fix: download binaries from JP01 :3000

connMap type changed from map[string]net.Conn to map[string]io.Closer.
RoutedPacketConnection now registers in connMap + uuidConns, so
CloseByUUID and CloseByID tear down both TCP and UDP.

.github/workflows/ci.yml

@@ -56,12 +56,18 @@ jobs:
             make build-linux-arm64
           fi
 
-      - name: Upload Artifact
+      - name: Upload xboard-node artifact
         uses: actions/upload-artifact@v4
         with:
           name: xboard-node-${{ matrix.goos }}-${{ matrix.goarch }}
           path: xboard-node-${{ matrix.goos }}-${{ matrix.goarch }}
 
+      - name: Upload xbctl artifact
+        uses: actions/upload-artifact@v4
+        with:
+          name: xbctl-${{ matrix.goos }}-${{ matrix.goarch }}
+          path: xbctl-${{ matrix.goos }}-${{ matrix.goarch }}
+
   build-docker:
     name: Build & Push Docker Image
     runs-on: ubuntu-latest
@@ -118,3 +124,5 @@ jobs:
           files: |
             dist/xboard-node-linux-amd64
             dist/xboard-node-linux-arm64
+            dist/xbctl-linux-amd64
+            dist/xbctl-linux-arm64

.gitignore

@@ -22,3 +22,13 @@ Thumbs.db
 # Go
 vendor/
 config.json
+benchmark/
+third_party
+bench-results
+internal/kernel/singbox/config_bench_test.go
+internal/kernel/xray/config_bench_test.go
+internal/service/service_bench_test.go
+internal/tracker/tracker_bench_test.go
+cmd/xboard-node-bench-compare
+cmd/xboard-node-bench
+.env

CLAUDE.md

@@ -0,0 +1,106 @@
+# xboard-node fork 运维手册
+
+## 仓库说明
+
+JP01 Gitea 自维护 fork，**不是上游镜像**，独立分支。
+
+- 本仓库：`http://103.112.1.141:3000/will/xboard-node`（dev 分支）
+- 上游（仅供 cherry-pick 参考）：`https://github.com/cedar2025/Xboard-Node`
+- 工作目录：`/tmp/xboard-node`（JP01）
+
+---
+
+## 已应用补丁
+
+| 补丁 | 文件 | 说明 | 是否可移除 |
+|------|------|------|-----------|
+| #32 | `internal/service/service.go` | ban 用户前调 `CloseUserConnections()`，立即断连 | 待上游合并后移除 |
+| #55 | `internal/kernel/singbox/conntracker.go` | UDP PacketConn upload/download 计数对调 | 待上游合并后移除 |
+| 自改 | `cmd/xbctl/main.go` | `downloadBase` 指向 JP01 Gitea releases | 永久保留 |
+| 自改 | `cmd/xbctl/main.go` | `downloadToken` + `downloadFile` 加 Authorization header，支持 private 仓库下载 | 永久保留 |
+| 自改 | `install.sh` | binary 下载地址改为 JP01 Gitea releases | 永久保留 |
+
+### xbctl private 仓库下载（2026-06-24）
+
+**背景**：Gitea 仓库改为 private 后，`xbctl upgrade` 下载时 HTTP 401。
+
+**改动位置**：`cmd/xbctl/main.go`
+
+```go
+// 新增常量
+downloadToken = "cf6634b57d2c60c66030dbb29d77c4ed25370e94"
+
+// downloadFile 改用 NewRequest，加 Authorization header
+func downloadFile(url, dest string) error {
+    client := &http.Client{Timeout: 120 * time.Second}
+    req, err := http.NewRequest("GET", url, nil)
+    if err != nil {
+        return err
+    }
+    if downloadToken != "" {
+        req.Header.Set("Authorization", "token "+downloadToken)
+    }
+    resp, err := client.Do(req)
+    ...
+}
+```
+
+**注意**：token 是只读 `read:repository` 权限，泄露只能读仓库，不能写。
+
+---
+
+## 编译命令
+
+```bash
+cd /tmp/xboard-node
+VERSION=$(git describe --tags --always --dirty)
+
+# xboard-node（必须带这些 tags）
+CGO_ENABLED=0 /usr/local/go/bin/go build \
+  -ldflags "-s -w -X main.version=${VERSION}" \
+  -tags "with_quic with_utls with_wireguard with_acme with_clash_api" \
+  -o xboard-node-linux-amd64 ./cmd/xboard-node/
+
+# xbctl
+CGO_ENABLED=0 /usr/local/go/bin/go build \
+  -ldflags "-s -w" \
+  -o xbctl-linux-amd64 ./cmd/xbctl/
+```
+
+---
+
+## 发布流程
+
+```bash
+TOKEN=cf6634b57d2c60c66030dbb29d77c4ed25370e94
+RELEASE_ID=14  # v1.13-patched
+
+# 删旧 asset
+OLD_ID=$(curl -s "http://localhost:3000/api/v1/repos/will/xboard-node/releases/$RELEASE_ID/assets" \
+  -H "Authorization: token $TOKEN" | python3 -c "
+import sys,json
+for a in json.load(sys.stdin):
+    if a['name'] == 'xboard-node-linux-amd64': print(a['id'])")
+curl -s -X DELETE "http://localhost:3000/api/v1/repos/will/xboard-node/releases/$RELEASE_ID/assets/$OLD_ID" \
+  -H "Authorization: token $TOKEN"
+
+# 上传新 asset
+curl -s -X POST "http://localhost:3000/api/v1/repos/will/xboard-node/releases/$RELEASE_ID/assets?name=xboard-node-linux-amd64" \
+  -H "Authorization: token $TOKEN" -H "Content-Type: application/octet-stream" \
+  --data-binary @xboard-node-linux-amd64
+```
+
+---
+
+## 节点部署
+
+| 节点 | IP | MCP |
+|------|----|-----|
+| dmit-tyo | 154.31.113.157 | mcp__dmit-tyo__exec |
+| us-unlock | 103.152.113.30 | mcp__us-unlock__exec |
+| vmiss-us01 | 38.47.112.162 | mcp__vmiss-us04__exec |
+
+节点上执行升级：
+```bash
+xbctl upgrade  # 自动带 token，直接用
+```

Dockerfile

@@ -1,5 +1,5 @@
 # Build stage
-FROM golang:1.25-alpine AS builder
+FROM golang:1.26-alpine AS builder
 
 RUN apk add --no-cache git
 

Makefile

@@ -2,19 +2,22 @@ VERSION ?= $(shell git describe --tags --always --dirty 2>/dev/null || echo dev)
 BUILD_TIME ?= $(shell date -u +%Y-%m-%dT%H:%M:%SZ)
 LDFLAGS := -s -w -X main.version=$(VERSION) -X main.buildTime=$(BUILD_TIME) -X main.commit=$(shell git rev-parse --short HEAD 2>/dev/null || echo unknown)
 
-.PHONY: build clean test docker install
+.PHONY: build clean test docker install build-linux build-linux-arm64 build-all
 
 # Build for current platform
 build:
 	go build -ldflags "$(LDFLAGS)" -tags "with_quic with_utls with_wireguard with_clash_api" -o xboard-node ./cmd/xboard-node
+	go build -ldflags "$(LDFLAGS)" -o xbctl ./cmd/xbctl
 
 # Build for Linux amd64
 build-linux:
 	CGO_ENABLED=0 GOOS=linux GOARCH=amd64 go build -ldflags "$(LDFLAGS)" -tags "with_quic with_utls with_wireguard with_acme with_clash_api" -o xboard-node-linux-amd64 ./cmd/xboard-node
+	CGO_ENABLED=0 GOOS=linux GOARCH=amd64 go build -ldflags "$(LDFLAGS)" -o xbctl-linux-amd64 ./cmd/xbctl
 
 # Build for Linux arm64
 build-linux-arm64:
 	CGO_ENABLED=0 GOOS=linux GOARCH=arm64 go build -ldflags "$(LDFLAGS)" -tags "with_quic with_utls with_wireguard with_acme with_clash_api" -o xboard-node-linux-arm64 ./cmd/xboard-node
+	CGO_ENABLED=0 GOOS=linux GOARCH=arm64 go build -ldflags "$(LDFLAGS)" -o xbctl-linux-arm64 ./cmd/xbctl
 
 # Build all platforms
 build-all: build-linux build-linux-arm64
@@ -25,7 +28,7 @@ test:
 
 # Clean build artifacts
 clean:
-	rm -f xboard-node xboard-node-linux-*
+	rm -f xboard-node xbctl xboard-node-linux-* xbctl-linux-*
 
 # Build Docker image
 docker:
@@ -34,6 +37,7 @@ docker:
 # Install to system (single node, legacy compat)
 install: build
 	sudo cp xboard-node /usr/local/bin/
+	sudo cp xbctl /usr/local/bin/
 	sudo mkdir -p /etc/xboard-node
 	@if [ ! -f /etc/xboard-node/config.yml ]; then \
 		sudo cp config.yml.example /etc/xboard-node/config.yml; \

README.md

@@ -1,45 +1,70 @@
 # xboard-node
 
-Dedicated node backend for [Xboard](https://github.com/cedar2025/Xboard). Fully compatible with Xboard API.
+Node backend for [Xboard](https://github.com/cedar2025/Xboard). Supports `sing-box` / `xray-core` dual kernels.
 
 > **Disclaimer**: This project is for educational and learning purposes only.
 
+## Features
+
+- Protocols: V2Ray family, Trojan, Shadowsocks, Hysteria2, TUIC, AnyTLS
+- Sync: WebSocket push + REST polling dual channel
+- User controls: speed limit, device limit, alive-IP tracking, hot update
+- Deploy modes: node mode, machine mode, standalone mode
+- Multi-instance: single process binding multiple panels / nodes
+
 ## Install
 
-### Script
-```bash
-bash <(curl -Ls https://raw.githubusercontent.com/cedar2025/xboard-node/dev/install.sh) \
-  -a https://panel.com -t TOKEN -n 1
-```
-
 ### Docker
+
 ```bash
 docker run -d --restart=always --network=host \
-  -e apiHost=https://panel.com \
-  -e apiKey=TOKEN \
-  -e nodeID=1 \
+  -e apiHost=https://panel.com -e apiKey=TOKEN -e nodeID=1 \
   ghcr.io/cedar2025/xboard-node:latest
 ```
 
-## Features
+### Docker Compose
 
-- **Kernels**: sing-box (default) / Xray-core.
-- **Protocols**: Full coverage (V2Ray, Trojan, SS, Hysteria2, TUIC, Naive).
-- **Speed**: Kernel-level rate limiting.
-- **Sync**: Real-time WebSocket + REST fallback.
-- **Dev**: Single Go binary.
+```bash
+git clone -b compose --depth 1 https://github.com/cedar2025/xboard-node.git
+cd xboard-node
+vim config/config.yml   # set panel.url / token / node_id
+docker compose up -d
+```
+
+### Installer (Linux systemd)
+
+```bash
+# Node mode
+curl -fsSL https://raw.githubusercontent.com/cedar2025/xboard-node/dev/install.sh | \
+  sudo bash -s -- --mode node --panel https://panel.example.com --token TOKEN --node-id 1
+
+# Machine mode
+curl -fsSL https://raw.githubusercontent.com/cedar2025/xboard-node/dev/install.sh | \
+  sudo bash -s -- --mode machine --panel https://panel.example.com --token TOKEN --machine-id 1
+
+## xbctl
+
+Run `xbctl` after installation for help. Common commands:
+
+```bash
+xbctl list                          # list all instances
+xbctl status                        # running status
+xbctl bind add-node --panel URL --token TOKEN --node-id 1
+xbctl bind add-machine --panel URL --token TOKEN --machine-id 1
+xbctl bind remove-node --panel URL --node-id 1
+xbctl service restart
+```
 
 ## Configuration
 
-`config.yml`:
-```yaml
-panel:
-  url: "https://panel.com"
-  token: "token"
-  node_id: 1
-```
+Legacy single-panel config is fully compatible. Appending bindings auto-migrates to `instances` format. See `config.yml.example`.
+
+## Extensions
+
+- Custom routes: [docs-custom-routes.md](docs-custom-routes.md)
+- Custom outbounds: [docs-custom-outbounds.md](docs-custom-outbounds.md)
+- DNS providers (ACME DNS-01): [docs-dns-providers.md](docs-dns-providers.md)
 
 ## License
 
 MPL-2.0.
-

cmd/xboard-node/main.go

@@ -4,7 +4,6 @@ import (
 	"context"
 	"flag"
 	"fmt"
-	"log/slog"
 	"net"
 	"net/http"
 	"os"
@@ -16,6 +15,8 @@ import (
 	"time"
 
 	"github.com/cedar2025/xboard-node/internal/config"
+	"github.com/cedar2025/xboard-node/internal/machine"
+	"github.com/cedar2025/xboard-node/internal/nlog"
 	"github.com/cedar2025/xboard-node/internal/service"
 )
 
@@ -34,51 +35,36 @@ func main() {
 		os.Exit(0)
 	}
 
-	cfg, err := config.Load(*configPath)
+	rootCfg, err := config.LoadRoot(*configPath)
 	if err != nil {
 		fmt.Fprintf(os.Stderr, "failed to load config: %v\n", err)
 		os.Exit(1)
 	}
 
-	config.InitLogger(cfg.Log)
+	instances, err := rootCfg.NormalizeInstances()
+	if err != nil {
+		fmt.Fprintf(os.Stderr, "failed to normalize config: %v\n", err)
+		os.Exit(1)
+	}
+	config.InitLogger(instances[0].Log)
 
 	// Apply runtime memory tuning before anything else allocates.
-	applyRuntimeConfig(cfg.Runtime)
+	applyRuntimeConfig(instances[0].Runtime)
 
-	nodes := cfg.ExpandNodes()
-	slog.Info("xboard-node starting",
-		"version", version,
-		"build_time", buildTime,
-		"nodes", len(nodes),
-	)
+	runWithReload(rootCfg, *configPath)
+}
 
-	ctx, cancel := context.WithCancel(context.Background())
-	defer cancel()
-
-	sigCh := make(chan os.Signal, 1)
-	signal.Notify(sigCh, syscall.SIGINT, syscall.SIGTERM)
-
-	go func() {
-		sig := <-sigCh
-		slog.Info("received signal, shutting down gracefully", "signal", sig)
-		cancel()
-
-		// Force-exit on second signal or after timeout.
-		select {
-		case sig = <-sigCh:
-			slog.Warn("received second signal, forcing exit", "signal", sig)
-			os.Exit(1)
-		case <-time.After(15 * time.Second):
-			slog.Error("shutdown timed out after 15s, forcing exit")
-			os.Exit(2)
+// runWithReload restarts all node services when the config file changes.
+func runWithReload(initialRoot *config.RootConfig, configPath string) {
+	var healthSrv *http.Server
+	var healthPort int
+	startHealth := func(port int) {
+		if port <= 0 {
+			return
 		}
-	}()
-
-	// Optional health-check endpoint for container orchestrators.
-	if cfg.HealthPort > 0 {
-		ln, err := net.Listen("tcp", fmt.Sprintf(":%d", cfg.HealthPort))
+		ln, err := net.Listen("tcp", fmt.Sprintf(":%d", port))
 		if err != nil {
-			slog.Error("failed to start health check listener", "port", cfg.HealthPort, "error", err)
+			nlog.Core().Error("failed to start health check listener", "port", port, "error", err)
 			os.Exit(1)
 		}
 		mux := http.NewServeMux()
@@ -87,46 +73,165 @@ func main() {
 			w.WriteHeader(http.StatusOK)
 			w.Write([]byte(`{"status":"ok"}`))
 		})
-		srv := &http.Server{Handler: mux, ReadHeaderTimeout: 5 * time.Second}
+		healthSrv = &http.Server{Handler: mux, ReadHeaderTimeout: 5 * time.Second}
+		healthPort = port
 		go func() {
-			slog.Info("health check listening", "addr", ln.Addr())
-			if err := srv.Serve(ln); err != nil && err != http.ErrServerClosed {
-				slog.Warn("health check server stopped", "error", err)
-			}
-		}()
-		go func() {
-			<-ctx.Done()
-			srv.Close()
-		}()
-	}
-
-	// Run one service per node. If any node fails, cancel all others.
-	errCh := make(chan error, len(nodes))
-	var wg sync.WaitGroup
-	for _, nodeCfg := range nodes {
-		nodeCfg := nodeCfg // capture
-		wg.Add(1)
-		go func() {
-			defer wg.Done()
-			svc := service.New(nodeCfg)
-			if err := svc.Run(ctx); err != nil {
-				slog.Error("node service exited with error",
-					"node_id", nodeCfg.Panel.NodeID, "error", err)
-				errCh <- err
-				cancel() // bring down all other nodes
-			} else {
-				slog.Info("node service stopped", "node_id", nodeCfg.Panel.NodeID)
+			nlog.Core().Debug(fmt.Sprintf("health check listening on :%d", port))
+			if err := healthSrv.Serve(ln); err != nil && err != http.ErrServerClosed {
+				nlog.Core().Warn("health check server stopped", "error", err)
 			}
 		}()
 	}
 
-	wg.Wait()
-	close(errCh)
-
-	if err := <-errCh; err != nil {
+	initialInstances, err := initialRoot.NormalizeInstances()
+	if err != nil {
+		nlog.Core().Error("failed to normalize initial config", "error", err)
 		os.Exit(1)
 	}
-	slog.Info("xboard-node stopped")
+	startHealth(initialInstances[0].HealthPort)
+	defer func() {
+		if healthSrv != nil {
+			healthSrv.Close()
+		}
+	}()
+
+	for root := initialRoot; ; {
+		ctx, cancel := context.WithCancel(context.Background())
+
+		sigCh := make(chan os.Signal, 1)
+		signal.Notify(sigCh, syscall.SIGINT, syscall.SIGTERM)
+
+		reloadCh := make(chan *config.RootConfig, 1)
+
+		watcher, err := config.WatchConfigRoot(ctx, configPath, func(newCfg *config.RootConfig) {
+			select {
+			case reloadCh <- newCfg:
+			default:
+			}
+		})
+		if err != nil {
+			nlog.Core().Warn("config watcher unavailable, hot-reload disabled", "error", err)
+		}
+
+		go func() {
+			select {
+			case sig := <-sigCh:
+				nlog.Core().Info(fmt.Sprintf("received %v, shutting down...", sig))
+				cancel()
+
+				select {
+				case sig = <-sigCh:
+					nlog.Core().Warn("received second signal, forcing exit", "signal", sig)
+					os.Exit(1)
+				case <-time.After(15 * time.Second):
+					nlog.Core().Error("shutdown timed out after 15s, forcing exit")
+					os.Exit(2)
+				}
+			case <-ctx.Done():
+			}
+		}()
+
+		instances, err := root.NormalizeInstances()
+		if err != nil {
+			nlog.Core().Error("failed to normalize config", "error", err)
+			os.Exit(1)
+		}
+		if err := config.ValidateStartupLayout(instances); err != nil {
+			nlog.Core().Error("startup layout validation failed", "error", err)
+			os.Exit(1)
+		}
+
+		if instances[0].HealthPort != healthPort {
+			if healthSrv != nil {
+				healthSrv.Close()
+				healthSrv = nil
+			}
+			startHealth(instances[0].HealthPort)
+		}
+
+		errCh := make(chan error, len(instances))
+		doneCh := make(chan struct{})
+		var wg sync.WaitGroup
+		for _, instanceCfg := range instances {
+			instanceCfg := instanceCfg
+			wg.Add(1)
+			go func() {
+				defer wg.Done()
+				if instanceCfg.IsMachineMode() {
+					nlog.Core().Info("starting machine instance", "instance", instanceCfg.InstanceID, "machine_id", instanceCfg.Machine.MachineID, "panel_url", instanceCfg.Panel.URL)
+					orch := machine.New(instanceCfg)
+					if err := orch.Run(ctx); err != nil {
+						nlog.Core().Error("machine instance exited with error", "instance", instanceCfg.InstanceID, "error", err)
+						errCh <- err
+						cancel()
+					}
+					return
+				}
+				nodes := instanceCfg.ExpandNodes()
+				nlog.Core().Info("starting node instance", "instance", instanceCfg.InstanceID, "nodes", len(nodes), "panel_url", instanceCfg.Panel.URL)
+				var instanceWG sync.WaitGroup
+				for idx, nodeCfg := range nodes {
+					nodeCfg := nodeCfg
+					instanceWG.Add(1)
+					go func(idx int) {
+						defer instanceWG.Done()
+						if idx > 0 {
+							delay := time.Duration(idx) * 250 * time.Millisecond
+							if delay > 2*time.Second {
+								delay = 2 * time.Second
+							}
+							select {
+							case <-time.After(delay):
+							case <-ctx.Done():
+								return
+							}
+						}
+						svc := service.New(nodeCfg)
+						if err := svc.Run(ctx); err != nil {
+							nlog.Core().Error("node service exited with error", "instance", nodeCfg.InstanceID, "node_id", nodeCfg.Panel.NodeID, "error", err)
+							errCh <- err
+							cancel()
+						}
+					}(idx)
+				}
+				instanceWG.Wait()
+			}()
+		}
+		go func() { wg.Wait(); close(doneCh) }()
+
+		var newRoot *config.RootConfig
+		select {
+		case newRoot = <-reloadCh:
+			nlog.Core().Info("config changed, restarting all services...")
+			cancel()
+			<-doneCh
+		case <-doneCh:
+		}
+
+		signal.Stop(sigCh)
+		if watcher != nil {
+			watcher.Stop()
+		}
+
+		if newRoot == nil {
+			close(errCh)
+			if err := <-errCh; err != nil {
+				os.Exit(1)
+			}
+			nlog.Core().Info("stopped")
+			return
+		}
+
+		newInstances, err := newRoot.NormalizeInstances()
+		if err != nil {
+			nlog.Core().Error("failed to normalize reloaded config", "error", err)
+			os.Exit(1)
+		}
+		config.InitLogger(newInstances[0].Log)
+		applyRuntimeConfig(newInstances[0].Runtime)
+		root = newRoot
+		nlog.Core().Info("reload complete, services restarting with new config")
+	}
 }
 
 // applyRuntimeConfig wires up Go runtime memory limits from the config file.
@@ -138,17 +243,17 @@ func applyRuntimeConfig(rt config.RuntimeConfig) {
 	// GOGC
 	if rt.GoGCPercent > 0 {
 		prev := debug.SetGCPercent(rt.GoGCPercent)
-		slog.Info("runtime: GOGC set", "gogc", rt.GoGCPercent, "prev", prev)
+		nlog.Core().Info("runtime: GOGC set", "gogc", rt.GoGCPercent, "prev", prev)
 	}
 
 	// GOMEMLIMIT — parse human-readable size string (e.g. "30MiB")
 	if rt.GoMemLimit != "" {
 		limit, err := parseMemLimit(rt.GoMemLimit)
 		if err != nil {
-			slog.Warn("runtime: invalid gomemlimit, ignoring", "value", rt.GoMemLimit, "error", err)
+			nlog.Core().Warn("runtime: invalid gomemlimit, ignoring", "value", rt.GoMemLimit, "error", err)
 		} else {
 			prev := debug.SetMemoryLimit(limit)
-			slog.Info("runtime: GOMEMLIMIT set",
+			nlog.Core().Info("runtime: GOMEMLIMIT set",
 				"limit", rt.GoMemLimit,
 				"bytes", limit,
 				"prev_bytes", prev,

config.yml.example

@@ -1,92 +1,26 @@
 # Xboard Node Configuration
-# Copy this file to config.yml and modify as needed
+# Top-level settings are inherited by all instances. Instance overrides take precedence.
 
-# ══════════════════════════════════════════════════════════════════════════════
-# MINIMAL CONFIGURATION (Recommended)
-# Most settings (Node ID, Kernel, TLS, Rules) can be managed via the Panel.
-# ══════════════════════════════════════════════════════════════════════════════
+# ── Single node (simplest) ──
+# panel:
+#   url: "https://panel.example.com"
+#   token: "your-api-key"
+#   node_id: 1
 
-# Panel connection settings
-panel:
-  # Xboard panel URL (required)
-  url: "https://panel.example.com"
-  # Server communication token (required, must match panel settings)
-  token: "your-server-token"
-  # Node ID as configured in the panel
-  node_id: 1
-
-# Kernel settings (optional defaults)
-kernel:
-  # Kernel type: "singbox" (default) or "xray"
-  type: "singbox"
-  # Directory to store generated configs (default: /etc/xboard-node)
-  config_dir: "/etc/xboard-node"
-  # Kernel log level: debug/info/warn/error (default: warn, can be overridden by panel)
-  log_level: "warn"
-
-# TLS certificate settings (optional, can be managed by panel)
-# cert_mode: "http" | "dns" | "self" | "file" | "none"
-#   http  - ACME HTTP-01 challenge (requires port 80 open to internet)
-#   dns   - ACME DNS-01 challenge (supports wildcard certs, requires dns_provider)
-#   self  - self-signed certificate (valid 10 years, no external dependency)
-#   file  - manually specify cert_file / key_file
-#   none  - no TLS (default)
-cert:
-  cert_mode: "none"
-  # Domain for ACME or self-signed cert
-  # domain: "node.example.com"
-  # Email for ACME registration
-  # email: "admin@example.com"
-
-  # ── HTTP-01 mode ──
-  # http_port: 80               # local port for ACME challenge (default: 80)
-
-  # ── DNS-01 mode ──
-  # dns_provider: "cloudflare"   # supported: cloudflare, alidns
-  # dns_env:                     # provider-specific credentials
-  #   CF_API_TOKEN: "your-cloudflare-api-token"
-
-  # ── File mode ──
-  # cert_file: "/etc/ssl/certs/cert.pem"
-  # key_file: "/etc/ssl/private/key.pem"
-
-  # Storage directory for auto certs (default: {config_dir}/certs)
-  # cert_dir: "/etc/xboard-node/certs"
-
-# Logging settings
-log:
-  # Node log level: debug/info/warn/error (default: info)
-  level: "info"
-  # Log output: stdout/stderr or file path (default: stdout)
-  output: "stdout"
-
-# ══════════════════════════════════════════════════════════════════════════════
-# ADVANCED / MULTI-NODE MODE
-# Run multiple nodes with one configuration file.
-# Each node gets its own isolated working directory.
-# ══════════════════════════════════════════════════════════════════════════════
-# When `nodes:` is present, the top-level `panel.node_id` is ignored.
-#
-# nodes:
-#   - node_id: 1
-#
-#   - node_id: 2
-#     # Optional overrides per node
-#     kernel_type: "xray"
-#     cert:
-#       auto_tls: true
-#       domain: "node2.example.com"
-#
-#   - node_id: 3
-#     kernel:
-#       config_dir: "/etc/xboard-node/custom-node-3"
-
-# Runtime memory tuning (optional)
-# runtime:
-#   # Soft memory limit (e.g. "30MiB", "512MiB")
-#   gomemlimit: ""
-#   # GOGC target percentage (default: 100)
-#   gogc: 0
-
-# Health check endpoint for Docker/K8s (optional, 0 = disabled)
-# health_port: 65530
+# ── Multiple instances ──
+# kernel:
+#   type: "singbox"
+# log:
+#   level: "info"
+# instances:
+#   - panel:
+#       url: "https://panel.example.com"
+#       token: "api-key-1"
+#       node_id: 1
+#   - panel:
+#       url: "https://panel.example.com"
+#     machine:
+#       machine_id: 2
+#       token: "machine-token-2"
+#     log:
+#       level: "debug"   # override for this instance only

docs-custom-outbounds.md

@@ -0,0 +1,55 @@
+# Custom Outbounds
+
+## Quick Example
+
+```json
+{
+  "custom_outbounds": [
+    {
+      "tag": "warp",
+      "protocol": "wireguard",
+      "settings": {
+        "peers": [{"address": "162.159.195.1"}],
+        "private_key": "aMzA..."
+      }
+    },
+    {
+      "tag": "us-proxy",
+      "protocol": "socks",
+      "settings": {"servers": [{"address": "1.2.3.4", "port": 1080}]},
+      "proxy_tag": "warp"
+    }
+  ]
+}
+```
+
+## Supported Protocols
+
+| Protocol | Xray | Sing-box |
+|----------|------|----------|
+| vmess | ✅ | ✅ |
+| vless | ✅ | ✅ |
+| trojan | ✅ | ✅ |
+| shadowsocks | ✅ | ✅ |
+| socks | ✅ | ✅ |
+| http | ✅ | ✅ |
+| wireguard | ✅ | ✅ |
+| tuic | ❌ | ✅ |
+| hysteria2 | ❌ | ✅ |
+| anytls | ❌ | ✅ |
+| naive | ❌ | ✅ |
+| mieru | ❌ | ✅ |
+
+## Core Fields
+
+| Field | Required | Description |
+|-------|----------|-------------|
+| `tag` | ✅ | Outbound tag, referenced in routes |
+| `protocol` | ✅ | Protocol type |
+| `settings` | ✅ | Protocol-specific configuration |
+| `proxy_tag` | ❌ | Chain to another outbound (references its tag) |
+
+## Best Practices
+
+- **Prefer** `custom_outbounds`: panel-managed, cross-kernel compatible
+- **Use** `kernel.custom_outbound` only: when native fields are needed

docs-custom-routes.md

@@ -0,0 +1,64 @@
+# Custom Routes
+
+## Quick Example
+
+```json
+{
+  "custom_route_rules": [
+    {
+      "name": "direct-example",
+      "match": {"domain_suffixes": ["example.com"]},
+      "action": {"type": "direct"}
+    },
+    {
+      "name": "block-ads",
+      "match": {"domains": ["ads.example.com"]},
+      "action": {"type": "block"}
+    },
+    {
+      "name": "route-warp",
+      "match": {"ip_cidrs": ["1.1.1.0/24"], "ports": ["80", "443"]},
+      "action": {"type": "route", "target": "warp-out"}
+    }
+  ]
+}
+```
+
+## Match Conditions
+
+| Condition | Description | Example |
+|-----------|-------------|---------|
+| `domains` | Exact domain match | `["api.example.com"]` |
+| `domain_suffixes` | Suffix match | `["example.com"]` |
+| `ip_cidrs` | IP CIDR ranges | `["10.0.0.0/8"]` |
+| `ports` | Port (single or range) | `["443", "8000-9000"]` |
+| `networks` | Protocol | `["tcp"]` or `["udp"]` |
+| `source_cidrs` | Source IP CIDR | `["192.168.1.0/24"]` |
+| `source_ports` | Source port | `["1024-65535"]` |
+
+## Action Types
+
+| Action | Description |
+|--------|-------------|
+| `{"type": "direct"}` | Direct connection, bypass proxy |
+| `{"type": "block"}` | Block connection |
+| `{"type": "route", "target": "tag"}` | Route to specified outbound (by tag) |
+
+## Application Order
+
+1. Structured `custom_route_rules` (highest priority)
+2. Raw `custom_routes`
+3. Built-in blocklist rules
+4. Panel routes
+
+## Kernel Compatibility
+
+| Feature | Xray | Sing-box |
+|---------|------|----------|
+| All match conditions | ✅ | ✅ |
+| direct / block / route | ✅ | ✅ |
+
+## Best Practices
+
+- **Prefer** `custom_route_rules`: cross-kernel compatible, panel-managed
+- **Use** `custom_routes` only: when native features are needed (e.g., load balancing)

docs-dns-providers.md

@@ -0,0 +1,35 @@
+# DNS Providers
+
+Used by `cert_config.cert_mode: "dns"` for ACME DNS-01.
+
+```yaml
+cert_config:
+  cert_mode: "dns"
+  dns_provider: "cloudflare"
+  dns_env:
+    CLOUDFLARE_DNS_API_TOKEN: "xxx"
+```
+
+| provider | aliases | env |
+|---|---|---|
+| cloudflare | cf | `CLOUDFLARE_DNS_API_TOKEN` |
+| alidns | aliyun | `ALICLOUD_ACCESS_KEY_ID`, `ALICLOUD_ACCESS_KEY_SECRET` |
+| tencentcloud | tencent | `TENCENTCLOUD_SECRET_ID`, `TENCENTCLOUD_SECRET_KEY` |
+| route53 | aws | AWS credential chain (env / IAM role / profile) |
+| godaddy | | `GODADDY_API_KEY`, `GODADDY_API_SECRET` |
+| namecheap | | `NAMECHEAP_API_KEY`, `NAMECHEAP_API_USER` |
+| namesilo | | `NAMESILO_API_TOKEN` |
+| digitalocean | do | `DO_AUTH_TOKEN` |
+| linode | | `LINODE_TOKEN` |
+| hetzner | | `HETZNER_API_TOKEN` |
+| gandi | | `GANDI_BEARER_TOKEN` |
+| porkbun | | `PORKBUN_API_KEY`, `PORKBUN_API_SECRET_KEY` |
+| netlify | | `NETLIFY_PERSONAL_ACCESS_TOKEN` |
+| azure | | `AZURE_SUBSCRIPTION_ID`, `AZURE_RESOURCE_GROUP_NAME` |
+| googleclouddns | gcp, gcloud | `GCE_PROJECT` |
+| huaweicloud | huawei | `HUAWEICLOUD_ACCESS_KEY_ID`, `HUAWEICLOUD_SECRET_ACCESS_KEY` |
+| bunny | | `BUNNY_ACCESS_KEY` |
+| duckdns | | `DUCKDNS_TOKEN` |
+| ovh | | `OVH_APPLICATION_KEY`, `OVH_APPLICATION_SECRET`, `OVH_CONSUMER_KEY` |
+| vultr | | `VULTR_API_TOKEN` |
+| desec | | `DESEC_TOKEN` |

go.mod

@@ -1,31 +1,80 @@
 module github.com/cedar2025/xboard-node
 
-go 1.25.7
+go 1.26
 
 require (
 	github.com/caddyserver/certmagic v0.25.2
+	github.com/fsnotify/fsnotify v1.9.0
 	github.com/go-viper/mapstructure/v2 v2.5.0
 	github.com/gorilla/websocket v1.5.3
 	github.com/libdns/alidns v1.0.6
+	github.com/libdns/azure v0.5.0
+	github.com/libdns/bunny v1.6.1
 	github.com/libdns/cloudflare v0.2.2
+	github.com/libdns/desec v1.1.1
+	github.com/libdns/digitalocean v0.0.0-20250606071607-dfa7af5c2e31
+	github.com/libdns/duckdns v0.3.0
+	github.com/libdns/gandi v1.1.0
+	github.com/libdns/godaddy v1.1.0
+	github.com/libdns/googleclouddns v1.2.0
+	github.com/libdns/hetzner v1.0.0
+	github.com/libdns/huaweicloud v1.0.1
+	github.com/libdns/linode v0.5.0
+	github.com/libdns/namecheap v1.0.0
+	github.com/libdns/namesilo v1.0.0
+	github.com/libdns/netlify v1.2.0
+	github.com/libdns/ovh v1.1.0
+	github.com/libdns/porkbun v1.1.0
+	github.com/libdns/route53 v1.6.0
+	github.com/libdns/tencentcloud v1.4.3
+	github.com/libdns/vultr/v2 v2.0.4
 	github.com/sagernet/sing v0.8.2
 	github.com/sagernet/sing-box v1.13.2
 	github.com/shirou/gopsutil/v4 v4.26.2
-	github.com/xtls/xray-core v1.260206.0
+	github.com/xtls/xray-core v1.260327.0
+	golang.org/x/net v0.52.0
+	golang.org/x/term v0.41.0
 	golang.org/x/time v0.12.0
 	gopkg.in/yaml.v3 v3.0.1
 )
 
 require (
+	cloud.google.com/go/auth v0.16.5 // indirect
+	cloud.google.com/go/auth/oauth2adapt v0.2.8 // indirect
+	cloud.google.com/go/compute/metadata v0.9.0 // indirect
 	filippo.io/edwards25519 v1.1.0 // indirect
+	github.com/Azure/azure-sdk-for-go/sdk/azcore v1.18.0 // indirect
+	github.com/Azure/azure-sdk-for-go/sdk/azidentity v1.9.0 // indirect
+	github.com/Azure/azure-sdk-for-go/sdk/internal v1.11.1 // indirect
+	github.com/Azure/azure-sdk-for-go/sdk/resourcemanager/dns/armdns v1.2.0 // indirect
+	github.com/AzureAD/microsoft-authentication-library-for-go v1.4.2 // indirect
+	github.com/PuerkitoBio/purell v1.1.1 // indirect
+	github.com/PuerkitoBio/urlesc v0.0.0-20170810143723-de5bf2ad4578 // indirect
 	github.com/ajg/form v1.5.1 // indirect
 	github.com/akutz/memconn v0.1.0 // indirect
 	github.com/alexbrainman/sspi v0.0.0-20231016080023-1a75b4708caa // indirect
 	github.com/andybalholm/brotli v1.1.0 // indirect
 	github.com/anthropics/anthropic-sdk-go v1.26.0 // indirect
 	github.com/anytls/sing-anytls v0.0.11 // indirect
-	github.com/apernet/quic-go v0.57.2-0.20260111184307-eec823306178 // indirect
+	github.com/apernet/quic-go v0.59.1-0.20260217092621-db4786c77a22 // indirect
+	github.com/asaskevich/govalidator v0.0.0-20200907205600-7a23bdc65eef // indirect
+	github.com/aws/aws-sdk-go-v2 v1.39.1 // indirect
+	github.com/aws/aws-sdk-go-v2/config v1.31.10 // indirect
+	github.com/aws/aws-sdk-go-v2/credentials v1.18.14 // indirect
+	github.com/aws/aws-sdk-go-v2/feature/ec2/imds v1.18.8 // indirect
+	github.com/aws/aws-sdk-go-v2/internal/configsources v1.4.8 // indirect
+	github.com/aws/aws-sdk-go-v2/internal/endpoints/v2 v2.7.8 // indirect
+	github.com/aws/aws-sdk-go-v2/internal/ini v1.8.3 // indirect
+	github.com/aws/aws-sdk-go-v2/service/internal/accept-encoding v1.13.1 // indirect
+	github.com/aws/aws-sdk-go-v2/service/internal/presigned-url v1.13.8 // indirect
+	github.com/aws/aws-sdk-go-v2/service/route53 v1.58.3 // indirect
+	github.com/aws/aws-sdk-go-v2/service/sso v1.29.4 // indirect
+	github.com/aws/aws-sdk-go-v2/service/ssooidc v1.35.0 // indirect
+	github.com/aws/aws-sdk-go-v2/service/sts v1.38.5 // indirect
+	github.com/aws/smithy-go v1.23.0 // indirect
 	github.com/caddyserver/zerossl v0.1.5 // indirect
+	github.com/cespare/xxhash/v2 v2.3.0 // indirect
+	github.com/cilium/ebpf v0.21.0 // indirect
 	github.com/cloudflare/circl v1.6.3 // indirect
 	github.com/coder/websocket v1.8.14 // indirect
 	github.com/coreos/go-iptables v0.7.1-0.20240112124308-65c67c9f46e6 // indirect
@@ -33,52 +82,82 @@ require (
 	github.com/database64128/netx-go v0.1.1 // indirect
 	github.com/database64128/tfo-go/v2 v2.3.2 // indirect
 	github.com/dblohm7/wingoes v0.0.0-20240119213807-a09d6be7affa // indirect
+	github.com/digitalocean/godo v1.148.0 // indirect
 	github.com/ebitengine/purego v0.10.0 // indirect
 	github.com/enfein/mieru/v3 v3.29.0 // indirect
+	github.com/felixge/httpsnoop v1.0.4 // indirect
 	github.com/florianl/go-nfqueue/v2 v2.0.2 // indirect
-	github.com/fsnotify/fsnotify v1.7.0 // indirect
 	github.com/fxamacker/cbor/v2 v2.7.0 // indirect
 	github.com/gaissmai/bart v0.18.0 // indirect
 	github.com/ghodss/yaml v1.0.1-0.20220118164431-d8423dcdf344 // indirect
 	github.com/go-chi/chi/v5 v5.2.5 // indirect
 	github.com/go-chi/render v1.0.3 // indirect
 	github.com/go-json-experiment/json v0.0.0-20250813024750-ebf49471dced // indirect
+	github.com/go-logr/logr v1.4.3 // indirect
+	github.com/go-logr/stdr v1.2.2 // indirect
 	github.com/go-ole/go-ole v1.3.0 // indirect
+	github.com/go-openapi/analysis v0.19.16 // indirect
+	github.com/go-openapi/errors v0.19.9 // indirect
+	github.com/go-openapi/jsonpointer v0.19.5 // indirect
+	github.com/go-openapi/jsonreference v0.19.5 // indirect
+	github.com/go-openapi/loads v0.20.0 // indirect
+	github.com/go-openapi/runtime v0.19.24 // indirect
+	github.com/go-openapi/spec v0.20.0 // indirect
+	github.com/go-openapi/strfmt v0.19.11 // indirect
+	github.com/go-openapi/swag v0.19.12 // indirect
+	github.com/go-openapi/validate v0.20.0 // indirect
+	github.com/go-resty/resty/v2 v2.16.5 // indirect
+	github.com/go-stack/stack v1.8.0 // indirect
 	github.com/gobwas/httphead v0.1.0 // indirect
 	github.com/gobwas/pool v0.2.1 // indirect
 	github.com/godbus/dbus/v5 v5.2.2 // indirect
 	github.com/gofrs/uuid/v5 v5.4.0 // indirect
+	github.com/golang-jwt/jwt/v5 v5.2.2 // indirect
 	github.com/golang/groupcache v0.0.0-20210331224755-41bb18bfe9da // indirect
 	github.com/google/btree v1.1.3 // indirect
 	github.com/google/go-cmp v0.7.0 // indirect
+	github.com/google/go-querystring v1.1.0 // indirect
 	github.com/google/nftables v0.2.1-0.20240414091927-5e242ec57806 // indirect
+	github.com/google/s2a-go v0.1.9 // indirect
 	github.com/google/uuid v1.6.0 // indirect
+	github.com/googleapis/enterprise-certificate-proxy v0.3.6 // indirect
+	github.com/googleapis/gax-go/v2 v2.15.0 // indirect
+	github.com/hashicorp/go-cleanhttp v0.5.2 // indirect
+	github.com/hashicorp/go-retryablehttp v0.7.7 // indirect
 	github.com/hashicorp/yamux v0.1.2 // indirect
 	github.com/hdevalence/ed25519consensus v0.2.0 // indirect
 	github.com/insomniacslk/dhcp v0.0.0-20260220084031-5adc3eb26f91 // indirect
+	github.com/josharian/intern v1.0.0 // indirect
 	github.com/jsimonetti/rtnetlink v1.4.0 // indirect
 	github.com/juju/ratelimit v1.0.2 // indirect
 	github.com/keybase/go-keychain v0.0.1 // indirect
 	github.com/klauspost/compress v1.18.0 // indirect
 	github.com/klauspost/cpuid/v2 v2.3.0 // indirect
+	github.com/kylelemons/godebug v1.1.0 // indirect
 	github.com/libdns/acmedns v0.5.0 // indirect
 	github.com/libdns/libdns v1.1.1 // indirect
+	github.com/linode/linodego v1.56.0 // indirect
 	github.com/logrusorgru/aurora v2.0.3+incompatible // indirect
 	github.com/lufia/plan9stats v0.0.0-20211012122336-39d0f177ccd0 // indirect
+	github.com/mailru/easyjson v0.7.7 // indirect
 	github.com/mdlayher/netlink v1.9.0 // indirect
 	github.com/mdlayher/socket v0.5.1 // indirect
 	github.com/metacubex/utls v1.8.4 // indirect
 	github.com/mholt/acmez/v3 v3.1.6 // indirect
 	github.com/miekg/dns v1.1.72 // indirect
 	github.com/mitchellh/go-ps v1.0.0 // indirect
+	github.com/mitchellh/mapstructure v1.4.0 // indirect
+	github.com/netlify/open-api/v2 v2.40.0 // indirect
 	github.com/openai/openai-go/v3 v3.26.0 // indirect
+	github.com/ovh/go-ovh v1.7.0 // indirect
 	github.com/pelletier/go-toml v1.9.5 // indirect
 	github.com/pierrec/lz4/v4 v4.1.21 // indirect
-	github.com/pires/go-proxyproto v0.9.2 // indirect
+	github.com/pires/go-proxyproto v0.11.0 // indirect
+	github.com/pkg/browser v0.0.0-20240102092130-5ac0b6a4141c // indirect
 	github.com/power-devops/perfstat v0.0.0-20240221224432-82ca36839d55 // indirect
 	github.com/prometheus-community/pro-bing v0.4.0 // indirect
 	github.com/quic-go/qpack v0.6.0 // indirect
-	github.com/refraction-networking/utls v1.8.2 // indirect
+	github.com/refraction-networking/utls v1.8.3-0.20260301010127-aa6edf4b11af // indirect
 	github.com/safchain/ethtool v0.3.0 // indirect
 	github.com/sagernet/bbolt v0.0.0-20231014093535-ea5cb2fe9f0a // indirect
 	github.com/sagernet/cors v1.2.1 // indirect
@@ -145,31 +224,38 @@ require (
 	github.com/u-root/uio v0.0.0-20240224005618-d2acac8f3701 // indirect
 	github.com/vishvananda/netlink v1.3.1 // indirect
 	github.com/vishvananda/netns v0.0.5 // indirect
+	github.com/vultr/govultr/v3 v3.20.0 // indirect
 	github.com/x448/float16 v0.8.4 // indirect
-	github.com/xtls/reality v0.0.0-20251014195629-e4eec4520535 // indirect
+	github.com/xtls/reality v0.0.0-20260322125925-9234c772ba8f // indirect
 	github.com/yusufpapurcu/wmi v1.2.4 // indirect
 	github.com/zeebo/blake3 v0.2.4 // indirect
+	go.mongodb.org/mongo-driver v1.4.4 // indirect
+	go.opentelemetry.io/auto/sdk v1.2.1 // indirect
+	go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp v0.61.0 // indirect
+	go.opentelemetry.io/otel v1.39.0 // indirect
+	go.opentelemetry.io/otel/metric v1.39.0 // indirect
+	go.opentelemetry.io/otel/trace v1.39.0 // indirect
 	go.uber.org/multierr v1.11.0 // indirect
 	go.uber.org/zap v1.27.1 // indirect
 	go.uber.org/zap/exp v0.3.0 // indirect
 	go4.org/mem v0.0.0-20240501181205-ae6ca9944745 // indirect
 	go4.org/netipx v0.0.0-20231129151722-fdeea329fbba // indirect
-	golang.org/x/crypto v0.48.0 // indirect
+	golang.org/x/crypto v0.49.0 // indirect
 	golang.org/x/exp v0.0.0-20251219203646-944ab1f22d93 // indirect
 	golang.org/x/mod v0.33.0 // indirect
-	golang.org/x/net v0.50.0 // indirect
 	golang.org/x/oauth2 v0.34.0 // indirect
-	golang.org/x/sync v0.19.0 // indirect
-	golang.org/x/sys v0.41.0 // indirect
-	golang.org/x/term v0.40.0 // indirect
-	golang.org/x/text v0.34.0 // indirect
+	golang.org/x/sync v0.20.0 // indirect
+	golang.org/x/sys v0.42.0 // indirect
+	golang.org/x/text v0.35.0 // indirect
 	golang.org/x/tools v0.42.0 // indirect
 	golang.zx2c4.com/wintun v0.0.0-20230126152724-0fa3db229ce2 // indirect
-	golang.zx2c4.com/wireguard v0.0.0-20231211153847-12269c276173 // indirect
+	golang.zx2c4.com/wireguard v0.0.0-20250521234502-f333402bd9cb // indirect
 	golang.zx2c4.com/wireguard/windows v0.5.3 // indirect
+	google.golang.org/api v0.249.0 // indirect
 	google.golang.org/genproto/googleapis/rpc v0.0.0-20251202230838-ff82c1b0f217 // indirect
-	google.golang.org/grpc v1.79.1 // indirect
+	google.golang.org/grpc v1.79.3 // indirect
 	google.golang.org/protobuf v1.36.11 // indirect
+	gopkg.in/ini.v1 v1.67.0 // indirect
 	gopkg.in/yaml.v2 v2.4.0 // indirect
 	gvisor.dev/gvisor v0.0.0-20260122175437-89a5d21be8f0 // indirect
 	lukechampine.com/blake3 v1.4.1 // indirect
@@ -177,4 +263,4 @@ require (
 
 replace github.com/sagernet/sing-box => github.com/cedar2025/sing-box v1.14.0-alpha.2.0.20260316103356-2e665cb7e295
 
-// replace github.com/sagernet/sing-box => ./third_party/sing-box
+replace github.com/xtls/xray-core => github.com/cedar2025/Xray-core v0.0.0-20260409213332-f47935539965

internal/cert/cert.go

@@ -5,24 +5,28 @@ import (
 	"crypto/ecdsa"
 	"crypto/elliptic"
 	"crypto/rand"
+	"crypto/sha256"
+	"crypto/tls"
 	"crypto/x509"
 	"crypto/x509/pkix"
+	"encoding/hex"
 	"encoding/pem"
 	"fmt"
-	"log/slog"
 	"math/big"
 	"net"
 	"os"
 	"path/filepath"
+	"sort"
 	"strings"
 	"sync/atomic"
 	"time"
 
 	"github.com/caddyserver/certmagic"
-	"github.com/libdns/alidns"
-	"github.com/libdns/cloudflare"
 
+	"github.com/cedar2025/xboard-node/internal/cert/dnsproviders"
 	"github.com/cedar2025/xboard-node/internal/config"
+	"github.com/cedar2025/xboard-node/internal/kernel"
+	"github.com/cedar2025/xboard-node/internal/nlog"
 )
 
 // Manager handles TLS certificate lifecycle.
@@ -30,80 +34,191 @@ import (
 // Supported modes (CertConfig.CertMode):
 //   - "http"    — ACME HTTP-01 challenge via certmagic (needs port 80).
 //   - "dns"     — ACME DNS-01 challenge via certmagic + libdns provider.
-//   - "self"    — Self-signed certificate. Automatically generated and stored in CertDir.
-//     Recommended for nodes behind a CDN or reverse proxy.
+//   - "self"    — Self-signed certificate generated in memory.
 //   - "file"    — User-provided certificate and key file paths (CertFile/KeyFile).
 //   - "content" — Certificate and key PEM content pushed from the panel.
-//     Written to disk in CertDir for the kernel to consume.
 //   - "none"    — No TLS. The node will handle plain connections.
 //
-// Priority Logic:
+// All modes normalise to in-memory PEM. Kernels never see file paths —
+// they always receive PEM bytes via TLSCert().
 //
-//	If CertMode is empty, the manager infers the mode in this order:
-//	1. "http" (if auto_tls is true)
-//	2. "content" (if both CertContent and KeyContent are provided)
-//	3. "file" (if both CertFile and KeyFile paths are provided)
-//	4. "none" (default)
+// Priority Logic (when CertMode is empty):
+//
+//  1. "http" (if auto_tls is true)
+//  2. "content" (if both CertContent and KeyContent are provided)
+//  3. "file" (if both CertFile and KeyFile paths are provided)
+//  4. "none" (default)
 type Manager struct {
-	cfg      config.CertConfig
-	certFile string
-	keyFile  string
+	cfg config.CertConfig
 
-	magic       *certmagic.Config
-	renewed     atomic.Bool
-	acmeStarted bool // prevents double-init of certmagic background goroutines
+	// Atomic so the ACME renewal goroutine can swap without racing readers.
+	mat atomic.Pointer[certMaterial]
+
+	magic           *certmagic.Config
+	renewed         atomic.Bool
+	acmeStarted     bool
+	acmeFingerprint string
+	acmeCancel      context.CancelFunc
+}
+
+// certMaterial is an immutable snapshot of PEM-encoded cert + key.
+type certMaterial struct {
+	certPEM []byte
+	keyPEM  []byte
 }
 
 // NewManager creates a certificate manager.
 func NewManager(cfg config.CertConfig) *Manager {
-	m := &Manager{cfg: cfg}
-	m.setCertPaths()
-	return m
+	return &Manager{cfg: cfg}
 }
 
-// setCertPaths derives certFile / keyFile from the current m.cfg.
-func (m *Manager) setCertPaths() {
-	switch m.resolveMode() {
-	case "file":
-		m.certFile = m.cfg.CertFile
-		m.keyFile = m.cfg.KeyFile
-	case "http", "dns", "self", "content":
-		dir := m.cfg.CertDir
-		if dir == "" {
-			dir = "/etc/xboard-node/certs"
-		}
-		domain := m.cfg.Domain
-		if domain == "" {
-			domain = "localhost"
-		}
-		m.certFile = filepath.Join(dir, domain+".crt")
-		m.keyFile = filepath.Join(dir, domain+".key")
-	default:
-		m.certFile = ""
-		m.keyFile = ""
+// storePEM atomically swaps the in-memory cert material and persists to disk
+// so that restarts can reload without re-generating or re-requesting.
+func (m *Manager) storePEM(cert, key []byte) {
+	m.mat.Store(&certMaterial{certPEM: cert, keyPEM: key})
+	m.persistPEM(cert, key)
+}
+
+// persistPEM writes cert/key to cert_dir for restart recovery.
+// Errors are logged but not returned — persistence is best-effort.
+func (m *Manager) persistPEM(cert, key []byte) {
+	dir := m.cfg.CertDir
+	if dir == "" {
+		return
+	}
+	if err := os.MkdirAll(dir, 0o755); err != nil {
+		nlog.Core().Warn("cert: failed to create cert_dir", "dir", dir, "error", err)
+		return
+	}
+	certPath := filepath.Join(dir, "cert.pem")
+	keyPath := filepath.Join(dir, "key.pem")
+	if err := atomicWriteFile(certPath, cert, 0o644); err != nil {
+		nlog.Core().Warn("cert: failed to persist cert", "path", certPath, "error", err)
+	}
+	if err := atomicWriteFile(keyPath, key, 0o600); err != nil {
+		nlog.Core().Warn("cert: failed to persist key", "path", keyPath, "error", err)
 	}
 }
 
