADE_MANAGER]goroutine panic:", r)
+			}
+		}()
+
 		err = this.restart()
 		if err != nil {
 			log.Println("[UPGRADE_MANAGER]" + err.Error())