+// atomicWriteFile writes data to a temp file and renames, preventing partial reads.
+func atomicWriteFile(path string, data []byte, perm os.FileMode) error {
+	tmp := path + ".tmp"
+	if err := os.WriteFile(tmp, data, perm); err != nil {
+		return err
+	}
+	return os.Rename(tmp, path)
+}
+
+// loadPersistedPEM loads previously persisted cert material from cert_dir.
+// Returns true if valid material was loaded.
+func (m *Manager) loadPersistedPEM() bool {
+	dir := m.cfg.CertDir
+	if dir == "" {
+		return false
+	}
+	certPEM, err := os.ReadFile(filepath.Join(dir, "cert.pem"))
+	if err != nil {
+		return false
+	}
+	keyPEM, err := os.ReadFile(filepath.Join(dir, "key.pem"))
+	if err != nil {
+		return false
+	}
+	if err := validateKeyPair(certPEM, keyPEM); err != nil {
+		nlog.Core().Warn("cert: persisted cert invalid, will regenerate", "error", err)
+		return false
+	}
+	// Store in memory only (don't re-persist what we just read).
+	m.mat.Store(&certMaterial{certPEM: certPEM, keyPEM: keyPEM})
+	nlog.Core().Info("cert: loaded persisted certificate from disk", "dir", dir)
+	return true
+}
+
 // Reconfigure applies a new cert configuration at runtime (e.g. from panel push).
-// It is safe to call after Start(). For ACME modes that are already running this
-// is a no-op — a process restart is required to change ACME settings.
-// Returns true when the on-disk cert paths changed (caller should restart kernel).
+// Returns true when the PEM material changed (caller should restart kernel).
 func (m *Manager) Reconfigure(ctx context.Context, newCfg config.CertConfig) (bool, error) {
-	// Preserve the storage directory so panel-pushed configs don't change
-	// where certs live on disk (only the operator controls this path).
 	if newCfg.CertDir == "" {
 		newCfg.CertDir = m.cfg.CertDir
 	}
 
-	oldCertFile, oldKeyFile := m.certFile, m.keyFile
+	// If ACME is running and the new config materially differs (or switches
+	// away from ACME), tear down the old certmagic instance first.
+	if m.acmeStarted {
+		newFp := acmeFingerprint(newCfg)
+		newMode := resolveModeFor(newCfg)
+		if newMode != "http" && newMode != "dns" || newFp != m.acmeFingerprint {
+			m.tearDownACME()
+		}
+	}
+
+	oldTLS := m.TLSCert()
 	m.cfg = newCfg
-	m.setCertPaths()
 
 	if err := m.Start(ctx); err != nil {
 		return false, fmt.Errorf("cert reconfigure: %w", err)
 	}
 
-	return m.certFile != oldCertFile || m.keyFile != oldKeyFile, nil
+	newTLS := m.TLSCert()
+	return !pemEqual(oldTLS, newTLS), nil
+}
+
+// tearDownACME cancels the running certmagic background goroutines and clears
+// related state so a fresh ACME init can run.
+func (m *Manager) tearDownACME() {
+	if m.acmeCancel != nil {
+		m.acmeCancel()
+		m.acmeCancel = nil
+	}
+	m.magic = nil
+	m.acmeStarted = false
+	m.acmeFingerprint = ""
+	m.mat.Store(nil)
+}
+
+// acmeFingerprint produces a stable string capturing every config field that
+// requires re-running ACME when changed.
+func acmeFingerprint(cfg config.CertConfig) string {
+	keys := make([]string, 0, len(cfg.DNSEnv))
+	for k := range cfg.DNSEnv {
+		keys = append(keys, k)
+	}
+	sort.Strings(keys)
+	var b strings.Builder
+	b.WriteString(strings.ToLower(strings.TrimSpace(cfg.CertMode)))
+	b.WriteByte('|')
+	b.WriteString(strings.TrimSpace(cfg.Domain))
+	b.WriteByte('|')
+	b.WriteString(strings.TrimSpace(cfg.Email))
+	b.WriteByte('|')
+	b.WriteString(strings.TrimSpace(cfg.DNSProvider))
+	b.WriteByte('|')
+	for _, k := range keys {
+		b.WriteString(k)
+		b.WriteByte('=')
+		b.WriteString(cfg.DNSEnv[k])
+		b.WriteByte(';')
+	}
+	sum := sha256.Sum256([]byte(b.String()))
+	return hex.EncodeToString(sum[:])
+}
+
+// resolveModeFor mirrors (*Manager).resolveMode() for an arbitrary cfg, used
+// by Reconfigure to decide tear-down without mutating manager state.
+func resolveModeFor(cfg config.CertConfig) string {
+	mode := strings.ToLower(strings.TrimSpace(cfg.CertMode))
+	if mode != "" {
+		return mode
+	}
+	if cfg.AutoTLS {
+		return "http"
+	}
+	if cfg.CertContent != "" && cfg.KeyContent != "" {
+		return "content"
+	}
+	if cfg.CertFile != "" && cfg.KeyFile != "" {
+		return "file"
+	}
+	return "none"
 }
 
 // resolveMode returns the effective cert mode, handling backward compat for auto_tls.
@@ -127,15 +242,29 @@ func (m *Manager) resolveMode() string {
 	return "none"
 }
 
-func (m *Manager) CertFile() string  { return m.certFile }
-func (m *Manager) KeyFile() string   { return m.keyFile }
-func (m *Manager) HasCert() bool     { return m.certFile != "" && m.keyFile != "" }
+func (m *Manager) HasCert() bool {
+	mat := m.mat.Load()
+	return mat != nil && len(mat.certPEM) > 0 && len(mat.keyPEM) > 0
+}
 func (m *Manager) CertRenewed() bool { return m.renewed.Swap(false) }
 
+// TLSCert returns the current PEM material as a kernel.TLSCert.
+func (m *Manager) TLSCert() kernel.TLSCert {
+	mat := m.mat.Load()
+	if mat == nil {
+		return kernel.TLSCert{}
+	}
+	return kernel.TLSCert{CertPEM: mat.certPEM, KeyPEM: mat.keyPEM}
+}
+
+// pemEqual reports whether two TLSCert values carry identical PEM content.
+func pemEqual(a, b kernel.TLSCert) bool {
+	return string(a.CertPEM) == string(b.CertPEM) && string(a.KeyPEM) == string(b.KeyPEM)
+}
+
 // Start initializes the cert manager based on the resolved mode.
 func (m *Manager) Start(ctx context.Context) error {
 	mode := m.resolveMode()
-	slog.Info("cert manager starting", "mode", mode, "domain", m.cfg.Domain)
 
 	switch mode {
 	case "none", "":
@@ -166,29 +295,30 @@ func (m *Manager) Stop() {}
 // ─── Mode: file ────────────────────────────────────────────────────────────
 
 func (m *Manager) startFile() error {
-	if _, err := os.Stat(m.cfg.CertFile); err != nil {
+	certPEM, err := os.ReadFile(m.cfg.CertFile)
+	if err != nil {
 		return fmt.Errorf("cert file: %w", err)
 	}
-	if _, err := os.Stat(m.cfg.KeyFile); err != nil {
+	keyPEM, err := os.ReadFile(m.cfg.KeyFile)
+	if err != nil {
 		return fmt.Errorf("key file: %w", err)
 	}
-	slog.Info("using manual TLS certificates", "cert", m.certFile, "key", m.keyFile)
+	if err := validateKeyPair(certPEM, keyPEM); err != nil {
+		return fmt.Errorf("invalid certificate pair: %w", err)
+	}
+	m.storePEM(certPEM, keyPEM)
+	nlog.Core().Debug("TLS certificate loaded from files", "cert", m.cfg.CertFile, "key", m.cfg.KeyFile)
 	return nil
 }
 
 // ─── Mode: self ────────────────────────────────────────────────────────────
 
 func (m *Manager) startSelfSigned() error {
-	// If files already exist and are valid, skip regeneration.
-	if fileExists(m.certFile) && fileExists(m.keyFile) {
-		slog.Info("self-signed cert already exists, reusing", "cert", m.certFile)
+	// Reuse persisted self-signed cert if available and valid.
+	if m.loadPersistedPEM() {
 		return nil
 	}
 
-	if err := os.MkdirAll(filepath.Dir(m.certFile), 0o755); err != nil {
-		return fmt.Errorf("create cert dir: %w", err)
-	}
-
 	domain := m.cfg.Domain
 	if domain == "" {
 		domain = "localhost"
@@ -232,46 +362,37 @@ func (m *Manager) startSelfSigned() error {
 	}
 	keyPEM := pem.EncodeToMemory(&pem.Block{Type: "EC PRIVATE KEY", Bytes: keyDER})
 
-	if err := atomicWriteFile(m.certFile, certPEM, 0o644); err != nil {
-		return fmt.Errorf("write cert: %w", err)
-	}
-	if err := atomicWriteFile(m.keyFile, keyPEM, 0o600); err != nil {
-		return fmt.Errorf("write key: %w", err)
-	}
-
-	slog.Info("self-signed certificate generated",
-		"domain", domain, "cert", m.certFile, "valid_years", 10)
+	m.storePEM(certPEM, keyPEM)
+	nlog.Core().Info("self-signed certificate generated (in-memory)", "domain", domain, "valid_years", 10)
 	return nil
 }
 
 // ─── Mode: content ────────────────────────────────────────────────────────
 
-// startContent writes panel-supplied PEM strings to disk so the kernel can
-// reference them as ordinary cert/key files.
+// startContent stores panel-supplied PEM strings in memory for inline use.
 func (m *Manager) startContent() error {
 	if m.cfg.CertContent == "" || m.cfg.KeyContent == "" {
+		// No fresh content from panel — try previously persisted material.
+		if m.loadPersistedPEM() {
+			return nil
+		}
 		return fmt.Errorf("cert_mode 'content' requires both cert_content and key_content")
 	}
-	if err := os.MkdirAll(filepath.Dir(m.certFile), 0o755); err != nil {
-		return fmt.Errorf("create cert dir: %w", err)
+	certPEM := []byte(m.cfg.CertContent)
+	keyPEM := []byte(m.cfg.KeyContent)
+	if err := validateKeyPair(certPEM, keyPEM); err != nil {
+		return fmt.Errorf("invalid certificate content: %w", err)
 	}
-	if err := atomicWriteFile(m.certFile, []byte(m.cfg.CertContent), 0o644); err != nil {
-		return fmt.Errorf("write cert content: %w", err)
-	}
-	if err := atomicWriteFile(m.keyFile, []byte(m.cfg.KeyContent), 0o600); err != nil {
-		return fmt.Errorf("write key content: %w", err)
-	}
-	slog.Info("TLS certificate written from panel content", "cert", m.certFile)
+	m.storePEM(certPEM, keyPEM)
+	nlog.Core().Info("TLS certificate loaded from panel content (in-memory)")
 	return nil
 }
 
 // ─── Mode: http / dns (ACME) ──────────────────────────────────────────────
 
 func (m *Manager) startACME(ctx context.Context, dnsSolver *certmagic.DNS01Solver) error {
-	// Guard: certmagic's background goroutines must only be started once.
-	// A process restart is required to change ACME settings at runtime.
+	// Idempotent for unchanged config: Reconfigure already tore down stale ACME state.
 	if m.acmeStarted {
-		slog.Info("cert: ACME already active, skipping re-init (restart to change ACME settings)")
 		return nil
 	}
 
@@ -295,9 +416,23 @@ func (m *Manager) startACME(ctx context.Context, dnsSolver *certmagic.DNS01Solve
 	magic = certmagic.New(cache, certmagic.Config{
 		Storage: storage,
 		OnEvent: func(evtCtx context.Context, event string, data map[string]any) error {
-			if event == "cert_obtained" {
-				m.syncCertFiles(evtCtx, storage, data)
+			// Only react to renewals; initial load happens explicitly after ObtainCertSync.
+			if event != "cert_obtained" {
+				return nil
 			}
+			if renewed, _ := data["renewal"].(bool); !renewed {
+				return nil
+			}
+			issuerKey, _ := data["issuer"].(string)
+			if issuerKey == "" {
+				return nil
+			}
+			if err := m.loadPEMFromStorage(evtCtx, storage, issuerKey, m.cfg.Domain); err != nil {
+				nlog.Core().Error("failed to reload cert after renewal", "error", err)
+				return nil
+			}
+			m.renewed.Store(true)
+			nlog.Core().Info("TLS certificate reloaded after renewal", "domain", m.cfg.Domain)
 			return nil
 		},
 	})
@@ -327,18 +462,38 @@ func (m *Manager) startACME(ctx context.Context, dnsSolver *certmagic.DNS01Solve
 	}
 	m.magic = magic
 
-	if err := magic.ObtainCertSync(ctx, m.cfg.Domain); err != nil {
+	// Derived ctx so Reconfigure can cancel ACME background goroutines.
+	acmeCtx, acmeCancel := context.WithCancel(ctx)
+
+	if err := magic.ObtainCertSync(acmeCtx, m.cfg.Domain); err != nil {
+		acmeCancel()
 		return fmt.Errorf("obtain certificate: %w", err)
 	}
 
-	if err := magic.ManageAsync(ctx, []string{m.cfg.Domain}); err != nil {
+	issuerKey := magic.Issuers[0].IssuerKey()
+	if err := m.loadPEMFromStorage(acmeCtx, storage, issuerKey, m.cfg.Domain); err != nil {
+		acmeCancel()
+		return fmt.Errorf("load cert from storage: %w", err)
+	}
+
+	if err := magic.ManageAsync(acmeCtx, []string{m.cfg.Domain}); err != nil {
+		acmeCancel()
 		return fmt.Errorf("start cert manager: %w", err)
 	}
 
+	m.acmeCancel = acmeCancel
+	m.acmeFingerprint = acmeFingerprint(m.cfg)
 	m.acmeStarted = true
 	return nil
 }
 
+func validateKeyPair(certPEM, keyPEM []byte) error {
+	if _, err := tls.X509KeyPair(certPEM, keyPEM); err != nil {
+		return err
+	}
+	return nil
+}
+
 // ─── DNS Provider Factory ──────────────────────────────────────────────────
 
 // buildDNSSolver builds a certmagic DNS01Solver from the configured provider.
@@ -351,111 +506,44 @@ func (m *Manager) buildDNSSolver() (*certmagic.DNS01Solver, error) {
 }
 
 func (m *Manager) newDNSProvider() (certmagic.DNSProvider, error) {
-	name := strings.ToLower(strings.TrimSpace(m.cfg.DNSProvider))
+	name := strings.TrimSpace(m.cfg.DNSProvider)
+	if name == "" {
+		return nil, fmt.Errorf("dns_provider is required for cert_mode=dns")
+	}
 	env := m.cfg.DNSEnv
 	if env == nil {
 		env = map[string]string{}
 	}
-
-	switch name {
-	case "cloudflare", "cf":
-		token := firstOf(env, "CF_API_TOKEN", "CLOUDFLARE_API_TOKEN")
-		if token == "" {
-			return nil, fmt.Errorf("cloudflare requires CF_API_TOKEN in dns_env")
-		}
-		return &cloudflare.Provider{APIToken: token}, nil
-
-	case "alidns", "aliyun":
-		keyID := firstOf(env, "ALICLOUD_ACCESS_KEY_ID", "ALI_ACCESS_KEY_ID")
-		keySecret := firstOf(env, "ALICLOUD_ACCESS_KEY_SECRET", "ALI_ACCESS_KEY_SECRET")
-		if keyID == "" || keySecret == "" {
-			return nil, fmt.Errorf("alidns requires ALICLOUD_ACCESS_KEY_ID and ALICLOUD_ACCESS_KEY_SECRET in dns_env")
-		}
-		return &alidns.Provider{
-			CredentialInfo: alidns.CredentialInfo{
-				AccessKeyID:     keyID,
-				AccessKeySecret: keySecret,
-			},
-		}, nil
-
-	default:
-		return nil, fmt.Errorf("unsupported dns_provider: %q (supported: cloudflare, alidns)", name)
+	p, ok := dnsproviders.Get(name)
+	if !ok {
+		return nil, fmt.Errorf("unsupported dns_provider: %q (supported: %s)",
+			name, strings.Join(dnsproviders.CanonicalNames(), ", "))
 	}
+	return p.Build(env)
 }
 
 // ─── Helpers ───────────────────────────────────────────────────────────────
 
-// syncCertFiles copies the cert and key from certmagic's storage to stable paths.
-func (m *Manager) syncCertFiles(ctx context.Context, storage certmagic.Storage, data map[string]any) {
-	certKey, _ := data["certificate_path"].(string)
-	keyKey, _ := data["private_key_path"].(string)
-	if certKey == "" || keyKey == "" {
-		slog.Warn("cert_obtained event missing storage paths", "data", data)
-		return
+// loadPEMFromStorage reads cert + key for domain from certmagic storage into
+// the in-memory cache. Single entry point for both initial load and renewal refresh.
+func (m *Manager) loadPEMFromStorage(ctx context.Context, storage certmagic.Storage, issuerKey, domain string) error {
+	if issuerKey == "" || domain == "" {
+		return fmt.Errorf("loadPEMFromStorage: issuerKey and domain are required")
 	}
+	certKey := certmagic.StorageKeys.SiteCert(issuerKey, domain)
+	keyKey := certmagic.StorageKeys.SitePrivateKey(issuerKey, domain)
 
 	certPEM, err := storage.Load(ctx, certKey)
 	if err != nil {
-		slog.Error("failed to load cert from storage", "error", err)
-		return
+		return fmt.Errorf("load cert %q: %w", certKey, err)
 	}
 	keyPEM, err := storage.Load(ctx, keyKey)
 	if err != nil {
-		slog.Error("failed to load key from storage", "error", err)
-		return
+		return fmt.Errorf("load key %q: %w", keyKey, err)
 	}
-
-	if err := atomicWriteFile(m.certFile, certPEM, 0o644); err != nil {
-		slog.Error("failed to write cert file", "path", m.certFile, "error", err)
-		return
+	if err := validateKeyPair(certPEM, keyPEM); err != nil {
+		return fmt.Errorf("invalid cert/key pair on disk: %w", err)
 	}
-	if err := atomicWriteFile(m.keyFile, keyPEM, 0o600); err != nil {
-		slog.Error("failed to write key file", "path", m.keyFile, "error", err)
-		return
-	}
-
-	renewal, _ := data["renewal"].(bool)
-	slog.Info("TLS certificate synced", "domain", m.cfg.Domain, "renewal", renewal,
-		"cert", m.certFile)
-	m.renewed.Store(true)
-}
-
-func atomicWriteFile(path string, data []byte, perm os.FileMode) error {
-	dir := filepath.Dir(path)
-	tmp, err := os.CreateTemp(dir, ".tmp-*")
-	if err != nil {
-		return err
-	}
-	tmpName := tmp.Name()
-
-	if _, err := tmp.Write(data); err != nil {
-		tmp.Close()
-		os.Remove(tmpName)
-		return err
-	}
-	if err := tmp.Chmod(perm); err != nil {
-		tmp.Close()
-		os.Remove(tmpName)
-		return err
-	}
-	if err := tmp.Close(); err != nil {
-		os.Remove(tmpName)
-		return err
-	}
-	return os.Rename(tmpName, path)
-}
-
-func fileExists(path string) bool {
-	_, err := os.Stat(path)
-	return err == nil
-}
-
-// firstOf returns the first non-empty value for any of the given keys in the map.
-func firstOf(m map[string]string, keys ...string) string {
-	for _, k := range keys {
-		if v := m[k]; v != "" {
-			return v
-		}
-	}
-	return ""
+	m.storePEM(certPEM, keyPEM)
+	return nil
 }

internal/cert/cert_test.go

@@ -0,0 +1,236 @@
+package cert
+
+import (
+	"context"
+	"crypto/ecdsa"
+	"crypto/elliptic"
+	"crypto/rand"
+	"crypto/x509"
+	"crypto/x509/pkix"
+	"encoding/pem"
+	"math/big"
+	"path/filepath"
+	"testing"
+	"time"
+
+	"github.com/caddyserver/certmagic"
+
+	"github.com/cedar2025/xboard-node/internal/config"
+)
+
+// generateSelfSignedPair returns a fresh self-signed certificate and matching
+// private key in PEM form, suitable for storing as a libdns/certmagic site
+// resource in tests.
+func generateSelfSignedPair(t *testing.T, domain string) (certPEM, keyPEM []byte) {
+	t.Helper()
+	priv, err := ecdsa.GenerateKey(elliptic.P256(), rand.Reader)
+	if err != nil {
+		t.Fatalf("gen key: %v", err)
+	}
+	tmpl := &x509.Certificate{
+		SerialNumber: big.NewInt(1),
+		Subject:      pkix.Name{CommonName: domain},
+		DNSNames:     []string{domain},
+		NotBefore:    time.Now().Add(-time.Hour),
+		NotAfter:     time.Now().Add(24 * time.Hour),
+		KeyUsage:     x509.KeyUsageDigitalSignature,
+		ExtKeyUsage:  []x509.ExtKeyUsage{x509.ExtKeyUsageServerAuth},
+	}
+	der, err := x509.CreateCertificate(rand.Reader, tmpl, tmpl, &priv.PublicKey, priv)
+	if err != nil {
+		t.Fatalf("create cert: %v", err)
+	}
+	certPEM = pem.EncodeToMemory(&pem.Block{Type: "CERTIFICATE", Bytes: der})
+	keyDER, err := x509.MarshalECPrivateKey(priv)
+	if err != nil {
+		t.Fatalf("marshal key: %v", err)
+	}
+	keyPEM = pem.EncodeToMemory(&pem.Block{Type: "EC PRIVATE KEY", Bytes: keyDER})
+	return
+}
+
+// TestLoadPEMFromStorage_RestartScenario simulates the warm-restart bug fix:
+// certmagic.ObtainCertSync returns nil without firing cert_obtained when a
+// valid cert already exists in storage. The manager must still populate its
+// in-memory PEM cache by reading storage explicitly.
+func TestLoadPEMFromStorage_RestartScenario(t *testing.T) {
+	dir := t.TempDir()
+	storage := &certmagic.FileStorage{Path: dir}
+	const domain = "example.test"
+	const issuerKey = "test-issuer"
+
+	certPEM, keyPEM := generateSelfSignedPair(t, domain)
+
+	ctx := context.Background()
+	if err := storage.Store(ctx, certmagic.StorageKeys.SiteCert(issuerKey, domain), certPEM); err != nil {
+		t.Fatalf("seed cert: %v", err)
+	}
+	if err := storage.Store(ctx, certmagic.StorageKeys.SitePrivateKey(issuerKey, domain), keyPEM); err != nil {
+		t.Fatalf("seed key: %v", err)
+	}
+
+	m := NewManager(config.CertConfig{})
+	if m.HasCert() {
+		t.Fatal("manager should start with no cert in memory")
+	}
+
+	if err := m.loadPEMFromStorage(ctx, storage, issuerKey, domain); err != nil {
+		t.Fatalf("loadPEMFromStorage: %v", err)
+	}
+	if !m.HasCert() {
+		t.Fatal("expected HasCert() == true after explicit load")
+	}
+	got := m.TLSCert()
+	if string(got.CertPEM) != string(certPEM) {
+		t.Errorf("certPEM mismatch")
+	}
+	if string(got.KeyPEM) != string(keyPEM) {
+		t.Errorf("keyPEM mismatch")
+	}
+}
+
+// TestLoadPEMFromStorage_RenewalRefresh confirms that calling load a second
+// time with newer bytes on disk swaps the in-memory cache atomically — this
+// is the renewal hot-reload path triggered from the cert_obtained event.
+func TestLoadPEMFromStorage_RenewalRefresh(t *testing.T) {
+	dir := t.TempDir()
+	storage := &certmagic.FileStorage{Path: dir}
+	const domain = "example.test"
+	const issuerKey = "test-issuer"
+	ctx := context.Background()
+
+	certV1, keyV1 := generateSelfSignedPair(t, domain)
+	if err := storage.Store(ctx, certmagic.StorageKeys.SiteCert(issuerKey, domain), certV1); err != nil {
+		t.Fatalf("seed v1 cert: %v", err)
+	}
+	if err := storage.Store(ctx, certmagic.StorageKeys.SitePrivateKey(issuerKey, domain), keyV1); err != nil {
+		t.Fatalf("seed v1 key: %v", err)
+	}
+
+	m := NewManager(config.CertConfig{})
+	if err := m.loadPEMFromStorage(ctx, storage, issuerKey, domain); err != nil {
+		t.Fatalf("initial load: %v", err)
+	}
+
+	certV2, keyV2 := generateSelfSignedPair(t, domain)
+	if err := storage.Store(ctx, certmagic.StorageKeys.SiteCert(issuerKey, domain), certV2); err != nil {
+		t.Fatalf("seed v2 cert: %v", err)
+	}
+	if err := storage.Store(ctx, certmagic.StorageKeys.SitePrivateKey(issuerKey, domain), keyV2); err != nil {
+		t.Fatalf("seed v2 key: %v", err)
+	}
+
+	if err := m.loadPEMFromStorage(ctx, storage, issuerKey, domain); err != nil {
+		t.Fatalf("refresh load: %v", err)
+	}
+	got := m.TLSCert()
+	if string(got.CertPEM) != string(certV2) {
+		t.Error("expected v2 cert after refresh")
+	}
+	if string(got.KeyPEM) != string(keyV2) {
+		t.Error("expected v2 key after refresh")
+	}
+}
+
+// TestLoadPEMFromStorage_MissingFiles ensures we surface a clear error when
+// storage doesn't have the expected keys (defensive — should not happen in
+// practice because ObtainCertSync guarantees presence on success).
+func TestLoadPEMFromStorage_MissingFiles(t *testing.T) {
+	dir := t.TempDir()
+	storage := &certmagic.FileStorage{Path: dir}
+	m := NewManager(config.CertConfig{})
+	err := m.loadPEMFromStorage(context.Background(), storage, "test-issuer", "example.test")
+	if err == nil {
+		t.Fatal("expected error when storage is empty")
+	}
+}
+
+// TestLoadPEMFromStorage_RejectsMismatchedPair guards against silently loading
+// a cert and key that do not pair (e.g. partial write or external tampering).
+func TestLoadPEMFromStorage_RejectsMismatchedPair(t *testing.T) {
+	dir := t.TempDir()
+	storage := &certmagic.FileStorage{Path: dir}
+	const domain = "example.test"
+	const issuerKey = "test-issuer"
+	ctx := context.Background()
+
+	certA, _ := generateSelfSignedPair(t, domain)
+	_, keyB := generateSelfSignedPair(t, domain)
+	if err := storage.Store(ctx, certmagic.StorageKeys.SiteCert(issuerKey, domain), certA); err != nil {
+		t.Fatalf("seed cert: %v", err)
+	}
+	if err := storage.Store(ctx, certmagic.StorageKeys.SitePrivateKey(issuerKey, domain), keyB); err != nil {
+		t.Fatalf("seed key: %v", err)
+	}
+
+	m := NewManager(config.CertConfig{})
+	err := m.loadPEMFromStorage(ctx, storage, issuerKey, domain)
+	if err == nil {
+		t.Fatal("expected error for mismatched cert/key pair")
+	}
+	if m.HasCert() {
+		t.Error("manager should not store invalid pair")
+	}
+	_ = filepath.Join(dir, "stub")
+}
+
+// TestACMEFingerprint covers the config-equality logic that decides whether
+// Reconfigure must tear down a running certmagic instance.
+func TestACMEFingerprint(t *testing.T) {
+	base := config.CertConfig{
+		CertMode:    "dns",
+		Domain:      "a.example.com",
+		Email:       "ops@example.com",
+		DNSProvider: "cloudflare",
+		DNSEnv:      map[string]string{"CLOUDFLARE_API_TOKEN": "tok"},
+	}
+
+	cases := []struct {
+		name    string
+		mutate  func(*config.CertConfig)
+		differs bool
+	}{
+		{"same", func(*config.CertConfig) {}, false},
+		{"domain", func(c *config.CertConfig) { c.Domain = "b.example.com" }, true},
+		{"mode", func(c *config.CertConfig) { c.CertMode = "http" }, true},
+		{"email", func(c *config.CertConfig) { c.Email = "other@example.com" }, true},
+		{"provider", func(c *config.CertConfig) { c.DNSProvider = "alidns" }, true},
+		{"env_value", func(c *config.CertConfig) { c.DNSEnv = map[string]string{"CLOUDFLARE_API_TOKEN": "other"} }, true},
+		{"env_key_only", func(c *config.CertConfig) { c.DNSEnv = map[string]string{"OTHER": "tok"} }, true},
+	}
+
+	baseFp := acmeFingerprint(base)
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			cfg := base
+			if tc.mutate != nil {
+				// deep-copy DNSEnv before mutation to avoid sharing
+				cfg.DNSEnv = map[string]string{}
+				for k, v := range base.DNSEnv {
+					cfg.DNSEnv[k] = v
+				}
+				tc.mutate(&cfg)
+			}
+			got := acmeFingerprint(cfg)
+			if (got != baseFp) != tc.differs {
+				t.Errorf("differs=%v but got==base=%v", tc.differs, got == baseFp)
+			}
+		})
+	}
+}
+
+// TestACMEFingerprintEnvOrderStable ensures map iteration order does not
+// affect the fingerprint (must be sorted internally).
+func TestACMEFingerprintEnvOrderStable(t *testing.T) {
+	a := config.CertConfig{
+		CertMode: "dns", Domain: "x", DNSProvider: "p",
+		DNSEnv: map[string]string{"A": "1", "B": "2", "C": "3"},
+	}
+	b := config.CertConfig{
+		CertMode: "dns", Domain: "x", DNSProvider: "p",
+		DNSEnv: map[string]string{"C": "3", "A": "1", "B": "2"},
+	}
+	if acmeFingerprint(a) != acmeFingerprint(b) {
+		t.Fatal("fingerprint not order-independent")
+	}
+}

internal/cert/dnsproviders/alidns.go

@@ -0,0 +1,26 @@
+package dnsproviders
+
+import (
+	"github.com/caddyserver/certmagic"
+	"github.com/libdns/alidns"
+)
+
+func init() {
+	Register(&Provider{
+		Names:   []string{"alidns", "aliyun"},
+		EnvVars: []string{"ALICLOUD_ACCESS_KEY_ID", "ALICLOUD_ACCESS_KEY_SECRET"},
+		Build: func(env map[string]string) (certmagic.DNSProvider, error) {
+			keyID := firstOf(env, "ALICLOUD_ACCESS_KEY_ID", "ALI_ACCESS_KEY_ID", "ALIYUN_ACCESS_KEY_ID")
+			keySecret := firstOf(env, "ALICLOUD_ACCESS_KEY_SECRET", "ALI_ACCESS_KEY_SECRET", "ALIYUN_ACCESS_KEY_SECRET")
+			if keyID == "" || keySecret == "" {
+				return nil, missingEnvErr("alidns", "ALICLOUD_ACCESS_KEY_ID", "ALICLOUD_ACCESS_KEY_SECRET")
+			}
+			return &alidns.Provider{
+				CredentialInfo: alidns.CredentialInfo{
+					AccessKeyID:     keyID,
+					AccessKeySecret: keySecret,
+				},
+			}, nil
+		},
+	})
+}

internal/cert/dnsproviders/azure.go

@@ -0,0 +1,33 @@
+package dnsproviders
+
+import (
+	"github.com/caddyserver/certmagic"
+	"github.com/libdns/azure"
+)
+
+func init() {
+	Register(&Provider{
+		Names: []string{"azure"},
+		EnvVars: []string{
+			"AZURE_SUBSCRIPTION_ID", "AZURE_RESOURCE_GROUP_NAME",
+			"AZURE_TENANT_ID", "AZURE_CLIENT_ID", "AZURE_CLIENT_SECRET",
+		},
+		Build: func(env map[string]string) (certmagic.DNSProvider, error) {
+			sub := firstOf(env, "AZURE_SUBSCRIPTION_ID")
+			rg := firstOf(env, "AZURE_RESOURCE_GROUP_NAME", "AZURE_RESOURCE_GROUP")
+			if sub == "" || rg == "" {
+				return nil, missingEnvErr("azure", "AZURE_SUBSCRIPTION_ID", "AZURE_RESOURCE_GROUP_NAME")
+			}
+			// TenantId/ClientId/ClientSecret are optional — when omitted, libdns
+			// falls back to the default Azure credential chain (managed identity,
+			// CLI login, etc.).
+			return &azure.Provider{
+				SubscriptionId:    sub,
+				ResourceGroupName: rg,
+				TenantId:          firstOf(env, "AZURE_TENANT_ID"),
+				ClientId:          firstOf(env, "AZURE_CLIENT_ID"),
+				ClientSecret:      firstOf(env, "AZURE_CLIENT_SECRET"),
+			}, nil
+		},
+	})
+}

internal/cert/dnsproviders/bunny.go

@@ -0,0 +1,20 @@
+package dnsproviders
+
+import (
+	"github.com/caddyserver/certmagic"
+	"github.com/libdns/bunny"
+)
+
+func init() {
+	Register(&Provider{
+		Names:   []string{"bunny"},
+		EnvVars: []string{"BUNNY_ACCESS_KEY"},
+		Build: func(env map[string]string) (certmagic.DNSProvider, error) {
+			key := firstOf(env, "BUNNY_ACCESS_KEY", "BUNNY_API_KEY")
+			if key == "" {
+				return nil, missingEnvErr("bunny", "BUNNY_ACCESS_KEY")
+			}
+			return &bunny.Provider{AccessKey: key}, nil
+		},
+	})
+}

internal/cert/dnsproviders/cloudflare.go

@@ -0,0 +1,20 @@
+package dnsproviders
+
+import (
+	"github.com/caddyserver/certmagic"
+	"github.com/libdns/cloudflare"
+)
+
+func init() {
+	Register(&Provider{
+		Names:   []string{"cloudflare", "cf"},
+		EnvVars: []string{"CLOUDFLARE_DNS_API_TOKEN", "CF_API_TOKEN", "CLOUDFLARE_API_TOKEN"},
+		Build: func(env map[string]string) (certmagic.DNSProvider, error) {
+			token := firstOf(env, "CLOUDFLARE_DNS_API_TOKEN", "CF_API_TOKEN", "CLOUDFLARE_API_TOKEN")
+			if token == "" {
+				return nil, missingEnvErr("cloudflare", "CLOUDFLARE_DNS_API_TOKEN", "CF_API_TOKEN")
+			}
+			return &cloudflare.Provider{APIToken: token}, nil
+		},
+	})
+}

internal/cert/dnsproviders/desec.go

@@ -0,0 +1,20 @@
+package dnsproviders
+
+import (
+	"github.com/caddyserver/certmagic"
+	"github.com/libdns/desec"
+)
+
+func init() {
+	Register(&Provider{
+		Names:   []string{"desec"},
+		EnvVars: []string{"DESEC_TOKEN"},
+		Build: func(env map[string]string) (certmagic.DNSProvider, error) {
+			token := firstOf(env, "DESEC_TOKEN", "DESEC_API_TOKEN")
+			if token == "" {
+				return nil, missingEnvErr("desec", "DESEC_TOKEN")
+			}
+			return &desec.Provider{Token: token}, nil
+		},
+	})
+}

internal/cert/dnsproviders/digitalocean.go

@@ -0,0 +1,20 @@
+package dnsproviders
+
+import (
+	"github.com/caddyserver/certmagic"
+	"github.com/libdns/digitalocean"
+)
+
+func init() {
+	Register(&Provider{
+		Names:   []string{"digitalocean", "do"},
+		EnvVars: []string{"DO_AUTH_TOKEN", "DIGITALOCEAN_TOKEN"},
+		Build: func(env map[string]string) (certmagic.DNSProvider, error) {
+			token := firstOf(env, "DO_AUTH_TOKEN", "DIGITALOCEAN_TOKEN", "DIGITALOCEAN_API_TOKEN")
+			if token == "" {
+				return nil, missingEnvErr("digitalocean", "DO_AUTH_TOKEN")
+			}
+			return &digitalocean.Provider{APIToken: token}, nil
+		},
+	})
+}

internal/cert/dnsproviders/duckdns.go

@@ -0,0 +1,23 @@
+package dnsproviders
+
+import (
+	"github.com/caddyserver/certmagic"
+	"github.com/libdns/duckdns"
+)
+
+func init() {
+	Register(&Provider{
+		Names:   []string{"duckdns"},
+		EnvVars: []string{"DUCKDNS_TOKEN"},
+		Build: func(env map[string]string) (certmagic.DNSProvider, error) {
+			token := firstOf(env, "DUCKDNS_TOKEN", "DUCKDNS_API_TOKEN")
+			if token == "" {
+				return nil, missingEnvErr("duckdns", "DUCKDNS_TOKEN")
+			}
+			return &duckdns.Provider{
+				APIToken:       token,
+				OverrideDomain: firstOf(env, "DUCKDNS_OVERRIDE_DOMAIN"),
+			}, nil
+		},
+	})
+}

internal/cert/dnsproviders/gandi.go

@@ -0,0 +1,20 @@
+package dnsproviders
+
+import (
+	"github.com/caddyserver/certmagic"
+	"github.com/libdns/gandi"
+)
+
+func init() {
+	Register(&Provider{
+		Names:   []string{"gandi"},
+		EnvVars: []string{"GANDI_BEARER_TOKEN", "GANDI_API_TOKEN"},
+		Build: func(env map[string]string) (certmagic.DNSProvider, error) {
+			token := firstOf(env, "GANDI_BEARER_TOKEN", "GANDI_API_TOKEN", "GANDI_PERSONAL_ACCESS_TOKEN")
+			if token == "" {
+				return nil, missingEnvErr("gandi", "GANDI_BEARER_TOKEN")
+			}
+			return &gandi.Provider{BearerToken: token}, nil
+		},
+	})
+}

internal/cert/dnsproviders/godaddy.go

@@ -0,0 +1,30 @@
+package dnsproviders
+
+import (
+	"github.com/caddyserver/certmagic"
+	"github.com/libdns/godaddy"
+)
+
+func init() {
+	Register(&Provider{
+		Names:   []string{"godaddy"},
+		EnvVars: []string{"GODADDY_API_TOKEN"},
+		Build: func(env map[string]string) (certmagic.DNSProvider, error) {
+			// GoDaddy auth header format: "sso-key KEY:SECRET"
+			// Pass the full "KEY:SECRET" pair via GODADDY_API_TOKEN, OR set
+			// GODADDY_API_KEY + GODADDY_API_SECRET and we'll join them.
+			token := firstOf(env, "GODADDY_API_TOKEN")
+			if token == "" {
+				key := firstOf(env, "GODADDY_API_KEY")
+				secret := firstOf(env, "GODADDY_API_SECRET")
+				if key != "" && secret != "" {
+					token = key + ":" + secret
+				}
+			}
+			if token == "" {
+				return nil, missingEnvErr("godaddy", "GODADDY_API_TOKEN", "GODADDY_API_KEY+GODADDY_API_SECRET")
+			}
+			return &godaddy.Provider{APIToken: token}, nil
+		},
+	})
+}

internal/cert/dnsproviders/googleclouddns.go

@@ -0,0 +1,25 @@
+package dnsproviders
+
+import (
+	"github.com/caddyserver/certmagic"
+	"github.com/libdns/googleclouddns"
+)
+
+func init() {
+	Register(&Provider{
+		Names:   []string{"googleclouddns", "gcp", "gcloud"},
+		EnvVars: []string{"GCE_PROJECT", "GCE_SERVICE_ACCOUNT_FILE"},
+		Build: func(env map[string]string) (certmagic.DNSProvider, error) {
+			project := firstOf(env, "GCE_PROJECT", "GOOGLE_PROJECT_ID", "GCP_PROJECT")
+			if project == "" {
+				return nil, missingEnvErr("googleclouddns", "GCE_PROJECT")
+			}
+			// ServiceAccountJSON may be a file path OR a raw JSON blob; when
+			// empty, the default Google credential chain is used.
+			return &googleclouddns.Provider{
+				Project:            project,
+				ServiceAccountJSON: firstOf(env, "GCE_SERVICE_ACCOUNT_FILE", "GOOGLE_APPLICATION_CREDENTIALS"),
+			}, nil
+		},
+	})
+}

internal/cert/dnsproviders/helpers.go

@@ -0,0 +1,22 @@
+package dnsproviders
+
+import (
+	"fmt"
+	"strings"
+)
+
+// firstOf returns the value of the first non-empty env key from candidates.
+// Lookup is case-sensitive (env keys are conventionally upper-case).
+func firstOf(env map[string]string, keys ...string) string {
+	for _, k := range keys {
+		if v := strings.TrimSpace(env[k]); v != "" {
+			return v
+		}
+	}
+	return ""
+}
+
+// missingEnvErr formats a "missing required env" error consistently.
+func missingEnvErr(provider string, keys ...string) error {
+	return fmt.Errorf("dns provider %q requires one of %s in dns_env", provider, strings.Join(keys, ", "))
+}

internal/cert/dnsproviders/hetzner.go

@@ -0,0 +1,20 @@
+package dnsproviders
+
+import (
+	"github.com/caddyserver/certmagic"
+	"github.com/libdns/hetzner"
+)
+
+func init() {
+	Register(&Provider{
+		Names:   []string{"hetzner"},
+		EnvVars: []string{"HETZNER_API_TOKEN"},
+		Build: func(env map[string]string) (certmagic.DNSProvider, error) {
+			token := firstOf(env, "HETZNER_API_TOKEN", "HETZNER_DNS_API_TOKEN")
+			if token == "" {
+				return nil, missingEnvErr("hetzner", "HETZNER_API_TOKEN")
+			}
+			return &hetzner.Provider{AuthAPIToken: token}, nil
+		},
+	})
+}

internal/cert/dnsproviders/huaweicloud.go

@@ -0,0 +1,25 @@
+package dnsproviders
+
+import (
+	"github.com/caddyserver/certmagic"
+	"github.com/libdns/huaweicloud"
+)
+
+func init() {
+	Register(&Provider{
+		Names:   []string{"huaweicloud", "huawei"},
+		EnvVars: []string{"HUAWEICLOUD_ACCESS_KEY_ID", "HUAWEICLOUD_SECRET_ACCESS_KEY", "HUAWEICLOUD_REGION_ID"},
+		Build: func(env map[string]string) (certmagic.DNSProvider, error) {
+			id := firstOf(env, "HUAWEICLOUD_ACCESS_KEY_ID", "HUAWEI_ACCESS_KEY_ID")
+			secret := firstOf(env, "HUAWEICLOUD_SECRET_ACCESS_KEY", "HUAWEI_SECRET_ACCESS_KEY")
+			if id == "" || secret == "" {
+				return nil, missingEnvErr("huaweicloud", "HUAWEICLOUD_ACCESS_KEY_ID", "HUAWEICLOUD_SECRET_ACCESS_KEY")
+			}
+			return &huaweicloud.Provider{
+				AccessKeyId:     id,
+				SecretAccessKey: secret,
+				RegionId:        firstOf(env, "HUAWEICLOUD_REGION_ID", "HUAWEI_REGION"),
+			}, nil
+		},
+	})
+}

internal/cert/dnsproviders/linode.go

@@ -0,0 +1,20 @@
+package dnsproviders
+
+import (
+	"github.com/caddyserver/certmagic"
+	"github.com/libdns/linode"
+)
+
+func init() {
+	Register(&Provider{
+		Names:   []string{"linode"},
+		EnvVars: []string{"LINODE_TOKEN"},
+		Build: func(env map[string]string) (certmagic.DNSProvider, error) {
+			token := firstOf(env, "LINODE_TOKEN", "LINODE_API_TOKEN")
+			if token == "" {
+				return nil, missingEnvErr("linode", "LINODE_TOKEN")
+			}
+			return &linode.Provider{APIToken: token}, nil
+		},
+	})
+}

internal/cert/dnsproviders/namecheap.go

@@ -0,0 +1,26 @@
+package dnsproviders
+
+import (
+	"github.com/caddyserver/certmagic"
+	"github.com/libdns/namecheap"
+)
+
+func init() {
+	Register(&Provider{
+		Names:   []string{"namecheap"},
+		EnvVars: []string{"NAMECHEAP_API_KEY", "NAMECHEAP_API_USER", "NAMECHEAP_CLIENT_IP"},
+		Build: func(env map[string]string) (certmagic.DNSProvider, error) {
+			apiKey := firstOf(env, "NAMECHEAP_API_KEY")
+			user := firstOf(env, "NAMECHEAP_API_USER", "NAMECHEAP_USER")
+			if apiKey == "" || user == "" {
+				return nil, missingEnvErr("namecheap", "NAMECHEAP_API_KEY", "NAMECHEAP_API_USER")
+			}
+			return &namecheap.Provider{
+				APIKey:      apiKey,
+				User:        user,
+				APIEndpoint: firstOf(env, "NAMECHEAP_API_ENDPOINT"),
+				ClientIP:    firstOf(env, "NAMECHEAP_CLIENT_IP"),
+			}, nil
+		},
+	})
+}

internal/cert/dnsproviders/namesilo.go

@@ -0,0 +1,20 @@
+package dnsproviders
+
+import (
+	"github.com/caddyserver/certmagic"
+	"github.com/libdns/namesilo"
+)
+
+func init() {
+	Register(&Provider{
+		Names:   []string{"namesilo"},
+		EnvVars: []string{"NAMESILO_API_TOKEN"},
+		Build: func(env map[string]string) (certmagic.DNSProvider, error) {
+			token := firstOf(env, "NAMESILO_API_TOKEN")
+			if token == "" {
+				return nil, missingEnvErr("namesilo", "NAMESILO_API_TOKEN")
+			}
+			return &namesilo.Provider{APIToken: token}, nil
+		},
+	})
+}

internal/cert/dnsproviders/netlify.go

@@ -0,0 +1,20 @@
+package dnsproviders
+
+import (
+	"github.com/caddyserver/certmagic"
+	"github.com/libdns/netlify"
+)
+
+func init() {
+	Register(&Provider{
+		Names:   []string{"netlify"},
+		EnvVars: []string{"NETLIFY_PERSONAL_ACCESS_TOKEN"},
+		Build: func(env map[string]string) (certmagic.DNSProvider, error) {
+			token := firstOf(env, "NETLIFY_PERSONAL_ACCESS_TOKEN", "NETLIFY_AUTH_TOKEN", "NETLIFY_API_TOKEN")
+			if token == "" {
+				return nil, missingEnvErr("netlify", "NETLIFY_PERSONAL_ACCESS_TOKEN")
+			}
+			return &netlify.Provider{PersonalAccessToken: token}, nil
+		},
+	})
+}

internal/cert/dnsproviders/ovh.go

@@ -0,0 +1,29 @@
+package dnsproviders
+
+import (
+	"github.com/caddyserver/certmagic"
+	"github.com/libdns/ovh"
+)
+
+func init() {
+	Register(&Provider{
+		Names: []string{"ovh"},
+		EnvVars: []string{
+			"OVH_ENDPOINT", "OVH_APPLICATION_KEY", "OVH_APPLICATION_SECRET", "OVH_CONSUMER_KEY",
+		},
+		Build: func(env map[string]string) (certmagic.DNSProvider, error) {
+			appKey := firstOf(env, "OVH_APPLICATION_KEY")
+			appSecret := firstOf(env, "OVH_APPLICATION_SECRET")
+			consumerKey := firstOf(env, "OVH_CONSUMER_KEY")
+			if appKey == "" || appSecret == "" || consumerKey == "" {
+				return nil, missingEnvErr("ovh", "OVH_APPLICATION_KEY", "OVH_APPLICATION_SECRET", "OVH_CONSUMER_KEY")
+			}
+			return &ovh.Provider{
+				Endpoint:          firstOf(env, "OVH_ENDPOINT"),
+				ApplicationKey:    appKey,
+				ApplicationSecret: appSecret,
+				ConsumerKey:       consumerKey,
+			}, nil
+		},
+	})
+}

internal/cert/dnsproviders/porkbun.go

@@ -0,0 +1,21 @@
+package dnsproviders
+
+import (
+	"github.com/caddyserver/certmagic"
+	"github.com/libdns/porkbun"
+)
+
+func init() {
+	Register(&Provider{
+		Names:   []string{"porkbun"},
+		EnvVars: []string{"PORKBUN_API_KEY", "PORKBUN_API_SECRET_KEY"},
+		Build: func(env map[string]string) (certmagic.DNSProvider, error) {
+			key := firstOf(env, "PORKBUN_API_KEY")
+			secret := firstOf(env, "PORKBUN_API_SECRET_KEY", "PORKBUN_SECRET_API_KEY")
+			if key == "" || secret == "" {
+				return nil, missingEnvErr("porkbun", "PORKBUN_API_KEY", "PORKBUN_API_SECRET_KEY")
+			}
+			return &porkbun.Provider{APIKey: key, APISecretKey: secret}, nil
+		},
+	})
+}

internal/cert/dnsproviders/registry.go

@@ -0,0 +1,93 @@
+// Package dnsproviders is a registry of DNS providers usable for ACME DNS-01
+// challenges. Each provider is implemented as a separate file that registers
+// itself in init(); to add a new provider, drop a file in this package — no
+// changes are needed elsewhere.
+//
+// All providers wrap the libdns ecosystem (https://github.com/libdns) so they
+// plug directly into certmagic's DNSManager.
+package dnsproviders
+
+import (
+	"fmt"
+	"sort"
+	"strings"
+	"sync"
+
+	"github.com/caddyserver/certmagic"
+)
+
+// Builder constructs a certmagic.DNSProvider from a flat env-style map.
+// The map mirrors the cert_config.dns_env block from config.yml.
+type Builder func(env map[string]string) (certmagic.DNSProvider, error)
+
+// Provider describes a single DNS provider entry in the registry.
+type Provider struct {
+	// Names is the list of accepted dns_provider values; the first is canonical,
+	// the rest are aliases (case-insensitive match).
+	Names []string
+	// EnvVars lists the environment variable names this provider may consume.
+	// Used for documentation and error messages only.
+	EnvVars []string
+	// Build constructs the underlying libdns provider.
+	Build Builder
+}
+
+var (
+	mu       sync.RWMutex
+	registry = map[string]*Provider{}
+	ordered  []*Provider
+)
+
+// Register adds a provider to the global registry. It panics on duplicate
+// names so that registration errors surface at process start.
+func Register(p *Provider) {
+	if p == nil || len(p.Names) == 0 || p.Build == nil {
+		panic("dnsproviders: invalid provider registration")
+	}
+	mu.Lock()
+	defer mu.Unlock()
+	for _, n := range p.Names {
+		key := strings.ToLower(strings.TrimSpace(n))
+		if key == "" {
+			panic("dnsproviders: empty provider name")
+		}
+		if _, exists := registry[key]; exists {
+			panic(fmt.Sprintf("dnsproviders: duplicate name %q", key))
+		}
+		registry[key] = p
+	}
+	ordered = append(ordered, p)
+}
+
+// Get returns the provider registered under the given (case-insensitive) name
+// or alias.
+func Get(name string) (*Provider, bool) {
+	mu.RLock()
+	defer mu.RUnlock()
+	p, ok := registry[strings.ToLower(strings.TrimSpace(name))]
+	return p, ok
+}
+
+// CanonicalNames returns the canonical (first-listed) name of every registered
+// provider, sorted alphabetically. Useful for error messages.
+func CanonicalNames() []string {
+	mu.RLock()
+	defer mu.RUnlock()
+	names := make([]string, 0, len(ordered))
+	for _, p := range ordered {
+		names = append(names, p.Names[0])
+	}
+	sort.Strings(names)
+	return names
+}
+
+// List returns all registered providers (canonical entries only), sorted by
+// canonical name.
+func List() []*Provider {
+	mu.RLock()
+	defer mu.RUnlock()
+	out := make([]*Provider, len(ordered))
+	copy(out, ordered)
+	sort.Slice(out, func(i, j int) bool { return out[i].Names[0] < out[j].Names[0] })
+	return out
+}

internal/cert/dnsproviders/registry_test.go

@@ -0,0 +1,90 @@
+package dnsproviders
+
+import (
+	"strings"
+	"testing"
+)
+
+// TestRegistryHasMainstreamProviders verifies that all expected mainstream
+// providers are registered with their canonical names (and key aliases).
+func TestRegistryHasMainstreamProviders(t *testing.T) {
+	cases := map[string][]string{
+		"cloudflare":     {"cf"},
+		"alidns":         {"aliyun"},
+		"tencentcloud":   {"tencent"},
+		"route53":        {"aws"},
+		"godaddy":        nil,
+		"namecheap":      nil,
+		"namesilo":       nil,
+		"digitalocean":   {"do"},
+		"linode":         nil,
+		"hetzner":        nil,
+		"gandi":          nil,
+		"porkbun":        nil,
+		"netlify":        nil,
+		"azure":          nil,
+		"googleclouddns": {"gcp", "gcloud"},
+		"huaweicloud":    {"huawei"},
+		"bunny":          nil,
+		"duckdns":        nil,
+		"ovh":            nil,
+		"vultr":          nil,
+		"desec":          nil,
+	}
+	for canonical, aliases := range cases {
+		if _, ok := Get(canonical); !ok {
+			t.Errorf("provider %q not registered", canonical)
+		}
+		// case-insensitive lookup must work
+		if _, ok := Get(strings.ToUpper(canonical)); !ok {
+			t.Errorf("provider %q not found case-insensitively", canonical)
+		}
+		for _, a := range aliases {
+			if _, ok := Get(a); !ok {
+				t.Errorf("alias %q for %q not registered", a, canonical)
+			}
+		}
+	}
+}
+
+func TestRegistryRejectsUnknown(t *testing.T) {
+	if _, ok := Get("definitely-not-a-real-provider"); ok {
+		t.Fatal("Get returned ok for unknown provider")
+	}
+}
+
+// TestProvidersValidateMissingEnv ensures every provider's Build returns a
+// helpful error when no credentials are supplied (rather than panicking or
+// silently building an unusable client).
+//
+// route53/azure/googleclouddns are excluded because they intentionally fall
+// back to an ambient credential chain (IAM role, managed identity, ADC) and
+// produce a usable provider even with empty config.
+func TestProvidersValidateMissingEnv(t *testing.T) {
+	excluded := map[string]bool{
+		"route53":        true,
+		"azure":          true, // partially excluded — actually still requires sub+rg
+		"googleclouddns": true, // requires GCE_PROJECT
+	}
+	for _, p := range List() {
+		canonical := p.Names[0]
+		if excluded[canonical] {
+			continue
+		}
+		t.Run(canonical, func(t *testing.T) {
+			_, err := p.Build(map[string]string{})
+			if err == nil {
+				t.Errorf("provider %q built successfully with empty env; expected missing-credential error", canonical)
+			}
+		})
+	}
+}
+
+func TestCanonicalNamesSorted(t *testing.T) {
+	names := CanonicalNames()
+	for i := 1; i < len(names); i++ {
+		if names[i-1] > names[i] {
+			t.Fatalf("CanonicalNames not sorted: %v", names)
+		}
+	}
+}

internal/cert/dnsproviders/route53.go

@@ -0,0 +1,26 @@
+package dnsproviders
+
+import (
+	"github.com/caddyserver/certmagic"
+	"github.com/libdns/route53"
+)
+
+func init() {
+	Register(&Provider{
+		Names:   []string{"route53", "aws"},
+		EnvVars: []string{"AWS_ACCESS_KEY_ID", "AWS_SECRET_ACCESS_KEY", "AWS_REGION", "AWS_SESSION_TOKEN", "AWS_HOSTED_ZONE_ID", "AWS_PROFILE"},
+		Build: func(env map[string]string) (certmagic.DNSProvider, error) {
+			// route53 supports the full AWS credential chain (env, IAM role, profile,
+			// shared credentials file). We deliberately do not require any field —
+			// users running on EC2/EKS may rely entirely on the instance role.
+			return &route53.Provider{
+				AccessKeyId:     firstOf(env, "AWS_ACCESS_KEY_ID"),
+				SecretAccessKey: firstOf(env, "AWS_SECRET_ACCESS_KEY"),
+				SessionToken:    firstOf(env, "AWS_SESSION_TOKEN"),
+				Region:          firstOf(env, "AWS_REGION", "AWS_DEFAULT_REGION"),
+				Profile:         firstOf(env, "AWS_PROFILE"),
+				HostedZoneID:    firstOf(env, "AWS_HOSTED_ZONE_ID"),
+			}, nil
+		},
+	})
+}

internal/cert/dnsproviders/tencentcloud.go

@@ -0,0 +1,25 @@
+package dnsproviders
+
+import (
+	"github.com/caddyserver/certmagic"
+	"github.com/libdns/tencentcloud"
+)
+
+func init() {
+	Register(&Provider{
+		Names:   []string{"tencentcloud", "tencent"},
+		EnvVars: []string{"TENCENTCLOUD_SECRET_ID", "TENCENTCLOUD_SECRET_KEY", "TENCENTCLOUD_REGION"},
+		Build: func(env map[string]string) (certmagic.DNSProvider, error) {
+			id := firstOf(env, "TENCENTCLOUD_SECRET_ID", "TENCENT_SECRET_ID")
+			key := firstOf(env, "TENCENTCLOUD_SECRET_KEY", "TENCENT_SECRET_KEY")
+			if id == "" || key == "" {
+				return nil, missingEnvErr("tencentcloud", "TENCENTCLOUD_SECRET_ID", "TENCENTCLOUD_SECRET_KEY")
+			}
+			return &tencentcloud.Provider{
+				SecretId:  id,
+				SecretKey: key,
+				Region:    firstOf(env, "TENCENTCLOUD_REGION"),
+			}, nil
+		},
+	})
+}

internal/cert/dnsproviders/vultr.go

@@ -0,0 +1,20 @@
+package dnsproviders
+
+import (
+	"github.com/caddyserver/certmagic"
+	"github.com/libdns/vultr/v2"
+)
+
+func init() {
+	Register(&Provider{
+		Names:   []string{"vultr"},
+		EnvVars: []string{"VULTR_API_TOKEN"},
+		Build: func(env map[string]string) (certmagic.DNSProvider, error) {
+			token := firstOf(env, "VULTR_API_TOKEN", "VULTR_API_KEY")
+			if token == "" {
+				return nil, missingEnvErr("vultr", "VULTR_API_TOKEN")
+			}
+			return &vultr.Provider{APIToken: token}, nil
+		},
+	})
+}

internal/config/config.go

@@ -1,23 +1,34 @@
 package config
 
 import (
+	"crypto/sha1"
+	"encoding/hex"
 	"fmt"
 	"io"
 	"log/slog"
+	"net/url"
 	"os"
 	"path/filepath"
+	"sort"
 	"strconv"
+	"strings"
 
+	"github.com/cedar2025/xboard-node/internal/nlog"
+	"golang.org/x/term"
 	"gopkg.in/yaml.v3"
 )
 
 type Config struct {
+	InstanceID string `yaml:"-"`
 	Panel   PanelConfig   `yaml:"panel"`
 	Node    NodeConfig    `yaml:"node"`
 	Kernel  KernelConfig  `yaml:"kernel"`
 	Cert    CertConfig    `yaml:"cert"`
 	Log     LogConfig     `yaml:"log"`
 	Runtime RuntimeConfig `yaml:"runtime"`
+	WS      WSConfig      `yaml:"ws"`
+	// Standalone enables a local-only node that never contacts the panel.
+	Standalone *StandaloneConfig `yaml:"standalone,omitempty"`
 	// HealthPort enables a lightweight HTTP health-check endpoint on the
 	// given port (e.g. 65530). 0 = disabled (default).
 	HealthPort int `yaml:"health_port"`
@@ -25,6 +36,20 @@ type Config struct {
 	// one service instance is started per entry. All entries share the same
 	// panel URL/token, kernel type, log settings and runtime tuning.
 	Nodes []NodeEntry `yaml:"nodes,omitempty"`
+
+	// Machine enables machine mode: a single process manages all nodes
+	// bound to this machine on the panel. Nodes are discovered dynamically
+	// via GET /api/v2/server/machine/nodes. When set, Panel.NodeID, Nodes
+	// and Panel.Token are ignored; the machine token is used instead.
+	Machine *MachineConfig `yaml:"machine,omitempty"`
+}
+
+// MachineConfig identifies this process as a panel-managed machine that
+// dynamically discovers and runs all nodes bound to it.
+type MachineConfig struct {
+	MachineID int    `yaml:"machine_id"`
+	Token     string `yaml:"token"`
+	TokenEnv  string `yaml:"token_env,omitempty"`
 }
 
 // NodeEntry describes a single node in multi-node mode.
@@ -68,15 +93,28 @@ type RuntimeConfig struct {
 }
 
 type PanelConfig struct {
-	URL      string `yaml:"url"`
-	Token    string `yaml:"token"`
-	NodeID   int    `yaml:"node_id"`
-	NodeType string `yaml:"node_type"`
+	URL       string `yaml:"url"`
+	Token     string `yaml:"token"`
+	TokenEnv  string `yaml:"token_env,omitempty"`
+	NodeID    int    `yaml:"node_id"`
+	NodeType  string `yaml:"node_type"`
+	MachineID int    `yaml:"-"`
 }
 
 type NodeConfig struct {
-	PushInterval int `yaml:"push_interval"`
-	PullInterval int `yaml:"pull_interval"`
+	PushInterval         int `yaml:"push_interval"`
+	PullInterval         int `yaml:"pull_interval"`
+	TrackInterval        int `yaml:"track_interval"`         // sec, default 10
+	DeviceReportInterval int `yaml:"device_report_interval"` // sec, default 30
+}
+
+// WSConfig holds WebSocket client tuning options.
+type WSConfig struct {
+	StatusInterval    int `yaml:"status_interval"`    // node.status interval (sec), default 10
+	HandshakeTimeout  int `yaml:"handshake_timeout"`  // WS handshake timeout (sec), default 15
+	BackoffInitial    int `yaml:"backoff_initial"`    // initial reconnect delay (sec), default 1
+	BackoffMax        int `yaml:"backoff_max"`        // max reconnect delay (sec), default 60
+	DiscoveryInterval int `yaml:"discovery_interval"` // WS discovery interval (sec), default 300
 }
 
 type KernelConfig struct {
@@ -145,6 +183,174 @@ type LogConfig struct {
 	Output string `yaml:"output"`
 }
 
+type RootConfig struct {
+	Instances []Config `yaml:"instances,omitempty"`
+	Config    `yaml:",inline"`
+}
+
+func LoadRoot(path string) (*RootConfig, error) {
+	rc := &RootConfig{}
+
+	data, err := os.ReadFile(path)
+	if err == nil {
+		if err := yaml.Unmarshal(data, rc); err != nil {
+			return nil, fmt.Errorf("parse config: %w", err)
+		}
+		if len(rc.Instances) == 0 {
+			legacy := &Config{}
+			if err := yaml.Unmarshal(data, legacy); err != nil {
+				return nil, fmt.Errorf("parse legacy config: %w", err)
+			}
+			rc.Config = *legacy
+		}
+	} else if !os.IsNotExist(err) {
+		return nil, fmt.Errorf("read config file: %w", err)
+	}
+
+	// In multi-instance mode, top-level settings (log, kernel, node, ws, etc.)
+	// serve as defaults that each instance inherits. Instance-level settings
+	// take precedence over top-level ones.
+	if len(rc.Instances) > 0 {
+		for i := range rc.Instances {
+			rc.Instances[i].inheritFrom(&rc.Config)
+		}
+	}
+
+	// Derive default base directory from the config file's location so that
+	// data (certs, sing-box cache, …) lives next to the config file when
+	// config_dir is not explicitly set. This makes non-root / dev deployments
+	// work without any extra configuration.
+	baseDir := configBaseDir(path)
+
+	rc.applyEnvOverrides()
+	rc.resolveEnvRefs()
+
+	// Compute stable instance IDs early — they're content-based (derived from
+	// panel URL + machine/node ID), so reordering instances in the YAML doesn't
+	// cause data-directory mix-ups. Must run before setDefaultsFrom which uses
+	// InstanceID to derive per-instance config_dir.
+	if err := rc.assignInstanceIDs(); err != nil {
+		return nil, err
+	}
+
+	rc.setDefaultsFrom(baseDir)
+
+	if err := rc.validateRoot(); err != nil {
+		return nil, fmt.Errorf("invalid config: %w", err)
+	}
+
+	return rc, nil
+}
+
+func (rc *RootConfig) applyEnvOverrides() {
+	if len(rc.Instances) == 0 {
+		rc.Config.applyEnvOverrides()
+		return
+	}
+	for i := range rc.Instances {
+		rc.Instances[i].applyEnvOverrides()
+	}
+}
+
+func (rc *RootConfig) resolveEnvRefs() {
+	if len(rc.Instances) == 0 {
+		rc.Config.resolveEnvRefs()
+		return
+	}
+	for i := range rc.Instances {
+		rc.Instances[i].resolveEnvRefs()
+	}
+}
+
+func (rc *RootConfig) setDefaultsFrom(baseDir string) {
+	if len(rc.Instances) == 0 {
+		rc.Config.setDefaultsFrom(baseDir)
+		return
+	}
+	for i := range rc.Instances {
+		instBase := baseDir
+		// When multiple instances share the same base dir, derive a unique
+		// sub-directory per instance to avoid config_dir collisions.
+		// InstanceID is always populated by assignInstanceIDs() before this runs.
+		if len(rc.Instances) > 1 && rc.Instances[i].Kernel.ConfigDir == "" {
+			instBase = filepath.Join(baseDir, rc.Instances[i].InstanceID)
+		}
+		rc.Instances[i].setDefaultsFrom(instBase)
+	}
+}
+
+// assignInstanceIDs computes a stable, content-based InstanceID for each
+// instance that doesn't already have one. The ID is derived from the panel URL
+// and machine/node ID, so reordering instances in the YAML file doesn't cause
+// data directories to swap.
+func (rc *RootConfig) assignInstanceIDs() error {
+	for i := range rc.Instances {
+		if rc.Instances[i].InstanceID != "" {
+			continue
+		}
+		id, err := rc.Instances[i].AutoInstanceID()
+		if err != nil {
+			return fmt.Errorf("instances[%d]: %w", i, err)
+		}
+		rc.Instances[i].InstanceID = id
+	}
+	return nil
+}
+
+// configBaseDir resolves the canonical base directory from the config file path.
+// This directory is used as the default for config_dir when not explicitly set,
+// so that data (certs, sing-box cache, …) lives next to the config file.
+func configBaseDir(configPath string) string {
+	abs, err := filepath.Abs(configPath)
+	if err != nil {
+		return "/etc/xboard-node"
+	}
+	return filepath.Dir(abs)
+}
+
+func (rc *RootConfig) validateRoot() error {
+	if len(rc.Instances) == 0 {
+		return rc.Config.validate()
+	}
+	seen := map[string]struct{}{}
+	for i := range rc.Instances {
+		// InstanceID is already computed by assignInstanceIDs().
+		id := rc.Instances[i].InstanceID
+		if err := rc.Instances[i].validate(); err != nil {
+			return fmt.Errorf("instances[%d]: %w", i, err)
+		}
+		if _, ok := seen[id]; ok {
+			return fmt.Errorf("instances[%d]: duplicate instance id %q", i, id)
+		}
+		seen[id] = struct{}{}
+	}
+	return nil
+}
+
+func (rc *RootConfig) NormalizeInstances() ([]*Config, error) {
+	if len(rc.Instances) == 0 {
+		id, err := rc.Config.AutoInstanceID()
+		if err != nil {
+			return nil, err
+		}
+		rc.Config.InstanceID = id
+		return []*Config{&rc.Config}, nil
+	}
+	out := make([]*Config, 0, len(rc.Instances))
+	for i := range rc.Instances {
+		cfg := rc.Instances[i]
+		if cfg.InstanceID == "" {
+			id, err := cfg.AutoInstanceID()
+			if err != nil {
+				return nil, err
+			}
+			cfg.InstanceID = id
+		}
+		out = append(out, &cfg)
+	}
+	return out, nil
+}
+
 // Load reads configuration from a YAML file, then applies environment variable
 // overrides. If the config file does not exist, a config is built entirely from
 // environment variables (useful for Docker deployment with -e flags).
@@ -160,8 +366,11 @@ func Load(path string) (*Config, error) {
 		return nil, fmt.Errorf("read config file: %w", err)
 	}
 
+	baseDir := configBaseDir(path)
+
 	cfg.applyEnvOverrides()
-	cfg.setDefaults()
+	cfg.resolveEnvRefs()
+	cfg.setDefaultsFrom(baseDir)
 
 	if err := cfg.validate(); err != nil {
 		return nil, fmt.Errorf("invalid config: %w", err)
@@ -211,14 +420,145 @@ func (c *Config) applyEnvOverrides() {
 	if v := envFirst("logLevel", "LOG_LEVEL"); v != "" {
 		c.Log.Level = v
 	}
+	if v := envFirst("MACHINE_ID"); v != "" {
+		if id, err := strconv.Atoi(v); err == nil {
+			if c.Machine == nil {
+				c.Machine = &MachineConfig{}
+			}
+			c.Machine.MachineID = id
+		}
+	}
+	if v := envFirst("MACHINE_TOKEN"); v != "" {
+		if c.Machine == nil {
+			c.Machine = &MachineConfig{}
+		}
+		c.Machine.Token = v
+	}
 }
 
-func (c *Config) setDefaults() {
+func (c *Config) resolveEnvRefs() {
+	if c.Panel.Token == "" && c.Panel.TokenEnv != "" {
+		c.Panel.Token = os.Getenv(c.Panel.TokenEnv)
+	}
+	if c.Machine != nil && c.Machine.Token == "" && c.Machine.TokenEnv != "" {
+		c.Machine.Token = os.Getenv(c.Machine.TokenEnv)
+	}
+}
+
+// inheritFrom copies non-zero fields from the parent config into c when c's
+// corresponding field is zero-valued. This lets top-level settings serve as
+// defaults for each instance in multi-instance configs.
+// Fields that must be unique per instance (config_dir, cert_dir, instance_id)
+// are intentionally excluded.
+func (c *Config) inheritFrom(parent *Config) {
+	// Log
+	if c.Log.Level == "" {
+		c.Log.Level = parent.Log.Level
+	}
+	if c.Log.Output == "" {
+		c.Log.Output = parent.Log.Output
+	}
+	// Node intervals
+	if c.Node.PushInterval == 0 {
+		c.Node.PushInterval = parent.Node.PushInterval
+	}
+	if c.Node.PullInterval == 0 {
+		c.Node.PullInterval = parent.Node.PullInterval
+	}
+	if c.Node.TrackInterval == 0 {
+		c.Node.TrackInterval = parent.Node.TrackInterval
+	}
+	if c.Node.DeviceReportInterval == 0 {
+		c.Node.DeviceReportInterval = parent.Node.DeviceReportInterval
+	}
+	// WS
+	if c.WS.StatusInterval == 0 {
+		c.WS.StatusInterval = parent.WS.StatusInterval
+	}
+	if c.WS.HandshakeTimeout == 0 {
+		c.WS.HandshakeTimeout = parent.WS.HandshakeTimeout
+	}
+	if c.WS.BackoffInitial == 0 {
+		c.WS.BackoffInitial = parent.WS.BackoffInitial
+	}
+	if c.WS.BackoffMax == 0 {
+		c.WS.BackoffMax = parent.WS.BackoffMax
+	}
+	if c.WS.DiscoveryInterval == 0 {
+		c.WS.DiscoveryInterval = parent.WS.DiscoveryInterval
+	}
+	// Runtime
+	if c.Runtime.GoGCPercent == 0 {
+		c.Runtime.GoGCPercent = parent.Runtime.GoGCPercent
+	}
+	if c.Runtime.GoMemLimit == "" {
+		c.Runtime.GoMemLimit = parent.Runtime.GoMemLimit
+	}
+	// Kernel (NOT config_dir — each instance needs unique dir)
+	if c.Kernel.Type == "" {
+		c.Kernel.Type = parent.Kernel.Type
+	}
+	if c.Kernel.LogLevel == "" {
+		c.Kernel.LogLevel = parent.Kernel.LogLevel
+	}
+	if c.Kernel.GeoDataDir == "" {
+		c.Kernel.GeoDataDir = parent.Kernel.GeoDataDir
+	}
+	if c.Kernel.CustomConfig == "" {
+		c.Kernel.CustomConfig = parent.Kernel.CustomConfig
+	}
+	if len(c.Kernel.CustomOutbound) == 0 {
+		c.Kernel.CustomOutbound = parent.Kernel.CustomOutbound
+	}
+	if len(c.Kernel.CustomRoute) == 0 {
+		c.Kernel.CustomRoute = parent.Kernel.CustomRoute
+	}
+	// Cert (NOT cert_dir — derived from config_dir later)
+	if c.Cert.CertMode == "" {
+		c.Cert.CertMode = parent.Cert.CertMode
+	}
+	if c.Cert.Domain == "" {
+		c.Cert.Domain = parent.Cert.Domain
+	}
+	if c.Cert.Email == "" {
+		c.Cert.Email = parent.Cert.Email
+	}
+	if c.Cert.CertFile == "" {
+		c.Cert.CertFile = parent.Cert.CertFile
+	}
+	if c.Cert.KeyFile == "" {
+		c.Cert.KeyFile = parent.Cert.KeyFile
+	}
+	if c.Cert.DNSProvider == "" {
+		c.Cert.DNSProvider = parent.Cert.DNSProvider
+	}
+	if len(c.Cert.DNSEnv) == 0 {
+		c.Cert.DNSEnv = parent.Cert.DNSEnv
+	}
+	if c.Cert.HTTPPort == 0 {
+		c.Cert.HTTPPort = parent.Cert.HTTPPort
+	}
+	if c.Cert.CertContent == "" {
+		c.Cert.CertContent = parent.Cert.CertContent
+	}
+	if c.Cert.KeyContent == "" {
+		c.Cert.KeyContent = parent.Cert.KeyContent
+	}
+	// Only inherit AutoTLS when the child has no explicit cert configuration.
+	// A child with cert_mode, cert_file, or cert_content explicitly chooses
+	// its own cert strategy — inheriting auto_tls would override that choice.
+	childHasCertConfig := c.Cert.CertMode != "" || c.Cert.CertFile != "" || c.Cert.CertContent != ""
+	if !childHasCertConfig && !c.Cert.AutoTLS && parent.Cert.AutoTLS {
+		c.Cert.AutoTLS = parent.Cert.AutoTLS
+	}
+}
+
+func (c *Config) setDefaultsFrom(baseDir string) {
 	if c.Kernel.Type == "" {
 		c.Kernel.Type = "singbox"
 	}
 	if c.Kernel.ConfigDir == "" {
-		c.Kernel.ConfigDir = "/etc/xboard-node"
+		c.Kernel.ConfigDir = baseDir
 	}
 	if c.Kernel.GeoDataDir == "" {
 		c.Kernel.GeoDataDir = c.Kernel.ConfigDir
@@ -238,22 +578,145 @@ func (c *Config) setDefaults() {
 	if c.Cert.HTTPPort == 0 {
 		c.Cert.HTTPPort = 80
 	}
+	// WS defaults
+	if c.WS.StatusInterval == 0 {
+		c.WS.StatusInterval = 10
+	}
+	if c.WS.HandshakeTimeout == 0 {
+		c.WS.HandshakeTimeout = 15
+	}
+	if c.WS.BackoffInitial == 0 {
+		c.WS.BackoffInitial = 1
+	}
+	if c.WS.BackoffMax == 0 {
+		c.WS.BackoffMax = 60
+	}
+	if c.WS.DiscoveryInterval == 0 {
+		c.WS.DiscoveryInterval = 300
+	}
+	// Node defaults
+	if c.Node.TrackInterval == 0 {
+		c.Node.TrackInterval = 10
+	}
+	if c.Node.DeviceReportInterval == 0 {
+		c.Node.DeviceReportInterval = 30
+	}
+}
+
+func (c *Config) IsMachineMode() bool {
+	return c.Machine != nil && c.Machine.MachineID > 0
+}
+
+func (c *Config) AutoInstanceID() (string, error) {
+	mode := "node"
+	target := "0"
+	baseURL := strings.TrimSpace(c.Panel.URL)
+	if c.IsStandalone() {
+		mode = "standalone"
+		target = "local"
+		baseURL = "standalone"
+	} else if c.IsMachineMode() {
+		mode = "machine"
+		target = strconv.Itoa(c.Machine.MachineID)
+	} else if len(c.Nodes) > 0 {
+		// Multi-node mode: include sorted node IDs in the hash to avoid collisions.
+		ids := make([]string, len(c.Nodes))
+		for i, n := range c.Nodes {
+			ids[i] = strconv.Itoa(n.NodeID)
+		}
+		sort.Strings(ids)
+		target = strings.Join(ids, ",")
+	} else {
+		target = strconv.Itoa(c.Panel.NodeID)
+	}
+	normalized, slug, err := normalizeBaseURL(baseURL)
+	if err != nil {
+		return "", err
+	}
+	key := normalized + "|" + mode + "|" + target
+	sum := sha1.Sum([]byte(key))
+	short := hex.EncodeToString(sum[:])[:6]
+	return fmt.Sprintf("%s-%s-%s-%s", slug, mode, target, short), nil
+}
+
+func normalizeBaseURL(raw string) (string, string, error) {
+	if raw == "standalone" {
+		return raw, raw, nil
+	}
+	u, err := url.Parse(strings.TrimSpace(raw))
+	if err != nil {
+		return "", "", fmt.Errorf("invalid panel.url %q: %w", raw, err)
+	}
+	if u.Scheme == "" || u.Host == "" {
+		return "", "", fmt.Errorf("invalid panel.url %q", raw)
+	}
+	u.Scheme = strings.ToLower(u.Scheme)
+	u.Host = strings.ToLower(u.Host)
+	u.RawQuery = ""
+	u.Fragment = ""
+	pathPart := strings.TrimSuffix(u.EscapedPath(), "/")
+	normalized := u.Scheme + "://" + u.Host
+	if pathPart != "" && pathPart != "/" {
+		normalized += pathPart
+	}
+	slugBase := u.Host
+	if pathPart != "" && pathPart != "/" {
+		slugBase += "-" + strings.Trim(pathPart, "/")
+	}
+	slugBase = strings.ToLower(slugBase)
+	var b strings.Builder
+	lastDash := false
+	for _, r := range slugBase {
+		if (r >= 'a' && r <= 'z') || (r >= '0' && r <= '9') {
+			b.WriteRune(r)
+			lastDash = false
+			continue
+		}
+		if !lastDash {
+			b.WriteByte('-')
+			lastDash = true
+		}
+	}
+	slug := strings.Trim(b.String(), "-")
+	if slug == "" {
+		slug = "panel"
+	}
+	if len(slug) > 48 {
+		slug = strings.Trim(slug[:48], "-")
+	}
+	return normalized, slug, nil
 }
 
 func (c *Config) validate() error {
-	if c.Panel.URL == "" {
-		return fmt.Errorf("panel.url is required")
-	}
-	if c.Panel.Token == "" {
-		return fmt.Errorf("panel.token is required")
-	}
-	// In multi-node mode panel.node_id is optional; validate each NodeEntry instead.
-	if len(c.Nodes) == 0 && c.Panel.NodeID <= 0 {
-		return fmt.Errorf("panel.node_id must be positive (or use 'nodes:' for multi-node)")
-	}
-	for i, n := range c.Nodes {
-		if n.NodeID <= 0 {
-			return fmt.Errorf("nodes[%d].node_id must be positive", i)
+	if c.IsStandalone() {
+		if err := c.validateStandalone(); err != nil {
+			return err
+		}
+	} else if c.IsMachineMode() {
+		if c.Panel.URL == "" {
+			return fmt.Errorf("panel.url is required")
+		}
+		if c.Machine.Token == "" {
+			return fmt.Errorf("machine.token is required")
+		}
+		if len(c.Nodes) > 0 {
+			return fmt.Errorf("machine mode and nodes: are mutually exclusive")
+		}
+	} else {
+		if c.Panel.URL == "" {
+			return fmt.Errorf("panel.url is required")
+		}
+		if c.Panel.Token == "" {
+			return fmt.Errorf("panel.token is required")
+		}
+		// In multi-node mode panel.node_id is optional; validate each NodeEntry instead.
+		if len(c.Nodes) == 0 && c.Panel.NodeID <= 0 {
+			return fmt.Errorf("panel.node_id must be positive (or use 'nodes:' for multi-node)")
+		}
+		for i, n := range c.Nodes {
+			if n.NodeID <= 0 {
+				return fmt.Errorf("nodes[%d].node_id must be positive", i)
+			}
 		}
 	}
 	switch c.Kernel.Type {
@@ -330,41 +793,114 @@ func (c *Config) ExpandNodes() []*Config {
 	return result
 }
 
+func (c *Config) ExpandMachineNode(nodeID int, nodeType string) *Config {
+	nodeCfg := *c
+	nodeCfg.Nodes = nil
+	nodeCfg.Panel.NodeID = nodeID
+	nodeCfg.Panel.NodeType = nodeType
+	nodeCfg.Panel.Token = c.Machine.Token
+	nodeCfg.Panel.MachineID = c.Machine.MachineID
+
+	nodeCfg.Kernel.ConfigDir = fmt.Sprintf("%s/node-%d", c.Kernel.ConfigDir, nodeID)
+	if nodeCfg.Kernel.GeoDataDir == c.Kernel.ConfigDir {
+		nodeCfg.Kernel.GeoDataDir = c.Kernel.GeoDataDir
+	}
+	nodeCfg.Cert.CertDir = filepath.Join(nodeCfg.Kernel.ConfigDir, "certs")
+
+	return &nodeCfg
+}
+
 func InitLogger(cfg LogConfig) {
-	var level slog.Level
+	var minLevel slog.Level
 	switch cfg.Level {
 	case "debug":
-		level = slog.LevelDebug
+		minLevel = slog.LevelDebug
 	case "warn":
-		level = slog.LevelWarn
+		minLevel = slog.LevelWarn
 	case "error":
-		level = slog.LevelError
+		minLevel = slog.LevelError
 	default:
-		level = slog.LevelInfo
+		minLevel = slog.LevelInfo
 	}
 
 	var w io.Writer
+	useColor := false
 	switch cfg.Output {
 	case "stdout", "":
 		w = os.Stdout
+		useColor = term.IsTerminal(int(os.Stdout.Fd()))
 	case "stderr":
 		w = os.Stderr
+		useColor = term.IsTerminal(int(os.Stderr.Fd()))
 	default:
 		dir := filepath.Dir(cfg.Output)
 		if err := os.MkdirAll(dir, 0o755); err != nil {
-			slog.Error("failed to create log dir, falling back to stdout", "error", err)
+			fmt.Fprintf(os.Stderr, "failed to create log dir, falling back to stdout: %v\n", err)
 			w = os.Stdout
+			useColor = term.IsTerminal(int(os.Stdout.Fd()))
 		} else {
 			f, err := os.OpenFile(cfg.Output, os.O_CREATE|os.O_WRONLY|os.O_APPEND, 0o644)
 			if err != nil {
-				slog.Error("failed to open log file, falling back to stdout", "error", err)
+				fmt.Fprintf(os.Stderr, "failed to open log file, falling back to stdout: %v\n", err)
 				w = os.Stdout
+				useColor = term.IsTerminal(int(os.Stdout.Fd()))
 			} else {
 				w = f
+				useColor = false
 			}
 		}
 	}
 
-	handler := slog.NewTextHandler(w, &slog.HandlerOptions{Level: level})
-	slog.SetDefault(slog.New(handler))
+	nlog.Init(w, minLevel, useColor)
+	// Application logging goes through nlog; silence slog.Default for stray library use.
+	slog.SetDefault(slog.New(slog.DiscardHandler))
+}
+
+// ValidateStartupLayout checks that multiple instances do not conflict on
+// health ports, kernel config directories, or node bindings.
+func ValidateStartupLayout(instances []*Config) error {
+	healthPorts := make(map[int]string)
+	configDirs := make(map[string]string)
+	nodeBindings := make(map[string]string)
+	for _, instance := range instances {
+		if instance == nil {
+			continue
+		}
+		owner := instance.InstanceID
+		if instance.HealthPort > 0 {
+			if other, ok := healthPorts[instance.HealthPort]; ok {
+				return fmt.Errorf("health_port %d is used by both %s and %s", instance.HealthPort, other, owner)
+			}
+			healthPorts[instance.HealthPort] = owner
+		}
+		if dir := strings.TrimSpace(instance.Kernel.ConfigDir); dir != "" {
+			if other, ok := configDirs[dir]; ok && other != owner {
+				return fmt.Errorf("kernel config_dir %q is shared by %s and %s", dir, other, owner)
+			}
+			configDirs[dir] = owner
+		}
+		// Machine mode discovers nodes dynamically — skip static binding check,
+		// but still validate custom_config if set at the instance level.
+		if instance.IsMachineMode() {
+			if path := strings.TrimSpace(instance.Kernel.CustomConfig); path != "" {
+				if _, err := os.Stat(path); err != nil {
+					return fmt.Errorf("kernel.custom_config %q for %s: %w", path, owner, err)
+				}
+			}
+			continue
+		}
+		for _, node := range instance.ExpandNodes() {
+			binding := fmt.Sprintf("%s/%d", strings.TrimSpace(node.Panel.URL), node.Panel.NodeID)
+			if other, ok := nodeBindings[binding]; ok {
+				return fmt.Errorf("node binding %s is declared by both %s and %s", binding, other, owner)
+			}
+			nodeBindings[binding] = owner
+			if path := strings.TrimSpace(node.Kernel.CustomConfig); path != "" {
+				if _, err := os.Stat(path); err != nil {
+					return fmt.Errorf("kernel.custom_config %q for %s: %w", path, owner, err)
+				}
+			}
+		}
+	}
+	return nil
 }

internal/config/config_test.go

@@ -1,6 +1,7 @@
 package config
 
 import (
+	"fmt"
 	"os"
 	"path/filepath"
 	"testing"
@@ -67,8 +68,10 @@ panel:
 	if cfg.Kernel.Type != "singbox" {
 		t.Errorf("default kernel.type: got %q, want singbox", cfg.Kernel.Type)
 	}
-	if cfg.Kernel.ConfigDir != "/etc/xboard-node" {
-		t.Errorf("default config_dir: got %q", cfg.Kernel.ConfigDir)
+	// config_dir should default to the directory containing the config file.
+	expectedDir := filepath.Dir(path)
+	if cfg.Kernel.ConfigDir != expectedDir {
+		t.Errorf("default config_dir: got %q, want %q", cfg.Kernel.ConfigDir, expectedDir)
 	}
 	if cfg.Kernel.LogLevel != "warn" {
 		t.Errorf("default kernel log_level: got %q, want warn", cfg.Kernel.LogLevel)
@@ -82,7 +85,7 @@ panel:
 	if cfg.Cert.HTTPPort != 80 {
 		t.Errorf("default http_port: got %d, want 80", cfg.Cert.HTTPPort)
 	}
-	expectedCertDir := filepath.Join("/etc/xboard-node", "certs")
+	expectedCertDir := filepath.Join(expectedDir, "certs")
 	if cfg.Cert.CertDir != expectedCertDir {
 		t.Errorf("default cert_dir: got %q, want %q", cfg.Cert.CertDir, expectedCertDir)
 	}
@@ -267,3 +270,276 @@ node:
 		t.Errorf("pull_interval: got %d", cfg.Node.PullInterval)
 	}
 }
+
+func TestLoad_StandaloneConfig(t *testing.T) {
+	path := writeTemp(t, `
+standalone:
+  enabled: true
+  node:
+    protocol: "vless"
+    server_port: 8443
+    network: "ws"
+    tls: 1
+    network_settings:
+      path: "/ws"
+  users:
+    - id: 1
+      uuid: "11111111-1111-1111-1111-111111111111"
+kernel:
+  type: xray
+cert:
+  cert_mode: none
+`)
+	cfg, err := Load(path)
+	if err != nil {
+		t.Fatalf("Load: %v", err)
+	}
+	if !cfg.IsStandalone() {
+		t.Fatal("expected standalone mode to be enabled")
+	}
+	if cfg.Standalone.Node.Protocol != "vless" {
+		t.Errorf("standalone.node.protocol: got %q", cfg.Standalone.Node.Protocol)
+	}
+	if cfg.Standalone.Node.ServerPort != 8443 {
+		t.Errorf("standalone.node.server_port: got %d", cfg.Standalone.Node.ServerPort)
+	}
+	if len(cfg.Standalone.Users) != 1 {
+		t.Fatalf("standalone.users: got %d", len(cfg.Standalone.Users))
+	}
+	if cfg.Standalone.Users[0].UUID != "11111111-1111-1111-1111-111111111111" {
+		t.Errorf("standalone.users[0].uuid: got %q", cfg.Standalone.Users[0].UUID)
+	}
+}
+
+func TestLoad_StandaloneRequiresUsers(t *testing.T) {
+	path := writeTemp(t, `
+standalone:
+  enabled: true
+  node:
+    protocol: "trojan"
+    server_port: 443
+kernel:
+  type: singbox
+`)
+	_, err := Load(path)
+	if err == nil {
+		t.Fatal("expected error for standalone mode without users")
+	}
+}
+
+
+func TestLoadRoot_LegacyConfigNormalizesToSingleInstance(t *testing.T) {
+	path := writeTemp(t, `
+panel:
+  url: "https://panel.example.com"
+  token: "tok"
+  node_id: 1
+`)
+	root, err := LoadRoot(path)
+	if err != nil {
+		t.Fatalf("LoadRoot: %v", err)
+	}
+	instances, err := root.NormalizeInstances()
+	if err != nil {
+		t.Fatalf("NormalizeInstances: %v", err)
+	}
+	if len(instances) != 1 {
+		t.Fatalf("instances: got %d, want 1", len(instances))
+	}
+	if instances[0].Panel.URL != "https://panel.example.com" {
+		t.Errorf("instance url: got %q", instances[0].Panel.URL)
+	}
+	if instances[0].InstanceID == "" {
+		t.Fatal("expected auto instance id")
+	}
+}
+
+func TestLoadRoot_InstancesConfig(t *testing.T) {
+	path := writeTemp(t, `
+instances:
+  - panel:
+      url: "https://panel-a.example.com"
+      token_env: "PANEL_A_TOKEN"
+      node_id: 1
+    kernel:
+      type: singbox
+  - panel:
+      url: "https://panel-b.example.com"
+    machine:
+      machine_id: 2
+      token_env: "PANEL_B_MACHINE_TOKEN"
+    kernel:
+      type: singbox
+`)
+	t.Setenv("PANEL_A_TOKEN", "token-a")
+	t.Setenv("PANEL_B_MACHINE_TOKEN", "token-b")
+	root, err := LoadRoot(path)
+	if err != nil {
+		t.Fatalf("LoadRoot: %v", err)
+	}
+	instances, err := root.NormalizeInstances()
+	if err != nil {
+		t.Fatalf("NormalizeInstances: %v", err)
+	}
+	if len(instances) != 2 {
+		t.Fatalf("instances: got %d, want 2", len(instances))
+	}
+	if instances[0].Panel.Token != "token-a" {
+		t.Errorf("panel a token: got %q", instances[0].Panel.Token)
+	}
+	if instances[1].Machine == nil || instances[1].Machine.Token != "token-b" {
+		t.Fatalf("machine token: got %+v", instances[1].Machine)
+	}
+	if instances[0].InstanceID == instances[1].InstanceID {
+		t.Fatal("expected unique instance ids")
+	}
+}
+
+func TestConfig_AutoInstanceIDStable(t *testing.T) {
+	cfg := &Config{Panel: PanelConfig{URL: "https://Panel.Example.com/", NodeID: 1, Token: "tok"}}
+	cfg.setDefaultsFrom("/etc/xboard-node")
+	id1, err := cfg.AutoInstanceID()
+	if err != nil {
+		t.Fatalf("AutoInstanceID: %v", err)
+	}
+	id2, err := cfg.AutoInstanceID()
+	if err != nil {
+		t.Fatalf("AutoInstanceID second: %v", err)
+	}
+	if id1 != id2 {
+		t.Fatalf("ids differ: %q vs %q", id1, id2)
+	}
+}
+
+func TestLoadRoot_InheritanceFromTopLevel(t *testing.T) {
+	path := writeTemp(t, `
+log:
+  level: "debug"
+  output: "stderr"
+kernel:
+  type: xray
+  log_level: "error"
+node:
+  push_interval: 42
+  pull_interval: 99
+instances:
+  - panel:
+      url: "https://panel.example.com"
+      token: "tok-a"
+      node_id: 1
+  - panel:
+      url: "https://panel.example.com"
+      token: "tok-b"
+      node_id: 2
+    log:
+      level: "warn"
+`)
+	root, err := LoadRoot(path)
+	if err != nil {
+		t.Fatalf("LoadRoot: %v", err)
+	}
+	if len(root.Instances) != 2 {
+		t.Fatalf("instances: got %d, want 2", len(root.Instances))
+	}
+	// Instance 0: inherits everything from top-level.
+	inst0 := root.Instances[0]
+	if inst0.Log.Level != "debug" {
+		t.Errorf("inst0 log.level: got %q, want %q", inst0.Log.Level, "debug")
+	}
+	if inst0.Log.Output != "stderr" {
+		t.Errorf("inst0 log.output: got %q, want %q", inst0.Log.Output, "stderr")
+	}
+	if inst0.Kernel.Type != "xray" {
+		t.Errorf("inst0 kernel.type: got %q, want %q", inst0.Kernel.Type, "xray")
+	}
+	if inst0.Kernel.LogLevel != "error" {
+		t.Errorf("inst0 kernel.log_level: got %q, want %q", inst0.Kernel.LogLevel, "error")
+	}
+	if inst0.Node.PushInterval != 42 {
+		t.Errorf("inst0 push_interval: got %d, want 42", inst0.Node.PushInterval)
+	}
+	if inst0.Node.PullInterval != 99 {
+		t.Errorf("inst0 pull_interval: got %d, want 99", inst0.Node.PullInterval)
+	}
+	// Instance 1: overrides log.level, inherits the rest.
+	inst1 := root.Instances[1]
+	if inst1.Log.Level != "warn" {
+		t.Errorf("inst1 log.level: got %q, want %q", inst1.Log.Level, "warn")
+	}
+	if inst1.Log.Output != "stderr" {
+		t.Errorf("inst1 log.output: got %q, want %q", inst1.Log.Output, "stderr")
+	}
+	if inst1.Kernel.Type != "xray" {
+		t.Errorf("inst1 kernel.type: got %q, want %q", inst1.Kernel.Type, "xray")
+	}
+}
+
+func TestLoadRoot_InstanceOrderDoesNotAffectConfigDir(t *testing.T) {
+	yamlTmpl := func(first, second int) string {
+		return fmt.Sprintf(`
+instances:
+  - panel:
+      url: "https://panel.example.com"
+      token: "tok-a"
+      node_id: %d
+  - panel:
+      url: "https://panel.example.com"
+      token: "tok-b"
+      node_id: %d
+`, first, second)
+	}
+
+	path1 := writeTemp(t, yamlTmpl(1, 2))
+	path2 := writeTemp(t, yamlTmpl(2, 1))
+	root1, err := LoadRoot(path1)
+	if err != nil {
+		t.Fatalf("LoadRoot order1: %v", err)
+	}
+	root2, err := LoadRoot(path2)
+	if err != nil {
+		t.Fatalf("LoadRoot order2: %v", err)
+	}
+
+	// Collect the config_dir suffix (relative to base) by InstanceID.
+	suffix := func(instances []Config) map[string]string {
+		m := map[string]string{}
+		for _, inst := range instances {
+			// The config_dir ends with .../{instanceID}, extract last component.
+			m[inst.InstanceID] = filepath.Base(inst.Kernel.ConfigDir)
+		}
+		return m
+	}
+	dirs1 := suffix(root1.Instances)
+	dirs2 := suffix(root2.Instances)
+
+	for id, s1 := range dirs1 {
+		s2, ok := dirs2[id]
+		if !ok {
+			t.Fatalf("instance %q missing in reversed order", id)
+		}
+		if s1 != s2 {
+			t.Errorf("instance %q config_dir suffix differs: %q vs %q", id, s1, s2)
+		}
+	}
+}
+
+func TestInheritFrom_AutoTLSNotForcedWhenChildHasCertMode(t *testing.T) {
+	parent := &Config{Cert: CertConfig{AutoTLS: true, Domain: "example.com"}}
+	child := &Config{Cert: CertConfig{CertMode: "file", CertFile: "/etc/cert.pem"}}
+	child.inheritFrom(parent)
+	if child.Cert.AutoTLS {
+		t.Error("auto_tls should NOT be inherited when child has explicit cert_mode")
+	}
+	if child.Cert.Domain != "example.com" {
+		t.Errorf("domain: got %q, want %q", child.Cert.Domain, "example.com")
+	}
+}
+
+func TestInheritFrom_AutoTLSInheritedWhenChildHasNoCertConfig(t *testing.T) {
+	parent := &Config{Cert: CertConfig{AutoTLS: true, Domain: "example.com"}}
+	child := &Config{}
+	child.inheritFrom(parent)
+	if !child.Cert.AutoTLS {
+		t.Error("auto_tls should be inherited when child has no cert config")
+	}
+}

internal/config/standalone.go

@@ -0,0 +1,152 @@
+package config
+
+import "fmt"
+
+// StandaloneConfig describes a locally-managed node that runs without panel
+// handshake, polling, reporting, or WebSocket sync.
+type StandaloneConfig struct {
+	Enabled bool                 `yaml:"enabled"`
+	Node    StandaloneNodeConfig `yaml:"node"`
+	Users   []StandaloneUser     `yaml:"users"`
+}
+
+// StandaloneNodeConfig captures the subset of panel node config that is needed
+// to generate a working local inbound.
+type StandaloneNodeConfig struct {
+	Protocol string `yaml:"protocol"`
+	ListenIP string `yaml:"listen_ip,omitempty"`
+
+	ServerPort int    `yaml:"server_port"`
+	Network    string `yaml:"network,omitempty"`
+
+	NetworkSettings  map[string]any              `yaml:"network_settings,omitempty"`
+	Routes           []StandaloneRouteRule       `yaml:"routes,omitempty"`
+	CustomRouteRules []StandaloneCustomRouteRule `yaml:"custom_route_rules,omitempty"`
+
+	Cipher    string `yaml:"cipher,omitempty"`
+	Plugin    string `yaml:"plugin,omitempty"`
+	PluginOpt string `yaml:"plugin_opts,omitempty"`
+	ServerKey string `yaml:"server_key,omitempty"`
+
+	TLS         int            `yaml:"tls,omitempty"`
+	Flow        string         `yaml:"flow,omitempty"`
+	Decryption  string         `yaml:"decryption,omitempty"`
+	TLSSettings map[string]any `yaml:"tls_settings,omitempty"`
+
+	Host       string `yaml:"host,omitempty"`
+	ServerName string `yaml:"server_name,omitempty"`
+
+	Version      int    `yaml:"version,omitempty"`
+	UpMbps       int    `yaml:"up_mbps,omitempty"`
+	DownMbps     int    `yaml:"down_mbps,omitempty"`
+	Obfs         string `yaml:"obfs,omitempty"`
+	ObfsPassword string `yaml:"obfs_password,omitempty"`
+
+	CongestionControl string `yaml:"congestion_control,omitempty"`
+	PaddingScheme     string `yaml:"padding_scheme,omitempty"`
+	Transport         string `yaml:"transport,omitempty"`
+	TrafficPattern    string `yaml:"traffic_pattern,omitempty"`
+
+	Multiplex           *StandaloneMultiplexConfig `yaml:"multiplex,omitempty"`
+	AcceptProxyProtocol bool                       `yaml:"accept_proxy_protocol,omitempty"`
+}
+
+type StandaloneRouteRule struct {
+	ID          int      `yaml:"id,omitempty"`
+	Match       []string `yaml:"match,omitempty"`
+	Action      string   `yaml:"action,omitempty"`
+	ActionValue string   `yaml:"action_value,omitempty"`
+}
+
+type StandaloneCustomRouteRule struct {
+	Name     string                `yaml:"name,omitempty"`
+	Disabled bool                  `yaml:"disabled,omitempty"`
+	Match    StandaloneRouteMatch  `yaml:"match,omitempty"`
+	Action   StandaloneRouteAction `yaml:"action,omitempty"`
+}
+
+type StandaloneRouteMatch struct {
+	Domains        []string `yaml:"domains,omitempty"`
+	DomainSuffixes []string `yaml:"domain_suffixes,omitempty"`
+	IPCIDRs        []string `yaml:"ip_cidrs,omitempty"`
+	Ports          []string `yaml:"ports,omitempty"`
+	Networks       []string `yaml:"networks,omitempty"`
+	SourceCIDRs    []string `yaml:"source_cidrs,omitempty"`
+	SourcePorts    []string `yaml:"source_ports,omitempty"`
+}
+
+type StandaloneRouteAction struct {
+	Type   string `yaml:"type,omitempty"`
+	Target string `yaml:"target,omitempty"`
+}
+
+type StandaloneMultiplexConfig struct {
+	Enabled        bool                    `yaml:"enabled"`
+	Protocol       string                  `yaml:"protocol,omitempty"`
+	MaxConnections int                     `yaml:"max_connections,omitempty"`
+	MinStreams     int                     `yaml:"min_streams,omitempty"`
+	MaxStreams     int                     `yaml:"max_streams,omitempty"`
+	Padding        bool                    `yaml:"padding,omitempty"`
+	Brutal         *StandaloneBrutalConfig `yaml:"brutal,omitempty"`
+}
+
+type StandaloneBrutalConfig struct {
+	Enabled  bool `yaml:"enabled"`
+	UpMbps   int  `yaml:"up_mbps,omitempty"`
+	DownMbps int  `yaml:"down_mbps,omitempty"`
+}
+
+type StandaloneUser struct {
+	ID          int    `yaml:"id"`
+	UUID        string `yaml:"uuid"`
+	SpeedLimit  int    `yaml:"speed_limit,omitempty"`
+	DeviceLimit int    `yaml:"device_limit,omitempty"`
+}
+
+func (c *Config) IsStandalone() bool {
+	return c.Standalone != nil && c.Standalone.Enabled
+}
+
+func (c *Config) validateStandalone() error {
+	if c.Standalone == nil {
+		return fmt.Errorf("standalone config is required when standalone mode is enabled")
+	}
+	if len(c.Nodes) > 0 {
+		return fmt.Errorf("standalone mode does not support top-level 'nodes'")
+	}
+
+	node := c.Standalone.Node
+	if node.Protocol == "" {
+		return fmt.Errorf("standalone.node.protocol is required")
+	}
+	if node.ServerPort <= 0 || node.ServerPort > 65535 {
+		return fmt.Errorf("standalone.node.server_port must be between 1 and 65535")
+	}
+	if node.TLS < 0 || node.TLS > 2 {
+		return fmt.Errorf("standalone.node.tls must be 0 (off), 1 (tls), or 2 (reality)")
+	}
+	if len(c.Standalone.Users) == 0 {
+		return fmt.Errorf("standalone.users must not be empty")
+	}
+
+	ids := make(map[int]struct{}, len(c.Standalone.Users))
+	uuids := make(map[string]struct{}, len(c.Standalone.Users))
+	for i, user := range c.Standalone.Users {
+		if user.ID <= 0 {
+			return fmt.Errorf("standalone.users[%d].id must be positive", i)
+		}
+		if user.UUID == "" {
+			return fmt.Errorf("standalone.users[%d].uuid is required", i)
+		}
+		if _, exists := ids[user.ID]; exists {
+			return fmt.Errorf("standalone.users[%d].id duplicates %d", i, user.ID)
+		}
+		if _, exists := uuids[user.UUID]; exists {
+			return fmt.Errorf("standalone.users[%d].uuid duplicates %q", i, user.UUID)
+		}
+		ids[user.ID] = struct{}{}
+		uuids[user.UUID] = struct{}{}
+	}
+
+	return nil
+}

internal/config/watcher.go

@@ -0,0 +1,160 @@
+package config
+
+import (
+	"context"
+	"path/filepath"
+	"sync"
+	"sync/atomic"
+	"time"
+
+	"github.com/cedar2025/xboard-node/internal/nlog"
+	"github.com/fsnotify/fsnotify"
+)
+
+// Watcher reloads the config file on change (debounced) and calls onChange.
+type Watcher struct {
+	path         string
+	debounce     time.Duration
+	onChange     func(*Config)
+	onChangeRoot func(*RootConfig)
+	watcher      *fsnotify.Watcher
+
+	stopOnce sync.Once
+	stopCh   chan struct{}
+	stopped  atomic.Bool
+}
+
+// WatchConfig watches path; invalid reloads are logged and ignored.
+// Stop via Stop() or cancel ctx.
+func WatchConfig(ctx context.Context, path string, onChange func(*Config)) (*Watcher, error) {
+	absPath, err := filepath.Abs(path)
+	if err != nil {
+		return nil, err
+	}
+
+	fsw, err := fsnotify.NewWatcher()
+	if err != nil {
+		return nil, err
+	}
+
+	// Watch parent dir so rename/atomic saves are visible.
+	dir := filepath.Dir(absPath)
+	if err := fsw.Add(dir); err != nil {
+		fsw.Close()
+		return nil, err
+	}
+
+	w := &Watcher{
+		path:     absPath,
+		debounce: 1 * time.Second,
+		onChange: onChange,
+		watcher:  fsw,
+		stopCh:   make(chan struct{}),
+	}
+
+	go w.loop(ctx)
+	nlog.Core().Info("config watcher started", "path", absPath)
+	return w, nil
+}
+
+
+// WatchConfigRoot watches path and reloads the root config model.
+func WatchConfigRoot(ctx context.Context, path string, onChange func(*RootConfig)) (*Watcher, error) {
+	absPath, err := filepath.Abs(path)
+	if err != nil {
+		return nil, err
+	}
+
+	fsw, err := fsnotify.NewWatcher()
+	if err != nil {
+		return nil, err
+	}
+
+	dir := filepath.Dir(absPath)
+	if err := fsw.Add(dir); err != nil {
+		fsw.Close()
+		return nil, err
+	}
+
+	w := &Watcher{
+		path:         absPath,
+		debounce:     1 * time.Second,
+		onChangeRoot: onChange,
+		watcher:      fsw,
+		stopCh:       make(chan struct{}),
+	}
+
+	go w.loop(ctx)
+	nlog.Core().Info("config watcher started", "path", absPath)
+	return w, nil
+}
+
+func (w *Watcher) loop(ctx context.Context) {
+	var timer *time.Timer
+	defer func() {
+		if timer != nil {
+			timer.Stop()
+		}
+		w.watcher.Close()
+	}()
+
+	for {
+		select {
+		case <-ctx.Done():
+			return
+		case <-w.stopCh:
+			return
+
+		case ev, ok := <-w.watcher.Events:
+			if !ok {
+				return
+			}
+			if filepath.Clean(ev.Name) != w.path {
+				continue
+			}
+			if !ev.Has(fsnotify.Write) && !ev.Has(fsnotify.Create) {
+				continue
+			}
+
+			if timer == nil {
+				timer = time.AfterFunc(w.debounce, w.reload)
+			} else {
+				timer.Reset(w.debounce)
+			}
+
+		case err, ok := <-w.watcher.Errors:
+			if !ok {
+				return
+			}
+			nlog.Core().Warn("config watcher error", "error", err)
+		}
+	}
+}
+
+func (w *Watcher) reload() {
+	if w.stopped.Load() {
+		return
+	}
+	if w.onChangeRoot != nil {
+		root, err := LoadRoot(w.path)
+		if err != nil {
+			nlog.Core().Error("config reload failed, keeping current config", "error", err)
+			return
+		}
+		nlog.Core().Info("config reloaded successfully")
+		w.onChangeRoot(root)
+		return
+	}
+	cfg, err := Load(w.path)
+	if err != nil {
+		nlog.Core().Error("config reload failed, keeping current config", "error", err)
+		return
+	}
+	nlog.Core().Info("config reloaded successfully")
+	w.onChange(cfg)
+}
+
+func (w *Watcher) Stop() {
+	w.stopped.Store(true)
+	w.stopOnce.Do(func() { close(w.stopCh) })
+}

internal/controlplane/local.go

@@ -0,0 +1,61 @@
+package controlplane
+
+import (
+	"context"
+	"fmt"
+
+	"github.com/cedar2025/xboard-node/internal/config"
+	"github.com/cedar2025/xboard-node/internal/model"
+)
+
+type LocalControlPlane struct {
+	cfg *config.Config
+}
+
+func NewLocalControlPlane(cfg *config.Config) *LocalControlPlane { return &LocalControlPlane{cfg: cfg} }
+func (l *LocalControlPlane) SupportsPolling() bool               { return false }
+func (l *LocalControlPlane) SupportsDiscovery() bool             { return false }
+func (l *LocalControlPlane) SupportsReporting() bool             { return false }
+func (l *LocalControlPlane) SupportsDeviceReports() bool         { return false }
+
+func (l *LocalControlPlane) Initial(ctx context.Context, _ func() map[string]interface{}, _ chan<- Event, _ chan<- StatusChange) (Bootstrap, error) {
+	select {
+	case <-ctx.Done():
+		return Bootstrap{}, ctx.Err()
+	default:
+	}
+	nodeSpec, err := model.NodeSpecFromStandaloneValidated(l.cfg)
+	if err != nil {
+		return Bootstrap{}, fmt.Errorf("load standalone node config: %w", err)
+	}
+	return Bootstrap{
+		PushInterval: l.cfg.Node.PushInterval,
+		PullInterval: l.cfg.Node.PullInterval,
+		Config:       nodeSpec,
+		Users:        model.UserSpecsFromStandalone(l.cfg),
+	}, nil
+}
+
+func (l *LocalControlPlane) Poll(ctx context.Context) (Snapshot, error) {
+	select {
+	case <-ctx.Done():
+		return Snapshot{}, ctx.Err()
+	default:
+		return Snapshot{}, nil
+	}
+}
+
+func (l *LocalControlPlane) Discover(ctx context.Context, _ func() map[string]interface{}, _ chan<- Event, _ chan<- StatusChange) (PushClient, error) {
+	select {
+	case <-ctx.Done():
+		return nil, ctx.Err()
+	default:
+		return nil, nil
+	}
+}
+
+func (l *LocalControlPlane) Report(payload ReportPayload) error { _ = payload; return nil }
+func (l *LocalControlPlane) ReportDevices(push PushClient, devices map[int][]string) {
+	_, _ = push, devices
+}
+func (l *LocalControlPlane) Metrics() APIMetrics { return APIMetrics{} }

internal/controlplane/local_test.go

@@ -0,0 +1,32 @@
+package controlplane
+
+import (
+	"context"
+	"testing"
+
+	"github.com/cedar2025/xboard-node/internal/config"
+)
+
+func TestLocalControlPlaneInitial(t *testing.T) {
+	cp := NewLocalControlPlane(&config.Config{
+		Kernel: config.KernelConfig{Type: "singbox", LogLevel: "warn"},
+		Standalone: &config.StandaloneConfig{
+			Enabled: true,
+			Node: config.StandaloneNodeConfig{Protocol: "shadowsocks", ServerPort: 8388, Cipher: "aes-128-gcm"},
+			Users: []config.StandaloneUser{{ID: 1, UUID: "secret"}},
+		},
+	})
+	bootstrap, err := cp.Initial(context.Background(), nil, nil, nil)
+	if err != nil {
+		t.Fatalf("Initial: %v", err)
+	}
+	if bootstrap.Push != nil {
+		t.Fatal("local control plane should not create push client")
+	}
+	if bootstrap.Config == nil || bootstrap.Config.Protocol != "shadowsocks" {
+		t.Fatalf("config: %#v", bootstrap.Config)
+	}
+	if len(bootstrap.Users) != 1 || bootstrap.Users[0].UUID != "secret" {
+		t.Fatalf("users: %#v", bootstrap.Users)
+	}
+}

internal/controlplane/machine.go

@@ -0,0 +1,145 @@
+package controlplane
+
+import (
+	"context"
+	"fmt"
+
+	"github.com/cedar2025/xboard-node/internal/config"
+	"github.com/cedar2025/xboard-node/internal/model"
+	"github.com/cedar2025/xboard-node/internal/nlog"
+	"github.com/cedar2025/xboard-node/internal/panel"
+)
+
+// MachinePanelControlPlane implements ControlPlane for a single node running
+// under the machine orchestrator. REST calls go through a per-node panel.Client
+// with machine-level auth; WS events are delivered via an externally-managed
+// PushClient supplied by the machine WS mux.
+type MachinePanelControlPlane struct {
+	client *panel.Client
+	kcfg   config.KernelConfig
+	push   PushClient // virtual push client from WS mux (may be nil)
+
+	registerFn func(statuses chan<- StatusChange) *NodeMailbox
+}
+
+// NewMachinePanelControlPlane creates a per-node control plane for machine mode.
+// push may be nil if WS is unavailable.
+// registerFn is called during Initial() to wire machine mailbox access and status
+// updates into the service loop. It may be nil when WS is not used.
+func NewMachinePanelControlPlane(
+	client *panel.Client,
+	kcfg config.KernelConfig,
+	push PushClient,
+	registerFn func(statuses chan<- StatusChange) *NodeMailbox,
+) *MachinePanelControlPlane {
+	return &MachinePanelControlPlane{
+		client:     client,
+		kcfg:       kcfg,
+		push:       push,
+		registerFn: registerFn,
+	}
+}
+
+func (p *MachinePanelControlPlane) SupportsPolling() bool       { return true }
+func (p *MachinePanelControlPlane) SupportsDiscovery() bool     { return false }
+func (p *MachinePanelControlPlane) SupportsReporting() bool     { return true }
+func (p *MachinePanelControlPlane) SupportsDeviceReports() bool { return p.push != nil }
+
+func (p *MachinePanelControlPlane) Initial(
+	ctx context.Context,
+	metricsFn func() map[string]interface{},
+	events chan<- Event,
+	statuses chan<- StatusChange,
+) (Bootstrap, error) {
+	// In machine mode the orchestrator handles the shared WS handshake, but each
+	// node service still needs one initial REST snapshot so newly discovered nodes
+	// can start immediately instead of waiting for a subsequent WS full-sync.
+	hs, err := p.client.Handshake()
+	if err != nil {
+		nlog.Core().Debug("machine node handshake skipped, using REST", "error", err)
+	}
+
+	bootstrap := Bootstrap{}
+	if hs != nil {
+		bootstrap.PushInterval = hs.Settings.PushInterval
+		bootstrap.PullInterval = hs.Settings.PullInterval
+	}
+
+	configSnapshot, err := p.client.GetConfig()
+	if err != nil {
+		return Bootstrap{}, fmt.Errorf("machine initial config: %w", err)
+	}
+	if configSnapshot == nil {
+		return Bootstrap{}, fmt.Errorf("machine initial config is nil")
+	}
+	users, err := p.client.GetUsers()
+	if err != nil {
+		return Bootstrap{}, fmt.Errorf("machine initial users: %w", err)
+	}
+	bootstrap.Config, err = model.NodeSpecFromPanelValidated(configSnapshot, p.kcfg)
+	if err != nil {
+		return Bootstrap{}, fmt.Errorf("machine initial config normalize: %w", err)
+	}
+	bootstrap.Users = model.UserSpecsFromPanel(users)
+
+	// Register mailbox access after the initial snapshot is prepared, so the
+	// Service can start the kernel immediately and only then drain coalesced WS
+	// state from the machine mailbox.
+	if p.push != nil {
+		if p.registerFn != nil {
+			bootstrap.Mailbox = p.registerFn(statuses)
+		}
+		bootstrap.Push = p.push
+	}
+
+	return bootstrap, nil
+}
+
+func (p *MachinePanelControlPlane) Poll(ctx context.Context) (Snapshot, error) {
+	configSnapshot, err := p.client.GetConfig()
+	if err != nil {
+		return Snapshot{}, fmt.Errorf("machine poll config: %w", err)
+	}
+	users, err := p.client.GetUsers()
+	if err != nil {
+		return Snapshot{}, fmt.Errorf("machine poll users: %w", err)
+	}
+	select {
+	case <-ctx.Done():
+		return Snapshot{}, ctx.Err()
+	default:
+	}
+	nodeSpec, err := model.NodeSpecFromPanelValidated(configSnapshot, p.kcfg)
+	if err != nil {
+		return Snapshot{}, fmt.Errorf("machine poll normalize: %w", err)
+	}
+	return Snapshot{Config: nodeSpec, Users: model.UserSpecsFromPanel(users)}, nil
+}
+
+func (p *MachinePanelControlPlane) Discover(
+	ctx context.Context,
+	metricsFn func() map[string]interface{},
+	events chan<- Event,
+	statuses chan<- StatusChange,
+) (PushClient, error) {
+	return nil, nil
+}
+
+func (p *MachinePanelControlPlane) Report(payload ReportPayload) error {
+	return p.client.Report(
+		payload.Traffic, payload.Alive, payload.Online,
+		payload.CPU, payload.Mem, payload.Swap, payload.Disk,
+		payload.Metrics,
+	)
+}
+
+func (p *MachinePanelControlPlane) ReportDevices(push PushClient, devices map[int][]string) {
+	if push != nil {
+		push.SendDeviceReport(devices)
+	}
+}
+
+func (p *MachinePanelControlPlane) Metrics() APIMetrics {
+	api := p.client.SnapshotMetrics()
+	return APIMetrics{Success: api.Success, Failure: api.Failure}
+}

internal/controlplane/mailbox.go

@@ -0,0 +1,261 @@
+package controlplane
+
+import (
+	"sync"
+
+	"github.com/cedar2025/xboard-node/internal/model"
+)
+
+// MailboxState is the coalesced snapshot drained from a NodeMailbox.
+type MailboxState struct {
+	Config         *model.NodeSpec
+	Users          []model.UserSpec
+	DeviceUsers    map[int][]string
+	HasConfig      bool
+	HasUsers       bool
+	HasDevices     bool
+	NeedsReconcile bool
+}
+
+// NodeMailbox buffers WS events for a machine-mode node service that is not
+// yet ready to process them. Events are coalesced (latest-wins) and drained
+// as a single snapshot once the service marks the mailbox as ready.
+type NodeMailbox struct {
+	mu               sync.Mutex
+	ready            bool
+	config           *model.NodeSpec
+	users            []model.UserSpec
+	deviceUsers      map[int][]string
+	dirtyConfig      bool
+	dirtyUsers       bool
+	dirtyDevices     bool
+	needsReconcile   bool
+	hasFullUserState bool
+	notifyCh         chan struct{}
+}
+
+func NewNodeMailbox() *NodeMailbox {
+	return &NodeMailbox{notifyCh: make(chan struct{}, 1)}
+}
+
+func (m *NodeMailbox) MarkReady() {
+	m.mu.Lock()
+	m.ready = true
+	m.mu.Unlock()
+	m.notify()
+}
+
+// SeedBaseline initializes the mailbox with the REST-derived baseline so that
+// subsequent delta events can be applied incrementally instead of triggering
+// a full REST reconciliation.
+func (m *NodeMailbox) SeedBaseline(users []model.UserSpec, config *model.NodeSpec) {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	if users != nil && !m.hasFullUserState {
+		m.users = cloneUsers(users)
+		m.hasFullUserState = true
+	}
+	if config != nil && m.config == nil {
+		m.config = cloneNodeSpec(config)
+	}
+}
+
+func (m *NodeMailbox) Apply(event Event) {
+	m.mu.Lock()
+	changed := false
+
+	switch event.Type {
+	case EventSyncConfig:
+		if event.Config != nil {
+			m.config = cloneNodeSpec(event.Config)
+			m.dirtyConfig = true
+			changed = true
+		}
+	case EventSyncUsers:
+		if event.Users != nil {
+			m.users = cloneUsers(event.Users)
+			m.dirtyUsers = true
+			m.hasFullUserState = true
+			changed = true
+		}
+	case EventSyncDevices:
+		if event.DeviceUsers != nil {
+			m.deviceUsers = cloneDeviceUsers(event.DeviceUsers)
+			m.dirtyDevices = true
+			changed = true
+		}
+	case EventSyncUserDelta:
+		if len(event.DeltaUsers) > 0 {
+			if !m.hasFullUserState {
+				m.needsReconcile = true
+				changed = true
+			} else if !applyUserDelta(&m.users, event.DeltaAction, event.DeltaUsers) {
+				m.needsReconcile = true
+				changed = true
+			} else {
+				m.dirtyUsers = true
+				changed = true
+			}
+		}
+	}
+	m.mu.Unlock()
+
+	if changed {
+		m.notify()
+	}
+}
+
+func (m *NodeMailbox) notify() {
+	select {
+	case m.notifyCh <- struct{}{}:
+	default:
+	}
+}
+
+func (m *NodeMailbox) NotifyCh() <-chan struct{} { return m.notifyCh }
+
+func (m *NodeMailbox) DrainIfReady() MailboxState {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	if !m.ready {
+		return MailboxState{}
+	}
+	state := MailboxState{NeedsReconcile: m.needsReconcile}
+	if m.dirtyConfig && m.config != nil {
+		state.Config = cloneNodeSpec(m.config)
+		state.HasConfig = true
+		m.dirtyConfig = false
+	}
+	if m.dirtyUsers {
+		state.Users = cloneUsers(m.users)
+		state.HasUsers = true
+		m.dirtyUsers = false
+	}
+	if m.dirtyDevices && m.deviceUsers != nil {
+		state.DeviceUsers = cloneDeviceUsers(m.deviceUsers)
+		state.HasDevices = true
+		m.dirtyDevices = false
+	}
+	m.needsReconcile = false
+	return state
+}
+
+func cloneNodeSpec(spec *model.NodeSpec) *model.NodeSpec {
+	if spec == nil {
+		return nil
+	}
+	clone := *spec
+	if spec.NetworkSettings != nil {
+		clone.NetworkSettings = make(map[string]interface{}, len(spec.NetworkSettings))
+		for k, v := range spec.NetworkSettings {
+			clone.NetworkSettings[k] = v
+		}
+	}
+	if spec.TLSSettings != nil {
+		clone.TLSSettings = make(map[string]interface{}, len(spec.TLSSettings))
+		for k, v := range spec.TLSSettings {
+			clone.TLSSettings[k] = v
+		}
+	}
+	if spec.Routes != nil {
+		clone.Routes = append([]model.RouteRule(nil), spec.Routes...)
+	}
+	if spec.CustomOutbounds != nil {
+		clone.CustomOutbounds = append([]model.OutboundConfig(nil), spec.CustomOutbounds...)
+	}
+	if spec.CustomRoutes != nil {
+		clone.CustomRoutes = make([]map[string]any, len(spec.CustomRoutes))
+		for i, route := range spec.CustomRoutes {
+			if route == nil {
+				continue
+			}
+			copied := make(map[string]any, len(route))
+			for k, v := range route {
+				copied[k] = v
+			}
+			clone.CustomRoutes[i] = copied
+		}
+	}
+	if spec.CustomRouteRules != nil {
+		clone.CustomRouteRules = append([]model.CustomRouteRule(nil), spec.CustomRouteRules...)
+	}
+	if spec.CertConfig != nil {
+		certCopy := *spec.CertConfig
+		if spec.CertConfig.DNSEnv != nil {
+			certCopy.DNSEnv = make(map[string]string, len(spec.CertConfig.DNSEnv))
+			for k, v := range spec.CertConfig.DNSEnv {
+				certCopy.DNSEnv[k] = v
+			}
+		}
+		clone.CertConfig = &certCopy
+	}
+	if spec.Multiplex != nil {
+		muxCopy := *spec.Multiplex
+		if spec.Multiplex.Brutal != nil {
+			brutalCopy := *spec.Multiplex.Brutal
+			muxCopy.Brutal = &brutalCopy
+		}
+		clone.Multiplex = &muxCopy
+	}
+	return &clone
+}
+
+func cloneUsers(users []model.UserSpec) []model.UserSpec {
+	if users == nil {
+		return nil
+	}
+	cloned := make([]model.UserSpec, len(users))
+	copy(cloned, users)
+	return cloned
+}
+
+func cloneDeviceUsers(devices map[int][]string) map[int][]string {
+	if devices == nil {
+		return nil
+	}
+	cloned := make(map[int][]string, len(devices))
+	for uid, ips := range devices {
+		dup := make([]string, len(ips))
+		copy(dup, ips)
+		cloned[uid] = dup
+	}
+	return cloned
+}
+
+func applyUserDelta(users *[]model.UserSpec, action string, delta []model.UserSpec) bool {
+	current := cloneUsers(*users)
+	index := make(map[int]int, len(current))
+	for i, user := range current {
+		index[user.ID] = i
+	}
+
+	switch action {
+	case "add":
+		for _, user := range delta {
+			if idx, ok := index[user.ID]; ok {
+				current[idx] = user
+				continue
+			}
+			index[user.ID] = len(current)
+			current = append(current, user)
+		}
+	case "remove":
+		removeIDs := make(map[int]struct{}, len(delta))
+		for _, user := range delta {
+			removeIDs[user.ID] = struct{}{}
+		}
+		filtered := current[:0]
+		for _, user := range current {
+			if _, ok := removeIDs[user.ID]; ok {
+				continue
+			}
+			filtered = append(filtered, user)
+		}
+		current = append([]model.UserSpec(nil), filtered...)
+	default:
+		return false
+	}
+
+	*users = current
+	return true
+}

internal/controlplane/mailbox_test.go

@@ -0,0 +1,164 @@
+package controlplane
+
+import (
+	"testing"
+
+	"github.com/cedar2025/xboard-node/internal/model"
+)
+
+func TestNodeMailboxLatestStateAndNotifyCollapse(t *testing.T) {
+	mb := NewNodeMailbox()
+	cfg1 := &model.NodeSpec{Protocol: "vmess", ServerPort: 10010}
+	cfg2 := &model.NodeSpec{Protocol: "vless", ServerPort: 10020}
+
+	mb.Apply(Event{Type: EventSyncConfig, Config: cfg1})
+	mb.Apply(Event{Type: EventSyncConfig, Config: cfg2})
+
+	select {
+	case <-mb.NotifyCh():
+	default:
+		t.Fatal("expected collapsed notify signal")
+	}
+
+	mb.MarkReady()
+	state := mb.DrainIfReady()
+	if !state.HasConfig || state.Config == nil {
+		t.Fatal("expected config state")
+	}
+	if state.Config.Protocol != "vless" || state.Config.ServerPort != 10020 {
+		t.Fatalf("unexpected config: %+v", state.Config)
+	}
+}
+
+func TestNodeMailboxFullUsersOverwriteAndDeltaBeforeBaseline(t *testing.T) {
+	mb := NewNodeMailbox()
+	mb.Apply(Event{
+		Type:        EventSyncUserDelta,
+		DeltaAction: "add",
+		DeltaUsers:  []model.UserSpec{{ID: 1, UUID: "u1"}},
+	})
+	mb.MarkReady()
+	state := mb.DrainIfReady()
+	if !state.NeedsReconcile {
+		t.Fatal("expected reconcile when delta arrives before baseline")
+	}
+
+	mb.Apply(Event{Type: EventSyncUsers, Users: []model.UserSpec{{ID: 1, UUID: "u1"}}})
+	mb.Apply(Event{Type: EventSyncUsers, Users: []model.UserSpec{{ID: 2, UUID: "u2"}}})
+	state = mb.DrainIfReady()
+	if !state.HasUsers {
+		t.Fatal("expected users snapshot")
+	}
+	if len(state.Users) != 1 || state.Users[0].ID != 2 {
+		t.Fatalf("unexpected users: %+v", state.Users)
+	}
+}
+
+func TestNodeMailboxDevicesLatestWins(t *testing.T) {
+	mb := NewNodeMailbox()
+	mb.Apply(Event{Type: EventSyncDevices, DeviceUsers: map[int][]string{1: {"1.1.1.1"}}})
+	mb.Apply(Event{Type: EventSyncDevices, DeviceUsers: map[int][]string{1: {"2.2.2.2"}}})
+	mb.MarkReady()
+	state := mb.DrainIfReady()
+	if !state.HasDevices {
+		t.Fatal("expected devices state")
+	}
+	if got := state.DeviceUsers[1][0]; got != "2.2.2.2" {
+		t.Fatalf("unexpected device state: %v", state.DeviceUsers)
+	}
+}
+
+func TestNodeMailboxReconcileNotifies(t *testing.T) {
+	mb := NewNodeMailbox()
+	mb.MarkReady()
+
+	// Drain the MarkReady notification
+	select {
+	case <-mb.NotifyCh():
+	default:
+	}
+
+	// Delta without baseline should trigger reconcile AND notify
+	mb.Apply(Event{
+		Type:        EventSyncUserDelta,
+		DeltaAction: "add",
+		DeltaUsers:  []model.UserSpec{{ID: 1, UUID: "u1"}},
+	})
+
+	select {
+	case <-mb.NotifyCh():
+	default:
+		t.Fatal("expected notify on reconcile path")
+	}
+
+	state := mb.DrainIfReady()
+	if !state.NeedsReconcile {
+		t.Fatal("expected NeedsReconcile")
+	}
+}
+
+func TestNodeMailboxSeedBaselineEnablesDelta(t *testing.T) {
+	mb := NewNodeMailbox()
+	mb.SeedBaseline([]model.UserSpec{{ID: 1, UUID: "u1"}, {ID: 2, UUID: "u2"}}, nil)
+	mb.MarkReady()
+
+	// Drain MarkReady notify
+	select {
+	case <-mb.NotifyCh():
+	default:
+	}
+
+	// Delta should now apply incrementally, not trigger reconcile
+	mb.Apply(Event{
+		Type:        EventSyncUserDelta,
+		DeltaAction: "add",
+		DeltaUsers:  []model.UserSpec{{ID: 3, UUID: "u3"}},
+	})
+
+	select {
+	case <-mb.NotifyCh():
+	default:
+		t.Fatal("expected notify after delta")
+	}
+
+	state := mb.DrainIfReady()
+	if state.NeedsReconcile {
+		t.Fatal("should not need reconcile after seeded baseline")
+	}
+	if !state.HasUsers {
+		t.Fatal("expected users")
+	}
+	if len(state.Users) != 3 {
+		t.Fatalf("expected 3 users, got %d: %+v", len(state.Users), state.Users)
+	}
+}
+
+func TestNodeMailboxDeltaFailedApplyNotifies(t *testing.T) {
+	mb := NewNodeMailbox()
+	mb.SeedBaseline([]model.UserSpec{{ID: 1, UUID: "u1"}}, nil)
+	mb.MarkReady()
+
+	// Drain MarkReady notify
+	select {
+	case <-mb.NotifyCh():
+	default:
+	}
+
+	// Invalid action should trigger reconcile and still notify
+	mb.Apply(Event{
+		Type:        EventSyncUserDelta,
+		DeltaAction: "invalid_action",
+		DeltaUsers:  []model.UserSpec{{ID: 2, UUID: "u2"}},
+	})
+
+	select {
+	case <-mb.NotifyCh():
+	default:
+		t.Fatal("expected notify on failed delta apply")
+	}
+
+	state := mb.DrainIfReady()
+	if !state.NeedsReconcile {
+		t.Fatal("expected NeedsReconcile for invalid delta action")
+	}
+}

internal/controlplane/panel.go

@@ -0,0 +1,184 @@
+package controlplane
+
+import (
+	"context"
+	"fmt"
+	"time"
+
+	"github.com/cedar2025/xboard-node/internal/config"
+	"github.com/cedar2025/xboard-node/internal/model"
+	"github.com/cedar2025/xboard-node/internal/nlog"
+	"github.com/cedar2025/xboard-node/internal/panel"
+)
+
+type PanelControlPlane struct {
+	cfg    config.PanelConfig
+	wsCfg  config.WSConfig
+	kcfg   config.KernelConfig
+	client *panel.Client
+}
+
+type panelPushClient struct {
+	inner *panel.WSClient
+}
+
+func NewPanelControlPlane(panelCfg config.PanelConfig, wsCfg config.WSConfig, kcfg config.KernelConfig) *PanelControlPlane {
+	return &PanelControlPlane{
+		cfg:    panelCfg,
+		wsCfg:  wsCfg,
+		kcfg:   kcfg,
+		client: panel.NewClient(panelCfg),
+	}
+}
+
+func (p *PanelControlPlane) SupportsPolling() bool       { return true }
+func (p *PanelControlPlane) SupportsDiscovery() bool     { return true }
+func (p *PanelControlPlane) SupportsReporting() bool     { return true }
+func (p *PanelControlPlane) SupportsDeviceReports() bool { return true }
+
+func (p *PanelControlPlane) Initial(ctx context.Context, metricsFn func() map[string]interface{}, events chan<- Event, statuses chan<- StatusChange) (Bootstrap, error) {
+	hs, err := p.client.Handshake()
+	if err != nil {
+		return Bootstrap{}, fmt.Errorf("handshake: %w", err)
+	}
+
+	bootstrap := Bootstrap{PushInterval: hs.Settings.PushInterval, PullInterval: hs.Settings.PullInterval}
+	if hs.WebSocket.Enabled && hs.WebSocket.WSURL != "" {
+		bootstrap.Push = p.newPushClient(metricsFn, events, statuses, hs.WebSocket.WSURL)
+		return bootstrap, nil
+	}
+
+	nlog.Core().Info("websocket disabled, using REST API")
+	configSnapshot, err := p.client.GetConfig()
+	if err != nil {
+		return Bootstrap{}, fmt.Errorf("initial config fetch: %w", err)
+	}
+	if configSnapshot == nil {
+		return Bootstrap{}, fmt.Errorf("initial config is nil")
+	}
+	users, err := p.client.GetUsers()
+	if err != nil {
+		return Bootstrap{}, fmt.Errorf("initial user fetch: %w", err)
+	}
+	bootstrap.Config, err = model.NodeSpecFromPanelValidated(configSnapshot, p.kcfg)
+	if err != nil {
+		return Bootstrap{}, fmt.Errorf("initial config normalize: %w", err)
+	}
+	bootstrap.Users = model.UserSpecsFromPanel(users)
+	return bootstrap, nil
+}
+
+func (p *PanelControlPlane) Poll(ctx context.Context) (Snapshot, error) {
+	configSnapshot, err := p.client.GetConfig()
+	if err != nil {
+		return Snapshot{}, fmt.Errorf("poll config: %w", err)
+	}
+	users, err := p.client.GetUsers()
+	if err != nil {
+		return Snapshot{}, fmt.Errorf("poll users: %w", err)
+	}
+	select {
+	case <-ctx.Done():
+		return Snapshot{}, ctx.Err()
+	default:
+	}
+	nodeSpec, err := model.NodeSpecFromPanelValidated(configSnapshot, p.kcfg)
+	if err != nil {
+		return Snapshot{}, fmt.Errorf("poll config normalize: %w", err)
+	}
+	return Snapshot{Config: nodeSpec, Users: model.UserSpecsFromPanel(users)}, nil
+}
+
+func (p *PanelControlPlane) Discover(ctx context.Context, metricsFn func() map[string]interface{}, events chan<- Event, statuses chan<- StatusChange) (PushClient, error) {
+	hs, err := p.client.Handshake()
+	if err != nil {
+		return nil, fmt.Errorf("handshake: %w", err)
+	}
+	select {
+	case <-ctx.Done():
+		return nil, ctx.Err()
+	default:
+	}
+	if hs.WebSocket.Enabled && hs.WebSocket.WSURL != "" {
+		return p.newPushClient(metricsFn, events, statuses, hs.WebSocket.WSURL), nil
+	}
+	return nil, nil
+}
+
+func (p *PanelControlPlane) Report(payload ReportPayload) error {
+	return p.client.Report(payload.Traffic, payload.Alive, payload.Online, payload.CPU, payload.Mem, payload.Swap, payload.Disk, payload.Metrics)
+}
+
+func (p *PanelControlPlane) ReportDevices(push PushClient, devices map[int][]string) {
+	if push != nil {
+		push.SendDeviceReport(devices)
+	}
+}
+
+func (p *PanelControlPlane) Metrics() APIMetrics {
+	api := p.client.SnapshotMetrics()
+	return APIMetrics{Success: api.Success, Failure: api.Failure}
+}
+
+func (p *PanelControlPlane) newPushClient(metricsFn func() map[string]interface{}, events chan<- Event, statuses chan<- StatusChange, wsURL string) PushClient {
+	cfg := panel.WSClientConfig{
+		StatusInterval:   time.Duration(p.wsCfg.StatusInterval) * time.Second,
+		HandshakeTimeout: time.Duration(p.wsCfg.HandshakeTimeout) * time.Second,
+		BackoffInitial:   time.Duration(p.wsCfg.BackoffInitial) * time.Second,
+		BackoffMax:       time.Duration(p.wsCfg.BackoffMax) * time.Second,
+	}
+	inner := panel.NewWSClient(
+		wsURL,
+		p.cfg.Token,
+		p.cfg.NodeID,
+		cfg,
+		func(event panel.WSEvent) {
+			translated, err := TranslateWSEvent(event, p.kcfg)
+			if err != nil {
+				nlog.Core().Warn("invalid ws event config, dropping event", "type", event.Type, "error", err)
+				return
+			}
+			select {
+			case events <- translated:
+			default:
+				nlog.Core().Warn("ws event channel full, dropping event", "type", translated.Type)
+				select {
+				case statuses <- StatusChange{Connected: true, NeedsResync: true}:
+				default:
+				}
+			}
+		},
+		func(status panel.WSStatusChange) {
+			select {
+			case statuses <- StatusChange{Connected: status.Connected}:
+			default:
+			}
+		},
+		metricsFn,
+	)
+	return &panelPushClient{inner: inner}
+}
+
+func TranslateWSEvent(event panel.WSEvent, kcfg config.KernelConfig) (Event, error) {
+	translated := Event{Type: EventType(event.Type), DeltaAction: event.DeltaAction, DeviceUsers: event.DeviceUsers}
+	if event.Config != nil {
+		var err error
+		translated.Config, err = model.NodeSpecFromPanelValidated(event.Config, kcfg)
+		if err != nil {
+			return Event{}, fmt.Errorf("translate node config: %w", err)
+		}
+	}
+	if event.Users != nil {
+		translated.Users = model.UserSpecsFromPanel(event.Users)
+	}
+	if event.DeltaUsers != nil {
+		translated.DeltaUsers = model.UserSpecsFromPanel(event.DeltaUsers)
+	}
+	return translated, nil
+}
+
+func (p *panelPushClient) Run(ctx context.Context) { p.inner.Run(ctx) }
+func (p *panelPushClient) IsConnected() bool       { return p.inner.IsConnected() }
+func (p *panelPushClient) SendDeviceReport(devices map[int][]string) {
+	p.inner.SendDeviceReport(devices)
+}

internal/controlplane/panel_test.go

@@ -0,0 +1,95 @@
+package controlplane
+
+import (
+	"context"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"github.com/cedar2025/xboard-node/internal/config"
+	panelapi "github.com/cedar2025/xboard-node/internal/panel"
+)
+
+func TestPanelControlPlaneInitialRejectsInvalidCustomOutbounds(t *testing.T) {
+	server := newPanelTestServer(`{"protocol":"shadowsocks","server_port":8388,"custom_outbounds":[{"tag":"proxy","protocol":"socks","proxy_tag":"missing","settings":{"server":"2.2.2.2","server_port":1080}}]}`)
+	defer server.Close()
+
+	cp := NewPanelControlPlane(config.PanelConfig{URL: server.URL, Token: "token", NodeID: 1}, config.WSConfig{}, config.KernelConfig{Type: "singbox"})
+	_, err := cp.Initial(context.Background(), nil, nil, nil)
+	if err == nil {
+		t.Fatal("expected error, got nil")
+	}
+	if !strings.Contains(err.Error(), `initial config normalize: validate custom outbounds: custom_outbounds[0].proxy_tag references unknown outbound "missing"`) {
+		t.Fatalf("unexpected error: %v", err)
+	}
+}
+
+func TestPanelControlPlanePollRejectsInvalidCustomOutbounds(t *testing.T) {
+	server := newPanelTestServer(`{"protocol":"shadowsocks","server_port":8388,"custom_outbounds":[{"tag":"proxy","protocol":"socks","proxy_tag":"missing","settings":{"server":"2.2.2.2","server_port":1080}}]}`)
+	defer server.Close()
+
+	cp := NewPanelControlPlane(config.PanelConfig{URL: server.URL, Token: "token", NodeID: 1}, config.WSConfig{}, config.KernelConfig{Type: "singbox"})
+	_, err := cp.Poll(context.Background())
+	if err == nil {
+		t.Fatal("expected error, got nil")
+	}
+	if !strings.Contains(err.Error(), `poll config normalize: validate custom outbounds: custom_outbounds[0].proxy_tag references unknown outbound "missing"`) {
+		t.Fatalf("unexpected error: %v", err)
+	}
+}
+
+func TestTranslateWSEventRejectsInvalidCustomOutbounds(t *testing.T) {
+	_, err := TranslateWSEvent(panelapi.WSEvent{
+		Type: panelapi.WSEventSyncConfig,
+		Config: &panelapi.NodeConfig{
+			Protocol:   "shadowsocks",
+			ServerPort: 8388,
+			CustomOutbounds: []panelapi.OutboundConfig{
+				{Tag: "proxy", Protocol: "socks", ProxyTag: "missing", Settings: map[string]any{"server": "2.2.2.2", "server_port": 1080}},
+			},
+		},
+	}, config.KernelConfig{Type: "singbox"})
+	if err == nil {
+		t.Fatal("expected error, got nil")
+	}
+	if !strings.Contains(err.Error(), `translate node config: validate custom outbounds: custom_outbounds[0].proxy_tag references unknown outbound "missing"`) {
+		t.Fatalf("unexpected error: %v", err)
+	}
+}
+
+func newPanelTestServer(configBody string) *httptest.Server {
+	mux := http.NewServeMux()
+	mux.HandleFunc("/api/v2/server/handshake", func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		_, _ = w.Write([]byte(`{"websocket":{"enabled":false},"settings":{"push_interval":60,"pull_interval":60}}`))
+	})
+	mux.HandleFunc("/api/v1/server/UniProxy/config", func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		_, _ = w.Write([]byte(configBody))
+	})
+	mux.HandleFunc("/api/v1/server/UniProxy/user", func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		_, _ = w.Write([]byte(`{"users":[{"id":1,"uuid":"11111111-1111-1111-1111-111111111111"}]}`))
+	})
+	return httptest.NewServer(mux)
+}
+
+func TestTranslateWSEventRejectsUnsupportedProtocolForKernel(t *testing.T) {
+	_, err := TranslateWSEvent(panelapi.WSEvent{
+		Type: panelapi.WSEventSyncConfig,
+		Config: &panelapi.NodeConfig{
+			Protocol:   "shadowsocks",
+			ServerPort: 8388,
+			CustomOutbounds: []panelapi.OutboundConfig{
+				{Tag: "hy2", Protocol: "hysteria2", Settings: map[string]any{"server": "2.2.2.2", "server_port": 8443}},
+			},
+		},
+	}, config.KernelConfig{Type: "xray"})
+	if err == nil {
+		t.Fatal("expected error, got nil")
+	}
+	if !strings.Contains(err.Error(), `translate node config: validate custom outbounds: custom_outbounds[0].protocol "hysteria2" is not supported by kernel "xray"`) {
+		t.Fatalf("unexpected error: %v", err)
+	}
+}

internal/controlplane/types.go

@@ -0,0 +1,87 @@
+package controlplane
+
+import (
+	"context"
+
+	"github.com/cedar2025/xboard-node/internal/model"
+)
+
+type EventType string
+
+const (
+	EventSyncConfig    EventType = "sync.config"
+	EventSyncUsers     EventType = "sync.users"
+	EventSyncUserDelta EventType = "sync.user.delta"
+	EventSyncDevices   EventType = "sync.devices"
+)
+
+type Event struct {
+	Type        EventType
+	Config      *model.NodeSpec
+	Users       []model.UserSpec
+	DeltaAction string
+	DeltaUsers  []model.UserSpec
+	DeviceUsers map[int][]string
+}
+
+type StatusChange struct {
+	Connected   bool
+	NeedsResync bool
+}
+
+type APIMetrics struct {
+	Success uint64
+	Failure uint64
+}
+
+type Bootstrap struct {
+	PushInterval int
+	PullInterval int
+	Push         PushClient
+	Config       *model.NodeSpec
+	Users        []model.UserSpec
+	Mailbox      *NodeMailbox
+}
+
+type Snapshot struct {
+	Config *model.NodeSpec
+	Users  []model.UserSpec
+}
+
+type ReportPayload struct {
+	Traffic map[int][2]int64
+	Alive   map[int][]string
+	Online  map[int]int
+	CPU     float64
+	Mem     [2]uint64
+	Swap    [2]uint64
+	Disk    [2]uint64
+	Metrics map[string]interface{}
+}
+
+type PushClient interface {
+	Run(ctx context.Context)
+	IsConnected() bool
+	SendDeviceReport(devices map[int][]string)
+}
+
+type Source interface {
+	Initial(ctx context.Context, metricsFn func() map[string]interface{}, events chan<- Event, statuses chan<- StatusChange) (Bootstrap, error)
+	Poll(ctx context.Context) (Snapshot, error)
+	Discover(ctx context.Context, metricsFn func() map[string]interface{}, events chan<- Event, statuses chan<- StatusChange) (PushClient, error)
+	Metrics() APIMetrics
+	SupportsPolling() bool
+	SupportsDiscovery() bool
+}
+
+type Sink interface {
+	Report(payload ReportPayload) error
+	ReportDevices(push PushClient, devices map[int][]string)
+	SupportsReporting() bool
+	SupportsDeviceReports() bool
+}
+
+type ControlPlane interface {
+	Source
+	Sink
+}

internal/kernel/custom_route_compile.go

@@ -0,0 +1,39 @@
+package kernel
+
+import (
+	"strings"
+
+	"github.com/cedar2025/xboard-node/internal/model"
+)
+
+func NeedsGeoIPRules(rules []model.CustomRouteRule) bool {
+	for _, r := range rules {
+		for _, v := range r.Match.IPCIDRs {
+			if strings.HasPrefix(v, "geoip:") {
+				return true
+			}
+		}
+		for _, v := range r.Match.SourceCIDRs {
+			if strings.HasPrefix(v, "geoip:") {
+				return true
+			}
+		}
+	}
+	return false
+}
+
+func NeedsGeoSiteRules(rules []model.CustomRouteRule) bool {
+	for _, r := range rules {
+		for _, v := range r.Match.Domains {
+			if strings.HasPrefix(v, "geosite:") {
+				return true
+			}
+		}
+		for _, v := range r.Match.DomainSuffixes {
+			if strings.HasPrefix(v, "geosite:") {
+				return true
+			}
+		}
+	}
+	return false
+}

internal/kernel/customcfg.go

@@ -3,10 +3,10 @@ package kernel
 import (
 	"encoding/json"
 	"fmt"
-	"log/slog"
 	"os"
 	"strings"
 
+	"github.com/cedar2025/xboard-node/internal/nlog"
 	"gopkg.in/yaml.v3"
 )
 
@@ -20,7 +20,7 @@ func LoadCustomConfig(path string) (map[string]any, error) {
 	data, err := os.ReadFile(path)
 	if err != nil {
 		if os.IsNotExist(err) {
-			slog.Warn("custom config file not found, skipping", "path", path)
+			nlog.Core().Warn("custom config file not found, skipping", "path", path)
 			return nil, nil
 		}
 		return nil, fmt.Errorf("read custom config %s: %w", path, err)
@@ -43,7 +43,7 @@ func LoadCustomConfig(path string) (map[string]any, error) {
 		}
 	}
 
-	slog.Info("loaded custom config", "path", path, "keys", mapKeys(result))
+	nlog.Core().Info("loaded custom config", "path", path, "keys", mapKeys(result))
 	return result, nil
 }
 

internal/kernel/geo.go

@@ -3,11 +3,11 @@ package kernel
 import (
 	"strings"
 
-	"github.com/cedar2025/xboard-node/internal/panel"
+	"github.com/cedar2025/xboard-node/internal/model"
 )
 
 // NeedsGeoIP returns true when any panel route rule contains a "geoip:" match entry.
-func NeedsGeoIP(routes []panel.RouteRule) bool {
+func NeedsGeoIP(routes []model.RouteRule) bool {
 	for _, r := range routes {
 		for _, m := range r.Match {
 			if strings.HasPrefix(m, "geoip:") {
@@ -19,7 +19,7 @@ func NeedsGeoIP(routes []panel.RouteRule) bool {
 }
 
 // NeedsGeoSite returns true when any panel route rule contains a "geosite:" match entry.
-func NeedsGeoSite(routes []panel.RouteRule) bool {
+func NeedsGeoSite(routes []model.RouteRule) bool {
 	for _, r := range routes {
 		for _, m := range r.Match {
 			if strings.HasPrefix(m, "geosite:") {

internal/kernel/geodata/geodata.go

@@ -3,11 +3,12 @@ package geodata
 import (
 	"fmt"
 	"io"
-	"log/slog"
 	"net/http"
 	"os"
 	"path/filepath"
 	"time"
+
+	"github.com/cedar2025/xboard-node/internal/nlog"
 )
 
 const (
@@ -49,7 +50,7 @@ func Ensure(dir string, needGeoIP, needGeoSite bool, kernelType string) error {
 	var firstErr error
 	for _, f := range files {
 		if err := ensureFile(dir, f.name, f.url); err != nil {
-			slog.Warn("geo database auto-download failed", "file", f.name, "error", err)
+			nlog.Core().Warn("geo database auto-download failed", "file", f.name, "error", err)
 			if firstErr == nil {
 				firstErr = err
 			}
@@ -63,12 +64,12 @@ func ensureFile(dir, name, url string) error {
 	if _, err := os.Stat(dst); err == nil {
 		return nil
 	}
-	slog.Info("geo database missing, downloading automatically", "file", name, "url", url)
+	nlog.Core().Info("geo database missing, downloading automatically", "file", name, "url", url)
 	if err := atomicDownload(dst, url); err != nil {
 		return err
 	}
 	if fi, err := os.Stat(dst); err == nil {
-		slog.Info("geo database ready", "file", name, "size_kb", fi.Size()/1024)
+		nlog.Core().Info("geo database ready", "file", name, "size_kb", fi.Size()/1024)
 	}
 	return nil
 }

internal/kernel/kernel.go

@@ -7,10 +7,32 @@ import (
 	"fmt"
 	"sort"
 
-	"github.com/cedar2025/xboard-node/internal/panel"
+	"github.com/cedar2025/xboard-node/internal/model"
 	"golang.org/x/time/rate"
 )
 
+type Capabilities struct {
+	PerUserSpeedLimit    bool
+	DeviceLimit          bool
+	BuiltInTrafficStats  bool
+	AliveIPTracking      bool
+	ForceCloseConnection bool
+	ForceCloseUser       bool
+}
+
+// TLSCert holds PEM-encoded TLS certificate material.
+// All cert modes (self-signed, file, ACME, panel-content) normalise to PEM
+// so kernels never need to deal with file paths.
+type TLSCert struct {
+	CertPEM []byte
+	KeyPEM  []byte
+}
+
+// HasCert reports whether a valid cert+key pair is available.
+func (t TLSCert) HasCert() bool {
+	return len(t.CertPEM) > 0 && len(t.KeyPEM) > 0
+}
+
 // Kernel is the interface for proxy kernel backends (sing-box, xray, etc.).
 //
 // The interface is split into lifecycle, user management, and observability
@@ -25,12 +47,14 @@ type Kernel interface {
 	Name() string
 	// Protocols returns the protocol names this kernel supports.
 	Protocols() []string
+	// Capabilities returns the explicit runtime semantics supported by the kernel.
+	Capabilities() Capabilities
 
 	// ─── Lifecycle ──────────────────────────────────────────────────────
 	// Start initialises the kernel with the given node config and initial
 	// user set, binds listeners, and begins accepting connections.
 	// Calling Start on an already-running kernel stops the old instance first.
-	Start(nodeConfig *panel.NodeConfig, users []panel.User, certFile, keyFile string) error
+	Start(nodeConfig *model.NodeSpec, users []model.UserSpec, tls TLSCert) error
 	// Stop gracefully shuts down the kernel, draining active connections.
 	Stop()
 	// IsRunning returns whether the kernel is currently accepting connections.
@@ -38,39 +62,54 @@ type Kernel interface {
 	// Reload re-generates the full config and hot-swaps listeners/routes.
 	// Use this when port, protocol, or TLS settings change.
 	// Existing connections MAY be briefly interrupted.
-	Reload(nodeConfig *panel.NodeConfig, users []panel.User, certFile, keyFile string) error
+	Reload(nodeConfig *model.NodeSpec, users []model.UserSpec, tls TLSCert) error
 
 	// ─── User management (non-disruptive) ───────────────────────────────
 	// AddUsers registers new users with the running kernel.
 	// Existing connections are unaffected. Returns the count actually added
 	// (duplicates are silently skipped).
-	AddUsers(users []panel.User) (added int, err error)
+	AddUsers(users []model.UserSpec) (added int, err error)
 	// RemoveUsers deregisters users from the running kernel.
 	// Active connections of removed users are terminated. Returns the count
 	// actually removed (unknown IDs are silently skipped).
-	RemoveUsers(users []panel.User) (removed int, err error)
+	RemoveUsers(users []model.UserSpec) (removed int, err error)
 	// UpdateUsers replaces the entire user set atomically.
 	// This is equivalent to computing the diff and calling Add/Remove, but
 	// may be more efficient for bulk changes. Returns (added, removed) counts.
-	UpdateUsers(users []panel.User) (added, removed int, err error)
+	UpdateUsers(users []model.UserSpec) (added, removed int, err error)
 
 	// ─── Observability ──────────────────────────────────────────────────
-	// GetConnections returns a snapshot of all active proxy connections.
-	GetConnections(ctx context.Context) ([]Connection, error)
+	// GetUserTraffic returns per-user cumulative traffic counters and
+	// per-user alive IP sets. The kernel maintains these internally using
+	// per-user atomic counters — no per-connection iteration needed.
+	// aliveIPs maps userID → set of source IPs currently connected.
+	// traffic maps userID → [upload, download] cumulative bytes.
+	// connCount is the total number of active connections (for metrics).
+	GetUserTraffic(ctx context.Context) (traffic map[int][2]int64, aliveIPs map[int]map[string]bool, connCount int, err error)
 	// CloseConnection terminates a specific connection by ID.
 	CloseConnection(ctx context.Context, connID string) error
+	// CloseUserConnections terminates all connections for the given user UUID.
+	CloseUserConnections(ctx context.Context, uuid string) error
 	// SetSpeedLimitFunc configures per-user bandwidth throttling.
 	// The function resolves a user UUID to a *rate.Limiter (nil = unlimited).
 	SetSpeedLimitFunc(fn func(uuid string) *rate.Limiter)
+	// SetDeviceLimitFunc configures per-user device limit gate-keeping.
+	// The function resolves a user UUID to (limit, hasLimit).
+	// Kernels that already gate-keep internally (e.g. xray) may no-op.
+	SetDeviceLimitFunc(fn func(uuid string) (int, bool))
+	// UpdateGlobalDevices updates the global device state from panel (for multi-node).
+	UpdateGlobalDevices(users map[int][]string)
+	// ClearGlobalDevices clears the global device state (on WS disconnect).
+	ClearGlobalDevices()
 }
 
 // ComputeHash returns a hash of config + user identities that would
 // require a kernel restart/reconstruction if changed.
-func ComputeHash(nc *panel.NodeConfig, users []panel.User) string {
+func ComputeHash(nc *model.NodeSpec, users []model.UserSpec) string {
 	h := sha256.New()
 	configData, _ := json.Marshal(nc)
 	h.Write(configData)
-	sorted := make([]panel.User, len(users))
+	sorted := make([]model.UserSpec, len(users))
 	copy(sorted, users)
 	sort.Slice(sorted, func(i, j int) bool { return sorted[i].ID < sorted[j].ID })
 	for _, u := range sorted {
@@ -79,24 +118,15 @@ func ComputeHash(nc *panel.NodeConfig, users []panel.User) string {
 	return fmt.Sprintf("%x", h.Sum(nil))
 }
 
-// Connection represents an active proxy connection reported by the kernel.
-type Connection struct {
-	ID       string
-	UserID   int    // extracted user ID (0 if unknown)
-	Upload   int64  // cumulative upload bytes
-	Download int64  // cumulative download bytes
-	SourceIP string // client source IP (cleaned, no port/brackets)
-}
-
 // UserDiff computes which users to add and which to remove when transitioning
 // from oldUsers to newUsers. This is a pure helper used by callers; kernels
 // may also use it internally.
-func UserDiff(oldUsers, newUsers []panel.User) (toAdd, toRemove []panel.User) {
-	oldMap := make(map[int]panel.User, len(oldUsers))
+func UserDiff(oldUsers, newUsers []model.UserSpec) (toAdd, toRemove []model.UserSpec) {
+	oldMap := make(map[int]model.UserSpec, len(oldUsers))
 	for _, u := range oldUsers {
 		oldMap[u.ID] = u
 	}
-	newMap := make(map[int]panel.User, len(newUsers))
+	newMap := make(map[int]model.UserSpec, len(newUsers))
 	for _, u := range newUsers {
 		newMap[u.ID] = u
 	}

internal/kernel/singbox/config.go

@@ -1,8 +1,8 @@
 package singbox
 
 import (
+	"encoding/base64"
 	"fmt"
-	"log/slog"
 	"net"
 	"path/filepath"
 	"strconv"
@@ -10,14 +10,15 @@ import (
 
 	"github.com/cedar2025/xboard-node/internal/config"
 	"github.com/cedar2025/xboard-node/internal/kernel"
-	"github.com/cedar2025/xboard-node/internal/panel"
+	"github.com/cedar2025/xboard-node/internal/model"
+	"github.com/cedar2025/xboard-node/internal/nlog"
 	"github.com/go-viper/mapstructure/v2"
 )
 
 // M is a shorthand for building JSON-like maps
 type M = map[string]interface{}
 
-func buildConfig(kcfg config.KernelConfig, nc *panel.NodeConfig, users []panel.User, certFile, keyFile string) M {
+func buildConfig(kcfg config.KernelConfig, nc *model.NodeSpec, users []model.UserSpec, tc kernel.TLSCert) M {
 	var outbounds []M
 	tags := make(map[string]bool)
 
@@ -52,18 +53,18 @@ func buildConfig(kcfg config.KernelConfig, nc *panel.NodeConfig, users []panel.U
 		"outbounds": outbounds,
 	}
 
-	inbound := buildInbound(nc, users, certFile, keyFile)
+	inbound := buildInbound(nc, users, tc)
 	if inbound != nil {
 		cfg["inbounds"] = []M{inbound}
 	}
 
 	// Merge panel routes and static config routes
-	cfg["route"] = buildRoutes(nc.Routes, mergeRouteList(nc.CustomRoutes, kcfg.CustomRoute))
+	cfg["route"] = buildRoutes(nc.Routes, nc.CustomRouteRules, mergeRouteList(nc.CustomRoutes, kcfg.CustomRoute))
 
 	// Automatically enable rule_set caching (cache_file) when panel routes
 	// reference geoip:/geosite: entries so that the downloaded .srs rule_set
 	// files survive across process restarts.
-	if kernel.NeedsGeoIP(nc.Routes) || kernel.NeedsGeoSite(nc.Routes) {
+	if kernel.NeedsGeoIP(nc.Routes) || kernel.NeedsGeoSite(nc.Routes) || kernel.NeedsGeoIPRules(nc.CustomRouteRules) || kernel.NeedsGeoSiteRules(nc.CustomRouteRules) {
 		cfg["experimental"] = M{
 			"cache_file": M{
 				"enabled": true,
@@ -79,7 +80,7 @@ func buildConfig(kcfg config.KernelConfig, nc *panel.NodeConfig, users []panel.U
 // outboundConfigToSingbox converts a structured OutboundConfig (from the panel)
 // into a sing-box outbound object. sing-box uses a flat layout where all
 // protocol-specific fields sit at the top level alongside "type" and "tag".
-func outboundConfigToSingbox(oc panel.OutboundConfig) M {
+func outboundConfigToSingbox(oc model.OutboundConfig) M {
 	m := M{
 		"type": oc.Protocol,
 		"tag":  oc.Tag,
@@ -143,100 +144,56 @@ func mergeRouteList(a, b []map[string]any) []map[string]any {
 	return res
 }
 
-func buildRoutes(panelRoutes []panel.RouteRule, custom []map[string]any) M {
+func buildRoutes(panelRoutes []model.RouteRule, customRules []model.CustomRouteRule, custom []map[string]any) M {
 	var rules []M
 
-	// Custom Routes (Panel-pushed or Local) go FIRST to have highest priority
+	// Structured custom routes now take the highest priority for panel-managed overrides.
+	for _, rule := range customRules {
+		if rule.Disabled {
+			continue
+		}
+		rules = append(rules, compileCustomRouteRule(rule)...)
+	}
+
+	// Raw custom routes: detect structured format (match+action) and compile them,
+	// otherwise pass through as raw sing-box rules.
 	for _, cr := range custom {
+		if _, hasMatch := cr["match"]; hasMatch {
+			if parsed, ok := tryParseAsCustomRouteRule(cr); ok {
+				rules = append(rules, compileCustomRouteRule(parsed)...)
+				continue
+			}
+		}
 		rules = append(rules, M(cr))
 	}
 
-	// Standard blocks for private/loopback
-	rules = append(rules, M{
-		"outbound": "block",
-		"ip_cidr": []string{
-			"10.0.0.0/8",
-			"100.64.0.0/10",
-			"127.0.0.0/8",
-			"169.254.0.0/16",
-			"172.16.0.0/12",
-			"192.0.0.0/24",
-			"192.168.0.0/16",
-			"198.18.0.0/15",
-			"fc00::/7",
-			"fe80::/10",
-			"::1/128",
+	// Standard blocks for private IPv4 and IPv6 ranges to prevent SSRF.
+	rules = append(rules,
+		M{
+			"outbound": "block",
+			"ip_cidr": []string{
+				"10.0.0.0/8",
+				"100.64.0.0/10",
+				"127.0.0.0/8",
+				"169.254.0.0/16",
+				"172.16.0.0/12",
+				"192.0.0.0/24",
+				"192.168.0.0/16",
+				"198.18.0.0/15",
+			},
 		},
-	}, M{
-		"outbound": "block",
-		"domain": []string{
-			"geoip:private",
+		M{
+			"outbound": "block",
+			"ip_cidr": []string{
+				"fc00::/7",
+				"fe80::/10",
+				"::1/128",
+			},
 		},
-	})
+	)
 
-	// Panel-defined routes (usually specific blocks/proxies)
 	for _, pr := range panelRoutes {
-		if len(pr.Match) == 0 {
-			continue
-		}
-
-		var domains, cidrs []string
-		for _, m := range pr.Match {
-			m = strings.TrimSpace(m)
-			if m == "" {
-				continue
-			}
-			// Strip leading wildcard for domain matching
-			if strings.HasPrefix(m, "*.") {
-				m = strings.TrimPrefix(m, "*.")
-			}
-			// Check if it's a CIDR block (contains /)
-			if strings.Contains(m, "/") {
-				cidrs = append(cidrs, m)
-			} else {
-				// Otherwise treat as domain
-				domains = append(domains, m)
-			}
-		}
-
-		// Determine outbound tag based on action
-		outbound := "block"
-		if pr.Action == "direct" {
-			outbound = "direct"
-		} else if pr.Action == "dns" {
-			// Special case for sing-box DNS routing:
-			// If action is "dns" and action_value is provided, we use it as the server name.
-			// Otherwise it defaults to "dns-out".
-			server := "dns-out"
-			if pr.ActionValue != "" {
-				server = pr.ActionValue
-			}
-			outbound = server
-		} else if pr.Action == "proxy" {
-			// If panel provides a specific outbound tag in action_value, use it.
-			// This allows routing to WARP_JP etc. via normal routes.
-			if pr.ActionValue != "" {
-				outbound = pr.ActionValue
-			}
-		}
-
-		// Create separate rule for domains
-		if len(domains) > 0 {
-			rule := M{
-				"domain_suffix": domains,
-				"outbound":      outbound,
-			}
-			rules = append(rules, rule)
-		}
-
-		// Create separate rule for CIDRs
-		if len(cidrs) > 0 {
-			rule := M{
-				"ip_cidr":  cidrs,
-				"outbound": outbound,
-			}
-			rules = append(rules, rule)
-		}
+		rules = append(rules, compilePanelRouteRule(pr)...)
 	}
 
 	return M{
@@ -245,10 +202,241 @@ func buildRoutes(panelRoutes []panel.RouteRule, custom []map[string]any) M {
 	}
 }
 
+func compilePanelRouteRule(pr model.RouteRule) []M {
+	if len(pr.Match) == 0 {
+		return nil
+	}
+
+	var domains, cidrs []string
+	for _, item := range pr.Match {
+		item = strings.TrimSpace(item)
+		if item == "" {
+			continue
+		}
+		item = strings.TrimPrefix(item, "*.")
+		if strings.Contains(item, "/") {
+			cidrs = append(cidrs, item)
+			continue
+		}
+		domains = append(domains, item)
+	}
+
+	outbound := "block"
+	switch pr.Action {
+	case "direct":
+		outbound = "direct"
+	case "dns":
+		if pr.ActionValue != "" {
+			outbound = pr.ActionValue
+		} else {
+			outbound = "dns-out"
+		}
+	case "proxy":
+		if pr.ActionValue != "" {
+			outbound = pr.ActionValue
+		}
+	}
+
+	var compiled []M
+	if len(domains) > 0 {
+		compiled = append(compiled, M{
+			"domain_suffix": copyStrings(domains),
+			"outbound":      outbound,
+		})
+	}
+	if len(cidrs) > 0 {
+		compiled = append(compiled, M{
+			"ip_cidr":  copyStrings(cidrs),
+			"outbound": outbound,
+		})
+	}
+	return compiled
+}
+
+func compileCustomRouteRule(rule model.CustomRouteRule) []M {
+	if rule.Disabled {
+		return nil
+	}
+
+	outbound := singboxOutboundForAction(rule.Action)
+	var compiled []M
+
+	if len(rule.Match.Domains) > 0 {
+		compiled = append(compiled, M{
+			"domain":   copyStrings(rule.Match.Domains),
+			"outbound": outbound,
+		})
+	}
+	if len(rule.Match.DomainSuffixes) > 0 {
+		compiled = append(compiled, M{
+			"domain_suffix": copyStrings(rule.Match.DomainSuffixes),
+			"outbound":      outbound,
+		})
+	}
+	if len(rule.Match.IPCIDRs) > 0 {
+		compiled = append(compiled, M{
+			"ip_cidr":  copyStrings(rule.Match.IPCIDRs),
+			"outbound": outbound,
+		})
+	}
+	if len(rule.Match.Ports) > 0 {
+		ports, portRanges := splitPorts(rule.Match.Ports)
+		entry := M{"outbound": outbound}
+		if len(ports) > 0 {
+			entry["port"] = ports
+		}
+		if len(portRanges) > 0 {
+			entry["port_range"] = portRanges
+		}
+		compiled = append(compiled, entry)
+	}
+	if len(rule.Match.Networks) > 0 {
+		compiled = append(compiled, M{
+			"network":  copyStrings(rule.Match.Networks),
+			"outbound": outbound,
+		})
+	}
+	if len(rule.Match.SourceCIDRs) > 0 {
+		compiled = append(compiled, M{
+			"source_ip_cidr": copyStrings(rule.Match.SourceCIDRs),
+			"outbound":       outbound,
+		})
+	}
+	if len(rule.Match.SourcePorts) > 0 {
+		ports, portRanges := splitPorts(rule.Match.SourcePorts)
+		entry := M{"outbound": outbound}
+		if len(ports) > 0 {
+			entry["source_port"] = ports
+		}
+		if len(portRanges) > 0 {
+			entry["source_port_range"] = portRanges
+		}
+		compiled = append(compiled, entry)
+	}
+	return compiled
+}
+
+
+// tryParseAsCustomRouteRule attempts to convert a raw map (from custom_routes)
+// into a structured CustomRouteRule. Returns false if the map doesn't match
+// the structured format.
+func tryParseAsCustomRouteRule(cr map[string]any) (model.CustomRouteRule, bool) {
+	var rule model.CustomRouteRule
+
+	// Parse name
+	if name, ok := cr["name"].(string); ok {
+		rule.Name = name
+	}
+	if disabled, ok := cr["disabled"].(bool); ok {
+		rule.Disabled = disabled
+	}
+
+	// Parse match
+	matchRaw, ok := cr["match"].(map[string]any)
+	if !ok {
+		return rule, false
+	}
+	rule.Match = parseRouteMatch(matchRaw)
+
+	// Parse action
+	switch action := cr["action"].(type) {
+	case string:
+		rule.Action = model.RouteAction{Type: action}
+	case map[string]any:
+		if t, ok := action["type"].(string); ok {
+			rule.Action.Type = t
+		}
+		if target, ok := action["target"].(string); ok {
+			rule.Action.Target = target
+		}
+	default:
+		return rule, false
+	}
+
+	return rule, true
+}
+
+func parseRouteMatch(m map[string]any) model.RouteMatch {
+	var match model.RouteMatch
+	if v, ok := m["domains"].([]any); ok {
+		match.Domains = toStringSlice(v)
+	}
+	if v, ok := m["domain_suffixes"].([]any); ok {
+		match.DomainSuffixes = toStringSlice(v)
+	}
+	if v, ok := m["ip_cidrs"].([]any); ok {
+		match.IPCIDRs = toStringSlice(v)
+	}
+	if v, ok := m["ports"].([]any); ok {
+		match.Ports = toStringSlice(v)
+	}
+	if v, ok := m["networks"].([]any); ok {
+		match.Networks = toStringSlice(v)
+	}
+	if v, ok := m["source_cidrs"].([]any); ok {
+		match.SourceCIDRs = toStringSlice(v)
+	}
+	if v, ok := m["source_ports"].([]any); ok {
+		match.SourcePorts = toStringSlice(v)
+	}
+	return match
+}
+
+func toStringSlice(arr []any) []string {
+	result := make([]string, 0, len(arr))
+	for _, v := range arr {
+		if s, ok := v.(string); ok {
+			result = append(result, s)
+		}
+	}
+	return result
+}
+func singboxOutboundForAction(action model.RouteAction) string {
+	switch action.Type {
+	case "direct":
+		return "direct"
+	case "route":
+		if action.Target != "" {
+			return action.Target
+		}
+		return "block"
+	default:
+		return "block"
+	}
+}
+
+func splitPorts(values []string) ([]int, []string) {
+	var ports []int
+	var ranges []string
+	for _, value := range values {
+		value = strings.TrimSpace(value)
+		if value == "" {
+			continue
+		}
+		if strings.ContainsAny(value, ":-") {
+			ranges = append(ranges, strings.ReplaceAll(value, "-", ":"))
+			continue
+		}
+		if port, err := strconv.Atoi(value); err == nil {
+			ports = append(ports, port)
+		}
+	}
+	return ports, ranges
+}
+
+func copyStrings(src []string) []string {
+	if len(src) == 0 {
+		return nil
+	}
+	out := make([]string, len(src))
+	copy(out, src)
+	return out
+}
+
 func mergeCustomSingbox(cfg M, kcfg config.KernelConfig) {
 	custom, err := kernel.LoadCustomConfig(kcfg.CustomConfig)
 	if err != nil {
-		slog.Error("failed to load custom sing-box config", "error", err)
+		nlog.Core().Error("failed to load custom sing-box config", "error", err)
 		return
 	}
 	if custom == nil {
@@ -323,7 +511,7 @@ func mergeCustomSingboxRoute(cfg M, customRoute map[string]any) {
 	}
 }
 
-func buildInbound(nc *panel.NodeConfig, users []panel.User, certFile, keyFile string) M {
+func buildInbound(nc *model.NodeSpec, users []model.UserSpec, tc kernel.TLSCert) M {
 	base := M{
 		"tag":         nc.Protocol + "-in",
 		"listen":      "::",
@@ -334,23 +522,23 @@ func buildInbound(nc *panel.NodeConfig, users []panel.User, certFile, keyFile st
 	case "shadowsocks":
 		return buildShadowsocks(base, nc, users)
 	case "vmess":
-		return buildVMess(base, nc, users, certFile, keyFile)
+		return buildVMess(base, nc, users, tc)
 	case "vless":
-		return buildVLESS(base, nc, users, certFile, keyFile)
+		return buildVLESS(base, nc, users, tc)
 	case "trojan":
-		return buildTrojan(base, nc, users, certFile, keyFile)
+		return buildTrojan(base, nc, users, tc)
 	case "hysteria":
-		return buildHysteria(base, nc, users, certFile, keyFile)
+		return buildHysteria(base, nc, users, tc)
 	case "tuic":
-		return buildTUIC(base, nc, users, certFile, keyFile)
+		return buildTUIC(base, nc, users, tc)
 	case "anytls":
-		return buildAnyTLS(base, nc, users, certFile, keyFile)
+		return buildAnyTLS(base, nc, users, tc)
 	case "naive":
-		return buildNaive(base, nc, users, certFile, keyFile)
+		return buildNaive(base, nc, users, tc)
 	case "socks":
 		return buildSocks(base, users)
 	case "http":
-		return buildHTTP(base, nc, users, certFile, keyFile)
+		return buildHTTP(base, nc, users, tc)
 	case "mieru":
 		return buildMieru(base, nc, users)
 	default:
@@ -358,7 +546,7 @@ func buildInbound(nc *panel.NodeConfig, users []panel.User, certFile, keyFile st
 	}
 }
 
-func buildMieru(base M, nc *panel.NodeConfig, users []panel.User) M {
+func buildMieru(base M, nc *model.NodeSpec, users []model.UserSpec) M {
 	base["type"] = "mieru"
 	if nc.Transport != "" {
 		base["transport"] = nc.Transport
@@ -378,31 +566,59 @@ func buildMieru(base M, nc *panel.NodeConfig, users []panel.User) M {
 	return base
 }
 
-func buildShadowsocks(base M, nc *panel.NodeConfig, users []panel.User) M {
+type ss2022Config struct {
+	method string
+	size   int
+}
+
+var ss2022Methods = map[string]ss2022Config{
+	"2022-blake3-aes-128-gcm":       {"2022-blake3-aes-128-gcm", 16},
+	"2022-blake3-aes-256-gcm":       {"2022-blake3-aes-256-gcm", 32},
+	"2022-blake3-chacha20-poly1305": {"2022-blake3-chacha20-poly1305", 32},
+}
+
+func buildShadowsocks(base M, nc *model.NodeSpec, users []model.UserSpec) M {
 	base["type"] = "shadowsocks"
 	base["method"] = nc.Cipher
 
-	if strings.HasPrefix(nc.Cipher, "2022-blake3-") {
+	ss2022, isSS2022 := ss2022Methods[nc.Cipher]
+	if isSS2022 {
 		base["password"] = nc.ServerKey
 	}
 
-	userList := make([]M, 0, len(users))
-	for _, u := range users {
-		userList = append(userList, M{
+	userList := make([]M, len(users))
+	var rawBuf []byte
+	if isSS2022 {
+		rawBuf = make([]byte, ss2022.size)
+	}
+
+	for i := range users {
+		u := &users[i]
+		user := M{
 			"name":     u.UUID,
 			"password": u.UUID,
-		})
+		}
+
+		if isSS2022 {
+			// Reuse buffer and clear it to maintain SS2022 key integrity
+			for j := range rawBuf {
+				rawBuf[j] = 0
+			}
+			copy(rawBuf, u.UUID)
+			user["password"] = base64.StdEncoding.EncodeToString(rawBuf)
+		}
+		userList[i] = user
 	}
 	base["users"] = userList
 
 	if nc.Plugin != "" {
-		slog.Warn("sing-box shadowsocks inbound does not support plugin, ignoring", "plugin", nc.Plugin)
+		nlog.Core().Warn("sing-box shadowsocks inbound does not support plugin, ignoring", "plugin", nc.Plugin)
 	}
 
 	return base
 }
 
-func buildVMess(base M, nc *panel.NodeConfig, users []panel.User, certFile, keyFile string) M {
+func buildVMess(base M, nc *model.NodeSpec, users []model.UserSpec, tc kernel.TLSCert) M {
 	base["type"] = "vmess"
 
 	userList := make([]M, 0, len(users))
@@ -416,16 +632,23 @@ func buildVMess(base M, nc *panel.NodeConfig, users []panel.User, certFile, keyF
 	base["users"] = userList
 
 	applyTransport(base, nc)
+	applyProxyProtocol(base, nc)
 	applyMultiplex(base, nc)
 
 	if nc.TLS == 1 {
-		base["tls"] = buildTLSConfig(nc, certFile, keyFile)
+		if tls := buildTLSConfig(nc, tc); tls != nil {
+			base["tls"] = tls
+		}
 	}
 
 	return base
 }
 
-func buildVLESS(base M, nc *panel.NodeConfig, users []panel.User, certFile, keyFile string) M {
+func buildVLESS(base M, nc *model.NodeSpec, users []model.UserSpec, tc kernel.TLSCert) M {
+	if nc.Decryption != "" && nc.Decryption != "none" {
+		nlog.Core().Warn("sing-box does not support VLESS encryption (decryption), use xray kernel for this feature")
+	}
+
 	base["type"] = "vless"
 
 	userList := make([]M, 0, len(users))
@@ -442,10 +665,13 @@ func buildVLESS(base M, nc *panel.NodeConfig, users []panel.User, certFile, keyF
 	base["users"] = userList
 
 	applyTransport(base, nc)
+	applyProxyProtocol(base, nc)
 	applyMultiplex(base, nc)
 
 	if nc.TLS == 1 {
-		base["tls"] = buildTLSConfig(nc, certFile, keyFile)
+		if tls := buildTLSConfig(nc, tc); tls != nil {
+			base["tls"] = tls
+		}
 	} else if nc.TLS == 2 {
 		base["tls"] = buildRealityConfig(nc)
 	}
@@ -453,30 +679,46 @@ func buildVLESS(base M, nc *panel.NodeConfig, users []panel.User, certFile, keyF
 	return base
 }
 
-func buildTrojan(base M, nc *panel.NodeConfig, users []panel.User, certFile, keyFile string) M {
+func buildTrojan(base M, nc *model.NodeSpec, users []model.UserSpec, tc kernel.TLSCert) M {
 	base["type"] = "trojan"
 
-	userList := make([]M, 0, len(users))
-	for _, u := range users {
-		userList = append(userList, M{
+	userList := make([]M, len(users))
+	for i := range users {
+		u := &users[i]
+		userList[i] = M{
 			"name":     u.UUID,
 			"password": u.UUID,
-		})
+		}
 	}
 	base["users"] = userList
 
 	applyTransport(base, nc)
+	applyProxyProtocol(base, nc)
 	applyMultiplex(base, nc)
 
 	if nc.TLS == 1 {
-		base["tls"] = buildTLSConfig(nc, certFile, keyFile)
+		if tls := buildTLSConfig(nc, tc); tls != nil {
+			base["tls"] = tls
+		}
 	} else if nc.TLS == 2 {
 		base["tls"] = buildRealityConfig(nc)
 	}
+
+	// If still no TLS/Reality but cert paths exist, enable TLS (panel may omit tls=1).
+	if _, ok := base["tls"]; !ok {
+		if tls := buildTLSConfig(nc, tc); tls != nil {
+			base["tls"] = tls
+		}
+	}
+
+	if base["tls"] == nil {
+		nlog.Core().Warn("trojan inbound has no TLS (no certificate paths and no Reality); sing-box may refuse to start until cert_mode provides cert/key or Reality is configured")
+	}
+
 	return base
 }
 
-func buildHysteria(base M, nc *panel.NodeConfig, users []panel.User, certFile, keyFile string) M {
+func buildHysteria(base M, nc *model.NodeSpec, users []model.UserSpec, tc kernel.TLSCert) M {
 	if nc.Version == 2 {
 		base["type"] = "hysteria2"
 
@@ -514,7 +756,11 @@ func buildHysteria(base M, nc *panel.NodeConfig, users []panel.User, certFile, k
 		}
 	}
 
-	tls := buildTLSConfig(nc, certFile, keyFile)
+	tls := buildTLSConfig(nc, tc)
+	if tls == nil {
+		nlog.Core().Warn("hysteria requires TLS certificate files on disk; configure cert_mode (self, file, http, dns, or content). Sing-box will not start this inbound without tls.")
+		return base
+	}
 	// Hysteria/Hysteria2 uses QUIC and requires ALPN; default to h3 if not set.
 	if _, ok := tls["alpn"]; !ok {
 		tls["alpn"] = []string{"h3"}
@@ -523,7 +769,7 @@ func buildHysteria(base M, nc *panel.NodeConfig, users []panel.User, certFile, k
 	return base
 }
 
-func buildTUIC(base M, nc *panel.NodeConfig, users []panel.User, certFile, keyFile string) M {
+func buildTUIC(base M, nc *model.NodeSpec, users []model.UserSpec, tc kernel.TLSCert) M {
 	base["type"] = "tuic"
 
 	userList := make([]M, 0, len(users))
@@ -540,7 +786,11 @@ func buildTUIC(base M, nc *panel.NodeConfig, users []panel.User, certFile, keyFi
 		base["congestion_control"] = nc.CongestionControl
 	}
 
-	tls := buildTLSConfig(nc, certFile, keyFile)
+	tls := buildTLSConfig(nc, tc)
+	if tls == nil {
+		nlog.Core().Warn("tuic requires TLS certificate files on disk; configure cert_mode (self, file, http, dns, or content). Sing-box will not start this inbound without tls.")
+		return base
+	}
 	// TUIC requires ALPN for QUIC negotiation; default to h3 if not set by panel.
 	if _, ok := tls["alpn"]; !ok {
 		tls["alpn"] = []string{"h3"}
@@ -549,7 +799,7 @@ func buildTUIC(base M, nc *panel.NodeConfig, users []panel.User, certFile, keyFi
 	return base
 }
 
-func buildAnyTLS(base M, nc *panel.NodeConfig, users []panel.User, certFile, keyFile string) M {
+func buildAnyTLS(base M, nc *model.NodeSpec, users []model.UserSpec, tc kernel.TLSCert) M {
 	base["type"] = "anytls"
 
 	userList := make([]M, 0, len(users))
@@ -562,14 +812,18 @@ func buildAnyTLS(base M, nc *panel.NodeConfig, users []panel.User, certFile, key
 	base["users"] = userList
 
 	if nc.PaddingScheme != "" {
-		base["padding_scheme"] = string(nc.PaddingScheme)
+		base["padding_scheme"] = nc.PaddingScheme
 	}
 
-	base["tls"] = buildTLSConfig(nc, certFile, keyFile)
+	if tls := buildTLSConfig(nc, tc); tls != nil {
+		base["tls"] = tls
+	} else {
+		nlog.Core().Warn("anytls requires TLS certificate files on disk; configure cert_mode (self, file, http, dns, or content). Sing-box will not start this inbound without tls.")
+	}
 	return base
 }
 
-func buildNaive(base M, nc *panel.NodeConfig, users []panel.User, certFile, keyFile string) M {
+func buildNaive(base M, nc *model.NodeSpec, users []model.UserSpec, tc kernel.TLSCert) M {
 	base["type"] = "naive"
 
 	userList := make([]M, 0, len(users))
@@ -582,12 +836,14 @@ func buildNaive(base M, nc *panel.NodeConfig, users []panel.User, certFile, keyF
 	base["users"] = userList
 
 	if nc.TLS == 1 {
-		base["tls"] = buildTLSConfig(nc, certFile, keyFile)
+		if tls := buildTLSConfig(nc, tc); tls != nil {
+			base["tls"] = tls
+		}
 	}
 	return base
 }
 
-func buildSocks(base M, users []panel.User) M {
+func buildSocks(base M, users []model.UserSpec) M {
 	base["type"] = "socks"
 
 	userList := make([]M, 0, len(users))
@@ -601,7 +857,7 @@ func buildSocks(base M, users []panel.User) M {
 	return base
 }
 
-func buildHTTP(base M, nc *panel.NodeConfig, users []panel.User, certFile, keyFile string) M {
+func buildHTTP(base M, nc *model.NodeSpec, users []model.UserSpec, tc kernel.TLSCert) M {
 	base["type"] = "http"
 
 	userList := make([]M, 0, len(users))
@@ -613,13 +869,16 @@ func buildHTTP(base M, nc *panel.NodeConfig, users []panel.User, certFile, keyFi
 	}
 	base["users"] = userList
 
+	applyProxyProtocol(base, nc)
 	if nc.TLS == 1 {
-		base["tls"] = buildTLSConfig(nc, certFile, keyFile)
+		if tls := buildTLSConfig(nc, tc); tls != nil {
+			base["tls"] = tls
+		}
 	}
 	return base
 }
 
-func applyTransport(base M, nc *panel.NodeConfig) {
+func applyTransport(base M, nc *model.NodeSpec) {
 	if nc.Network == "" || nc.Network == "tcp" {
 		return
 	}
@@ -647,7 +906,9 @@ func applyTransport(base M, nc *panel.NodeConfig) {
 				transport["early_data_header_name"] = v
 			}
 		case "grpc":
-			if v, ok := nc.NetworkSettings["service_name"]; ok {
+			if v, ok := nc.NetworkSettings["serviceName"]; ok {
+				transport["service_name"] = v
+			} else if v, ok := nc.NetworkSettings["service_name"]; ok {
 				transport["service_name"] = v
 			}
 		case "httpupgrade":
@@ -671,40 +932,43 @@ func applyTransport(base M, nc *panel.NodeConfig) {
 	base["transport"] = transport
 }
 
-func buildTLSConfig(nc *panel.NodeConfig, certFile, keyFile string) M {
-	tls := M{"enabled": true}
+// buildTLSConfig returns sing-box inbound TLS options when certificate material
+// is available. Returns nil if no TLS material is provided.
+func buildTLSConfig(nc *model.NodeSpec, tc kernel.TLSCert) M {
+	if !tc.HasCert() {
+		return nil
+	}
+
+	t := M{"enabled": true}
 
 	serverName := nc.ServerName
 	if serverName == "" && nc.Host != "" {
 		serverName = nc.Host
 	}
 	if serverName != "" {
-		tls["server_name"] = serverName
+		t["server_name"] = serverName
 	}
 
 	if nc.TLSSettings != nil {
 		if sn, ok := nc.TLSSettings["server_name"]; ok && sn != "" {
-			tls["server_name"] = sn
+			t["server_name"] = sn
 		}
 		if alpn, ok := nc.TLSSettings["alpn"]; ok {
-			tls["alpn"] = alpn
+			t["alpn"] = alpn
+		}
+		// ECH server-side: only key/key_path needed for inbound
+		if ech := extractECHInbound(nc.TLSSettings); ech != nil {
+			t["ech"] = ech
 		}
 	}
 
-	if certFile != "" && keyFile != "" {
-		tls["certificate_path"] = certFile
-		tls["key_path"] = keyFile
-	} else {
-		// If no real certificates are provided, but TLS is requested,
-		// use a self-signed certificate as fallback to prevent sing-box
-		// from failing with "missing certificate".
-		tls["certificate_path"] = "self-signed"
-	}
+	t["certificate"] = []string{string(tc.CertPEM)}
+	t["key"] = []string{string(tc.KeyPEM)}
 
-	return tls
+	return t
 }
 
-func buildRealityConfig(nc *panel.NodeConfig) M {
+func buildRealityConfig(nc *model.NodeSpec) M {
 	tls := M{"enabled": true}
 	if nc.TLSSettings == nil {
 		return tls
@@ -770,7 +1034,7 @@ func buildRealityConfig(nc *panel.NodeConfig) M {
 	return tls
 }
 
-func applyMultiplex(base M, nc *panel.NodeConfig) {
+func applyMultiplex(base M, nc *model.NodeSpec) {
 	if nc.Multiplex == nil || !nc.Multiplex.Enabled {
 		return
 	}
@@ -797,3 +1061,41 @@ func applyMultiplex(base M, nc *panel.NodeConfig) {
 
 	base["multiplex"] = mux
 }
+
+func applyProxyProtocol(base M, nc *model.NodeSpec) {
+	// if !nc.GetProxyProtocol() {
+	// 	return
+	// }
+	// base["proxy_protocol"] = true
+}
+
+// extractECHInbound extracts ECH config for sing-box server (inbound).
+// sing-box InboundECHOptions needs: enabled, key (PEM array), key_path.
+func extractECHInbound(tlsSettings map[string]interface{}) M {
+	echRaw, ok := tlsSettings["ech"]
+	if !ok {
+		return nil
+	}
+	echMap, ok := echRaw.(map[string]interface{})
+	if !ok {
+		return nil
+	}
+	enabled, _ := echMap["enabled"].(bool)
+	if !enabled {
+		return nil
+	}
+	ech := M{"enabled": true}
+	hasSource := false
+	if key, _ := echMap["key"].(string); key != "" {
+		ech["key"] = []string{key} // sing-box expects PEM as-is
+		hasSource = true
+	}
+	if keyPath, _ := echMap["key_path"].(string); keyPath != "" {
+		ech["key_path"] = keyPath
+		hasSource = true
+	}
+	if !hasSource {
+		nlog.Core().Warn("ECH enabled but no key or key_path provided")
+	}
+	return ech // let sing-box report a clear error if key is missing
+}

internal/kernel/singbox/config_test.go

@@ -2,26 +2,40 @@ package singbox
 
 import (
 	"encoding/json"
+	"reflect"
 	"testing"
 
 	"github.com/cedar2025/xboard-node/internal/config"
+	"github.com/cedar2025/xboard-node/internal/kernel"
+	"github.com/cedar2025/xboard-node/internal/model"
 	"github.com/cedar2025/xboard-node/internal/panel"
 )
 
-var testUsers = []panel.User{
+var testUsersPanel = []panel.User{
 	{ID: 1, UUID: "aaaaaaaa-1111-2222-3333-444444444444", SpeedLimit: 0, DeviceLimit: 0},
 	{ID: 2, UUID: "bbbbbbbb-5555-6666-7777-888888888888", SpeedLimit: 3, DeviceLimit: 2},
 }
 
 // --- Shadowsocks ---
 
+var testUsers = model.UserSpecsFromPanel(testUsersPanel)
+
+func testNodeSpec(nc *panel.NodeConfig) *model.NodeSpec { return model.NodeSpecFromPanel(nc) }
+
+func testRouteRules(r []panel.RouteRule) []model.RouteRule {
+	if r == nil {
+		return nil
+	}
+	return model.NodeSpecFromPanel(&panel.NodeConfig{Routes: r}).Routes
+}
+
 func TestBuildInbound_Shadowsocks(t *testing.T) {
 	nc := &panel.NodeConfig{
 		Protocol:   "shadowsocks",
 		ServerPort: 111,
 		Cipher:     "aes-128-gcm",
 	}
-	inbound := buildInbound(nc, testUsers, "", "")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{})
 	assertMapValue(t, inbound, "type", "shadowsocks")
 	assertMapValue(t, inbound, "method", "aes-128-gcm")
 	assertMapValue(t, inbound, "listen_port", 111)
@@ -41,7 +55,7 @@ func TestBuildInbound_Shadowsocks2022(t *testing.T) {
 		Cipher:     "2022-blake3-aes-128-gcm",
 		ServerKey:  "base64serverkey==",
 	}
-	inbound := buildInbound(nc, testUsers, "", "")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{})
 	assertMapValue(t, inbound, "method", "2022-blake3-aes-128-gcm")
 	assertMapValue(t, inbound, "password", "base64serverkey==")
 }
@@ -53,7 +67,7 @@ func TestBuildInbound_VMess(t *testing.T) {
 		Protocol:   "vmess",
 		ServerPort: 443,
 	}
-	inbound := buildInbound(nc, testUsers, "", "")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{})
 	assertMapValue(t, inbound, "type", "vmess")
 	assertMapValue(t, inbound, "tag", "vmess-in")
 
@@ -69,12 +83,27 @@ func TestBuildInbound_VMess_WithTLS(t *testing.T) {
 		TLS:        1,
 		ServerName: "example.com",
 	}
-	inbound := buildInbound(nc, testUsers, "/path/cert.pem", "/path/key.pem")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{CertPEM: []byte("CERT"), KeyPEM: []byte("KEY")})
 	tls := inbound["tls"].(M)
 	assertMapValue(t, tls, "enabled", true)
 	assertMapValue(t, tls, "server_name", "example.com")
-	assertMapValue(t, tls, "certificate_path", "/path/cert.pem")
-	assertMapValue(t, tls, "key_path", "/path/key.pem")
+	assertMapValue(t, tls, "certificate", []string{"CERT"})
+	assertMapValue(t, tls, "key", []string{"KEY"})
+}
+
+// Panel tls=1 with cert_mode none (no paths): inbound must be plain — TLS terminated at nginx/CDN.
+func TestBuildInbound_VMess_TLS_PanelOn_NoCertFiles(t *testing.T) {
+	nc := &panel.NodeConfig{
+		Protocol:   "vmess",
+		ServerPort: 8888,
+		Network:    "ws",
+		TLS:        1,
+		ServerName: "example.com",
+	}
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{})
+	if _, exists := inbound["tls"]; exists {
+		t.Fatal("expected no tls block when cert/key paths are empty (nginx offload)")
+	}
 }
 
 func TestBuildInbound_VMess_NoTLS(t *testing.T) {
@@ -83,7 +112,7 @@ func TestBuildInbound_VMess_NoTLS(t *testing.T) {
 		ServerPort: 80,
 		TLS:        0,
 	}
-	inbound := buildInbound(nc, testUsers, "", "")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{})
 	if _, exists := inbound["tls"]; exists {
 		t.Error("should not have TLS when tls=0")
 	}
@@ -101,7 +130,7 @@ func TestBuildInbound_VMess_WithWebSocket(t *testing.T) {
 			},
 		},
 	}
-	inbound := buildInbound(nc, testUsers, "", "")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{})
 	transport := inbound["transport"].(M)
 	assertMapValue(t, transport, "type", "ws")
 	assertMapValue(t, transport, "path", "/ws")
@@ -113,15 +142,30 @@ func TestBuildInbound_VMess_WithGRPC(t *testing.T) {
 		ServerPort: 443,
 		Network:    "grpc",
 		NetworkSettings: map[string]interface{}{
-			"service_name": "mygrpc",
+			"serviceName": "mygrpc",
 		},
 	}
-	inbound := buildInbound(nc, testUsers, "", "")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{})
 	transport := inbound["transport"].(M)
 	assertMapValue(t, transport, "type", "grpc")
 	assertMapValue(t, transport, "service_name", "mygrpc")
 }
 
+func TestBuildInbound_VMess_WithGRPCSnakeCaseServiceName(t *testing.T) {
+	nc := &panel.NodeConfig{
+		Protocol:   "vmess",
+		ServerPort: 443,
+		Network:    "grpc",
+		NetworkSettings: map[string]interface{}{
+			"service_name": "mygrpc-snake",
+		},
+	}
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{})
+	transport := inbound["transport"].(M)
+	assertMapValue(t, transport, "type", "grpc")
+	assertMapValue(t, transport, "service_name", "mygrpc-snake")
+}
+
 func TestBuildInbound_VMess_WithH2(t *testing.T) {
 	nc := &panel.NodeConfig{
 		Protocol:   "vmess",
@@ -132,7 +176,7 @@ func TestBuildInbound_VMess_WithH2(t *testing.T) {
 			"host": "example.com",
 		},
 	}
-	inbound := buildInbound(nc, testUsers, "", "")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{})
 	transport := inbound["transport"].(M)
 	assertMapValue(t, transport, "type", "http")
 	assertMapValue(t, transport, "path", "/h2path")
@@ -148,7 +192,7 @@ func TestBuildInbound_VMess_WithHTTPUpgrade(t *testing.T) {
 			"host": "example.com",
 		},
 	}
-	inbound := buildInbound(nc, testUsers, "", "")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{})
 	transport := inbound["transport"].(M)
 	assertMapValue(t, transport, "type", "httpupgrade")
 	assertMapValue(t, transport, "path", "/upgrade")
@@ -162,7 +206,7 @@ func TestBuildInbound_VLESS(t *testing.T) {
 		Protocol:   "vless",
 		ServerPort: 443,
 	}
-	inbound := buildInbound(nc, testUsers, "", "")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{})
 	assertMapValue(t, inbound, "type", "vless")
 
 	users := inbound["users"].([]M)
@@ -175,7 +219,7 @@ func TestBuildInbound_VLESS_WithFlow(t *testing.T) {
 		ServerPort: 443,
 		Flow:       "xtls-rprx-vision",
 	}
-	inbound := buildInbound(nc, testUsers, "", "")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{})
 	users := inbound["users"].([]M)
 	assertMapValue(t, users[0], "flow", "xtls-rprx-vision")
 }
@@ -192,7 +236,7 @@ func TestBuildInbound_VLESS_Reality(t *testing.T) {
 			"server_name": "www.example.com",
 		},
 	}
-	inbound := buildInbound(nc, testUsers, "", "")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{})
 	tls := inbound["tls"].(M)
 	assertMapValue(t, tls, "enabled", true)
 
@@ -216,7 +260,7 @@ func TestBuildInbound_VLESS_Reality_ShortIDArray(t *testing.T) {
 			"dest":        "example.com",
 		},
 	}
-	inbound := buildInbound(nc, testUsers, "", "")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{})
 	reality := inbound["tls"].(M)["reality"].(M)
 	ids := reality["short_id"].([]string)
 	if len(ids) != 2 {
@@ -226,16 +270,29 @@ func TestBuildInbound_VLESS_Reality_ShortIDArray(t *testing.T) {
 
 // --- Trojan ---
 
-func TestBuildInbound_Trojan(t *testing.T) {
+func TestBuildInbound_Trojan_Reality(t *testing.T) {
 	nc := &panel.NodeConfig{
 		Protocol:   "trojan",
 		ServerPort: 443,
+		TLS:        2,
+		TLSSettings: map[string]interface{}{
+			"private_key": "test-private-key",
+			"dest":        "gateway.example.com:8443",
+			"server_name": "cdn.example.com",
+		},
 	}
-	inbound := buildInbound(nc, testUsers, "", "")
-	assertMapValue(t, inbound, "type", "trojan")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{})
+	tls := inbound["tls"].(M)
+	assertMapValue(t, tls, "enabled", true)
+	assertMapValue(t, tls, "server_name", "cdn.example.com")
 
-	users := inbound["users"].([]M)
-	assertMapValue(t, users[0], "password", "aaaaaaaa-1111-2222-3333-444444444444")
+	reality := tls["reality"].(M)
+	assertMapValue(t, reality, "enabled", true)
+	assertMapValue(t, reality, "private_key", "test-private-key")
+
+	handshake := reality["handshake"].(M)
+	assertMapValue(t, handshake, "server", "gateway.example.com")
+	assertMapValue(t, handshake, "server_port", 8443)
 }
 
 func TestBuildInbound_Trojan_NoTLS(t *testing.T) {
@@ -244,9 +301,10 @@ func TestBuildInbound_Trojan_NoTLS(t *testing.T) {
 		ServerPort: 80,
 		TLS:        0,
 	}
-	inbound := buildInbound(nc, testUsers, "", "")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{})
+	// Without certificate paths there is no TLS layer; sing-box trojan needs certs or Reality.
 	if _, exists := inbound["tls"]; exists {
-		t.Error("trojan with tls=0 should not have TLS config")
+		t.Fatal("expected no tls when cert paths are empty and tls=0")
 	}
 }
 
@@ -257,7 +315,7 @@ func TestBuildInbound_Trojan_WithTLS(t *testing.T) {
 		TLS:        1,
 		ServerName: "example.com",
 	}
-	inbound := buildInbound(nc, testUsers, "/c.pem", "/k.pem")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{CertPEM: []byte("CERT"), KeyPEM: []byte("KEY")})
 	tls := inbound["tls"].(M)
 	assertMapValue(t, tls, "enabled", true)
 }
@@ -270,7 +328,7 @@ func TestBuildInbound_Hysteria2(t *testing.T) {
 		ServerPort: 444,
 		Version:    2,
 	}
-	inbound := buildInbound(nc, testUsers, "/c.pem", "/k.pem")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{CertPEM: []byte("CERT"), KeyPEM: []byte("KEY")})
 	assertMapValue(t, inbound, "type", "hysteria2")
 
 	users := inbound["users"].([]M)
@@ -288,7 +346,7 @@ func TestBuildInbound_Hysteria2_WithObfs(t *testing.T) {
 		Obfs:         "salamander",
 		ObfsPassword: "secret",
 	}
-	inbound := buildInbound(nc, testUsers, "/c.pem", "/k.pem")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{CertPEM: []byte("CERT"), KeyPEM: []byte("KEY")})
 	obfs := inbound["obfs"].(M)
 	assertMapValue(t, obfs, "type", "salamander")
 	assertMapValue(t, obfs, "password", "secret")
@@ -302,7 +360,7 @@ func TestBuildInbound_Hysteria1(t *testing.T) {
 		UpMbps:     100,
 		DownMbps:   200,
 	}
-	inbound := buildInbound(nc, testUsers, "/c.pem", "/k.pem")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{CertPEM: []byte("CERT"), KeyPEM: []byte("KEY")})
 	assertMapValue(t, inbound, "type", "hysteria")
 	assertMapValue(t, inbound, "up_mbps", 100)
 	assertMapValue(t, inbound, "down_mbps", 200)
@@ -319,7 +377,7 @@ func TestBuildInbound_TUIC(t *testing.T) {
 		ServerPort:        555,
 		CongestionControl: "bbr",
 	}
-	inbound := buildInbound(nc, testUsers, "/c.pem", "/k.pem")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{CertPEM: []byte("CERT"), KeyPEM: []byte("KEY")})
 	assertMapValue(t, inbound, "type", "tuic")
 	assertMapValue(t, inbound, "congestion_control", "bbr")
 
@@ -336,7 +394,7 @@ func TestBuildInbound_AnyTLS(t *testing.T) {
 		ServerPort:    443,
 		PaddingScheme: "stop=8\n0=30-30",
 	}
-	inbound := buildInbound(nc, testUsers, "/c.pem", "/k.pem")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{CertPEM: []byte("CERT"), KeyPEM: []byte("KEY")})
 	assertMapValue(t, inbound, "type", "anytls")
 	assertMapValue(t, inbound, "padding_scheme", "stop=8\n0=30-30")
 
@@ -349,7 +407,7 @@ func TestBuildInbound_AnyTLS_NoPaddingScheme(t *testing.T) {
 		Protocol:   "anytls",
 		ServerPort: 443,
 	}
-	inbound := buildInbound(nc, testUsers, "/c.pem", "/k.pem")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{CertPEM: []byte("CERT"), KeyPEM: []byte("KEY")})
 	if _, exists := inbound["padding_scheme"]; exists {
 		t.Error("should not have padding_scheme when empty")
 	}
@@ -363,7 +421,7 @@ func TestBuildInbound_Naive(t *testing.T) {
 		ServerPort: 443,
 		TLS:        1,
 	}
-	inbound := buildInbound(nc, testUsers, "/c.pem", "/k.pem")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{CertPEM: []byte("CERT"), KeyPEM: []byte("KEY")})
 	assertMapValue(t, inbound, "type", "naive")
 
 	users := inbound["users"].([]M)
@@ -381,7 +439,7 @@ func TestBuildInbound_Socks(t *testing.T) {
 		Protocol:   "socks",
 		ServerPort: 1080,
 	}
-	inbound := buildInbound(nc, testUsers, "", "")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{})
 	assertMapValue(t, inbound, "type", "socks")
 
 	users := inbound["users"].([]M)
@@ -400,7 +458,7 @@ func TestBuildInbound_HTTP(t *testing.T) {
 		Protocol:   "http",
 		ServerPort: 8080,
 	}
-	inbound := buildInbound(nc, testUsers, "", "")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{})
 	assertMapValue(t, inbound, "type", "http")
 
 	users := inbound["users"].([]M)
@@ -413,7 +471,7 @@ func TestBuildInbound_HTTP_WithTLS(t *testing.T) {
 		ServerPort: 443,
 		TLS:        1,
 	}
-	inbound := buildInbound(nc, testUsers, "/c.pem", "/k.pem")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{CertPEM: []byte("CERT"), KeyPEM: []byte("KEY")})
 	tls := inbound["tls"].(M)
 	assertMapValue(t, tls, "enabled", true)
 }
@@ -425,7 +483,7 @@ func TestBuildInbound_Unknown(t *testing.T) {
 		Protocol:   "unknown-proto",
 		ServerPort: 999,
 	}
-	inbound := buildInbound(nc, testUsers, "", "")
+	inbound := buildInbound(testNodeSpec(nc), testUsers, kernel.TLSCert{})
 	if inbound != nil {
 		t.Errorf("unknown protocol should return nil, got %v", inbound)
 	}
@@ -440,7 +498,7 @@ func TestBuildConfig(t *testing.T) {
 		ServerPort: 111,
 		Cipher:     "aes-128-gcm",
 	}
-	cfg := buildConfig(kcfg, nc, testUsers, "", "")
+	cfg := buildConfig(kcfg, testNodeSpec(nc), testUsers, kernel.TLSCert{})
 
 	data, err := json.Marshal(cfg)
 	if err != nil {
@@ -484,7 +542,7 @@ func TestBuildConfig_OutboundPriority(t *testing.T) {
 		},
 	}
 
-	cfg := buildConfig(kcfg, nc, testUsers, "", "")
+	cfg := buildConfig(kcfg, testNodeSpec(nc), testUsers, kernel.TLSCert{})
 	outbounds := cfg["outbounds"].([]M)
 
 	// We have 2 overrides in input, so we should have exactly 2 outbounds total
@@ -537,7 +595,7 @@ func TestBuildConfig_AllProtocols_ValidJSON(t *testing.T) {
 
 	for _, tc := range protocols {
 		t.Run(tc.name, func(t *testing.T) {
-			cfg := buildConfig(config.KernelConfig{LogLevel: "warn"}, tc.nc, testUsers, "/c.pem", "/k.pem")
+			cfg := buildConfig(config.KernelConfig{LogLevel: "warn"}, testNodeSpec(tc.nc), testUsers, kernel.TLSCert{CertPEM: []byte("CERT"), KeyPEM: []byte("KEY")})
 			data, err := json.Marshal(cfg)
 			if err != nil {
 				t.Fatalf("marshal %s: %v", tc.name, err)
@@ -553,7 +611,7 @@ func TestBuildConfig_AllProtocols_ValidJSON(t *testing.T) {
 // --- Routes ---
 
 func TestBuildRoutes_Default(t *testing.T) {
-	route := buildRoutes(nil, nil)
+	route := buildRoutes(nil, nil, nil)
 	assertMapValue(t, route, "final", "direct")
 
 	rules := route["rules"].([]M)
@@ -570,7 +628,7 @@ func TestBuildRoutes_WithCustomRules(t *testing.T) {
 		{ID: 2, Match: []string{"10.0.0.0/8"}, Action: "block"},
 		{ID: 3, Match: []string{"allowed.com"}, Action: "direct"},
 	}
-	route := buildRoutes(rules, nil)
+	route := buildRoutes(testRouteRules(rules), nil, nil)
 	allRules := route["rules"].([]M)
 
 	if len(allRules) != 5 {
@@ -595,7 +653,7 @@ func TestBuildRoutes_MultiMatch(t *testing.T) {
 		{ID: 1, Match: []string{"*.evil.com", "bad.org", "192.168.1.0/24"}, Action: "block"},
 		{ID: 2, Match: []string{"*.bypass.com"}, Action: "direct"},
 	}
-	route := buildRoutes(rules, nil)
+	route := buildRoutes(testRouteRules(rules), nil, nil)
 	allRules := route["rules"].([]M)
 
 	// 2 default private-IP rules + 1 domain rule + 1 CIDR rule + 1 domain rule = 5
@@ -625,28 +683,88 @@ func TestBuildRoutes_MultiMatch(t *testing.T) {
 	assertMapValue(t, allRules[4], "outbound", "direct")
 }
 
+func TestBuildRoutes_WithCustomRouteRules(t *testing.T) {
+	customRules := []model.CustomRouteRule{
+		{
+			Name: "direct-mixed",
+			Match: model.RouteMatch{
+				Domains:        []string{"full.example.com"},
+				DomainSuffixes: []string{"example.org"},
+				IPCIDRs:        []string{"1.1.1.0/24"},
+				Ports:          []string{"53", "1000-1002"},
+				Networks:       []string{"tcp"},
+				SourceCIDRs:    []string{"10.10.0.0/16"},
+				SourcePorts:    []string{"2000-2001"},
+			},
+			Action: model.RouteAction{Type: "direct"},
+		},
+	}
+	route := buildRoutes(nil, customRules, nil)
+	allRules := route["rules"].([]M)
+	if len(allRules) != 9 {
+		t.Fatalf("rules count: got %d, want 9", len(allRules))
+	}
+	if allRules[0]["domain"].([]string)[0] != "full.example.com" {
+		t.Fatalf("unexpected exact domain rule: %v", allRules[0])
+	}
+	if allRules[1]["domain_suffix"].([]string)[0] != "example.org" {
+		t.Fatalf("unexpected domain suffix rule: %v", allRules[1])
+	}
+	if allRules[2]["ip_cidr"].([]string)[0] != "1.1.1.0/24" {
+		t.Fatalf("unexpected ip cidr rule: %v", allRules[2])
+	}
+	if allRules[3]["port"].([]int)[0] != 53 {
+		t.Fatalf("unexpected port rule: %v", allRules[3])
+	}
+	if allRules[4]["network"].([]string)[0] != "tcp" {
+		t.Fatalf("unexpected network rule: %v", allRules[4])
+	}
+	if allRules[5]["source_ip_cidr"].([]string)[0] != "10.10.0.0/16" {
+		t.Fatalf("unexpected source cidr rule: %v", allRules[5])
+	}
+	if allRules[6]["source_port_range"].([]string)[0] != "2000:2001" {
+		t.Fatalf("unexpected source port rule: %v", allRules[6])
+	}
+}
+
+func TestBuildRoutes_StructuredCustomRulesRemainFirst(t *testing.T) {
+	raw := []map[string]any{{"domain": []string{"raw.example.com"}, "outbound": "raw-tag"}}
+	custom := []model.CustomRouteRule{{
+		Match:  model.RouteMatch{DomainSuffixes: []string{"structured.example"}},
+		Action: model.RouteAction{Type: "direct"},
+	}}
+	route := buildRoutes(nil, custom, raw)
+	allRules := route["rules"].([]M)
+	if allRules[0]["outbound"] != "direct" {
+		t.Fatalf("expected structured route first, got %v", allRules[0]["outbound"])
+	}
+	if allRules[1]["outbound"] != "raw-tag" {
+		t.Fatalf("expected raw custom route second, got %v", allRules[1]["outbound"])
+	}
+}
+
 // --- TLS Config ---
 
 func TestBuildTLSConfig_WithCert(t *testing.T) {
 	nc := &panel.NodeConfig{ServerName: "example.com"}
-	tls := buildTLSConfig(nc, "/cert.pem", "/key.pem")
+	tls := buildTLSConfig(testNodeSpec(nc), kernel.TLSCert{CertPEM: []byte("CERT"), KeyPEM: []byte("KEY")})
 	assertMapValue(t, tls, "enabled", true)
 	assertMapValue(t, tls, "server_name", "example.com")
-	assertMapValue(t, tls, "certificate_path", "/cert.pem")
-	assertMapValue(t, tls, "key_path", "/key.pem")
+	assertMapValue(t, tls, "certificate", []string{"CERT"})
+	assertMapValue(t, tls, "key", []string{"KEY"})
 }
 
 func TestBuildTLSConfig_NoCert(t *testing.T) {
 	nc := &panel.NodeConfig{ServerName: "example.com"}
-	tls := buildTLSConfig(nc, "", "")
-	assertMapValue(t, tls, "enabled", true)
-	// Now we fallback to self-signed certificates when both cert and key are empty
-	assertMapValue(t, tls, "certificate_path", "self-signed")
+	tls := buildTLSConfig(testNodeSpec(nc), kernel.TLSCert{})
+	if tls != nil {
+		t.Fatalf("expected nil tls config when cert paths are empty, got %+v", tls)
+	}
 }
 
 func TestBuildTLSConfig_FallbackToHost(t *testing.T) {
 	nc := &panel.NodeConfig{Host: "fallback.com"}
-	tls := buildTLSConfig(nc, "", "")
+	tls := buildTLSConfig(testNodeSpec(nc), kernel.TLSCert{CertPEM: []byte("CERT"), KeyPEM: []byte("KEY")})
 	assertMapValue(t, tls, "server_name", "fallback.com")
 }
 
@@ -657,7 +775,7 @@ func TestBuildTLSConfig_TLSSettingsOverride(t *testing.T) {
 			"server_name": "override.com",
 		},
 	}
-	tls := buildTLSConfig(nc, "", "")
+	tls := buildTLSConfig(testNodeSpec(nc), kernel.TLSCert{CertPEM: []byte("CERT"), KeyPEM: []byte("KEY")})
 	assertMapValue(t, tls, "server_name", "override.com")
 }
 
@@ -666,7 +784,7 @@ func TestBuildTLSConfig_TLSSettingsOverride(t *testing.T) {
 func TestApplyTransport_TCP(t *testing.T) {
 	base := M{}
 	nc := &panel.NodeConfig{Network: "tcp"}
-	applyTransport(base, nc)
+	applyTransport(base, testNodeSpec(nc))
 	if _, exists := base["transport"]; exists {
 		t.Error("tcp should not add transport")
 	}
@@ -675,7 +793,7 @@ func TestApplyTransport_TCP(t *testing.T) {
 func TestApplyTransport_Empty(t *testing.T) {
 	base := M{}
 	nc := &panel.NodeConfig{Network: ""}
-	applyTransport(base, nc)
+	applyTransport(base, testNodeSpec(nc))
 	if _, exists := base["transport"]; exists {
 		t.Error("empty network should not add transport")
 	}
@@ -691,7 +809,7 @@ func TestApplyTransport_WS_MaxEarlyData(t *testing.T) {
 			"early_data_header_name": "Sec-WebSocket-Protocol",
 		},
 	}
-	applyTransport(base, nc)
+	applyTransport(base, testNodeSpec(nc))
 	transport := base["transport"].(M)
 	assertMapValue(t, transport, "max_early_data", 2048)
 	assertMapValue(t, transport, "early_data_header_name", "Sec-WebSocket-Protocol")
@@ -721,8 +839,51 @@ func assertMapValue(t *testing.T, m M, key string, expected interface{}) {
 			t.Errorf("%s: got %v (type %T), want %d", key, val, val, e)
 		}
 	default:
-		if val != expected {
+		if !reflect.DeepEqual(val, expected) {
 			t.Errorf("%s: got %v, want %v", key, val, expected)
 		}
 	}
 }
+
+func TestExtractECHInbound(t *testing.T) {
+	tests := []struct {
+		name      string
+		tls       map[string]interface{}
+		expectNil bool
+		expectKey []string
+	}{
+		{"no ech", map[string]interface{}{}, true, nil},
+		{"disabled", map[string]interface{}{"ech": map[string]interface{}{"enabled": false}}, true, nil},
+		{"enabled with key", map[string]interface{}{"ech": map[string]interface{}{
+			"enabled": true,
+			"key":     "-----BEGIN ECH KEYS-----\nABCD\n-----END ECH KEYS-----\n",
+		}}, false, []string{"-----BEGIN ECH KEYS-----\nABCD\n-----END ECH KEYS-----\n"}},
+		{"enabled with key_path", map[string]interface{}{"ech": map[string]interface{}{
+			"enabled":  true,
+			"key_path": "/etc/ech/keys.pem",
+		}}, false, nil},
+		{"enabled no key", map[string]interface{}{"ech": map[string]interface{}{"enabled": true}}, false, nil},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := extractECHInbound(tt.tls)
+			if tt.expectNil {
+				if result != nil {
+					t.Errorf("expected nil, got %v", result)
+				}
+				return
+			}
+			if result == nil {
+				t.Fatal("expected non-nil result")
+			}
+			if result["enabled"] != true {
+				t.Errorf("expected enabled=true, got %v", result["enabled"])
+			}
+			if tt.expectKey != nil {
+				if !reflect.DeepEqual(result["key"], tt.expectKey) {
+					t.Errorf("key: got %v, want %v", result["key"], tt.expectKey)
+				}
+			}
+		})
+	}
+}

internal/kernel/singbox/singbox.go

@@ -1,10 +1,10 @@
 package singbox
 
 import (
+	"bytes"
 	"context"
 	"encoding/json"
 	"fmt"
-	"log/slog"
 	"sync"
 	"time"
 
@@ -20,7 +20,8 @@ import (
 
 	"github.com/cedar2025/xboard-node/internal/config"
 	"github.com/cedar2025/xboard-node/internal/kernel"
-	"github.com/cedar2025/xboard-node/internal/panel"
+	"github.com/cedar2025/xboard-node/internal/model"
+	"github.com/cedar2025/xboard-node/internal/nlog"
 )
 
 // drainTimeout is how long stop() waits for in-flight connections to finish
@@ -35,15 +36,14 @@ const drainTimeout = 5 * time.Second
 type SingBox struct {
 	cfg config.KernelConfig
 
-	mu     sync.Mutex
+	mu     sync.RWMutex
 	box    *box.Box
 	ctx    context.Context
 	cancel context.CancelFunc
 
-	users      []panel.User
-	nodeConfig *panel.NodeConfig
-	certFile   string
-	keyFile    string
+	users      []model.UserSpec
+	nodeConfig *model.NodeSpec
+	tls        kernel.TLSCert
 
 	// connTracker is our lightweight in-process byte/IP tracker.
 	// Created fresh on every Start (full restart).
@@ -54,6 +54,10 @@ type SingBox struct {
 	// Set once by SetSpeedLimitFunc and forwarded to every new ConnTracker.
 	speedLimitFunc func(string) *rate.Limiter
 
+	// deviceLimitFunc resolves a user UUID to (limit, hasLimit) for gate-keeping.
+	// Set once by SetDeviceLimitFunc and forwarded to every new ConnTracker.
+	deviceLimitFunc func(string) (int, bool)
+
 	// trackerRegistered prevents duplicate AppendTracker calls on the same
 	// Router instance during Reload. Reset to false on full restart.
 	trackerRegistered bool
@@ -67,24 +71,35 @@ var _ kernel.Kernel = (*SingBox)(nil)
 
 func (s *SingBox) Name() string { return "sing-box" }
 
-func (s *SingBox) Protocols() []string {
-	return []string{
-		"vmess", "vless", "trojan", "shadowsocks",
-		"hysteria2", "tuic", "naive", "socks", "http", "anytls", "mieru",
+func (s *SingBox) Capabilities() kernel.Capabilities {
+	return kernel.Capabilities{
+		PerUserSpeedLimit:    true,
+		DeviceLimit:          true,
+		BuiltInTrafficStats:  false,
+		AliveIPTracking:      true,
+		ForceCloseConnection: false,
+		ForceCloseUser:       true,
 	}
 }
 
-func (s *SingBox) Start(nodeConfig *panel.NodeConfig, users []panel.User, certFile, keyFile string) error {
+func (s *SingBox) Protocols() []string {
+	return []string{
+		"vmess", "vless", "trojan", "shadowsocks",
+		"hysteria", "hysteria2", "tuic", "naive", "socks", "http", "anytls", "mieru",
+	}
+}
+
+func (s *SingBox) Start(nodeConfig *model.NodeSpec, users []model.UserSpec, tls kernel.TLSCert) error {
 	s.mu.Lock()
 	defer s.mu.Unlock()
 
-	cfgMap := buildConfig(s.cfg, nodeConfig, users, certFile, keyFile)
+	cfgMap := buildConfig(s.cfg, nodeConfig, users, tls)
 	data, err := json.Marshal(cfgMap)
 	if err != nil {
 		return fmt.Errorf("marshal config: %w", err)
 	}
 
-	slog.Debug("sing-box config generated", "len", len(data))
+	nlog.Core().Debug("sing-box config generated", "len", len(data))
 
 	ctx, cancel := context.WithCancel(context.Background())
 	ctx = include.Context(ctx)
@@ -95,7 +110,11 @@ func (s *SingBox) Start(nodeConfig *panel.NodeConfig, users []panel.User, certFi
 		return fmt.Errorf("parse sing-box options: %w", err)
 	}
 
-	s.stop()
+	// Save old state before creating the new instance.
+	oldBox := s.box
+	oldCancel := s.cancel
+	oldCtx := s.ctx
+	oldTracker := s.connTracker
 
 	instance, err := box.New(box.Options{
 		Context: ctx,
@@ -112,32 +131,68 @@ func (s *SingBox) Start(nodeConfig *panel.NodeConfig, users []panel.User, certFi
 		return fmt.Errorf("start sing-box: %w", err)
 	}
 
+	// New instance started successfully — swap state.
 	s.box = instance
 	s.ctx = ctx
 	s.cancel = cancel
 	s.users = users
 	s.nodeConfig = nodeConfig
-	s.certFile = certFile
-	s.keyFile = keyFile
+	s.tls = tls
 
-	// Fresh tracker on full restart (all connections were terminated by s.stop).
+	// Fresh tracker on full restart.
 	s.connTracker = NewConnTracker(0)
 	s.connTracker.SetUserMap(buildUserMap(users))
 	if s.speedLimitFunc != nil {
 		s.connTracker.SetSpeedLimitFunc(s.speedLimitFunc)
 	}
+	if s.deviceLimitFunc != nil {
+		s.connTracker.SetDeviceLimitFunc(s.deviceLimitFunc)
+	}
 
 	s.trackerRegistered = false
 	s.registerTracker(ctx)
 
-	slog.Info("sing-box started", "users", len(users))
+	// Recycle old instance in background — drain then close.
+	if oldBox != nil {
+		go recycleOldBox(oldBox, oldCancel, oldCtx, oldTracker)
+	}
+
+	nlog.Core().Debug("sing-box started", "users", len(users))
 	return nil
 }
 
+// recycleOldBox gracefully shuts down a previous sing-box instance in the
+// background. It closes listen sockets first, waits for connections to drain,
+// then hard-closes. This avoids blocking the new instance's startup.
+func recycleOldBox(oldBox *box.Box, oldCancel context.CancelFunc, oldCtx context.Context, oldTracker *ConnTracker) {
+	// Step 1: close listen sockets so no new connections arrive on old ports.
+	if im := service.FromContext[adapter.InboundManager](oldCtx); im != nil {
+		_ = im.Close()
+	}
+
+	// Step 2: drain in-flight connections (best-effort).
+	if oldTracker != nil {
+		deadline := time.Now().Add(drainTimeout)
+		for time.Now().Before(deadline) {
+			if oldTracker.ActiveCount() == 0 {
+				break
+			}
+			time.Sleep(100 * time.Millisecond)
+		}
+	}
+
+	// Step 3: hard-close everything.
+	oldBox.Close()
+	if oldCancel != nil {
+		oldCancel()
+	}
+	nlog.Core().Debug("sing-box: old instance recycled")
+}
+
 // Reload hot-swaps the inbound users and routing rules without restarting the box.
 // Routes, outbounds, and the connTracker stay alive so in-flight connections
 // continue to be tracked correctly.
-func (s *SingBox) Reload(nodeConfig *panel.NodeConfig, users []panel.User, certFile, keyFile string) error {
+func (s *SingBox) Reload(nodeConfig *model.NodeSpec, users []model.UserSpec, tls kernel.TLSCert) error {
 	s.mu.Lock()
 	defer s.mu.Unlock()
 
@@ -145,7 +200,7 @@ func (s *SingBox) Reload(nodeConfig *panel.NodeConfig, users []panel.User, certF
 		return fmt.Errorf("not running")
 	}
 
-	cfgMap := buildConfig(s.cfg, nodeConfig, users, certFile, keyFile)
+	cfgMap := buildConfig(s.cfg, nodeConfig, users, tls)
 	data, err := json.Marshal(cfgMap)
 	if err != nil {
 		return fmt.Errorf("marshal config: %w", err)
@@ -168,15 +223,16 @@ func (s *SingBox) Reload(nodeConfig *panel.NodeConfig, users []panel.User, certF
 
 	// Update routing rules
 	if err := router.UpdateRules(opts.Route.Rules, opts.Route.RuleSet); err != nil {
-		slog.Warn("routing reload failed", "error", err)
+		nlog.Core().Debug("routing reload failed", "error", err)
 	} else {
-		slog.Info("sing-box routing reloaded")
+		nlog.Core().Debug("sing-box routing reloaded")
 	}
 
 	nopFactory := singLog.NewNOPFactory()
 
 	// Configuration hash check for inbound reconstruction
-	configChanged := s.nodeConfig == nil || kernel.ComputeHash(nodeConfig, users) != kernel.ComputeHash(s.nodeConfig, s.users)
+	tlsChanged := !bytes.Equal(s.tls.CertPEM, tls.CertPEM) || !bytes.Equal(s.tls.KeyPEM, tls.KeyPEM)
+	configChanged := tlsChanged || s.nodeConfig == nil || kernel.ComputeHash(nodeConfig, users) != kernel.ComputeHash(s.nodeConfig, s.users)
 
 	for _, inb := range opts.Inbounds {
 		tag := inb.Tag
@@ -212,6 +268,10 @@ func (s *SingBox) Reload(nodeConfig *panel.NodeConfig, users []panel.User, certF
 					if opts, ok := inb.Options.(*option.AnyTLSInboundOptions); ok {
 						err = v.UpdateUsers(opts.Users)
 					}
+				case adapter.UpdatableInbound[option.MieruUser]:
+					if opts, ok := inb.Options.(*option.MieruInboundOptions); ok {
+						err = v.UpdateUsers(opts.Users)
+					}
 				case adapter.UpdatableInbound[auth.User]:
 					switch opts := inb.Options.(type) {
 					case *option.NaiveInboundOptions:
@@ -225,7 +285,7 @@ func (s *SingBox) Reload(nodeConfig *panel.NodeConfig, users []panel.User, certF
 				if err == nil {
 					continue
 				}
-				slog.Warn("incremental update failed, falling back to recreate", "tag", tag, "error", err)
+				nlog.Core().Warn("incremental update failed, falling back to recreate", "tag", tag, "error", err)
 			}
 		}
 
@@ -247,11 +307,10 @@ func (s *SingBox) Reload(nodeConfig *panel.NodeConfig, users []panel.User, certF
 		s.connTracker.SetUserMap(buildUserMap(users))
 	}
 
-	slog.Info("sing-box reloaded (users and routes hot-swapped)", "users", len(users))
+	nlog.Core().Debug("sing-box reloaded", "users", len(users))
 	s.users = users
 	s.nodeConfig = nodeConfig
-	s.certFile = certFile
-	s.keyFile = keyFile
+	s.tls = tls
 	return nil
 }
 
@@ -327,9 +386,6 @@ func (s *SingBox) connTrackerSafe() *ConnTracker {
 }
 
 // SetSpeedLimitFunc configures per-user bandwidth throttling.
-// The function is forwarded to the ConnTracker, which embeds the rate limiter
-// in each tracked connection wrapper. This merges byte counting and rate limiting
-// into a single CountFunc callback, allowing splice/zero-copy paths.
 func (s *SingBox) SetSpeedLimitFunc(fn func(uuid string) *rate.Limiter) {
 	s.mu.Lock()
 	defer s.mu.Unlock()
@@ -339,10 +395,39 @@ func (s *SingBox) SetSpeedLimitFunc(fn func(uuid string) *rate.Limiter) {
 	}
 }
 
+// SetDeviceLimitFunc configures per-user device limit gate-keeping.
+// Connections exceeding the limit are rejected at connect time.
+func (s *SingBox) SetDeviceLimitFunc(fn func(uuid string) (int, bool)) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	s.deviceLimitFunc = fn
+	if s.connTracker != nil {
+		s.connTracker.SetDeviceLimitFunc(fn)
+	}
+}
+
+// UpdateGlobalDevices updates the global device state from panel (for multi-node).
+func (s *SingBox) UpdateGlobalDevices(users map[int][]string) {
+	s.mu.RLock()
+	defer s.mu.RUnlock()
+	if s.connTracker != nil {
+		s.connTracker.UpdateGlobalDevices(users)
+	}
+}
+
+// ClearGlobalDevices clears the global device state (on WS disconnect).
+func (s *SingBox) ClearGlobalDevices() {
+	s.mu.RLock()
+	defer s.mu.RUnlock()
+	if s.connTracker != nil {
+		s.connTracker.ClearGlobalDevices()
+	}
+}
+
 // ─── User management (non-disruptive) ───────────────────────────────────────
 
 // AddUsers hot-swaps users into running inbounds. Zero connection disruption.
-func (s *SingBox) AddUsers(users []panel.User) (int, error) {
+func (s *SingBox) AddUsers(users []model.UserSpec) (int, error) {
 	s.mu.Lock()
 	defer s.mu.Unlock()
 
@@ -354,7 +439,7 @@ func (s *SingBox) AddUsers(users []panel.User) (int, error) {
 	for _, u := range s.users {
 		existing[u.ID] = struct{}{}
 	}
-	var toAdd []panel.User
+	var toAdd []model.UserSpec
 	for _, u := range users {
 		if _, dup := existing[u.ID]; !dup {
 			toAdd = append(toAdd, u)
@@ -364,7 +449,7 @@ func (s *SingBox) AddUsers(users []panel.User) (int, error) {
 		return 0, nil
 	}
 
-	merged := append(append([]panel.User{}, s.users...), toAdd...)
+	merged := append(append([]model.UserSpec{}, s.users...), toAdd...)
 	if err := s.reloadInboundsLocked(merged); err != nil {
 		return 0, err
 	}
@@ -374,7 +459,7 @@ func (s *SingBox) AddUsers(users []panel.User) (int, error) {
 
 // RemoveUsers hot-swaps users out of running inbounds. Zero connection disruption
 // for remaining users.
-func (s *SingBox) RemoveUsers(users []panel.User) (int, error) {
+func (s *SingBox) RemoveUsers(users []model.UserSpec) (int, error) {
 	s.mu.Lock()
 	defer s.mu.Unlock()
 
@@ -386,7 +471,7 @@ func (s *SingBox) RemoveUsers(users []panel.User) (int, error) {
 	for _, u := range users {
 		removeSet[u.ID] = struct{}{}
 	}
-	var kept []panel.User
+	var kept []model.UserSpec
 	removed := 0
 	for _, u := range s.users {
 		if _, rm := removeSet[u.ID]; rm {
@@ -407,7 +492,7 @@ func (s *SingBox) RemoveUsers(users []panel.User) (int, error) {
 }
 
 // UpdateUsers replaces the entire user set atomically via hot-swap.
-func (s *SingBox) UpdateUsers(users []panel.User) (added, removed int, err error) {
+func (s *SingBox) UpdateUsers(users []model.UserSpec) (added, removed int, err error) {
 	s.mu.Lock()
 	defer s.mu.Unlock()
 
@@ -436,8 +521,8 @@ func (s *SingBox) UpdateUsers(users []panel.User) (added, removed int, err error
 
 // reloadInboundsLocked hot-swaps inbound users using UpdatableInbound.
 // Must be called with s.mu held.
-func (s *SingBox) reloadInboundsLocked(users []panel.User) error {
-	cfgMap := buildConfig(s.cfg, s.nodeConfig, users, s.certFile, s.keyFile)
+func (s *SingBox) reloadInboundsLocked(users []model.UserSpec) error {
+	cfgMap := buildConfig(s.cfg, s.nodeConfig, users, s.tls)
 	data, err := json.Marshal(cfgMap)
 	if err != nil {
 		return fmt.Errorf("marshal config: %w", err)
@@ -493,6 +578,10 @@ func (s *SingBox) reloadInboundsLocked(users []panel.User) error {
 				if opts, ok := inb.Options.(*option.AnyTLSInboundOptions); ok {
 					err = v.UpdateUsers(opts.Users)
 				}
+			case adapter.UpdatableInbound[option.MieruUser]:
+				if opts, ok := inb.Options.(*option.MieruInboundOptions); ok {
+					err = v.UpdateUsers(opts.Users)
+				}
 			case adapter.UpdatableInbound[auth.User]:
 				switch opts := inb.Options.(type) {
 				case *option.NaiveInboundOptions:
@@ -506,7 +595,7 @@ func (s *SingBox) reloadInboundsLocked(users []panel.User) error {
 			if err == nil {
 				continue
 			}
-			slog.Warn("incremental update failed, recreating inbound", "tag", tag, "error", err)
+			nlog.Core().Warn("incremental update failed, recreating inbound", "tag", tag, "error", err)
 		}
 
 		_ = im.Remove(tag)
@@ -520,19 +609,18 @@ func (s *SingBox) reloadInboundsLocked(users []panel.User) error {
 		s.connTracker.SetUserMap(buildUserMap(users))
 	}
 
-	slog.Info("sing-box users hot-swapped", "users", len(users))
+	nlog.Core().Debug("sing-box users hot-swapped", "users", len(users))
 	return nil
 }
 
 // ─── Observability ──────────────────────────────────────────────────────────
-func (s *SingBox) GetConnections(_ context.Context) ([]kernel.Connection, error) {
+func (s *SingBox) GetUserTraffic(_ context.Context) (traffic map[int][2]int64, aliveIPs map[int]map[string]bool, connCount int, err error) {
 	ct := s.connTrackerSafe()
 	if ct == nil {
-		return nil, nil
+		return nil, nil, 0, nil
 	}
-	conns := ct.Snapshot()
-	slog.Debug("GetConnections snapshot", "count", len(conns))
-	return conns, nil
+	traffic, aliveIPs, connCount = ct.GetUserTraffic()
+	return traffic, aliveIPs, connCount, nil
 }
 
 // CloseConnection force-closes a specific connection by its ID.
@@ -542,15 +630,25 @@ func (s *SingBox) CloseConnection(_ context.Context, connID string) error {
 		return fmt.Errorf("not running")
 	}
 	if !ct.CloseByID(connID) {
-		slog.Debug("CloseConnection: connection not found (already closed?)", "id", connID)
+		nlog.Core().Debug("CloseConnection: connection not found (already closed?)", "id", connID)
 	}
 	return nil
 }
 
+// CloseUserConnections terminates all connections for a user UUID.
+func (s *SingBox) CloseUserConnections(_ context.Context, uuid string) error {
+	ct := s.connTrackerSafe()
+	if ct == nil {
+		return nil
+	}
+	ct.CloseByUUID(uuid)
+	return nil
+}
+
 // buildUserMap creates a UUID→userID mapping used by ConnTracker to attribute
 // connections to the correct user. All sing-box protocols use the user's UUID
 // as the inbound name/username, so this covers every protocol.
-func buildUserMap(users []panel.User) map[string]int {
+func buildUserMap(users []model.UserSpec) map[string]int {
 	m := make(map[string]int, len(users))
 	for _, u := range users {
 		m[u.UUID] = u.ID

internal/kernel/singbox/singbox_test.go

@@ -0,0 +1,215 @@
+package singbox
+
+import (
+	"context"
+	"errors"
+	"net"
+	"net/netip"
+	"testing"
+	"time"
+
+	"github.com/cedar2025/xboard-node/internal/config"
+	"github.com/sagernet/sing-box/adapter"
+	singM "github.com/sagernet/sing/common/metadata"
+	"golang.org/x/time/rate"
+)
+
+func TestSingBoxCapabilities(t *testing.T) {
+	s := New(config.KernelConfig{Type: "sing-box"})
+	caps := s.Capabilities()
+	if !caps.PerUserSpeedLimit || !caps.DeviceLimit || !caps.AliveIPTracking || !caps.ForceCloseUser {
+		t.Fatalf("unexpected sing-box capabilities: %+v", caps)
+	}
+	if caps.BuiltInTrafficStats || caps.ForceCloseConnection {
+		t.Fatalf("unexpected sing-box capabilities: %+v", caps)
+	}
+	protocols := s.Protocols()
+	found := false
+	for _, protocol := range protocols {
+		if protocol == "hysteria" {
+			found = true
+			break
+		}
+	}
+	if !found {
+		t.Fatalf("Protocols() missing hysteria: %v", protocols)
+	}
+}
+
+
+
+type testConn struct {
+	closed bool
+	reads  [][]byte
+	writes [][]byte
+}
+
+func (c *testConn) Read(b []byte) (int, error) {
+	if len(c.reads) == 0 {
+		return 0, errors.New("eof")
+	}
+	chunk := c.reads[0]
+	c.reads = c.reads[1:]
+	n := copy(b, chunk)
+	return n, nil
+}
+
+func (c *testConn) Write(b []byte) (int, error) {
+	cp := append([]byte(nil), b...)
+	c.writes = append(c.writes, cp)
+	return len(b), nil
+}
+
+func (c *testConn) Close() error { c.closed = true; return nil }
+func (c *testConn) LocalAddr() net.Addr { return &net.TCPAddr{} }
+func (c *testConn) RemoteAddr() net.Addr { return &net.TCPAddr{} }
+func (c *testConn) SetDeadline(time.Time) error { return nil }
+func (c *testConn) SetReadDeadline(time.Time) error { return nil }
+func (c *testConn) SetWriteDeadline(time.Time) error { return nil }
+
+func testInboundContext(uuid, ip string) adapter.InboundContext {
+	return adapter.InboundContext{
+		User:   uuid,
+		Source: singM.Socksaddr{Addr: netip.MustParseAddr(ip)},
+	}
+}
+
+func TestConnTrackerRoutedConnectionTracksTrafficAndAliveIPs(t *testing.T) {
+	tracker := NewConnTracker(0)
+	tracker.SetUserMap(map[string]int{"uuid-1": 1})
+	base := &testConn{reads: [][]byte{[]byte("hello")}}
+
+	wrapped := tracker.RoutedConnection(context.Background(), base, testInboundContext("uuid-1", "1.1.1.1"), nil, nil)
+
+	buf := make([]byte, 16)
+	n, err := wrapped.Read(buf)
+	if err != nil {
+		t.Fatalf("Read() error = %v", err)
+	}
+	if n != 5 {
+		t.Fatalf("Read() bytes = %d, want 5", n)
+	}
+	if _, err := wrapped.Write([]byte("bye")); err != nil {
+		t.Fatalf("Write() error = %v", err)
+	}
+
+	traffic, aliveIPs, connCount := tracker.GetUserTraffic()
+	if got := traffic[1]; got != [2]int64{5, 3} {
+		t.Fatalf("traffic[1] = %v, want [5 3]", got)
+	}
+	if !aliveIPs[1]["1.1.1.1"] {
+		t.Fatal("expected alive IP to include source address")
+	}
+	if connCount != 1 {
+		t.Fatalf("connCount = %d, want 1", connCount)
+	}
+
+	if err := wrapped.Close(); err != nil {
+		t.Fatalf("Close() error = %v", err)
+	}
+	_, aliveIPs, connCount = tracker.GetUserTraffic()
+	if aliveIPs[1] != nil {
+		t.Fatalf("aliveIPs after close = %v, want nil", aliveIPs[1])
+	}
+	if connCount != 0 {
+		t.Fatalf("connCount after close = %d, want 0", connCount)
+	}
+}
+
+func TestConnTrackerRoutedConnectionRejectsWhenDeviceLimitExceeded(t *testing.T) {
+	tracker := NewConnTracker(0)
+	tracker.SetUserMap(map[string]int{"uuid-1": 1})
+	tracker.SetDeviceLimitFunc(func(uuid string) (int, bool) {
+		if uuid != "uuid-1" {
+			return 0, false
+		}
+		return 1, true
+	})
+
+	first := &testConn{}
+	wrapped1 := tracker.RoutedConnection(context.Background(), first, testInboundContext("uuid-1", "1.1.1.1"), nil, nil)
+	if wrapped1 == first {
+		t.Fatal("expected first connection to be wrapped")
+	}
+
+	second := &testConn{}
+	wrapped2 := tracker.RoutedConnection(context.Background(), second, testInboundContext("uuid-1", "2.2.2.2"), nil, nil)
+	if wrapped2 != second {
+		t.Fatal("expected rejected connection to be returned unwrapped")
+	}
+	if !second.closed {
+		t.Fatal("expected rejected connection to be closed")
+	}
+}
+
+func TestConnTrackerCheckDeviceGateMergesFreshGlobalDevices(t *testing.T) {
+	tracker := NewConnTracker(0)
+	tracker.SetUserMap(map[string]int{"uuid-1": 1})
+	us := tracker.users[1]
+	us.addConn("1.1.1.1")
+	tracker.UpdateGlobalDevices(map[int][]string{1: {"9.9.9.9"}})
+
+	if tracker.checkDeviceGate(us, 1, "2.2.2.2", 2) {
+		t.Fatal("expected lexicographically earlier candidate to remain allowed")
+	}
+	if !tracker.checkDeviceGate(us, 1, "99.99.99.99", 2) {
+		t.Fatal("expected merged local+global device state to reject lexicographically later third device")
+	}
+	if tracker.checkDeviceGate(us, 1, "1.1.1.1", 2) {
+		t.Fatal("existing local IP should still be allowed")
+	}
+	if tracker.checkDeviceGate(us, 1, "9.9.9.9", 2) {
+		t.Fatal("existing global IP should still be allowed")
+	}
+}
+
+func TestConnTrackerCloseByIDClosesTrackedConnection(t *testing.T) {
+	tracker := NewConnTracker(0)
+	tracker.SetUserMap(map[string]int{"uuid-1": 1})
+	base := &testConn{}
+	wrapped := tracker.RoutedConnection(context.Background(), base, testInboundContext("uuid-1", "1.1.1.1"), nil, nil)
+	tracked, ok := wrapped.(*trackedConn)
+	if !ok {
+		t.Fatalf("wrapped type = %T, want *trackedConn", wrapped)
+	}
+	if !tracker.CloseByID(tracked.connID) {
+		t.Fatal("CloseByID() = false, want true")
+	}
+	if !base.closed {
+		t.Fatal("expected underlying connection to be closed")
+	}
+}
+
+func TestConnTrackerRateLimitHonorsContextCancellation(t *testing.T) {
+	tracker := NewConnTracker(0)
+	tracker.SetUserMap(map[string]int{"uuid-1": 1})
+	tracker.SetSpeedLimitFunc(func(uuid string) *rate.Limiter {
+		if uuid != "uuid-1" {
+			return nil
+		}
+		return rate.NewLimiter(rate.Limit(1), 1)
+	})
+
+	readCtx, cancelRead := context.WithCancel(context.Background())
+	readBase := &testConn{reads: [][]byte{[]byte("a"), []byte("a")}}
+	readTracked := tracker.RoutedConnection(readCtx, readBase, testInboundContext("uuid-1", "1.1.1.1"), nil, nil).(*trackedConn)
+	buf := make([]byte, 8)
+	if n, err := readTracked.Read(buf); err != nil || n != 1 {
+		t.Fatalf("first Read() = (%d, %v), want (1, nil)", n, err)
+	}
+	cancelRead()
+	if n, err := readTracked.Read(buf); !errors.Is(err, context.Canceled) || n != 1 {
+		t.Fatalf("second Read() = (%d, %v), want (1, context canceled)", n, err)
+	}
+
+	writeCtx, cancelWrite := context.WithCancel(context.Background())
+	writeBase := &testConn{}
+	writeTracked := tracker.RoutedConnection(writeCtx, writeBase, testInboundContext("uuid-1", "1.1.1.2"), nil, nil).(*trackedConn)
+	if n, err := writeTracked.Write([]byte("a")); err != nil || n != 1 {
+		t.Fatalf("first Write() = (%d, %v), want (1, nil)", n, err)
+	}
+	cancelWrite()
+	if n, err := writeTracked.Write([]byte("a")); !errors.Is(err, context.Canceled) || n != 0 {
+		t.Fatalf("second Write() = (%d, %v), want (0, context canceled)", n, err)
+	}
+}

internal/kernel/xray/config.go

@@ -1,19 +1,22 @@
 package xray
 
 import (
+	"encoding/base64"
+	"encoding/pem"
 	"fmt"
-	"log/slog"
+	"os"
 	"strings"
 
 	"github.com/cedar2025/xboard-node/internal/config"
 	"github.com/cedar2025/xboard-node/internal/kernel"
-	"github.com/cedar2025/xboard-node/internal/panel"
+	"github.com/cedar2025/xboard-node/internal/model"
+	"github.com/cedar2025/xboard-node/internal/nlog"
 )
 
 // M is a shorthand for building JSON-like maps
 type M = map[string]interface{}
 
-func buildConfig(kcfg config.KernelConfig, nc *panel.NodeConfig, users []panel.User, certFile, keyFile string) M {
+func buildConfig(kcfg config.KernelConfig, nc *model.NodeSpec, users []model.UserSpec, tc kernel.TLSCert) M {
 	var outbounds []M
 	tags := make(map[string]bool)
 
@@ -44,8 +47,8 @@ func buildConfig(kcfg config.KernelConfig, nc *panel.NodeConfig, users []panel.U
 	cfg := M{
 		"log": M{
 			"loglevel": xrayLogLevel(kcfg.LogLevel),
-			"error":    "stdout",
-			"access":   "stdout",
+			"error":    "",
+			"access":   "",
 		},
 		"stats": M{},
 		"policy": M{
@@ -65,17 +68,17 @@ func buildConfig(kcfg config.KernelConfig, nc *panel.NodeConfig, users []panel.U
 		"outbounds": outbounds,
 	}
 
-	inbound := buildInbound(nc, users, certFile, keyFile)
+	inbound := buildInbound(nc, users, tc)
 	if inbound != nil {
 		cfg["inbounds"] = []M{inbound}
 	} else {
-		slog.Warn("xray: unsupported protocol, no inbound configured — node will not accept connections",
+		nlog.Core().Warn("xray: unsupported protocol, no inbound configured — node will not accept connections",
 			"protocol", nc.Protocol,
-			"supported", "vmess, vless, trojan, shadowsocks, socks, http")
+			"supported", "vmess, vless, trojan, shadowsocks, hysteria, socks, http")
 	}
 
 	// Merge panel routes and static config routes
-	cfg["routing"] = buildRouting(nc.Routes, mergeRouteList(nc.CustomRoutes, kcfg.CustomRoute))
+	cfg["routing"] = buildRouting(nc.Routes, nc.CustomRouteRules, mergeRouteList(nc.CustomRoutes, kcfg.CustomRoute))
 
 	mergeCustomXray(cfg, kcfg)
 	return cfg
@@ -84,7 +87,7 @@ func buildConfig(kcfg config.KernelConfig, nc *panel.NodeConfig, users []panel.U
 // outboundConfigToXray converts a structured OutboundConfig (from the panel)
 // into an Xray outbound object. Xray uses a nested layout where protocol-
 // specific fields go inside a "settings" key, and chain proxying uses "proxySettings".
-func outboundConfigToXray(oc panel.OutboundConfig) M {
+func outboundConfigToXray(oc model.OutboundConfig) M {
 	m := M{
 		"protocol": oc.Protocol,
 		"tag":      oc.Tag,
@@ -115,7 +118,7 @@ func mergeRouteList(a, b []map[string]any) []map[string]any {
 func mergeCustomXray(cfg M, kcfg config.KernelConfig) {
 	custom, err := kernel.LoadCustomConfig(kcfg.CustomConfig)
 	if err != nil {
-		slog.Error("failed to load custom xray config", "error", err)
+		nlog.Core().Error("failed to load custom xray config", "error", err)
 		return
 	}
 	if custom == nil {
@@ -194,7 +197,7 @@ func xrayLogLevel(singboxLevel string) string {
 	}
 }
 
-func buildInbound(nc *panel.NodeConfig, users []panel.User, certFile, keyFile string) M {
+func buildInbound(nc *model.NodeSpec, users []model.UserSpec, tc kernel.TLSCert) M {
 	listenAddr := "::"
 	if nc.ListenIP != "" {
 		listenAddr = nc.ListenIP
@@ -213,17 +216,19 @@ func buildInbound(nc *panel.NodeConfig, users []panel.User, certFile, keyFile st
 
 	switch nc.Protocol {
 	case "vmess":
-		return buildVMess(base, nc, users, certFile, keyFile)
+		return buildVMess(base, nc, users, tc)
 	case "vless":
-		return buildVLESS(base, nc, users, certFile, keyFile)
+		return buildVLESS(base, nc, users, tc)
 	case "trojan":
-		return buildTrojan(base, nc, users, certFile, keyFile)
+		return buildTrojan(base, nc, users, tc)
 	case "shadowsocks":
 		return buildShadowsocks(base, nc, users)
 	case "socks":
 		return buildSocks(base, users)
 	case "http":
-		return buildHTTP(base, nc, users, certFile, keyFile)
+		return buildHTTP(base, nc, users, tc)
+	case "hysteria":
+		return buildHysteria(base, nc, users, tc)
 	default:
 		return nil
 	}
@@ -235,7 +240,7 @@ func userEmail(userID int) string {
 	return fmt.Sprintf("user@%d", userID)
 }
 
-func buildVMess(base M, nc *panel.NodeConfig, users []panel.User, certFile, keyFile string) M {
+func buildVMess(base M, nc *model.NodeSpec, users []model.UserSpec, tc kernel.TLSCert) M {
 	clients := make([]M, 0, len(users))
 	for _, u := range users {
 		clients = append(clients, M{
@@ -246,11 +251,11 @@ func buildVMess(base M, nc *panel.NodeConfig, users []panel.User, certFile, keyF
 	}
 	base["settings"] = M{"clients": clients}
 
-	applyStreamSettings(base, nc, certFile, keyFile)
+	applyStreamSettings(base, nc, tc)
 	return base
 }
 
-func buildVLESS(base M, nc *panel.NodeConfig, users []panel.User, certFile, keyFile string) M {
+func buildVLESS(base M, nc *model.NodeSpec, users []model.UserSpec, tc kernel.TLSCert) M {
 	clients := make([]M, 0, len(users))
 	for _, u := range users {
 		client := M{
@@ -262,61 +267,102 @@ func buildVLESS(base M, nc *panel.NodeConfig, users []panel.User, certFile, keyF
 		}
 		clients = append(clients, client)
 	}
+	decryption := "none"
+	if nc.Decryption != "" {
+		decryption = nc.Decryption
+	}
 	base["settings"] = M{
 		"clients":    clients,
-		"decryption": "none",
+		"decryption": decryption,
 	}
 
-	applyStreamSettings(base, nc, certFile, keyFile)
+	applyStreamSettings(base, nc, tc)
 	return base
 }
 
-func buildTrojan(base M, nc *panel.NodeConfig, users []panel.User, certFile, keyFile string) M {
-	clients := make([]M, 0, len(users))
-	for _, u := range users {
-		clients = append(clients, M{
+func buildTrojan(base M, nc *model.NodeSpec, users []model.UserSpec, tc kernel.TLSCert) M {
+	clients := make([]M, len(users))
+	for i := range users {
+		u := &users[i]
+		clients[i] = M{
 			"password": u.UUID,
 			"email":    userEmail(u.ID),
-		})
+		}
 	}
 	base["settings"] = M{"clients": clients}
 
-	applyStreamSettings(base, nc, certFile, keyFile)
+	applyStreamSettings(base, nc, tc)
+
+	// Trojan requires TLS or Reality to be enabled.
+	// If the panel didn't explicitly set TLS=1 or TLS=2, but we have certs,
+	// we should enable a default TLS config to ensure the inbound can start.
+	ss, _ := base["streamSettings"].(M)
+	if security, ok := ss["security"].(string); !ok || (security != "tls" && security != "reality") {
+		nc.TLS = 1 // Force internal state to trigger TLS build in applyStreamSettings
+		applyStreamSettings(base, nc, tc)
+	}
+
 	return base
 }
 
-func buildShadowsocks(base M, nc *panel.NodeConfig, users []panel.User) M {
-	// Single-user mode for traditional ciphers
-	if !strings.HasPrefix(nc.Cipher, "2022-blake3-") {
-		if len(users) > 0 {
-			base["settings"] = M{
-				"method":   nc.Cipher,
-				"password": users[0].UUID,
-				"email":    userEmail(users[0].ID),
-				"network":  "tcp,udp",
-			}
-		}
-		return base
-	}
+type ss2022Config struct {
+	method string
+	size   int
+}
+
+var ss2022Methods = map[string]ss2022Config{
+	"2022-blake3-aes-128-gcm":       {"2022-blake3-aes-128-gcm", 16},
+	"2022-blake3-aes-256-gcm":       {"2022-blake3-aes-256-gcm", 32},
+	"2022-blake3-chacha20-poly1305": {"2022-blake3-chacha20-poly1305", 32},
+}
+
+func buildShadowsocks(base M, nc *model.NodeSpec, users []model.UserSpec) M {
+	ss2022, isSS2022 := ss2022Methods[nc.Cipher]
 
-	// Multi-user mode for 2022-blake3 ciphers
 	clients := make([]M, 0, len(users))
-	for _, u := range users {
-		clients = append(clients, M{
-			"password": u.UUID,
-			"email":    userEmail(u.ID),
-		})
-	}
-	base["settings"] = M{
-		"method":   nc.Cipher,
-		"password": nc.ServerKey,
-		"clients":  clients,
-		"network":  "tcp,udp",
+
+	if isSS2022 {
+		// SS2022: server key at top level, per-user key must be Base64 of fixed-length raw bytes.
+		// Only blake3-aes-* supports multi-user in Xray; chacha20 variant is single-user only.
+		rawBuf := make([]byte, ss2022.size)
+		for i := range users {
+			u := &users[i]
+			for j := range rawBuf {
+				rawBuf[j] = 0
+			}
+			copy(rawBuf, u.UUID)
+			clients = append(clients, M{
+				"password": base64.StdEncoding.EncodeToString(rawBuf),
+				"email":    userEmail(u.ID),
+			})
+		}
+		base["settings"] = M{
+			"method":   nc.Cipher,
+			"password": nc.ServerKey,
+			"clients":  clients,
+			"network":  "tcp,udp",
+		}
+	} else {
+		// Traditional ciphers: multi-user via clients array.
+		// Each entry must carry its own "method" field for Xray to parse correctly.
+		for i := range users {
+			u := &users[i]
+			clients = append(clients, M{
+				"method":   nc.Cipher,
+				"password": u.UUID,
+				"email":    userEmail(u.ID),
+			})
+		}
+		base["settings"] = M{
+			"method":  nc.Cipher,
+			"clients": clients,
+			"network": "tcp,udp",
+		}
 	}
 	return base
 }
 
-func buildSocks(base M, users []panel.User) M {
+func buildSocks(base M, users []model.UserSpec) M {
 	base["protocol"] = "socks"
 	accounts := make([]M, 0, len(users))
 	for _, u := range users {
@@ -334,7 +380,7 @@ func buildSocks(base M, users []panel.User) M {
 	return base
 }
 
-func buildHTTP(base M, nc *panel.NodeConfig, users []panel.User, certFile, keyFile string) M {
+func buildHTTP(base M, nc *model.NodeSpec, users []model.UserSpec, tc kernel.TLSCert) M {
 	base["protocol"] = "http"
 	accounts := make([]M, 0, len(users))
 	for _, u := range users {
@@ -349,12 +395,71 @@ func buildHTTP(base M, nc *panel.NodeConfig, users []panel.User, certFile, keyFi
 	}
 
 	if nc.TLS == 1 {
-		applyStreamSettings(base, nc, certFile, keyFile)
+		applyStreamSettings(base, nc, tc)
 	}
 	return base
 }
 
-func applyStreamSettings(base M, nc *panel.NodeConfig, certFile, keyFile string) {
+// buildHysteria creates a Hysteria v2 inbound for xray-core.
+func buildHysteria(base M, nc *model.NodeSpec, users []model.UserSpec, tc kernel.TLSCert) M {
+	if nc.Version != 2 {
+		nlog.Core().Warn("xray: only supports hysteria v2, skipping v1 inbound")
+		return nil
+	}
+
+	clients := make([]M, 0, len(users))
+	for _, u := range users {
+		clients = append(clients, M{
+			"auth":  u.UUID,
+			"email": userEmail(u.ID),
+		})
+	}
+	base["settings"] = M{"clients": clients}
+
+	ss := M{
+		"network": "hysteria",
+		"hysteriaSettings": M{
+			"version":        nc.Version,
+			"udpIdleTimeout": 60,
+		},
+	}
+
+	if nc.UpMbps > 0 || nc.DownMbps > 0 {
+		quicParams := M{}
+		if nc.UpMbps > 0 {
+			quicParams["brutalUp"] = fmt.Sprintf("%d mbps", nc.UpMbps)
+		}
+		if nc.DownMbps > 0 {
+			quicParams["brutalDown"] = fmt.Sprintf("%d mbps", nc.DownMbps)
+		}
+		ss["finalMask"] = M{
+			"quicParams": quicParams,
+		}
+	}
+
+	if tc.HasCert() {
+		ss["security"] = "tls"
+		tlsCert := M{
+			"certificate": []string{string(tc.CertPEM)},
+			"key":         []string{string(tc.KeyPEM)},
+		}
+		ss["tlsSettings"] = M{
+			"certificates": []M{tlsCert},
+			"alpn":         []string{"h3"},
+		}
+	} else {
+		nlog.Core().Warn("hysteria requires TLS certificate files; configure cert_mode (self, file, http, dns, or content)")
+	}
+
+	if nc.GetProxyProtocol() {
+		ss["sockopt"] = M{"acceptProxyProtocol": true}
+	}
+
+	base["streamSettings"] = ss
+	return base
+}
+
+func applyStreamSettings(base M, nc *model.NodeSpec, tc kernel.TLSCert) {
 	ss := M{}
 
 	// Network / transport
@@ -391,7 +496,9 @@ func applyStreamSettings(base M, nc *panel.NodeConfig, certFile, keyFile string)
 	case "grpc":
 		grpcSettings := M{}
 		if nc.NetworkSettings != nil {
-			if v, ok := nc.NetworkSettings["service_name"]; ok {
+			if v, ok := nc.NetworkSettings["serviceName"]; ok {
+				grpcSettings["serviceName"] = v
+			} else if v, ok := nc.NetworkSettings["service_name"]; ok {
 				grpcSettings["serviceName"] = v
 			}
 		}
@@ -422,6 +529,31 @@ func applyStreamSettings(base M, nc *panel.NodeConfig, certFile, keyFile string)
 		}
 		ss["httpSettings"] = h2Settings
 
+	case "xhttp", "splithttp":
+		ss["network"] = "xhttp"
+		xhttpSettings := M{}
+		if nc.NetworkSettings != nil {
+			if v, ok := nc.NetworkSettings["path"]; ok {
+				xhttpSettings["path"] = v
+			}
+			if v, ok := nc.NetworkSettings["host"]; ok {
+				xhttpSettings["host"] = v
+			}
+			if v, ok := nc.NetworkSettings["mode"]; ok {
+				xhttpSettings["mode"] = v
+			}
+			if v, ok := nc.NetworkSettings["extra"]; ok {
+				// PHP sends empty arrays [] instead of {} for empty objects;
+				// xray rejects [] for fields that expect objects (e.g. sockopt, tlsSettings).
+				// Recursively strip empty arrays from the extra map before passing to xray.
+				if m, ok := v.(map[string]interface{}); ok && len(m) > 0 {
+					sanitizeEmptyArrays(m)
+					xhttpSettings["extra"] = m
+				}
+			}
+		}
+		ss["xhttpSettings"] = xhttpSettings
+
 	case "tcp":
 		// default, no extra settings
 	}
@@ -440,14 +572,17 @@ func applyStreamSettings(base M, nc *panel.NodeConfig, certFile, keyFile string)
 			if sn, ok := nc.TLSSettings["server_name"]; ok && sn != "" {
 				tlsSettings["serverName"] = sn
 			}
-		}
-		if certFile != "" && keyFile != "" {
-			tlsSettings["certificates"] = []M{
-				{
-					"certificateFile": certFile,
-					"keyFile":         keyFile,
-				},
+			// ECH server-side: extract key from tls_settings.ech
+			if echKeys := extractECHServerKeys(nc.TLSSettings); echKeys != "" {
+				tlsSettings["echServerKeys"] = echKeys
 			}
+		}
+		if tc.HasCert() {
+			tlsCert := M{
+				"certificate": []string{string(tc.CertPEM)},
+				"key":         []string{string(tc.KeyPEM)},
+			}
+			tlsSettings["certificates"] = []M{tlsCert}
 		} else {
 			// Fallback placeholder for auto-TLS environments.
 			// Xray allows empty certificates array in more cases than sing-box,
@@ -461,7 +596,7 @@ func applyStreamSettings(base M, nc *panel.NodeConfig, certFile, keyFile string)
 	}
 
 	// Proxy Protocol
-	if nc.AcceptProxyProtocol {
+	if nc.GetProxyProtocol() {
 		sockopt, ok := base["streamSettings"].(M)["sockopt"].(M)
 		if !ok {
 			sockopt = M{}
@@ -473,7 +608,7 @@ func applyStreamSettings(base M, nc *panel.NodeConfig, certFile, keyFile string)
 	base["streamSettings"] = ss
 }
 
-func buildRealitySettings(nc *panel.NodeConfig) M {
+func buildRealitySettings(nc *model.NodeSpec) M {
 	reality := M{"show": false}
 
 	if nc.TLSSettings == nil {
@@ -510,10 +645,18 @@ func buildRealitySettings(nc *panel.NodeConfig) M {
 	return reality
 }
 
-func buildRouting(rules []panel.RouteRule, customRules []map[string]any) M {
+func buildRouting(rules []model.RouteRule, customRouteRules []model.CustomRouteRule, customRules []map[string]any) M {
 	var xrayRules []M
 
-	// Custom route rules (from Panel CustomRoutes or local config) have HIGHEST priority
+	// Structured custom routes now take the highest priority for panel-managed overrides.
+	for _, rule := range customRouteRules {
+		if rule.Disabled {
+			continue
+		}
+		xrayRules = append(xrayRules, compileCustomRouteRule(rule)...)
+	}
+
+	// Raw custom routes remain the escape hatch, but no longer outrank structured rules.
 	for _, cr := range customRules {
 		xrayRules = append(xrayRules, M(cr))
 	}
@@ -537,49 +680,7 @@ func buildRouting(rules []panel.RouteRule, customRules []map[string]any) M {
 	})
 
 	for _, rule := range rules {
-		if len(rule.Match) == 0 {
-			continue
-		}
-
-		var ips, domains []string
-		for _, m := range rule.Match {
-			m = strings.TrimSpace(m)
-			if m == "" {
-				continue
-			}
-			if strings.HasPrefix(m, "geoip:") {
-				ips = append(ips, m) // "geoip:cn" → ip field, passed as-is
-			} else if strings.HasPrefix(m, "geosite:") {
-				domains = append(domains, m) // "geosite:google" → domain field, passed as-is
-			} else if strings.Contains(m, "/") {
-				ips = append(ips, m)
-			} else {
-				m = strings.TrimPrefix(m, "*.")
-				domains = append(domains, "domain:"+m)
-			}
-		}
-
-		outbound := "block"
-		if rule.Action == "direct" {
-			outbound = "direct"
-		} else if rule.Action == "proxy" && rule.ActionValue != "" {
-			outbound = rule.ActionValue
-		}
-
-		if len(domains) > 0 {
-			xrayRules = append(xrayRules, M{
-				"type":        "field",
-				"domain":      domains,
-				"outboundTag": outbound,
-			})
-		}
-		if len(ips) > 0 {
-			xrayRules = append(xrayRules, M{
-				"type":        "field",
-				"ip":          ips,
-				"outboundTag": outbound,
-			})
-		}
+		xrayRules = append(xrayRules, compilePanelRouteRule(rule)...)
 	}
 
 	return M{
@@ -587,3 +688,194 @@ func buildRouting(rules []panel.RouteRule, customRules []map[string]any) M {
 		"rules":          xrayRules,
 	}
 }
+
+func compilePanelRouteRule(rule model.RouteRule) []M {
+	if len(rule.Match) == 0 {
+		return nil
+	}
+	match := model.RouteMatch{}
+	for _, item := range rule.Match {
+		item = strings.TrimSpace(item)
+		if item == "" {
+			continue
+		}
+		if strings.HasPrefix(item, "geoip:") || strings.Contains(item, "/") {
+			match.IPCIDRs = append(match.IPCIDRs, item)
+			continue
+		}
+		if strings.HasPrefix(item, "geosite:") {
+			match.Domains = append(match.Domains, item)
+			continue
+		}
+		match.DomainSuffixes = append(match.DomainSuffixes, strings.TrimPrefix(item, "*."))
+	}
+	action := model.RouteAction{Type: "block"}
+	switch rule.Action {
+	case "direct":
+		action.Type = "direct"
+	case "proxy":
+		action.Type = "route"
+		action.Target = rule.ActionValue
+	}
+	return compileCustomRouteRule(model.CustomRouteRule{Match: match, Action: action})
+}
+
+func compileCustomRouteRule(rule model.CustomRouteRule) []M {
+	outbound := xrayOutboundForAction(rule.Action)
+	var compiled []M
+
+	if len(rule.Match.Domains) > 0 || len(rule.Match.DomainSuffixes) > 0 {
+		domains := make([]string, 0, len(rule.Match.Domains)+len(rule.Match.DomainSuffixes))
+		for _, value := range rule.Match.Domains {
+			value = strings.TrimSpace(value)
+			if value == "" {
+				continue
+			}
+			domains = append(domains, value)
+		}
+		for _, value := range rule.Match.DomainSuffixes {
+			value = strings.TrimSpace(value)
+			if value == "" {
+				continue
+			}
+			domains = append(domains, "domain:"+value)
+		}
+		if len(domains) > 0 {
+			compiled = append(compiled, M{
+				"type":        "field",
+				"domain":      domains,
+				"outboundTag": outbound,
+			})
+		}
+	}
+	if len(rule.Match.IPCIDRs) > 0 {
+		compiled = append(compiled, M{
+			"type":        "field",
+			"ip":          copyStrings(rule.Match.IPCIDRs),
+			"outboundTag": outbound,
+		})
+	}
+	if len(rule.Match.Ports) > 0 {
+		compiled = append(compiled, M{
+			"type":        "field",
+			"port":        strings.Join(rule.Match.Ports, ","),
+			"outboundTag": outbound,
+		})
+	}
+	if len(rule.Match.Networks) > 0 {
+		compiled = append(compiled, M{
+			"type":        "field",
+			"network":     strings.Join(rule.Match.Networks, ","),
+			"outboundTag": outbound,
+		})
+	}
+	if len(rule.Match.SourceCIDRs) > 0 {
+		compiled = append(compiled, M{
+			"type":        "field",
+			"source":      copyStrings(rule.Match.SourceCIDRs),
+			"outboundTag": outbound,
+		})
+	}
+	if len(rule.Match.SourcePorts) > 0 {
+		compiled = append(compiled, M{
+			"type":        "field",
+			"sourcePort":  strings.Join(rule.Match.SourcePorts, ","),
+			"outboundTag": outbound,
+		})
+	}
+	return compiled
+}
+
+func xrayOutboundForAction(action model.RouteAction) string {
+	switch action.Type {
+	case "direct":
+		return "direct"
+	case "route":
+		if action.Target != "" {
+			return action.Target
+		}
+		return "block"
+	default:
+		return "block"
+	}
+}
+
+func copyStrings(src []string) []string {
+	if len(src) == 0 {
+		return nil
+	}
+	out := make([]string, len(src))
+	copy(out, src)
+	return out
+}
+
+// sanitizeEmptyArrays recursively removes empty slices from a map.
+// PHP encodes empty objects as [] instead of {}; xray-core rejects []
+// for fields that expect struct types (e.g. sockopt, tlsSettings).
+func sanitizeEmptyArrays(m map[string]interface{}) {
+	for k, v := range m {
+		switch val := v.(type) {
+		case []interface{}:
+			if len(val) == 0 {
+				delete(m, k)
+			}
+		case map[string]interface{}:
+			sanitizeEmptyArrays(val)
+		}
+	}
+}
+
+// extractECHServerKeys extracts the ECH private key from tls_settings.ech
+// and converts it from PEM to base64 for xray-core's echServerKeys field.
+func extractECHServerKeys(tlsSettings map[string]interface{}) string {
+	echRaw, ok := tlsSettings["ech"]
+	if !ok {
+		return ""
+	}
+	echMap, ok := echRaw.(map[string]interface{})
+	if !ok {
+		return ""
+	}
+	enabled, _ := echMap["enabled"].(bool)
+	if !enabled {
+		return ""
+	}
+
+	// Try inline key first, then key_path
+	var pemData []byte
+	if key, _ := echMap["key"].(string); key != "" {
+		pemData = []byte(key)
+	} else if keyPath, _ := echMap["key_path"].(string); keyPath != "" {
+		data, err := os.ReadFile(keyPath)
+		if err != nil {
+			nlog.Core().Warn("ECH enabled but key_path read failed", "error", err)
+			return ""
+		}
+		pemData = data
+	} else {
+		nlog.Core().Warn("ECH enabled but no key or key_path provided")
+		return ""
+	}
+
+	return echPEMToBase64(pemData)
+}
+
+// echPEMToBase64 parses an "ECH KEYS" PEM block and returns base64 of the raw bytes.
+// If the input is not valid PEM or has the wrong type, it returns empty string.
+func echPEMToBase64(data []byte) string {
+	trimmed := strings.TrimSpace(string(data))
+	if !strings.Contains(trimmed, "-----") {
+		// Assume already raw base64
+		return trimmed
+	}
+	block, _ := pem.Decode(data)
+	if block == nil {
+		nlog.Core().Warn("ECH key: invalid PEM data")
+		return ""
+	}
+	if block.Type != "ECH KEYS" {
+		nlog.Core().Warn("ECH key: unexpected PEM type", "expected", "ECH KEYS", "got", block.Type)
+		return ""
+	}
+	return base64.StdEncoding.EncodeToString(block.Bytes)
+}

internal/kernel/xray/config_test.go

@@ -1,10 +1,14 @@
 package xray
 
 import (
+	"encoding/base64"
 	"encoding/json"
+	"encoding/pem"
 	"testing"
 
 	"github.com/cedar2025/xboard-node/internal/config"
+	"github.com/cedar2025/xboard-node/internal/kernel"
+	"github.com/cedar2025/xboard-node/internal/model"
 	"github.com/cedar2025/xboard-node/internal/panel"
 )
 
@@ -13,11 +17,22 @@ var testKernelCfg = config.KernelConfig{
 	LogLevel: "warn",
 }
 
-var testUsers = []panel.User{
+var testUsersPanel = []panel.User{
 	{ID: 1, UUID: "279d4f89-3a2c-488d-a67c-2d39a72acdde"},
 	{ID: 5, UUID: "4d5965c8-a60c-452a-a943-af83ec0bb0db"},
 }
 
+var testUsers = model.UserSpecsFromPanel(testUsersPanel)
+
+func testNodeSpec(nc *panel.NodeConfig) *model.NodeSpec { return model.NodeSpecFromPanel(nc) }
+
+func testRouteRules(r []panel.RouteRule) []model.RouteRule {
+	if r == nil {
+		return nil
+	}
+	return model.NodeSpecFromPanel(&panel.NodeConfig{Routes: r}).Routes
+}
+
 func TestBuildConfig_OutboundPriority(t *testing.T) {
 	kcfg := config.KernelConfig{
 		LogLevel: "info",
@@ -34,7 +49,7 @@ func TestBuildConfig_OutboundPriority(t *testing.T) {
 		},
 	}
 
-	cfg := buildConfig(kcfg, nc, testUsers, "", "")
+	cfg := buildConfig(kcfg, testNodeSpec(nc), testUsers, kernel.TLSCert{})
 	outbounds := cfg["outbounds"].([]M)
 
 	// Since we overrode both 'direct' and 'block', the result should contain
@@ -149,7 +164,7 @@ func TestBuildConfig_AllProtocols_ValidJSON(t *testing.T) {
 
 	for _, tc := range protocols {
 		t.Run(tc.name, func(t *testing.T) {
-			cfg := buildConfig(testKernelCfg, &tc.nc, testUsers, "/cert.pem", "/key.pem")
+			cfg := buildConfig(testKernelCfg, testNodeSpec(&tc.nc), testUsers, kernel.TLSCert{CertPEM: []byte("CERT"), KeyPEM: []byte("KEY")})
 
 			data, err := json.Marshal(cfg)
 			if err != nil {
@@ -182,7 +197,7 @@ func TestBuildConfig_VMess_Users(t *testing.T) {
 		Protocol:   "vmess",
 		ServerPort: 10086,
 	}
-	cfg := buildConfig(testKernelCfg, &nc, testUsers, "", "")
+	cfg := buildConfig(testKernelCfg, testNodeSpec(&nc), testUsers, kernel.TLSCert{})
 	data, _ := json.Marshal(cfg)
 
 	var parsed map[string]interface{}
@@ -222,7 +237,7 @@ func TestBuildConfig_VLESS_Flow(t *testing.T) {
 			"server_name": "example.com",
 		},
 	}
-	cfg := buildConfig(testKernelCfg, &nc, testUsers, "", "")
+	cfg := buildConfig(testKernelCfg, testNodeSpec(&nc), testUsers, kernel.TLSCert{})
 	data, _ := json.Marshal(cfg)
 
 	var parsed map[string]interface{}
@@ -245,7 +260,7 @@ func TestBuildConfig_VLESS_Flow(t *testing.T) {
 }
 
 func TestBuildRouting_Default(t *testing.T) {
-	routing := buildRouting(nil, nil)
+	routing := buildRouting(nil, nil, nil)
 	rules := routing["rules"].([]M)
 
 	if len(rules) != 1 {
@@ -275,7 +290,7 @@ func TestBuildRouting_WithRules(t *testing.T) {
 		},
 	}
 
-	routing := buildRouting(rules, nil)
+	routing := buildRouting(testRouteRules(rules), nil, nil)
 	xrayRules := routing["rules"].([]M)
 
 	// 1 default + 2 domain rules + 1 IP rule = 4
@@ -310,6 +325,63 @@ func TestBuildRouting_WithRules(t *testing.T) {
 	}
 }
 
+func TestBuildRouting_WithCustomRouteRules(t *testing.T) {
+	customRules := []model.CustomRouteRule{
+		{
+			Name: "proxy-web",
+			Match: model.RouteMatch{
+				Domains:        []string{"full.example.com"},
+				DomainSuffixes: []string{"example.org"},
+				Ports:          []string{"80", "443-445"},
+				Networks:       []string{"tcp", "udp"},
+				SourceCIDRs:    []string{"192.168.1.0/24"},
+				SourcePorts:    []string{"1000-1002"},
+			},
+			Action: model.RouteAction{Type: "route", Target: "warp-jp"},
+		},
+	}
+
+	routing := buildRouting(nil, customRules, nil)
+	xrayRules := routing["rules"].([]M)
+	if len(xrayRules) != 6 {
+		t.Fatalf("expected 6 rules, got %d", len(xrayRules))
+	}
+	if xrayRules[0]["outboundTag"] != "warp-jp" {
+		t.Fatalf("expected first custom outbound warp-jp, got %v", xrayRules[0]["outboundTag"])
+	}
+	if got := xrayRules[0]["domain"].([]string); len(got) != 2 || got[0] != "full.example.com" || got[1] != "domain:example.org" {
+		t.Fatalf("unexpected custom domains: %v", got)
+	}
+	if got := xrayRules[1]["port"]; got != "80,443-445" {
+		t.Fatalf("unexpected port matcher: %v", got)
+	}
+	if got := xrayRules[2]["network"]; got != "tcp,udp" {
+		t.Fatalf("unexpected network matcher: %v", got)
+	}
+	if got := xrayRules[3]["source"].([]string); len(got) != 1 || got[0] != "192.168.1.0/24" {
+		t.Fatalf("unexpected source cidr matcher: %v", got)
+	}
+	if got := xrayRules[4]["sourcePort"]; got != "1000-1002" {
+		t.Fatalf("unexpected source port matcher: %v", got)
+	}
+}
+
+func TestBuildRouting_StructuredCustomRulesRemainFirst(t *testing.T) {
+	raw := []map[string]any{{"type": "field", "domain": []string{"keyword:raw"}, "outboundTag": "raw-tag"}}
+	custom := []model.CustomRouteRule{{
+		Match:  model.RouteMatch{DomainSuffixes: []string{"structured.example"}},
+		Action: model.RouteAction{Type: "direct"},
+	}}
+	routing := buildRouting(nil, custom, raw)
+	xrayRules := routing["rules"].([]M)
+	if xrayRules[0]["outboundTag"] != "direct" {
+		t.Fatalf("expected structured rule first, got %v", xrayRules[0]["outboundTag"])
+	}
+	if xrayRules[1]["outboundTag"] != "raw-tag" {
+		t.Fatalf("expected raw custom rule second, got %v", xrayRules[1]["outboundTag"])
+	}
+}
+
 func TestBuildConfig_LogLevel(t *testing.T) {
 	tests := []struct {
 		input    string
@@ -337,7 +409,7 @@ func TestBuildConfig_StatsEnabled(t *testing.T) {
 		Protocol:   "vmess",
 		ServerPort: 10086,
 	}
-	cfg := buildConfig(testKernelCfg, &nc, testUsers, "", "")
+	cfg := buildConfig(testKernelCfg, testNodeSpec(&nc), testUsers, kernel.TLSCert{})
 	data, _ := json.Marshal(cfg)
 
 	var parsed map[string]interface{}
@@ -361,13 +433,13 @@ func TestBuildConfig_StatsEnabled(t *testing.T) {
 	}
 }
 
-func TestBuildConfig_Shadowsocks_SingleUser(t *testing.T) {
+func TestBuildConfig_Shadowsocks_MultiUserTraditional(t *testing.T) {
 	nc := panel.NodeConfig{
 		Protocol:   "shadowsocks",
 		ServerPort: 8388,
 		Cipher:     "aes-128-gcm",
 	}
-	cfg := buildConfig(testKernelCfg, &nc, testUsers, "", "")
+	cfg := buildConfig(testKernelCfg, testNodeSpec(&nc), testUsers, kernel.TLSCert{})
 	data, _ := json.Marshal(cfg)
 
 	var parsed map[string]interface{}
@@ -377,12 +449,19 @@ func TestBuildConfig_Shadowsocks_SingleUser(t *testing.T) {
 	ib := inbounds[0].(map[string]interface{})
 	settings := ib["settings"].(map[string]interface{})
 
-	// Single-user mode: password directly, no clients array
 	if settings["method"] != "aes-128-gcm" {
 		t.Errorf("expected method aes-128-gcm, got %v", settings["method"])
 	}
-	if settings["password"] != testUsers[0].UUID {
-		t.Errorf("expected password %s, got %v", testUsers[0].UUID, settings["password"])
+	clients := settings["clients"].([]interface{})
+	if len(clients) != len(testUsers) {
+		t.Fatalf("expected %d clients, got %d", len(testUsers), len(clients))
+	}
+	c0 := clients[0].(map[string]interface{})
+	if c0["method"] != "aes-128-gcm" {
+		t.Errorf("expected per-user method aes-128-gcm, got %v", c0["method"])
+	}
+	if c0["password"] != testUsers[0].UUID {
+		t.Errorf("expected password %s, got %v", testUsers[0].UUID, c0["password"])
 	}
 }
 
@@ -393,7 +472,7 @@ func TestBuildConfig_Shadowsocks_MultiUser(t *testing.T) {
 		Cipher:     "2022-blake3-aes-128-gcm",
 		ServerKey:  "server-key",
 	}
-	cfg := buildConfig(testKernelCfg, &nc, testUsers, "", "")
+	cfg := buildConfig(testKernelCfg, testNodeSpec(&nc), testUsers, kernel.TLSCert{})
 	data, _ := json.Marshal(cfg)
 
 	var parsed map[string]interface{}
@@ -417,7 +496,7 @@ func TestBuildConfig_SocksStats(t *testing.T) {
 		Protocol:   "socks",
 		ServerPort: 1080,
 	}
-	cfg := buildConfig(testKernelCfg, &nc, testUsers, "", "")
+	cfg := buildConfig(testKernelCfg, testNodeSpec(&nc), testUsers, kernel.TLSCert{})
 	data, _ := json.Marshal(cfg)
 
 	var parsed map[string]interface{}
@@ -427,7 +506,7 @@ func TestBuildConfig_SocksStats(t *testing.T) {
 	ib := inbounds[0].(map[string]interface{})
 	settings := ib["settings"].(map[string]interface{})
 	accounts := settings["accounts"].([]interface{})
-	
+
 	if len(accounts) == 0 {
 		t.Fatal("no accounts in socks config")
 	}
@@ -437,3 +516,59 @@ func TestBuildConfig_SocksStats(t *testing.T) {
 		t.Errorf("expected email user@1 for socks account, got %v", a1["email"])
 	}
 }
+
+func TestEchPEMToBase64(t *testing.T) {
+	// Valid ECH KEYS PEM
+	rawBytes := []byte{0x00, 0x04, 0xDE, 0xAD, 0xBE, 0xEF, 0x00, 0x03, 0xCA, 0xFE, 0x00}
+	pemBlock := &pem.Block{Type: "ECH KEYS", Bytes: rawBytes}
+	pemStr := string(pem.EncodeToMemory(pemBlock))
+
+	result := echPEMToBase64([]byte(pemStr))
+	expected := base64.StdEncoding.EncodeToString(rawBytes)
+	if result != expected {
+		t.Errorf("echPEMToBase64 got %q, want %q", result, expected)
+	}
+
+	// Wrong PEM type should return empty
+	wrongBlock := &pem.Block{Type: "ECH CONFIGS", Bytes: rawBytes}
+	wrongPEM := string(pem.EncodeToMemory(wrongBlock))
+	if got := echPEMToBase64([]byte(wrongPEM)); got != "" {
+		t.Errorf("echPEMToBase64 should reject ECH CONFIGS, got %q", got)
+	}
+
+	// Invalid PEM should return empty
+	if got := echPEMToBase64([]byte("-----BEGIN GARBAGE-----\nwhat\n-----END GARBAGE-----")); got != "" {
+		t.Errorf("echPEMToBase64 should reject invalid PEM, got %q", got)
+	}
+
+	// Raw base64 passthrough
+	raw := "AAQDQKDIAAMD"
+	if got := echPEMToBase64([]byte(raw)); got != raw {
+		t.Errorf("echPEMToBase64 raw passthrough: got %q, want %q", got, raw)
+	}
+}
+
+func TestExtractECHServerKeys(t *testing.T) {
+	rawBytes := []byte{0x00, 0x04, 0xDE, 0xAD, 0xBE, 0xEF, 0x00, 0x03, 0xCA, 0xFE, 0x00}
+	pemBlock := &pem.Block{Type: "ECH KEYS", Bytes: rawBytes}
+	pemStr := string(pem.EncodeToMemory(pemBlock))
+
+	tests := []struct {
+		name   string
+		tls    map[string]interface{}
+		expect string
+	}{
+		{"no ech", map[string]interface{}{}, ""},
+		{"disabled", map[string]interface{}{"ech": map[string]interface{}{"enabled": false, "key": pemStr}}, ""},
+		{"enabled with key", map[string]interface{}{"ech": map[string]interface{}{"enabled": true, "key": pemStr}}, base64.StdEncoding.EncodeToString(rawBytes)},
+		{"enabled no key", map[string]interface{}{"ech": map[string]interface{}{"enabled": true}}, ""},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			got := extractECHServerKeys(tt.tls)
+			if got != tt.expect {
+				t.Errorf("got %q, want %q", got, tt.expect)
+			}
+		})
+	}
+}

internal/kernel/xray/dispatcher.go

@@ -3,10 +3,8 @@ package xray
 import (
 	"context"
 	"errors"
-	"log/slog"
 	"reflect"
 	"sort"
-	"strconv"
 	"sync"
 	"sync/atomic"
 	_ "unsafe"
@@ -18,9 +16,8 @@ import (
 	"github.com/xtls/xray-core/common/session"
 	"github.com/xtls/xray-core/features/routing"
 	"github.com/xtls/xray-core/transport"
-	"golang.org/x/time/rate"
 
-	"github.com/cedar2025/xboard-node/internal/kernel"
+	"github.com/cedar2025/xboard-node/internal/nlog"
 )
 
 // Access xray's internal config creator registry so we can replace the
@@ -52,43 +49,55 @@ func limitDispatcherFactory(ctx context.Context, config interface{}) (interface{
 		return orig, nil
 	}
 	ld := &LimitDispatcher{
-		inner:     orig,
-		innerDisp: inner,
-		conns:     make(map[string]*dispatchedConn),
-		userIPs:   make(map[string]map[string]int),
+		inner:      orig,
+		innerDisp:  inner,
+		limitedIPs: make(map[string]map[string]int),
 	}
 	globalLimitDispatcher.Store(ld)
-	slog.Debug("xray: limit dispatcher installed")
+	nlog.Core().Debug("xray: limit dispatcher installed")
 	return ld, nil
 }
 
-// LimitDispatcher wraps xray's DefaultDispatcher to add per-user device
-// limit enforcement, speed limiting, and per-connection source IP tracking.
-// It hooks into Dispatch/DispatchLink which are called AFTER the protocol
-// handler has identified the user, so it works for ALL protocols.
+// LimitDispatcher wraps xray's DefaultDispatcher to enforce per-user
+// admission checks before a request is dispatched into xray-core.
+//
+// It intentionally does NOT mutate transport.Link.Reader/Writer. Xray's
+// mux/XUDP close path requires the original concrete *pipe.Reader to remain
+// intact, so the dispatcher is limited to gate-keeping and safe connection
+// lifecycle bookkeeping.
 type LimitDispatcher struct {
 	inner     interface{}        // original DefaultDispatcher (Feature + Dispatcher)
 	innerDisp routing.Dispatcher // same object, typed as Dispatcher
 
+	// limitedUsers: users with device limit > 0, protected by mu.
+	// Needs deterministic IP ordering for kick decisions.
 	mu           sync.RWMutex
-	conns        map[string]*dispatchedConn // connID → conn
-	userIPs      map[string]map[string]int  // email → sourceIP → count
-	deviceLimits map[string]int             // email → max devices
-	speedLimits  map[string]int             // email → Mbps
-	speedBuckets sync.Map                   // email → *rate.Limiter
-	emailToUID   map[string]int             // email → panel user ID
+	limitedIPs   map[string]map[string]int // email → sourceIP → refcount
+	deviceLimits map[string]int            // email → max devices
+	emailToUID   map[string]int            // email → panel user ID
 
-	connIDSeq atomic.Uint64
+	// unlimitedIPs: users without device limit — sync.Map for lock-free access.
+	// Each entry is *ipCounter{ips sync.Map}.
+	unlimitedIPs sync.Map // email → *ipCounter
+
+	connCount atomic.Int64 // total active connections tracked by dispatcher
 }
 
-type dispatchedConn struct {
-	id       string
-	email    string
-	sourceIP string
-	userID   int
-	upload   atomic.Int64
-	download atomic.Int64
-	closed   atomic.Bool
+// ipCounter tracks IPs for unlimited users without any lock.
+type ipCounter struct {
+	ips sync.Map // sourceIP → *atomic.Int64 (refcount)
+}
+
+// aliveIPs returns a snapshot of distinct IPs.
+func (ic *ipCounter) aliveIPs() map[string]bool {
+	result := make(map[string]bool)
+	ic.ips.Range(func(key, _ interface{}) bool {
+		if rv, ok := ic.ips.Load(key); ok && rv.(*atomic.Int64).Load() > 0 {
+			result[key.(string)] = true
+		}
+		return true
+	})
+	return result
 }
 
 // ─── routing.Dispatcher ──────────────────────────────────────────────────────
@@ -108,7 +117,7 @@ func (d *LimitDispatcher) Dispatch(ctx context.Context, dest net.Destination) (*
 	}
 
 	if email != "" {
-		d.wrapLink(ctx, link, email, sourceIP, isTCP)
+		d.trackLink(link, email, sourceIP, isTCP)
 	}
 	return link, nil
 }
@@ -120,7 +129,7 @@ func (d *LimitDispatcher) DispatchLink(ctx context.Context, dest net.Destination
 	}
 
 	if email != "" {
-		d.wrapLink(ctx, link, email, sourceIP, isTCP)
+		d.trackLink(link, email, sourceIP, isTCP)
 	}
 	return d.innerDisp.DispatchLink(ctx, dest, link)
 }
@@ -138,51 +147,27 @@ func (d *LimitDispatcher) identifyAndCheck(ctx context.Context, dest net.Destina
 	isTCP = dest.Network == net.Network_TCP
 
 	if d.checkDeviceLimit(email, sourceIP, isTCP) {
-		slog.Info("xray: device limit exceeded", "email", email, "ip", sourceIP)
+		nlog.Core().Debug("xray: device limit exceeded", "email", email, "ip", sourceIP)
 		return "", "", false, errors.New("device limit exceeded for " + email)
 	}
 	return email, sourceIP, isTCP, nil
 }
 
-// wrapLink instruments a transport.Link with per-connection byte counting,
-// rate limiting, and lifecycle tracking. Must only be called when email != "".
-func (d *LimitDispatcher) wrapLink(ctx context.Context, link *transport.Link, email, sourceIP string, isTCP bool) {
-	connID := "xd-" + strconv.FormatUint(d.connIDSeq.Add(1), 36)
-	dc := &dispatchedConn{
-		id:       connID,
-		email:    email,
-		sourceIP: sourceIP,
-		userID:   d.getUID(email),
-	}
-
-	d.mu.Lock()
-	d.conns[connID] = dc
-	d.mu.Unlock()
+// trackLink records connection lifecycle without mutating xray-core owned
+// transport primitives. This keeps mux/XUDP compatible while still allowing
+// the dispatcher to release device-limit state when the link closes.
+func (d *LimitDispatcher) trackLink(link *transport.Link, email, sourceIP string, isTCP bool) {
+	d.connCount.Add(1)
 
 	onClose := func() {
-		if dc.closed.CompareAndSwap(false, true) {
-			if isTCP {
-				d.delConn(email, sourceIP)
-			}
-			d.mu.Lock()
-			delete(d.conns, connID)
-			d.mu.Unlock()
+		if isTCP {
+			d.delConn(email, sourceIP)
 		}
+		d.connCount.Add(-1)
 	}
 
-	limiter := d.getBucket(email)
-
-	link.Reader = &statsCloseReader{
-		Reader:  link.Reader,
-		counter: &dc.upload,
-		limiter: limiter,
-		ctx:     ctx,
-	}
-	link.Writer = &statsCloseWriter{
+	link.Writer = &closeTrackingWriter{
 		Writer:  link.Writer,
-		counter: &dc.download,
-		limiter: limiter,
-		ctx:     ctx,
 		onClose: onClose,
 	}
 }
@@ -207,86 +192,137 @@ func (d *LimitDispatcher) Close() error {
 
 // ─── Limit management (called by Xray kernel) ──────────────────────────────
 
-func (d *LimitDispatcher) UpdateLimits(emailToUID map[string]int, deviceLimits, speedLimits map[string]int) {
+func (d *LimitDispatcher) UpdateLimits(emailToUID map[string]int, deviceLimits, _ map[string]int) {
 	d.mu.Lock()
 	d.emailToUID = emailToUID
 	d.deviceLimits = deviceLimits
-	d.speedLimits = speedLimits
 	d.mu.Unlock()
 
-	d.speedBuckets.Range(func(key, _ interface{}) bool {
-		if _, ok := speedLimits[key.(string)]; !ok {
-			d.speedBuckets.Delete(key)
-		}
-		return true
-	})
 }
 
 func (d *LimitDispatcher) ResetConns() {
 	d.mu.Lock()
-	d.conns = make(map[string]*dispatchedConn)
-	d.userIPs = make(map[string]map[string]int)
+	d.limitedIPs = make(map[string]map[string]int)
 	d.mu.Unlock()
 
-	d.speedBuckets.Range(func(key, _ interface{}) bool {
-		d.speedBuckets.Delete(key)
+	// Clear unlimited IPs
+	d.unlimitedIPs.Range(func(key, _ interface{}) bool {
+		d.unlimitedIPs.Delete(key)
 		return true
 	})
+
+	d.connCount.Store(0)
 }
 
-func (d *LimitDispatcher) Snapshot() []kernel.Connection {
+// GetConnectionState returns dispatcher-tracked alive IPs and connection count.
+// Traffic bytes are intentionally left to xray's built-in stats pipeline.
+func (d *LimitDispatcher) GetConnectionState() (aliveIPs map[int]map[string]bool, connCount int) {
 	d.mu.RLock()
-	defer d.mu.RUnlock()
-
-	result := make([]kernel.Connection, 0, len(d.conns))
-	for _, c := range d.conns {
-		result = append(result, kernel.Connection{
-			ID:       c.id,
-			UserID:   c.userID,
-			SourceIP: c.sourceIP,
-			Upload:   c.upload.Load(),
-			Download: c.download.Load(),
-		})
-	}
-	return result
-}
-
-func (d *LimitDispatcher) CloseConn(id string) bool {
-	d.mu.RLock()
-	c, ok := d.conns[id]
+	emailToUID := d.emailToUID
+	limitedIPs := d.limitedIPs
 	d.mu.RUnlock()
-	if !ok {
-		return false
+
+	aliveIPs = make(map[int]map[string]bool)
+
+	// Collect IPs from limited users (under RLock snapshot).
+	for email, ipsMap := range limitedIPs {
+		uid := emailToUID[email]
+		if uid == 0 {
+			continue
+		}
+		ipSet := make(map[string]bool, len(ipsMap))
+		for ip := range ipsMap {
+			ipSet[ip] = true
+		}
+		if len(ipSet) > 0 {
+			aliveIPs[uid] = ipSet
+		}
 	}
-	c.closed.Store(true)
-	d.delConn(c.email, c.sourceIP)
-	d.mu.Lock()
-	delete(d.conns, id)
-	d.mu.Unlock()
-	return true
+
+	// Collect IPs from unlimited users (lock-free).
+	d.unlimitedIPs.Range(func(key, value interface{}) bool {
+		email := key.(string)
+		uid := emailToUID[email]
+		if uid == 0 {
+			return true
+		}
+		ic := value.(*ipCounter)
+		if ips := ic.aliveIPs(); len(ips) > 0 {
+			// Merge with limited IPs if any
+			if existing, ok := aliveIPs[uid]; ok {
+				for ip := range ips {
+					existing[ip] = true
+				}
+			} else {
+				aliveIPs[uid] = ips
+			}
+		}
+		return true
+	})
+
+	connCount = int(d.connCount.Load())
+	return
 }
 
 // ─── Internal helpers ───────────────────────────────────────────────────────
 
+// checkDeviceLimit enforces per-user device limits.
+// Fast path: unlimited users use lock-free sync.Map.
+// Slow path: limited users use RWMutex with deterministic IP ordering.
 func (d *LimitDispatcher) checkDeviceLimit(email, sourceIP string, isTCP bool) bool {
-	d.mu.Lock()
-	defer d.mu.Unlock()
-
+	d.mu.RLock()
 	limit, hasLimit := d.deviceLimits[email]
+	d.mu.RUnlock()
+
+	// Fast path: no device limit — use lock-free sync.Map.
 	if !hasLimit || limit <= 0 {
 		if isTCP {
-			if d.userIPs[email] == nil {
-				d.userIPs[email] = make(map[string]int)
-			}
-			d.userIPs[email][sourceIP]++
+			v, _ := d.unlimitedIPs.LoadOrStore(email, &ipCounter{})
+			ic := v.(*ipCounter)
+
+			// Increment IP refcount atomically.
+			rv, _ := ic.ips.LoadOrStore(sourceIP, &atomic.Int64{})
+			rv.(*atomic.Int64).Add(1)
 		}
 		return false
 	}
 
-	ips := d.userIPs[email]
+	// Slow path: user has device limit — need deterministic ordering.
+	d.mu.RLock()
+	ips := d.limitedIPs[email]
+	if ips != nil && ips[sourceIP] > 0 {
+		d.mu.RUnlock()
+		if isTCP {
+			d.mu.Lock()
+			d.limitedIPs[email][sourceIP]++
+			d.mu.Unlock()
+		}
+		return false
+	}
+
+	if ips != nil && len(ips) < limit {
+		d.mu.RUnlock()
+		if isTCP {
+			d.mu.Lock()
+			if d.limitedIPs[email] == nil {
+				d.limitedIPs[email] = make(map[string]int)
+			}
+			d.limitedIPs[email][sourceIP]++
+			d.mu.Unlock()
+		}
+		return false
+	}
+	d.mu.RUnlock()
+
+	// Over limit — need write lock for deterministic check.
+	d.mu.Lock()
+	defer d.mu.Unlock()
+
+	// Re-check under write lock.
+	ips = d.limitedIPs[email]
 	if ips == nil {
 		ips = make(map[string]int)
-		d.userIPs[email] = ips
+		d.limitedIPs[email] = ips
 	}
 
 	if ips[sourceIP] > 0 {
@@ -303,7 +339,7 @@ func (d *LimitDispatcher) checkDeviceLimit(email, sourceIP string, isTCP bool) b
 		return false
 	}
 
-	// Over limit — deterministic: allow lowest IPs lexicographically
+	// Over limit — deterministic: allow lowest IPs lexicographically.
 	ipList := make([]string, 0, len(ips)+1)
 	for ip := range ips {
 		ipList = append(ipList, ip)
@@ -322,105 +358,48 @@ func (d *LimitDispatcher) checkDeviceLimit(email, sourceIP string, isTCP bool) b
 	return true
 }
 
+// delConn decrements the IP refcount when a connection closes.
 func (d *LimitDispatcher) delConn(email, sourceIP string) {
+	// Check if this is an unlimited user first (lock-free).
+	if v, ok := d.unlimitedIPs.Load(email); ok {
+		ic := v.(*ipCounter)
+		if rv, ok := ic.ips.Load(sourceIP); ok {
+			counter := rv.(*atomic.Int64)
+			if counter.Add(-1) <= 0 {
+				ic.ips.Delete(sourceIP)
+			}
+		}
+		return
+	}
+
+	// Limited user — use write lock.
 	d.mu.Lock()
 	defer d.mu.Unlock()
-	if ips, ok := d.userIPs[email]; ok {
+	if ips, ok := d.limitedIPs[email]; ok {
 		ips[sourceIP]--
 		if ips[sourceIP] <= 0 {
 			delete(ips, sourceIP)
 		}
 		if len(ips) == 0 {
-			delete(d.userIPs, email)
+			delete(d.limitedIPs, email)
 		}
 	}
 }
 
-func (d *LimitDispatcher) getUID(email string) int {
-	d.mu.RLock()
-	defer d.mu.RUnlock()
-	return d.emailToUID[email]
-}
-
-func (d *LimitDispatcher) getBucket(email string) *rate.Limiter {
-	d.mu.RLock()
-	mbps, ok := d.speedLimits[email]
-	d.mu.RUnlock()
-	if !ok || mbps <= 0 {
-		return nil
-	}
-	bytesPerSec := int(mbps) * 1_000_000 / 8
-	burst := bytesPerSec
-	if burst < 64*1024 {
-		burst = 64 * 1024
-	}
-	v, loaded := d.speedBuckets.LoadOrStore(email, rate.NewLimiter(rate.Limit(bytesPerSec), burst))
-	lim := v.(*rate.Limiter)
-	if loaded {
-		// Update existing limiter in case speed limit changed.
-		lim.SetLimit(rate.Limit(bytesPerSec))
-		lim.SetBurst(burst)
-	}
-	return lim
-}
-
-// ─── I/O wrappers ───────────────────────────────────────────────────────────
-
-type statsCloseReader struct {
-	buf.Reader
-	counter *atomic.Int64
-	limiter *rate.Limiter
-	ctx     context.Context
-}
-
-func (r *statsCloseReader) ReadMultiBuffer() (buf.MultiBuffer, error) {
-	mb, err := r.Reader.ReadMultiBuffer()
-	if n := int64(mb.Len()); n > 0 {
-		r.counter.Add(n)
-		if r.limiter != nil {
-			waitN := int(n)
-			if burst := r.limiter.Burst(); waitN > burst {
-				waitN = burst
-			}
-			_ = r.limiter.WaitN(r.ctx, waitN)
-		}
-	}
-	return mb, err
-}
-
-func (r *statsCloseReader) Close() error { return common.Close(r.Reader) }
-func (r *statsCloseReader) Interrupt()   { common.Interrupt(r.Reader) }
-
-type statsCloseWriter struct {
+type closeTrackingWriter struct {
 	buf.Writer
-	counter *atomic.Int64
-	limiter *rate.Limiter
-	ctx     context.Context
 	onClose func()
 	closed  atomic.Bool
 }
 
-func (w *statsCloseWriter) WriteMultiBuffer(mb buf.MultiBuffer) error {
-	n := int64(mb.Len())
-	if w.limiter != nil {
-		waitN := int(n)
-		if burst := w.limiter.Burst(); waitN > burst {
-			waitN = burst
-		}
-		_ = w.limiter.WaitN(w.ctx, waitN)
-	}
-	w.counter.Add(n)
-	return w.Writer.WriteMultiBuffer(mb)
-}
-
-func (w *statsCloseWriter) Close() error {
+func (w *closeTrackingWriter) Close() error {
 	if w.closed.CompareAndSwap(false, true) {
 		w.onClose()
 	}
 	return common.Close(w.Writer)
 }
 
-func (w *statsCloseWriter) Interrupt() {
+func (w *closeTrackingWriter) Interrupt() {
 	if w.closed.CompareAndSwap(false, true) {
 		w.onClose()
 	}

internal/kernel/xray/dispatcher_test.go

@@ -1,18 +1,28 @@
 package xray
 
 import (
+	"sync/atomic"
 	"testing"
 
-	"github.com/cedar2025/xboard-node/internal/panel"
+	"github.com/cedar2025/xboard-node/internal/model"
+	"github.com/xtls/xray-core/common/buf"
+	"github.com/xtls/xray-core/transport"
 )
 
-func TestLimitDispatcher_DeviceLimitCheck(t *testing.T) {
-	ld := &LimitDispatcher{
-		conns:   make(map[string]*dispatchedConn),
-		userIPs: make(map[string]map[string]int),
+func newTestDispatcher() *LimitDispatcher {
+	return &LimitDispatcher{
+		limitedIPs: make(map[string]map[string]int),
 	}
+}
 
-	users := []panel.User{
+type nopReader struct{}
+
+func (nopReader) ReadMultiBuffer() (buf.MultiBuffer, error) { return nil, nil }
+
+func TestLimitDispatcher_DeviceLimitCheck(t *testing.T) {
+	ld := newTestDispatcher()
+
+	users := []model.UserSpec{
 		{ID: 1, UUID: "uuid-1", DeviceLimit: 2, SpeedLimit: 0},
 		{ID: 2, UUID: "uuid-2", DeviceLimit: 0, SpeedLimit: 10},
 	}
@@ -65,10 +75,7 @@ func TestLimitDispatcher_DeviceLimitCheck(t *testing.T) {
 }
 
 func TestLimitDispatcher_DelConn(t *testing.T) {
-	ld := &LimitDispatcher{
-		conns:   make(map[string]*dispatchedConn),
-		userIPs: make(map[string]map[string]int),
-	}
+	ld := newTestDispatcher()
 
 	email := userEmail(1)
 	deviceLimits := map[string]int{email: 2}
@@ -92,113 +99,137 @@ func TestLimitDispatcher_DelConn(t *testing.T) {
 	}
 }
 
-func TestLimitDispatcher_SpeedBucket(t *testing.T) {
-	ld := &LimitDispatcher{
-		conns:   make(map[string]*dispatchedConn),
-		userIPs: make(map[string]map[string]int),
-	}
+func TestLimitDispatcher_GetConnectionState(t *testing.T) {
+	ld := newTestDispatcher()
 
 	email1 := userEmail(1)
 	email2 := userEmail(2)
+	ld.UpdateLimits(map[string]int{email1: 1, email2: 2}, nil, nil)
 
-	speedLimits := map[string]int{email1: 10} // 10 Mbps
-	ld.UpdateLimits(map[string]int{email1: 1, email2: 2}, nil, speedLimits)
+	ic1 := &ipCounter{}
+	r1 := &atomic.Int64{}
+	r1.Store(1)
+	ic1.ips.Store("1.1.1.1", r1)
+	r2 := &atomic.Int64{}
+	r2.Store(1)
+	ic1.ips.Store("2.2.2.2", r2)
+	ld.unlimitedIPs.Store(email1, ic1)
 
-	// User 1 has speed limit — should get a limiter
-	limiter := ld.getBucket(email1)
-	if limiter == nil {
-		t.Error("user with speed limit should get a limiter")
-	}
+	ic2 := &ipCounter{}
+	r3 := &atomic.Int64{}
+	r3.Store(1)
+	ic2.ips.Store("3.3.3.3", r3)
+	ld.unlimitedIPs.Store(email2, ic2)
 
-	// Same user should get the same limiter (cached)
-	limiter2 := ld.getBucket(email1)
-	if limiter != limiter2 {
-		t.Error("same user should get cached limiter")
-	}
+	ld.connCount.Store(5)
 
-	// User 2 has no speed limit — should get nil
-	limiter3 := ld.getBucket(email2)
-	if limiter3 != nil {
-		t.Error("user without speed limit should get nil")
-	}
-}
+	aliveIPs, connCount := ld.GetConnectionState()
 
-func TestLimitDispatcher_Snapshot(t *testing.T) {
-	ld := &LimitDispatcher{
-		conns:   make(map[string]*dispatchedConn),
-		userIPs: make(map[string]map[string]int),
+	if connCount != 5 {
+		t.Errorf("expected connCount=5, got %d", connCount)
 	}
-
-	// Add some connections
-	ld.conns["conn-1"] = &dispatchedConn{
-		id: "conn-1", email: "user@1", sourceIP: "1.1.1.1", userID: 1,
+	if len(aliveIPs[1]) != 2 {
+		t.Errorf("user 1 IPs: got %d, want 2", len(aliveIPs[1]))
 	}
-	ld.conns["conn-2"] = &dispatchedConn{
-		id: "conn-2", email: "user@2", sourceIP: "2.2.2.2", userID: 2,
-	}
-	ld.conns["conn-1"].upload.Store(1000)
-	ld.conns["conn-1"].download.Store(2000)
-
-	snapshot := ld.Snapshot()
-	if len(snapshot) != 2 {
-		t.Fatalf("expected 2 connections, got %d", len(snapshot))
-	}
-
-	found := false
-	for _, c := range snapshot {
-		if c.ID == "conn-1" {
-			found = true
-			if c.UserID != 1 || c.SourceIP != "1.1.1.1" || c.Upload != 1000 || c.Download != 2000 {
-				t.Errorf("unexpected connection data: %+v", c)
-			}
-		}
-	}
-	if !found {
-		t.Error("conn-1 not found in snapshot")
-	}
-}
-
-func TestLimitDispatcher_CloseConn(t *testing.T) {
-	ld := &LimitDispatcher{
-		conns:   make(map[string]*dispatchedConn),
-		userIPs: make(map[string]map[string]int),
-	}
-
-	email := userEmail(1)
-	ld.userIPs[email] = map[string]int{"1.1.1.1": 1}
-	ld.conns["conn-1"] = &dispatchedConn{
-		id: "conn-1", email: email, sourceIP: "1.1.1.1", userID: 1,
-	}
-
-	ok := ld.CloseConn("conn-1")
-	if !ok {
-		t.Error("CloseConn should return true for existing connection")
-	}
-
-	if len(ld.conns) != 0 {
-		t.Error("connection should be removed after close")
-	}
-
-	ok = ld.CloseConn("nonexistent")
-	if ok {
-		t.Error("CloseConn should return false for nonexistent connection")
+	if len(aliveIPs[2]) != 1 {
+		t.Errorf("user 2 IPs: got %d, want 1", len(aliveIPs[2]))
 	}
 }
 
 func TestLimitDispatcher_ResetConns(t *testing.T) {
-	ld := &LimitDispatcher{
-		conns:   make(map[string]*dispatchedConn),
-		userIPs: make(map[string]map[string]int),
-	}
-	ld.conns["c1"] = &dispatchedConn{id: "c1"}
-	ld.userIPs["user@1"] = map[string]int{"1.1.1.1": 1}
+	ld := newTestDispatcher()
+
+	ld.mu.Lock()
+	ld.limitedIPs["user@1"] = map[string]int{"1.1.1.1": 1}
+	ld.mu.Unlock()
+	ld.connCount.Store(3)
 
 	ld.ResetConns()
 
-	if len(ld.conns) != 0 {
-		t.Error("conns should be empty after reset")
+	ld.mu.RLock()
+	ipCount := len(ld.limitedIPs)
+	ld.mu.RUnlock()
+	if ipCount != 0 {
+		t.Error("limitedIPs should be empty after reset")
 	}
-	if len(ld.userIPs) != 0 {
-		t.Error("userIPs should be empty after reset")
+
+	if ld.connCount.Load() != 0 {
+		t.Error("connCount should be 0 after reset")
+	}
+}
+
+func TestLimitDispatcher_UnlimitedUserFastPath(t *testing.T) {
+	ld := newTestDispatcher()
+
+	email := userEmail(1)
+	// No device limit set for this user
+	ld.UpdateLimits(map[string]int{email: 1}, nil, nil)
+
+	// Should use fast path (sync.Map), no lock needed
+	for i := 0; i < 100; i++ {
+		ip := "10.0.0." + string(rune('0'+i%10))
+		if ld.checkDeviceLimit(email, ip, true) {
+			t.Errorf("unlimited user should always be allowed (ip=%s)", ip)
+		}
+	}
+
+	// Verify IPs are tracked in unlimitedIPs
+	v, ok := ld.unlimitedIPs.Load(email)
+	if !ok {
+		t.Error("unlimited user should have entry in unlimitedIPs")
+	}
+	ic := v.(*ipCounter)
+	ips := ic.aliveIPs()
+	if len(ips) == 0 {
+		t.Error("should have tracked some IPs")
+	}
+}
+
+
+func TestLimitDispatcher_TrackLinkPreservesReader(t *testing.T) {
+	ld := newTestDispatcher()
+	email := userEmail(1)
+	ld.UpdateLimits(map[string]int{email: 1}, map[string]int{email: 1}, nil)
+
+	origReader := nopReader{}
+	origWriter := &closeTrackingWriter{Writer: buf.Discard, onClose: func() {}}
+	link := &transport.Link{Reader: origReader, Writer: origWriter}
+
+	ld.trackLink(link, email, "1.1.1.1", true)
+
+	if link.Reader != origReader {
+		t.Fatal("trackLink must not replace link.Reader")
+	}
+	if link.Writer == origWriter {
+		t.Fatal("trackLink should wrap link.Writer for lifecycle callbacks")
+	}
+}
+
+func TestLimitDispatcher_CloseTrackingWriterReleasesConn(t *testing.T) {
+	ld := newTestDispatcher()
+	email := userEmail(1)
+	ld.UpdateLimits(map[string]int{email: 1}, map[string]int{email: 1}, nil)
+	if ld.checkDeviceLimit(email, "1.1.1.1", true) {
+		t.Fatal("first connection should be allowed")
+	}
+
+	link := &transport.Link{Reader: nopReader{}, Writer: buf.Discard}
+	ld.trackLink(link, email, "1.1.1.1", true)
+
+	if got := ld.connCount.Load(); got != 1 {
+		t.Fatalf("expected connCount=1 after tracking, got %d", got)
+	}
+	cw, ok := link.Writer.(*closeTrackingWriter)
+	if !ok {
+		t.Fatal("expected closeTrackingWriter wrapper")
+	}
+	if err := cw.Close(); err != nil {
+		t.Fatalf("closeTrackingWriter.Close() error = %v", err)
+	}
+	if got := ld.connCount.Load(); got != 0 {
+		t.Fatalf("expected connCount=0 after close, got %d", got)
+	}
+	if ld.checkDeviceLimit(email, "2.2.2.2", true) {
+		t.Fatal("device slot should be released after writer close")
 	}
 }

internal/kernel/xray/xray.go

@@ -6,7 +6,6 @@ import (
 	"crypto/sha256"
 	"encoding/json"
 	"fmt"
-	"log/slog"
 	"os"
 	"strings"
 	"sync"
@@ -18,6 +17,7 @@ import (
 	xrayCore "github.com/xtls/xray-core/core"
 	"github.com/xtls/xray-core/features/inbound"
 	"github.com/xtls/xray-core/features/stats"
+	featurebandwidth "github.com/xtls/xray-core/features/bandwidth"
 	"github.com/xtls/xray-core/infra/conf/serial"
 	xrayProxy "github.com/xtls/xray-core/proxy"
 	"github.com/xtls/xray-core/proxy/shadowsocks"
@@ -32,7 +32,8 @@ import (
 	"github.com/cedar2025/xboard-node/internal/config"
 	"github.com/cedar2025/xboard-node/internal/kernel"
 	"github.com/cedar2025/xboard-node/internal/kernel/geodata"
-	"github.com/cedar2025/xboard-node/internal/panel"
+	"github.com/cedar2025/xboard-node/internal/nlog"
+	"github.com/cedar2025/xboard-node/internal/model"
 )
 
 const (
@@ -59,14 +60,14 @@ type Xray struct {
 	mu              sync.Mutex
 	instance        *xrayCore.Instance
 	limitDispatcher *LimitDispatcher
-	users           []panel.User
-	nodeConfig      *panel.NodeConfig
-	certFile        string
-	keyFile         string
+	users           []model.UserSpec
+	nodeConfig      *model.NodeSpec
+	tls             kernel.TLSCert
 	protocol        string
 	inboundTag      string
 	lastKernelHash  string
 	cumTraffic      map[int][2]int64
+	speedLimitFunc  func(string) *rate.Limiter
 
 	// running is set after a successful Start and cleared before shutdown.
 	// Atomic so IsRunning / GetConnections never block.
@@ -82,29 +83,42 @@ func New(cfg config.KernelConfig) *Xray {
 
 func (x *Xray) Name() string { return "xray" }
 
+func (x *Xray) Capabilities() kernel.Capabilities {
+	return kernel.Capabilities{
+		PerUserSpeedLimit:    true,
+		DeviceLimit:          true,
+		BuiltInTrafficStats:  true,
+		AliveIPTracking:      true,
+		ForceCloseConnection: false,
+		ForceCloseUser:       false,
+	}
+}
+
 func (x *Xray) Protocols() []string {
 	return []string{
 		"vmess", "vless", "trojan", "shadowsocks",
-		"socks", "http", "dokodemo-door",
+		"hysteria",
 	}
 }
 
 // ─── Lifecycle ──────────────────────────────────────────────────────────────
 
 // Start builds a new xray-core instance and atomically replaces the old
-// one. The method is organised in four non-overlapping phases so that the
+// one. The method is organised in five non-overlapping phases so that the
 // kernel mutex is never held during slow operations (Start / Close).
+// Crucially, the old instance stays alive until the new one is confirmed
+// running — if StartNew fails, the old instance is untouched.
 //
-//	Phase 1 – Build:   generate protobuf config  (no lock, pure computation)
-//	Phase 2 – Create:  xrayCore.New + capture LD (brief global lock)
-//	Phase 3 – Swap:    stop old instance          (brief kernel lock)
-//	Phase 4 – Start:   instance.Start             (no lock, potentially slow)
-//	Phase 5 – Commit:  store new state            (brief kernel lock)
-func (x *Xray) Start(nodeConfig *panel.NodeConfig, users []panel.User, certFile, keyFile string) error {
+//	Phase 1 – Build:      generate protobuf config  (no lock, pure computation)
+//	Phase 2 – Create:     xrayCore.New + capture LD (brief global lock)
+//	Phase 3 – StartNew:   instance.Start            (no lock, potentially slow)
+//	Phase 4 – Swap:       store new, extract old     (brief kernel lock)
+//	Phase 5 – RecycleOld: close old in background    (non-blocking)
+func (x *Xray) Start(nodeConfig *model.NodeSpec, users []model.UserSpec, tls kernel.TLSCert) error {
 	// ── Phase 1: Build config (no shared state) ─────────────────────────
 	x.ensureGeoData(nodeConfig)
 
-	data, err := marshalConfig(x.cfg, nodeConfig, users, certFile, keyFile)
+	data, err := marshalConfig(x.cfg, nodeConfig, users, tls)
 	if err != nil {
 		return err
 	}
@@ -123,31 +137,21 @@ func (x *Xray) Start(nodeConfig *panel.NodeConfig, users []panel.User, certFile,
 		return fmt.Errorf("create xray: %w", err)
 	}
 
-	// ── Phase 3: Stop old (brief kernel lock, no drain) ─────────────────
-	x.mu.Lock()
-	x.running.Store(false)
-	old := x.instance
-	x.instance = nil
-	oldLD := x.limitDispatcher
-	x.limitDispatcher = nil
-	x.mu.Unlock()
-
-	closeOld(old, oldLD)
-
-	// ── Phase 4: Start new (no lock, potentially slow) ──────────────────
+	// ── Phase 3: Start new (no lock, potentially slow) ──────────────────
 	if err := startWithTimeout(inst, startTimeout); err != nil {
 		inst.Close()
 		return err
 	}
 
-	// ── Phase 5: Commit (brief kernel lock) ─────────────────────────────
+	// ── Phase 4: Swap old → new (brief kernel lock) ─────────────────────
 	x.mu.Lock()
+	old := x.instance
+	oldLD := x.limitDispatcher
 	x.instance = inst
 	x.limitDispatcher = ld
 	x.users = users
 	x.nodeConfig = nodeConfig
-	x.certFile = certFile
-	x.keyFile = keyFile
+	x.tls = tls
 	x.protocol = nodeConfig.Protocol
 	x.inboundTag = nodeConfig.Protocol + "-in"
 	x.cumTraffic = make(map[int][2]int64)
@@ -155,9 +159,13 @@ func (x *Xray) Start(nodeConfig *panel.NodeConfig, users []panel.User, certFile,
 	x.running.Store(true)
 	x.mu.Unlock()
 
-	x.updateDispatcherLimits(users)
+	// ── Phase 5: Recycle old (background, non-blocking) ─────────────────
+	closeOld(old, oldLD)
 
-	slog.Info("xray started",
+	x.updateDispatcherLimits(users)
+	x.updateBandwidthLimits(users)
+
+	nlog.Core().Info("xray started",
 		"users", len(users),
 		"protocol", nodeConfig.Protocol,
 	)
@@ -168,23 +176,24 @@ func (x *Xray) Start(nodeConfig *panel.NodeConfig, users []panel.User, certFile,
 // Since Xray does not support granular inbound reconstruction without an
 // instance restart for most transport/TLS settings, it triggers a full restart
 // if any kernel-affecting fields (hash mismatch) have changed.
-func (x *Xray) Reload(nodeConfig *panel.NodeConfig, users []panel.User, certFile, keyFile string) error {
+func (x *Xray) Reload(nodeConfig *model.NodeSpec, users []model.UserSpec, tls kernel.TLSCert) error {
 	x.updateDispatcherLimits(users)
+	x.updateBandwidthLimits(users)
 
 	newHash := kernel.ComputeHash(nodeConfig, users)
 
 	x.mu.Lock()
-	same := x.lastKernelHash == newHash
+	same := x.lastKernelHash == newHash && tlsEqual(x.tls, tls)
 	if same {
 		x.users = users
 		x.mu.Unlock()
-		slog.Debug("xray: limits updated, kernel configuration unchanged")
+		nlog.Core().Debug("xray: limits updated, kernel configuration unchanged")
 		return nil
 	}
 	x.mu.Unlock()
 
-	slog.Info("xray: configuration changed, performing full restart")
-	return x.Start(nodeConfig, users, certFile, keyFile)
+	nlog.Core().Info("xray: configuration changed, performing full restart")
+	return x.Start(nodeConfig, users, tls)
 }
 
 // Stop gracefully shuts down the kernel, draining active connections first.
@@ -206,42 +215,65 @@ func (x *Xray) Stop() {
 
 func (x *Xray) IsRunning() bool { return x.running.Load() }
 
-// ─── Connection queries ─────────────────────────────────────────────────────
-
-// GetConnections returns a snapshot of active proxy connections.
-// The lock is safe to take here because mu is never held during slow I/O.
-func (x *Xray) GetConnections(ctx context.Context) ([]kernel.Connection, error) {
-	if !x.running.Load() {
-		return nil, nil
-	}
-
-	x.mu.Lock()
-	defer x.mu.Unlock()
-
-	if x.instance == nil {
-		return nil, nil
-	}
-
-	if x.limitDispatcher != nil {
-		if conns := x.limitDispatcher.Snapshot(); len(conns) > 0 {
-			return conns, nil
-		}
-	}
-
-	return x.aggregateStats()
+func tlsEqual(a, b kernel.TLSCert) bool {
+	return bytes.Equal(a.CertPEM, b.CertPEM) && bytes.Equal(a.KeyPEM, b.KeyPEM)
 }
 
-func (x *Xray) CloseConnection(_ context.Context, connID string) error {
+// ─── Observability ──────────────────────────────────────────────────────────
+
+// GetUserTraffic returns per-user cumulative traffic from xray's built-in
+// stats pipeline and connection state from the admission dispatcher when
+// available. The dispatcher intentionally no longer wraps transport links,
+// so traffic accounting must come from xray-core itself.
+func (x *Xray) GetUserTraffic(_ context.Context) (traffic map[int][2]int64, aliveIPs map[int]map[string]bool, connCount int, err error) {
+	if !x.running.Load() {
+		return nil, nil, 0, nil
+	}
+
 	x.mu.Lock()
 	ld := x.limitDispatcher
+	traffic, err = x.aggregateStats()
 	x.mu.Unlock()
-	if ld != nil {
-		ld.CloseConn(connID)
+	if err != nil {
+		return nil, nil, 0, err
 	}
+
+	if ld != nil {
+		aliveIPs, connCount = ld.GetConnectionState()
+	}
+	return traffic, aliveIPs, connCount, nil
+}
+
+func (x *Xray) CloseConnection(_ context.Context, _ string) error {
+	// No-op: xray doesn't support force-closing individual connections.
+	// User removal goes through RemoveUsers which removes the inbound user.
 	return nil
 }
 
-func (x *Xray) SetSpeedLimitFunc(_ func(string) *rate.Limiter) {}
+func (x *Xray) CloseUserConnections(_ context.Context, _ string) error {
+	// No-op: handled by RemoveUsers at the xray core level.
+	return nil
+}
+
+// SetSpeedLimitFunc wires xray's patched bandwidth feature to the shared
+// per-user limiter callback used by the service layer. Unlike the old no-op
+// behavior, xray now consumes this callback to support dynamic speed updates.
+func (x *Xray) SetSpeedLimitFunc(fn func(string) *rate.Limiter) {
+	x.mu.Lock()
+	x.speedLimitFunc = fn
+	x.mu.Unlock()
+	x.updateBandwidthLimits(nil)
+}
+
+// SetDeviceLimitFunc is a no-op for xray — device limits are already
+// gate-kept by LimitDispatcher.checkDeviceLimit at Dispatch time.
+func (x *Xray) SetDeviceLimitFunc(_ func(string) (int, bool)) {}
+
+// UpdateGlobalDevices is a no-op for xray — xray handles device limits differently.
+func (x *Xray) UpdateGlobalDevices(_ map[int][]string) {}
+
+// ClearGlobalDevices is a no-op for xray.
+func (x *Xray) ClearGlobalDevices() {}
 
 // ─── User management (non-disruptive where possible) ────────────────────────
 
@@ -253,7 +285,7 @@ func (x *Xray) SetSpeedLimitFunc(_ func(string) *rate.Limiter) {}
 // Delta "add" events also carry property updates (speed/device limits) for
 // existing users, so this method always merges properties and refreshes the
 // dispatcher limits — even when no brand-new users need to be added.
-func (x *Xray) AddUsers(users []panel.User) (int, error) {
+func (x *Xray) AddUsers(users []model.UserSpec) (int, error) {
 	x.mu.Lock()
 	if x.instance == nil {
 		x.mu.Unlock()
@@ -261,18 +293,18 @@ func (x *Xray) AddUsers(users []panel.User) (int, error) {
 	}
 
 	// Merge: overwrite existing users' properties, collect truly new ones.
-	userMap := make(map[int]panel.User, len(x.users))
+	userMap := make(map[int]model.UserSpec, len(x.users))
 	for _, u := range x.users {
 		userMap[u.ID] = u
 	}
-	var toAdd []panel.User
+	var toAdd []model.UserSpec
 	for _, u := range users {
 		if _, exists := userMap[u.ID]; !exists {
 			toAdd = append(toAdd, u)
 		}
 		userMap[u.ID] = u // always overwrite properties
 	}
-	merged := make([]panel.User, 0, len(userMap))
+	merged := make([]model.UserSpec, 0, len(userMap))
 	for _, u := range userMap {
 		merged = append(merged, u)
 	}
@@ -282,16 +314,17 @@ func (x *Xray) AddUsers(users []panel.User) (int, error) {
 		x.users = merged
 		x.mu.Unlock()
 		x.updateDispatcherLimits(merged)
+	x.updateBandwidthLimits(merged)
 		return 0, nil
 	}
 
 	um, err := x.getUserManager()
 	if err != nil {
 		// Protocol doesn't support UserManager → full restart
-		nc, cf, kf := x.nodeConfig, x.certFile, x.keyFile
+		nc, t := x.nodeConfig, x.tls
 		x.mu.Unlock()
-		slog.Debug("xray: AddUsers fallback to restart", "reason", err)
-		if err := x.Start(nc, merged, cf, kf); err != nil {
+		nlog.Core().Debug("xray: AddUsers fallback to restart", "reason", err)
+		if err := x.Start(nc, merged, t); err != nil {
 			return 0, err
 		}
 		return len(toAdd), nil
@@ -306,11 +339,11 @@ func (x *Xray) AddUsers(users []panel.User) (int, error) {
 	for _, u := range toAdd {
 		mu, err := toMemoryUser(proto, nc, u)
 		if err != nil {
-			slog.Warn("xray: skip user, cannot build account", "user", u.ID, "error", err)
+			nlog.Core().Warn("xray: skip user, cannot build account", "user", u.ID, "error", err)
 			continue
 		}
 		if err := um.AddUser(ctx, mu); err != nil {
-			slog.Warn("xray: AddUser failed", "user", u.ID, "error", err)
+			nlog.Core().Warn("xray: AddUser failed", "user", u.ID, "error", err)
 			continue
 		}
 		added++
@@ -321,14 +354,15 @@ func (x *Xray) AddUsers(users []panel.User) (int, error) {
 	x.users = merged
 	x.mu.Unlock()
 	x.updateDispatcherLimits(merged)
+	x.updateBandwidthLimits(merged)
 
-	slog.Info("xray: users added via UserManager", "added", added, "total", len(merged))
+	nlog.Core().Info("xray: users added via UserManager", "added", added, "total", len(merged))
 	return added, nil
 }
 
 // RemoveUsers removes users from the running kernel via xray's UserManager API.
 // Truly hitless for supported protocols — remaining connections unaffected.
-func (x *Xray) RemoveUsers(users []panel.User) (int, error) {
+func (x *Xray) RemoveUsers(users []model.UserSpec) (int, error) {
 	x.mu.Lock()
 	if x.instance == nil {
 		x.mu.Unlock()
@@ -338,7 +372,7 @@ func (x *Xray) RemoveUsers(users []panel.User) (int, error) {
 	for _, u := range users {
 		removeSet[u.ID] = struct{}{}
 	}
-	var kept []panel.User
+	var kept []model.UserSpec
 	removed := 0
 	for _, u := range x.users {
 		if _, rm := removeSet[u.ID]; rm {
@@ -360,10 +394,10 @@ func (x *Xray) RemoveUsers(users []panel.User) (int, error) {
 
 	um, err := x.getUserManager()
 	if err != nil {
-		nc, cf, kf := x.nodeConfig, x.certFile, x.keyFile
+		nc, t := x.nodeConfig, x.tls
 		x.mu.Unlock()
-		slog.Debug("xray: RemoveUsers fallback to restart", "reason", err)
-		if err := x.Start(nc, kept, cf, kf); err != nil {
+		nlog.Core().Debug("xray: RemoveUsers fallback to restart", "reason", err)
+		if err := x.Start(nc, kept, t); err != nil {
 			return 0, err
 		}
 		return removed, nil
@@ -375,7 +409,7 @@ func (x *Xray) RemoveUsers(users []panel.User) (int, error) {
 	for _, u := range users {
 		email := userEmail(u.ID)
 		if err := um.RemoveUser(ctx, email); err != nil {
-			slog.Debug("xray: RemoveUser skipped", "user", u.ID, "error", err)
+			nlog.Core().Debug("xray: RemoveUser skipped", "user", u.ID, "error", err)
 			continue
 		}
 		actualRemoved++
@@ -385,15 +419,16 @@ func (x *Xray) RemoveUsers(users []panel.User) (int, error) {
 	x.users = kept
 	x.mu.Unlock()
 	x.updateDispatcherLimits(kept)
+	x.updateBandwidthLimits(kept)
 
-	slog.Info("xray: users removed via UserManager", "removed", actualRemoved, "total", len(kept))
+	nlog.Core().Info("xray: users removed via UserManager", "removed", actualRemoved, "total", len(kept))
 	return actualRemoved, nil
 }
 
 // UpdateUsers replaces the entire user set. If only speed/device limits
 // changed, updates the dispatcher without restarting. Otherwise uses
 // UserManager for hitless add/remove where supported.
-func (x *Xray) UpdateUsers(users []panel.User) (added, removed int, err error) {
+func (x *Xray) UpdateUsers(users []model.UserSpec) (added, removed int, err error) {
 	x.mu.Lock()
 	if x.instance == nil {
 		x.mu.Unlock()
@@ -407,17 +442,18 @@ func (x *Xray) UpdateUsers(users []panel.User) (added, removed int, err error) {
 		x.users = users
 		x.mu.Unlock()
 		x.updateDispatcherLimits(users)
-		slog.Info("xray: limits updated, kernel unchanged")
+		x.updateBandwidthLimits(users)
+		nlog.Core().Info("xray: limits updated, kernel unchanged")
 		return 0, 0, nil
 	}
 
 	um, umErr := x.getUserManager()
 	if umErr != nil {
 		// Protocol doesn't support UserManager → full restart
-		nc, cf, kf := x.nodeConfig, x.certFile, x.keyFile
+		nc, t := x.nodeConfig, x.tls
 		x.mu.Unlock()
-		slog.Debug("xray: UpdateUsers fallback to restart", "reason", umErr)
-		if err = x.Start(nc, users, cf, kf); err != nil {
+		nlog.Core().Debug("xray: UpdateUsers fallback to restart", "reason", umErr)
+		if err = x.Start(nc, users, t); err != nil {
 			return 0, 0, err
 		}
 		return
@@ -433,17 +469,17 @@ func (x *Xray) UpdateUsers(users []panel.User) (added, removed int, err error) {
 	for _, u := range toRemove {
 		email := userEmail(u.ID)
 		if err := um.RemoveUser(ctx, email); err != nil {
-			slog.Debug("xray: RemoveUser skipped in UpdateUsers", "user", u.ID, "error", err)
+			nlog.Core().Debug("xray: RemoveUser skipped in UpdateUsers", "user", u.ID, "error", err)
 		}
 	}
 	for _, u := range toAdd {
 		mu, err := toMemoryUser(proto, nc, u)
 		if err != nil {
-			slog.Warn("xray: skip user in UpdateUsers", "user", u.ID, "error", err)
+			nlog.Core().Warn("xray: skip user in UpdateUsers", "user", u.ID, "error", err)
 			continue
 		}
 		if err := um.AddUser(ctx, mu); err != nil {
-			slog.Warn("xray: AddUser failed in UpdateUsers", "user", u.ID, "error", err)
+			nlog.Core().Warn("xray: AddUser failed in UpdateUsers", "user", u.ID, "error", err)
 		}
 	}
 
@@ -451,8 +487,9 @@ func (x *Xray) UpdateUsers(users []panel.User) (added, removed int, err error) {
 	x.users = users
 	x.mu.Unlock()
 	x.updateDispatcherLimits(users)
+	x.updateBandwidthLimits(users)
 
-	slog.Info("xray: users updated via UserManager", "added", added, "removed", removed, "total", len(users))
+	nlog.Core().Info("xray: users updated via UserManager", "added", added, "removed", removed, "total", len(users))
 	return
 }
 
@@ -495,9 +532,9 @@ func (x *Xray) getUserManager() (xrayProxy.UserManager, error) {
 	return um, nil
 }
 
-// toMemoryUser converts a panel.User into an xray protocol.MemoryUser for the
+// toMemoryUser converts a model.UserSpec into an xray protocol.MemoryUser for the
 // given protocol. Each protocol needs a different Account type.
-func toMemoryUser(proto string, nc *panel.NodeConfig, u panel.User) (*protocol.MemoryUser, error) {
+func toMemoryUser(proto string, nc *model.NodeSpec, u model.UserSpec) (*protocol.MemoryUser, error) {
 	email := userEmail(u.ID)
 	mu := &protocol.MemoryUser{Email: email, Level: 0}
 
@@ -591,26 +628,26 @@ func hexEncode(dst, src []byte) {
 }
 
 // ensureGeoData downloads geo databases when routes reference geoip/geosite.
-func (x *Xray) ensureGeoData(nc *panel.NodeConfig) {
+func (x *Xray) ensureGeoData(nc *model.NodeSpec) {
 	needIP, needSite := kernel.NeedsGeoIP(nc.Routes), kernel.NeedsGeoSite(nc.Routes)
 	if !needIP && !needSite {
 		return
 	}
 	dir := x.cfg.GeoDataDir
 	if err := geodata.Ensure(dir, needIP, needSite, "xray"); err != nil {
-		slog.Warn("geo database unavailable", "error", err)
+		nlog.Core().Warn("geo database unavailable", "error", err)
 	}
 	os.Setenv("XRAY_LOCATION_ASSET", dir)
 }
 
 // marshalConfig builds the xray JSON config and returns the raw bytes.
-func marshalConfig(cfg config.KernelConfig, nc *panel.NodeConfig, users []panel.User, certFile, keyFile string) ([]byte, error) {
-	cfgMap := buildConfig(cfg, nc, users, certFile, keyFile)
-	data, err := json.Marshal(cfgMap)
+func marshalConfig(cfg config.KernelConfig, nc *model.NodeSpec, users []model.UserSpec, tls kernel.TLSCert) ([]byte, error) {
+	cfgMap := buildConfig(cfg, nc, users, tls)
+	data, err := json.MarshalIndent(cfgMap, "", "  ")
 	if err != nil {
 		return nil, fmt.Errorf("marshal config: %w", err)
 	}
-	slog.Debug("xray config generated", "len", len(data))
+	nlog.Core().Debug("xray config generated", "len", len(data))
 	return data, nil
 }
 
@@ -626,6 +663,7 @@ func startWithTimeout(inst *xrayCore.Instance, timeout time.Duration) error {
 		}
 		return nil
 	case <-time.After(timeout):
+		go inst.Close()
 		return fmt.Errorf("start xray: timeout after %v", timeout)
 	}
 }
@@ -649,7 +687,7 @@ func closeOld(inst *xrayCore.Instance, ld *LimitDispatcher) {
 		if ld != nil {
 			ld.ResetConns()
 		}
-		slog.Debug("xray: old instance recycled")
+		nlog.Core().Debug("xray: old instance recycled")
 	}()
 }
 
@@ -658,10 +696,7 @@ func closeOld(inst *xrayCore.Instance, ld *LimitDispatcher) {
 func drainConns(ld *LimitDispatcher, timeout time.Duration) {
 	deadline := time.Now().Add(timeout)
 	for time.Now().Before(deadline) {
-		ld.mu.RLock()
-		n := len(ld.conns)
-		ld.mu.RUnlock()
-		if n == 0 {
+		if ld.connCount.Load() == 0 {
 			return
 		}
 		time.Sleep(100 * time.Millisecond)
@@ -669,9 +704,9 @@ func drainConns(ld *LimitDispatcher, timeout time.Duration) {
 }
 
 // aggregateStats is a fallback path that reads xray's built-in stats counters
-// when LimitDispatcher has no active connections (e.g. short-lived UDP flows).
+// when LimitDispatcher is not available. Returns per-user cumulative traffic.
 // Must be called with x.mu held.
-func (x *Xray) aggregateStats() ([]kernel.Connection, error) {
+func (x *Xray) aggregateStats() (map[int][2]int64, error) {
 	sm := x.instance.GetFeature(stats.ManagerType())
 	if sm == nil {
 		return nil, nil
@@ -681,7 +716,7 @@ func (x *Xray) aggregateStats() ([]kernel.Connection, error) {
 		return nil, nil
 	}
 
-	var conns []kernel.Connection
+	traffic := make(map[int][2]int64)
 	for _, u := range x.users {
 		email := userEmail(u.ID)
 
@@ -701,20 +736,50 @@ func (x *Xray) aggregateStats() ([]kernel.Connection, error) {
 		}
 
 		if cum := x.cumTraffic[u.ID]; cum[0] > 0 || cum[1] > 0 {
-			conns = append(conns, kernel.Connection{
-				ID:       fmt.Sprintf("xray-%d", u.ID),
-				UserID:   u.ID,
-				Upload:   cum[0],
-				Download: cum[1],
-			})
+			traffic[u.ID] = cum
 		}
 	}
-	return conns, nil
+	return traffic, nil
+}
+
+// updateBandwidthLimits configures the patched xray-core bandwidth feature
+// with per-user speed limits in bytes per second.
+func (x *Xray) updateBandwidthLimits(users []model.UserSpec) {
+	x.mu.Lock()
+	inst := x.instance
+	fn := x.speedLimitFunc
+	currentUsers := x.users
+	x.mu.Unlock()
+	if inst == nil {
+		return
+	}
+	feat := inst.GetFeature(featurebandwidth.ManagerType())
+	if feat == nil {
+		return
+	}
+	bm, ok := feat.(featurebandwidth.Manager)
+	if !ok {
+		return
+	}
+	bm.Reset()
+	if users == nil {
+		users = currentUsers
+	}
+	for _, u := range users {
+		email := userEmail(u.ID)
+		if fn != nil {
+			bm.SetUserLimiter(email, fn(u.UUID))
+			continue
+		}
+		bps := int64(u.SpeedLimit) * 1_000_000 / 8
+		bm.SetUserLimit(email, bps)
+	}
 }
 
 // updateDispatcherLimits configures the LimitDispatcher with per-user
-// device and speed limits.
-func (x *Xray) updateDispatcherLimits(users []panel.User) {
+// device-limit admission metadata only. Speed limits are enforced by the
+// patched xray-core bandwidth feature.
+func (x *Xray) updateDispatcherLimits(users []model.UserSpec) {
 	x.mu.Lock()
 	ld := x.limitDispatcher
 	x.mu.Unlock()
@@ -724,7 +789,6 @@ func (x *Xray) updateDispatcherLimits(users []panel.User) {
 
 	emailToUID := make(map[string]int, len(users)*2)
 	deviceLimits := make(map[string]int)
-	speedLimits := make(map[string]int)
 
 	for _, u := range users {
 		email := userEmail(u.ID)
@@ -734,13 +798,9 @@ func (x *Xray) updateDispatcherLimits(users []panel.User) {
 			deviceLimits[email] = u.DeviceLimit
 			deviceLimits[u.UUID] = u.DeviceLimit
 		}
-		if u.SpeedLimit > 0 {
-			speedLimits[email] = u.SpeedLimit
-			speedLimits[u.UUID] = u.SpeedLimit
-		}
 	}
 
-	ld.UpdateLimits(emailToUID, deviceLimits, speedLimits)
+	ld.UpdateLimits(emailToUID, deviceLimits, nil)
 }
 
 // xrayCreationMu serialises xrayCore.New() + globalLimitDispatcher capture

internal/kernel/xray/xray_test.go

@@ -0,0 +1,278 @@
+package xray
+
+import (
+	"context"
+	"fmt"
+	"sync/atomic"
+	"testing"
+
+	"github.com/cedar2025/xboard-node/internal/config"
+	"github.com/cedar2025/xboard-node/internal/model"
+	appstats "github.com/xtls/xray-core/app/stats"
+	xrayCore "github.com/xtls/xray-core/core"
+	featurebandwidth "github.com/xtls/xray-core/features/bandwidth"
+	xraystats "github.com/xtls/xray-core/features/stats"
+	"golang.org/x/time/rate"
+)
+
+func newStatsManager(t *testing.T) xraystats.Manager {
+	t.Helper()
+	mgr, err := appstats.NewManager(context.Background(), &appstats.Config{})
+	if err != nil {
+		t.Fatalf("stats.NewManager() error = %v", err)
+	}
+	return mgr
+}
+
+func mustRegisterCounter(t *testing.T, mgr xraystats.Manager, name string) xraystats.Counter {
+	t.Helper()
+	counter, err := mgr.RegisterCounter(name)
+	if err != nil {
+		t.Fatalf("RegisterCounter(%q) error = %v", name, err)
+	}
+	return counter
+}
+
+func counterName(userID int, direction string) string {
+	return fmt.Sprintf("user>>>%s>>>traffic>>>%s", userEmail(userID), direction)
+}
+
+func newStatsBackedXray(t *testing.T, users []model.UserSpec, ld *LimitDispatcher) (*Xray, xraystats.Manager) {
+	t.Helper()
+	mgr := newStatsManager(t)
+	inst := new(xrayCore.Instance)
+	if err := inst.AddFeature(mgr); err != nil {
+		t.Fatalf("Instance.AddFeature(stats) error = %v", err)
+	}
+	x := New(config.KernelConfig{Type: "xray"})
+	x.instance = inst
+	x.limitDispatcher = ld
+	x.users = users
+	x.cumTraffic = make(map[int][2]int64)
+	x.running.Store(true)
+	return x, mgr
+}
+
+func TestXrayGetUserTrafficUsesBuiltInStatsAndDispatcherState(t *testing.T) {
+	users := []model.UserSpec{
+		{ID: 1, UUID: "uuid-1"},
+		{ID: 2, UUID: "uuid-2"},
+	}
+
+	ld := newTestDispatcher()
+	email1 := userEmail(1)
+	email2 := userEmail(2)
+	ld.UpdateLimits(map[string]int{email1: 1, email2: 2}, nil, nil)
+	ld.mu.Lock()
+	ld.limitedIPs[email1] = map[string]int{"1.1.1.1": 1}
+	ld.mu.Unlock()
+	ref := &atomic.Int64{}
+	ref.Store(1)
+	ic := &ipCounter{}
+	ic.ips.Store("2.2.2.2", ref)
+	ld.unlimitedIPs.Store(email2, ic)
+	ld.connCount.Store(7)
+
+	x, mgr := newStatsBackedXray(t, users, ld)
+	up1 := mustRegisterCounter(t, mgr, counterName(1, "uplink"))
+	down1 := mustRegisterCounter(t, mgr, counterName(1, "downlink"))
+	up2 := mustRegisterCounter(t, mgr, counterName(2, "uplink"))
+	down2 := mustRegisterCounter(t, mgr, counterName(2, "downlink"))
+	up1.Add(100)
+	down1.Add(200)
+	up2.Add(50)
+	down2.Add(80)
+
+	traffic, aliveIPs, connCount, err := x.GetUserTraffic(context.Background())
+	if err != nil {
+		t.Fatalf("GetUserTraffic() error = %v", err)
+	}
+	if connCount != 7 {
+		t.Fatalf("connCount = %d, want 7", connCount)
+	}
+	if got := traffic[1]; got != [2]int64{100, 200} {
+		t.Fatalf("traffic[1] = %v, want [100 200]", got)
+	}
+	if got := traffic[2]; got != [2]int64{50, 80} {
+		t.Fatalf("traffic[2] = %v, want [50 80]", got)
+	}
+	if !aliveIPs[1]["1.1.1.1"] {
+		t.Fatal("expected limited-user IP to be reported")
+	}
+	if !aliveIPs[2]["2.2.2.2"] {
+		t.Fatal("expected unlimited-user IP to be reported")
+	}
+
+	up1.Add(30)
+	down1.Add(40)
+	traffic, _, _, err = x.GetUserTraffic(context.Background())
+	if err != nil {
+		t.Fatalf("second GetUserTraffic() error = %v", err)
+	}
+	if got := traffic[1]; got != [2]int64{130, 240} {
+		t.Fatalf("traffic[1] after second poll = %v, want [130 240]", got)
+	}
+	if got := traffic[2]; got != [2]int64{50, 80} {
+		t.Fatalf("traffic[2] after second poll = %v, want [50 80]", got)
+	}
+}
+
+func TestXrayUpdateDispatcherLimitsPropagatesDeviceMetadata(t *testing.T) {
+	x := New(config.KernelConfig{Type: "xray"})
+	ld := newTestDispatcher()
+	x.limitDispatcher = ld
+
+	users := []model.UserSpec{
+		{ID: 7, UUID: "uuid-7", DeviceLimit: 2, SpeedLimit: 12},
+		{ID: 9, UUID: "uuid-9", DeviceLimit: 0, SpeedLimit: 0},
+	}
+
+	x.updateDispatcherLimits(users)
+
+	if got := ld.deviceLimits[userEmail(7)]; got != 2 {
+		t.Fatalf("deviceLimits[email] = %d, want 2", got)
+	}
+	if got := ld.deviceLimits["uuid-7"]; got != 2 {
+		t.Fatalf("deviceLimits[uuid] = %d, want 2", got)
+	}
+	if _, ok := ld.deviceLimits[userEmail(9)]; ok {
+		t.Fatal("unexpected device limit entry for unlimited user")
+	}
+}
+
+func TestXraySetSpeedLimitFuncUsesPatchedCorePath(t *testing.T) {
+	x := New(config.KernelConfig{Type: "xray"})
+	called := false
+
+	x.SetSpeedLimitFunc(func(string) *rate.Limiter {
+		called = true
+		return rate.NewLimiter(rate.Limit(1), 1)
+	})
+
+	if called {
+		t.Fatal("patched xray path should not consult external speed-limit callbacks")
+	}
+
+	mu, err := toMemoryUser("vless", &model.NodeSpec{Protocol: "vless"}, model.UserSpec{ID: 1, UUID: "11111111-1111-1111-1111-111111111111", SpeedLimit: 8})
+	if err != nil {
+		t.Fatalf("toMemoryUser() error = %v", err)
+	}
+	if mu.Level != 0 {
+		t.Fatalf("MemoryUser.Level = %d, want 0", mu.Level)
+	}
+}
+
+
+func TestXrayCapabilities(t *testing.T) {
+	x := New(config.KernelConfig{Type: "xray"})
+	caps := x.Capabilities()
+	if !caps.PerUserSpeedLimit || !caps.DeviceLimit || !caps.BuiltInTrafficStats || !caps.AliveIPTracking {
+		t.Fatalf("unexpected positive xray capabilities: %+v", caps)
+	}
+	if caps.ForceCloseConnection || caps.ForceCloseUser {
+		t.Fatalf("unexpected force-close xray capabilities: %+v", caps)
+	}
+}
+
+
+func TestXrayUpdateBandwidthLimitsWritesPatchedCoreFeature(t *testing.T) {
+	inst := new(xrayCore.Instance)
+	bm := featurebandwidth.New()
+	if err := inst.AddFeature(bm); err != nil {
+		t.Fatalf("Instance.AddFeature(bandwidth) error = %v", err)
+	}
+	x := New(config.KernelConfig{Type: "xray"})
+	x.instance = inst
+	x.updateBandwidthLimits([]model.UserSpec{{ID: 1, UUID: "uuid-1", SpeedLimit: 8}})
+	lim := bm.GetUserLimiter(userEmail(1))
+	if lim == nil {
+		t.Fatal("expected patched bandwidth feature to receive user limiter")
+	}
+}
+
+
+func TestXrayUpdateBandwidthLimitsUsesSpeedLimitFunc(t *testing.T) {
+	inst := new(xrayCore.Instance)
+	bm := featurebandwidth.New()
+	if err := inst.AddFeature(bm); err != nil {
+		t.Fatalf("Instance.AddFeature(bandwidth) error = %v", err)
+	}
+	x := New(config.KernelConfig{Type: "xray"})
+	x.instance = inst
+	x.users = []model.UserSpec{{ID: 1, UUID: "uuid-1", SpeedLimit: 8}}
+	shared := rate.NewLimiter(7, 7)
+	x.SetSpeedLimitFunc(func(uuid string) *rate.Limiter {
+		if uuid != "uuid-1" {
+			t.Fatalf("unexpected uuid: %s", uuid)
+		}
+		return shared
+	})
+	lim := bm.GetUserLimiter(userEmail(1))
+	if lim != shared {
+		t.Fatal("expected patched bandwidth feature to consume shared limiter from callback")
+	}
+}
+
+
+func TestXrayUpdateBandwidthLimitsFallsBackToUserSpeed(t *testing.T) {
+	inst := new(xrayCore.Instance)
+	bm := featurebandwidth.New()
+	if err := inst.AddFeature(bm); err != nil {
+		t.Fatalf("Instance.AddFeature(bandwidth) error = %v", err)
+	}
+	x := New(config.KernelConfig{Type: "xray"})
+	x.instance = inst
+	x.users = []model.UserSpec{{ID: 2, UUID: "uuid-2", SpeedLimit: 16}}
+	x.updateBandwidthLimits(nil)
+	if bm.GetUserLimiter(userEmail(2)) == nil {
+		t.Fatal("expected fallback limiter derived from user speed")
+	}
+}
+
+
+func TestXrayUpdateUsersLimitOnlyRefreshesDispatcherAndBandwidth(t *testing.T) {
+	x := New(config.KernelConfig{Type: "xray"})
+	ld := newTestDispatcher()
+	inst := new(xrayCore.Instance)
+	bm := featurebandwidth.New()
+	if err := inst.AddFeature(bm); err != nil {
+		t.Fatalf("Instance.AddFeature(bandwidth) error = %v", err)
+	}
+	x.instance = inst
+	x.limitDispatcher = ld
+	x.users = []model.UserSpec{{ID: 1, UUID: "uuid-1", DeviceLimit: 1, SpeedLimit: 8}}
+
+	added, removed, err := x.UpdateUsers([]model.UserSpec{{ID: 1, UUID: "uuid-1", DeviceLimit: 3, SpeedLimit: 16}})
+	if err != nil {
+		t.Fatalf("UpdateUsers() error = %v", err)
+	}
+	if added != 0 || removed != 0 {
+		t.Fatalf("UpdateUsers() counts = (%d, %d), want (0, 0)", added, removed)
+	}
+	if got := ld.deviceLimits[userEmail(1)]; got != 3 {
+		t.Fatalf("device limit after refresh = %d, want 3", got)
+	}
+	if bm.GetUserLimiter(userEmail(1)) == nil {
+		t.Fatal("expected bandwidth limiter to be refreshed for unchanged user set")
+	}
+}
+
+func TestXrayRemoveUsersStopsKernelWhenLastUserRemoved(t *testing.T) {
+	x := New(config.KernelConfig{Type: "xray"})
+	x.instance = new(xrayCore.Instance)
+	ld := newTestDispatcher()
+	x.limitDispatcher = ld
+	x.users = []model.UserSpec{{ID: 1, UUID: "uuid-1"}}
+	x.running.Store(true)
+
+	removed, err := x.RemoveUsers([]model.UserSpec{{ID: 1, UUID: "uuid-1"}})
+	if err != nil {
+		t.Fatalf("RemoveUsers() error = %v", err)
+	}
+	if removed != 1 {
+		t.Fatalf("removed = %d, want 1", removed)
+	}
+	if x.IsRunning() {
+		t.Fatal("expected xray to stop when last user is removed")
+	}
+}

internal/limiter/limiter.go

@@ -1,45 +1,38 @@
 package limiter
 
 import (
-	"fmt"
-	"log/slog"
-	"net/netip"
-	"sort"
 	"sync"
 	"sync/atomic"
 
-	"github.com/cedar2025/xboard-node/internal/kernel"
-	"github.com/cedar2025/xboard-node/internal/panel"
+	"github.com/cedar2025/xboard-node/internal/model"
 )
 
-// KickAction represents a connection that should be closed.
-type KickAction struct {
-	ConnID string
-	UserID int
-	Reason string
-}
-
 // Limiter enforces per-user device limits and detects removed users.
 type Limiter struct {
 	mu    sync.RWMutex
-	users map[int]panel.User
+	users map[int]model.UserSpec
+
+	// uuidDeviceLimit provides O(1) UUID→device-limit lookup for the
+	// kernel gate-keeping hot path (called per new connection).
+	uuidDeviceLimit map[string]int
 
 	deviceLimitEvents atomic.Uint64
 }
 
 func New() *Limiter {
 	return &Limiter{
-		users: make(map[int]panel.User),
+		users:           make(map[int]model.UserSpec),
+		uuidDeviceLimit: make(map[string]int),
 	}
 }
 
 // UpdateUsers refreshes the user limit configuration and returns
 // the IDs of users that were present before but are now removed.
-func (l *Limiter) UpdateUsers(users []panel.User) []int {
+func (l *Limiter) UpdateUsers(users []model.UserSpec) []int {
 	l.mu.Lock()
 	defer l.mu.Unlock()
 
-	newUsers := make(map[int]panel.User, len(users))
+	newUsers := make(map[int]model.UserSpec, len(users))
 	for _, u := range users {
 		newUsers[u.ID] = u
 	}
@@ -53,91 +46,19 @@ func (l *Limiter) UpdateUsers(users []panel.User) []int {
 	}
 
 	l.users = newUsers
+
+	// Rebuild UUID→device-limit index for O(1) lookups.
+	idx := make(map[string]int, len(newUsers))
+	for _, u := range newUsers {
+		if u.DeviceLimit > 0 {
+			idx[u.UUID] = u.DeviceLimit
+		}
+	}
+	l.uuidDeviceLimit = idx
+
 	return removed
 }
 
-// Check inspects connections against device limits.
-// Returns a list of connections that should be closed.
-func (l *Limiter) Check(conns []kernel.Connection) []KickAction {
-	l.mu.RLock()
-	defer l.mu.RUnlock()
-
-	var kicks []KickAction
-
-	// Device limit check
-	userIPs := make(map[int]map[string]bool)
-	userConnsByIP := make(map[int]map[string][]string) // user_id -> ip -> []conn_ids
-	for _, conn := range conns {
-		if conn.UserID == 0 || conn.SourceIP == "" {
-			continue
-		}
-		if userIPs[conn.UserID] == nil {
-			userIPs[conn.UserID] = make(map[string]bool)
-			userConnsByIP[conn.UserID] = make(map[string][]string)
-		}
-		userIPs[conn.UserID][conn.SourceIP] = true
-		userConnsByIP[conn.UserID][conn.SourceIP] = append(
-			userConnsByIP[conn.UserID][conn.SourceIP], conn.ID,
-		)
-	}
-
-	for uid, ips := range userIPs {
-		user, ok := l.users[uid]
-		if !ok || user.DeviceLimit == 0 {
-			continue
-		}
-		if len(ips) <= user.DeviceLimit {
-			continue
-		}
-
-		// Count a device-limit violation event for metrics reporting.
-		l.deviceLimitEvents.Add(1)
-
-		slog.Warn("user exceeded device limit",
-			"user_id", uid,
-			"devices", len(ips),
-			"limit", user.DeviceLimit,
-		)
-
-		// Sort IPs by numeric address so the "allowed" set is deterministic
-		// across calls. netip.Addr comparison avoids the lexicographic ordering
-		// quirks of string sort (e.g. "10.x" < "192.x", IPv6 always last).
-		// Parse all addresses up front to avoid O(n log n) repeated parsing.
-		type parsedIP struct {
-			raw  string
-			addr netip.Addr
-		}
-		ipList := make([]parsedIP, 0, len(ips))
-		for ip := range ips {
-			a, _ := netip.ParseAddr(ip)
-			ipList = append(ipList, parsedIP{raw: ip, addr: a})
-		}
-		sort.Slice(ipList, func(i, j int) bool {
-			return ipList[i].addr.Compare(ipList[j].addr) < 0
-		})
-
-		allowed := make(map[string]bool, user.DeviceLimit)
-		for i := 0; i < user.DeviceLimit && i < len(ipList); i++ {
-			allowed[ipList[i].raw] = true
-		}
-
-		for ip, connIDs := range userConnsByIP[uid] {
-			if allowed[ip] {
-				continue
-			}
-			for _, connID := range connIDs {
-				kicks = append(kicks, KickAction{
-					ConnID: connID,
-					UserID: uid,
-					Reason: fmt.Sprintf("devices %d > limit %d", len(ips), user.DeviceLimit),
-				})
-			}
-		}
-	}
-
-	return kicks
-}
-
 // LimiterMetrics holds aggregated limiter statistics.
 type LimiterMetrics struct {
 	// DeviceLimitEvents is the total number of times a user exceeded
@@ -152,26 +73,12 @@ func (l *Limiter) SnapshotMetrics() LimiterMetrics {
 	}
 }
 
-// KickUsers returns KickActions for all connections belonging to the given user IDs.
-func (l *Limiter) KickUsers(conns []kernel.Connection, userIDs []int) []KickAction {
-	if len(userIDs) == 0 {
-		return nil
-	}
-
-	remove := make(map[int]bool, len(userIDs))
-	for _, id := range userIDs {
-		remove[id] = true
-	}
-
-	var kicks []KickAction
-	for _, conn := range conns {
-		if remove[conn.UserID] {
-			kicks = append(kicks, KickAction{
-				ConnID: conn.ID,
-				UserID: conn.UserID,
-				Reason: "user removed from panel",
-			})
-		}
-	}
-	return kicks
+// GetDeviceLimitByUUID returns the device limit for a user identified by UUID.
+// Returns (limit, true) if a device limit is set, (0, false) otherwise.
+// This is designed to be passed as a function reference to kernel.SetDeviceLimitFunc.
+func (l *Limiter) GetDeviceLimitByUUID(uuid string) (int, bool) {
+	l.mu.RLock()
+	defer l.mu.RUnlock()
+	limit, ok := l.uuidDeviceLimit[uuid]
+	return limit, ok
 }

internal/limiter/limiter_test.go

@@ -4,8 +4,7 @@ import (
 	"sort"
 	"testing"
 
-	"github.com/cedar2025/xboard-node/internal/kernel"
-	"github.com/cedar2025/xboard-node/internal/panel"
+	"github.com/cedar2025/xboard-node/internal/model"
 )
 
 func TestNew(t *testing.T) {
@@ -17,7 +16,7 @@ func TestNew(t *testing.T) {
 
 func TestUpdateUsers(t *testing.T) {
 	l := New()
-	users := []panel.User{
+	users := []model.UserSpec{
 		{ID: 1, UUID: "u1", SpeedLimit: 3, DeviceLimit: 2},
 		{ID: 2, UUID: "u2", SpeedLimit: 0, DeviceLimit: 0},
 	}
@@ -26,25 +25,18 @@ func TestUpdateUsers(t *testing.T) {
 	if len(removed) != 0 {
 		t.Errorf("first update should have no removed users, got %v", removed)
 	}
-
-	// Verify by running a check
-	conns := []kernel.Connection{{ID: "c1", UserID: 1, SourceIP: "1.1.1.1"}}
-	kicks := l.Check(conns)
-	if len(kicks) != 0 {
-		t.Errorf("expected no kicks, got %d", len(kicks))
-	}
 }
 
 func TestUpdateUsers_DetectsRemoved(t *testing.T) {
 	l := New()
-	l.UpdateUsers([]panel.User{
+	l.UpdateUsers([]model.UserSpec{
 		{ID: 1, UUID: "u1"},
 		{ID: 2, UUID: "u2"},
 		{ID: 3, UUID: "u3"},
 	})
 
 	// Remove user 2
-	removed := l.UpdateUsers([]panel.User{
+	removed := l.UpdateUsers([]model.UserSpec{
 		{ID: 1, UUID: "u1"},
 		{ID: 3, UUID: "u3"},
 	})
@@ -56,14 +48,14 @@ func TestUpdateUsers_DetectsRemoved(t *testing.T) {
 
 func TestUpdateUsers_DetectsMultipleRemoved(t *testing.T) {
 	l := New()
-	l.UpdateUsers([]panel.User{
+	l.UpdateUsers([]model.UserSpec{
 		{ID: 1, UUID: "u1"},
 		{ID: 2, UUID: "u2"},
 		{ID: 3, UUID: "u3"},
 	})
 
 	// Remove users 1 and 3
-	removed := l.UpdateUsers([]panel.User{
+	removed := l.UpdateUsers([]model.UserSpec{
 		{ID: 2, UUID: "u2"},
 	})
 
@@ -75,12 +67,12 @@ func TestUpdateUsers_DetectsMultipleRemoved(t *testing.T) {
 
 func TestUpdateUsers_NoRemovals(t *testing.T) {
 	l := New()
-	l.UpdateUsers([]panel.User{
+	l.UpdateUsers([]model.UserSpec{
 		{ID: 1, UUID: "u1"},
 	})
 
 	// Add user, keep existing
-	removed := l.UpdateUsers([]panel.User{
+	removed := l.UpdateUsers([]model.UserSpec{
 		{ID: 1, UUID: "u1"},
 		{ID: 2, UUID: "u2"},
 	})
@@ -92,248 +84,46 @@ func TestUpdateUsers_NoRemovals(t *testing.T) {
 
 func TestUpdateUsers_ReplacesAll(t *testing.T) {
 	l := New()
-	l.UpdateUsers([]panel.User{{ID: 1, DeviceLimit: 1}})
+	l.UpdateUsers([]model.UserSpec{{ID: 1, DeviceLimit: 1}})
 
 	// Replace with different users
-	removed := l.UpdateUsers([]panel.User{{ID: 2, DeviceLimit: 1}})
+	removed := l.UpdateUsers([]model.UserSpec{{ID: 2, DeviceLimit: 1}})
 
 	if len(removed) != 1 || removed[0] != 1 {
 		t.Errorf("expected removed=[1], got %v", removed)
 	}
 }
 
-func TestDeviceLimit_NoKick_UnderLimit(t *testing.T) {
+func TestGetDeviceLimitByUUID(t *testing.T) {
 	l := New()
-	l.UpdateUsers([]panel.User{
-		{ID: 1, DeviceLimit: 3},
+	l.UpdateUsers([]model.UserSpec{
+		{ID: 1, UUID: "u1", DeviceLimit: 3},
+		{ID: 2, UUID: "u2", DeviceLimit: 0},
 	})
 
-	conns := []kernel.Connection{
-		{ID: "c1", UserID: 1, SourceIP: "1.1.1.1"},
-		{ID: "c2", UserID: 1, SourceIP: "2.2.2.2"},
+	// User with device limit
+	limit, ok := l.GetDeviceLimitByUUID("u1")
+	if !ok || limit != 3 {
+		t.Errorf("expected (3, true), got (%d, %v)", limit, ok)
 	}
-	kicks := l.Check(conns)
 
-	if len(kicks) != 0 {
-		t.Errorf("2 devices under limit 3, got %d kicks", len(kicks))
+	// User without device limit
+	_, ok = l.GetDeviceLimitByUUID("u2")
+	if ok {
+		t.Error("expected (_, false) for user with no device limit")
+	}
+
+	// Unknown UUID
+	_, ok = l.GetDeviceLimitByUUID("unknown")
+	if ok {
+		t.Error("expected (_, false) for unknown UUID")
 	}
 }
 
-func TestDeviceLimit_Kick_OverLimit(t *testing.T) {
+func TestSnapshotMetrics(t *testing.T) {
 	l := New()
-	l.UpdateUsers([]panel.User{
-		{ID: 1, DeviceLimit: 2},
-	})
-
-	conns := []kernel.Connection{
-		{ID: "c1", UserID: 1, SourceIP: "1.1.1.1"},
-		{ID: "c2", UserID: 1, SourceIP: "2.2.2.2"},
-		{ID: "c3", UserID: 1, SourceIP: "3.3.3.3"},
-	}
-	kicks := l.Check(conns)
-
-	if len(kicks) == 0 {
-		t.Fatal("expected kicks for exceeding device limit")
-	}
-	if len(kicks) < 1 {
-		t.Errorf("expected at least 1 kick, got %d", len(kicks))
-	}
-}
-
-func TestDeviceLimit_ExactlyAtLimit(t *testing.T) {
-	l := New()
-	l.UpdateUsers([]panel.User{
-		{ID: 1, DeviceLimit: 2},
-	})
-
-	conns := []kernel.Connection{
-		{ID: "c1", UserID: 1, SourceIP: "1.1.1.1"},
-		{ID: "c2", UserID: 1, SourceIP: "2.2.2.2"},
-	}
-	kicks := l.Check(conns)
-
-	if len(kicks) != 0 {
-		t.Errorf("exactly at limit should not kick, got %d", len(kicks))
-	}
-}
-
-func TestDeviceLimit_UnlimitedDevices(t *testing.T) {
-	l := New()
-	l.UpdateUsers([]panel.User{
-		{ID: 1, DeviceLimit: 0}, // 0 = unlimited
-	})
-
-	conns := []kernel.Connection{
-		{ID: "c1", UserID: 1, SourceIP: "1.1.1.1"},
-		{ID: "c2", UserID: 1, SourceIP: "2.2.2.2"},
-		{ID: "c3", UserID: 1, SourceIP: "3.3.3.3"},
-		{ID: "c4", UserID: 1, SourceIP: "4.4.4.4"},
-		{ID: "c5", UserID: 1, SourceIP: "5.5.5.5"},
-	}
-	kicks := l.Check(conns)
-
-	if len(kicks) != 0 {
-		t.Errorf("unlimited devices should not be kicked, got %d", len(kicks))
-	}
-}
-
-func TestDeviceLimit_SameIPMultipleConnections(t *testing.T) {
-	l := New()
-	l.UpdateUsers([]panel.User{
-		{ID: 1, DeviceLimit: 1},
-	})
-
-	conns := []kernel.Connection{
-		{ID: "c1", UserID: 1, SourceIP: "1.1.1.1"},
-		{ID: "c2", UserID: 1, SourceIP: "1.1.1.1"},
-		{ID: "c3", UserID: 1, SourceIP: "1.1.1.1"},
-	}
-	kicks := l.Check(conns)
-
-	if len(kicks) != 0 {
-		t.Errorf("same IP multiple conns should count as 1 device, got %d kicks", len(kicks))
-	}
-}
-
-func TestDeviceLimit_EmptySourceIP(t *testing.T) {
-	l := New()
-	l.UpdateUsers([]panel.User{
-		{ID: 1, DeviceLimit: 1},
-	})
-
-	conns := []kernel.Connection{
-		{ID: "c1", UserID: 1, SourceIP: ""},
-		{ID: "c2", UserID: 1, SourceIP: ""},
-	}
-	kicks := l.Check(conns)
-
-	if len(kicks) != 0 {
-		t.Errorf("empty source IP should be ignored, got %d kicks", len(kicks))
-	}
-}
-
-func TestMultipleUsers_DeviceLimits(t *testing.T) {
-	l := New()
-	l.UpdateUsers([]panel.User{
-		{ID: 1, DeviceLimit: 1},
-		{ID: 2, DeviceLimit: 1},
-	})
-
-	conns := []kernel.Connection{
-		{ID: "c1", UserID: 1, SourceIP: "1.1.1.1"},
-		{ID: "c2", UserID: 2, SourceIP: "2.2.2.2"},
-		{ID: "c3", UserID: 2, SourceIP: "3.3.3.3"},
-	}
-	kicks := l.Check(conns)
-
-	deviceKicks := 0
-	for _, k := range kicks {
-		if k.UserID == 2 {
-			deviceKicks++
-		}
-	}
-	if deviceKicks == 0 {
-		t.Error("expected device kick for user 2")
-	}
-
-	for _, k := range kicks {
-		if k.UserID == 1 {
-			t.Error("user 1 should not be kicked")
-		}
-	}
-}
-
-func TestNoSpeedKick(t *testing.T) {
-	// Speed limits should NOT cause kicks — only device limits and user removal
-	l := New()
-	l.UpdateUsers([]panel.User{
-		{ID: 1, SpeedLimit: 1, DeviceLimit: 0},
-	})
-
-	conns := []kernel.Connection{
-		{ID: "c1", UserID: 1, SourceIP: "1.1.1.1"},
-	}
-	kicks := l.Check(conns)
-	if len(kicks) != 0 {
-		t.Errorf("speed limits should not cause kicks, got %d", len(kicks))
-	}
-}
-
-func TestKickUsers_Basic(t *testing.T) {
-	l := New()
-
-	conns := []kernel.Connection{
-		{ID: "c1", UserID: 1, SourceIP: "1.1.1.1"},
-		{ID: "c2", UserID: 2, SourceIP: "2.2.2.2"},
-		{ID: "c3", UserID: 1, SourceIP: "1.1.1.2"},
-		{ID: "c4", UserID: 3, SourceIP: "3.3.3.3"},
-	}
-
-	kicks := l.KickUsers(conns, []int{1})
-
-	if len(kicks) != 2 {
-		t.Fatalf("expected 2 kicks for user 1, got %d", len(kicks))
-	}
-	for _, k := range kicks {
-		if k.UserID != 1 {
-			t.Errorf("expected kick for user 1, got user %d", k.UserID)
-		}
-		if k.Reason != "user removed from panel" {
-			t.Errorf("unexpected reason: %s", k.Reason)
-		}
-	}
-}
-
-func TestKickUsers_MultipleUsers(t *testing.T) {
-	l := New()
-
-	conns := []kernel.Connection{
-		{ID: "c1", UserID: 1, SourceIP: "1.1.1.1"},
-		{ID: "c2", UserID: 2, SourceIP: "2.2.2.2"},
-		{ID: "c3", UserID: 3, SourceIP: "3.3.3.3"},
-	}
-
-	kicks := l.KickUsers(conns, []int{1, 3})
-
-	if len(kicks) != 2 {
-		t.Fatalf("expected 2 kicks, got %d", len(kicks))
-	}
-	kickedUsers := map[int]bool{}
-	for _, k := range kicks {
-		kickedUsers[k.UserID] = true
-	}
-	if !kickedUsers[1] || !kickedUsers[3] {
-		t.Errorf("expected users 1 and 3 kicked, got %v", kickedUsers)
-	}
-}
-
-func TestKickUsers_NoUsers(t *testing.T) {
-	l := New()
-
-	conns := []kernel.Connection{
-		{ID: "c1", UserID: 1, SourceIP: "1.1.1.1"},
-	}
-
-	kicks := l.KickUsers(conns, nil)
-	if len(kicks) != 0 {
-		t.Errorf("expected no kicks for nil user list, got %d", len(kicks))
-	}
-
-	kicks = l.KickUsers(conns, []int{})
-	if len(kicks) != 0 {
-		t.Errorf("expected no kicks for empty user list, got %d", len(kicks))
-	}
-}
-
-func TestKickUsers_NoMatchingConnections(t *testing.T) {
-	l := New()
-
-	conns := []kernel.Connection{
-		{ID: "c1", UserID: 1, SourceIP: "1.1.1.1"},
-	}
-
-	kicks := l.KickUsers(conns, []int{99})
-	if len(kicks) != 0 {
-		t.Errorf("expected no kicks for non-existent user, got %d", len(kicks))
+	m := l.SnapshotMetrics()
+	if m.DeviceLimitEvents != 0 {
+		t.Errorf("expected 0 events initially, got %d", m.DeviceLimitEvents)
 	}
 }

internal/limiter/speedtracker.go

@@ -1,14 +1,16 @@
 package limiter
 
 import (
-	"log/slog"
 	"sync"
 	"sync/atomic"
 
-	"github.com/cedar2025/xboard-node/internal/panel"
+	"github.com/cedar2025/xboard-node/internal/model"
 	"golang.org/x/time/rate"
 )
 
+// SpeedTrackerLogCallback is called when bucket updates occur.
+type SpeedTrackerLogCallback func(msg string)
+
 // SpeedTracker manages per-user token-bucket rate limiters.
 // It does NOT wrap connections itself — instead, ConnTracker consults it
 // via GetLimiter to embed rate limiting in the same tracked connection
@@ -22,6 +24,9 @@ type SpeedTracker struct {
 	// Fast-path: when no users have a speed limit, GetLimiter returns nil
 	// immediately without any map lookup.
 	hasLimits atomic.Bool
+
+	// Optional callback for logging
+	logFunc SpeedTrackerLogCallback
 }
 
 // NewSpeedTracker creates a bucket manager for per-user bandwidth throttling.
@@ -33,75 +38,113 @@ func NewSpeedTracker(l *Limiter) *SpeedTracker {
 	}
 }
 
-// UpdateBuckets performs an incremental update of rate limiter buckets.
-// It reuses existing rate.Limiter instances to avoid burst resets and
-// minimizes memory allocations.
+// SetLogCallback sets the logging callback.
+func (t *SpeedTracker) SetLogCallback(f SpeedTrackerLogCallback) {
+	t.logFunc = f
+}
+
+// UpdateBuckets updates the UUID→userID mapping and syncs existing limiters.
 func (t *SpeedTracker) UpdateBuckets() {
+	currentUsers := make([]model.UserSpec, 0, 32)
 	t.limiter.mu.RLock()
-	currentUsers := make([]panel.User, 0, len(t.limiter.users))
 	for _, u := range t.limiter.users {
 		currentUsers = append(currentUsers, u)
 	}
 	t.limiter.mu.RUnlock()
 
-	t.mu.Lock()
-	defer t.mu.Unlock()
+	func() {
+		t.mu.Lock()
+		defer t.mu.Unlock()
 
-	newUUIDMap := make(map[string]int, len(currentUsers))
-	activeIDs := make(map[int]struct{}, len(currentUsers))
+		newUUIDMap := make(map[string]int, len(currentUsers))
+		activeIDs := make(map[int]struct{}, len(currentUsers))
 
-	for _, user := range currentUsers {
-		activeIDs[user.ID] = struct{}{}
-		if user.UUID != "" {
-			newUUIDMap[user.UUID] = user.ID
+		for _, user := range currentUsers {
+			activeIDs[user.ID] = struct{}{}
+			if user.UUID != "" {
+				newUUIDMap[user.UUID] = user.ID
+			}
+
+			// Update existing limiter if speed changed
+			if lim, ok := t.buckets[user.ID]; ok {
+				if user.SpeedLimit > 0 {
+					bytesPerSec := int(user.SpeedLimit) * 1_000_000 / 8
+					burst := bytesPerSec
+					if burst < 64*1024 {
+						burst = 64 * 1024
+					}
+					lim.SetLimit(rate.Limit(bytesPerSec))
+					lim.SetBurst(burst)
+				} else {
+					delete(t.buckets, user.ID)
+				}
+			}
 		}
 
-		if user.SpeedLimit <= 0 {
-			delete(t.buckets, user.ID)
-			continue
+		// Clean up buckets for removed users
+		for id := range t.buckets {
+			if _, ok := activeIDs[id]; !ok {
+				delete(t.buckets, id)
+			}
 		}
 
-		bytesPerSec := int(user.SpeedLimit) * 1_000_000 / 8
-		burst := bytesPerSec
-		if burst < 64*1024 {
-			burst = 64 * 1024
-		}
-		if cap4s := bytesPerSec * 4; cap4s > 64*1024 && burst > cap4s {
-			burst = cap4s
-		}
+		t.uuidMap = newUUIDMap
+		t.hasLimits.Store(len(t.buckets) > 0)
+	}()
 
-		if existing, ok := t.buckets[user.ID]; ok {
-			// Thread-safe in-place update. Existing connections continue
-			// using this limiter with the new rate immediately.
-			existing.SetLimit(rate.Limit(bytesPerSec))
-			existing.SetBurst(burst)
-		} else {
-			t.buckets[user.ID] = rate.NewLimiter(rate.Limit(bytesPerSec), burst)
-		}
+	if t.logFunc != nil {
+		t.logFunc("buckets updated")
 	}
-
-	// Remove limiters for users who are no longer in the list
-	for id := range t.buckets {
-		if _, ok := activeIDs[id]; !ok {
-			delete(t.buckets, id)
-		}
-	}
-
-	t.uuidMap = newUUIDMap
-	t.hasLimits.Store(len(t.buckets) > 0)
-	slog.Info("speedtracker: buckets updated", "active_limiters", len(t.buckets))
 }
 
 // GetLimiter returns the rate limiter for the given user UUID, or nil if
-// no limit applies. This is called by ConnTracker for every new connection.
+// no limit applies. Creates limiter on-demand if not exists.
 // Thread-safe.
 func (t *SpeedTracker) GetLimiter(user string) *rate.Limiter {
-	if !t.hasLimits.Load() {
+	t.mu.RLock()
+	uid, exists := t.uuidMap[user]
+	if !exists {
+		t.mu.RUnlock()
 		return nil
 	}
-	t.mu.RLock()
-	defer t.mu.RUnlock()
-	return t.buckets[t.uuidMap[user]]
+	if lim, ok := t.buckets[uid]; ok {
+		t.mu.RUnlock()
+		return lim
+	}
+	t.mu.RUnlock()
+
+	// Get user info from limiter
+	t.limiter.mu.RLock()
+	u, userExists := t.limiter.users[uid]
+	t.limiter.mu.RUnlock()
+
+	if !userExists || u.SpeedLimit <= 0 {
+		return nil
+	}
+
+	// Create limiter on-demand
+	bytesPerSec := int(u.SpeedLimit) * 1_000_000 / 8
+	burst := bytesPerSec
+	if burst < 64*1024 {
+		burst = 64 * 1024
+	}
+	if cap4s := bytesPerSec * 4; cap4s > 64*1024 && burst > cap4s {
+		burst = cap4s
+	}
+
+	lim := rate.NewLimiter(rate.Limit(bytesPerSec), burst)
+
+	t.mu.Lock()
+	// Double-check after acquiring write lock
+	if existing, ok := t.buckets[uid]; ok {
+		t.mu.Unlock()
+		return existing
+	}
+	t.buckets[uid] = lim
+	t.hasLimits.Store(true)
+	t.mu.Unlock()
+
+	return lim
 }
 
 // HasLimits returns true if any user currently has a speed limit configured.

internal/limiter/speedtracker_test.go

@@ -2,15 +2,16 @@ package limiter
 
 import (
 	"testing"
+	"time"
 
-	"github.com/cedar2025/xboard-node/internal/panel"
+	"github.com/cedar2025/xboard-node/internal/model"
 )
 
 func TestSpeedTracker_UpdateBuckets(t *testing.T) {
 	l := New()
 	st := NewSpeedTracker(l)
 
-	users := []panel.User{
+	users := []model.UserSpec{
 		{ID: 1, UUID: "u1", SpeedLimit: 3, DeviceLimit: 0},  // 3 Mbps
 		{ID: 2, UUID: "u2", SpeedLimit: 0, DeviceLimit: 0},  // unlimited
 		{ID: 3, UUID: "u3", SpeedLimit: 10, DeviceLimit: 2}, // 10 Mbps
@@ -59,7 +60,7 @@ func TestSpeedTracker_UpdateBuckets_PreservesExisting(t *testing.T) {
 	l := New()
 	st := NewSpeedTracker(l)
 
-	users := []panel.User{
+	users := []model.UserSpec{
 		{ID: 1, UUID: "u1", SpeedLimit: 3, DeviceLimit: 0},
 	}
 	l.UpdateUsers(users)
@@ -71,7 +72,7 @@ func TestSpeedTracker_UpdateBuckets_PreservesExisting(t *testing.T) {
 	}
 
 	// Update same user with different speed — should reuse same *rate.Limiter pointer
-	users2 := []panel.User{
+	users2 := []model.UserSpec{
 		{ID: 1, UUID: "u1", SpeedLimit: 5, DeviceLimit: 0},
 	}
 	l.UpdateUsers(users2)
@@ -96,7 +97,7 @@ func TestSpeedTracker_UpdateBuckets_RemovesUsers(t *testing.T) {
 	l := New()
 	st := NewSpeedTracker(l)
 
-	users := []panel.User{
+	users := []model.UserSpec{
 		{ID: 1, UUID: "u1", SpeedLimit: 3, DeviceLimit: 0},
 		{ID: 2, UUID: "u2", SpeedLimit: 5, DeviceLimit: 0},
 	}
@@ -108,7 +109,7 @@ func TestSpeedTracker_UpdateBuckets_RemovesUsers(t *testing.T) {
 	}
 
 	// Remove user 2
-	users2 := []panel.User{
+	users2 := []model.UserSpec{
 		{ID: 1, UUID: "u1", SpeedLimit: 3, DeviceLimit: 0},
 	}
 	l.UpdateUsers(users2)
@@ -121,3 +122,26 @@ func TestSpeedTracker_UpdateBuckets_RemovesUsers(t *testing.T) {
 		t.Error("expected no bucket for removed user 2")
 	}
 }
+
+// Regression: log callback must not run while UpdateBuckets holds t.mu.Lock,
+// otherwise callbacks that call LimitedUserCount() self-deadlock on RWMutex.
+func TestSpeedTracker_UpdateBuckets_LogCallbackMayCallLimitedUserCount(t *testing.T) {
+	l := New()
+	st := NewSpeedTracker(l)
+	l.UpdateUsers([]model.UserSpec{{ID: 1, UUID: "u1", SpeedLimit: 1}})
+	st.SetLogCallback(func(msg string) {
+		_ = st.LimitedUserCount()
+	})
+
+	done := make(chan struct{})
+	go func() {
+		st.UpdateBuckets()
+		close(done)
+	}()
+
+	select {
+	case <-done:
+	case <-time.After(2 * time.Second):
+		t.Fatal("deadlock: UpdateBuckets did not finish (log callback vs RWMutex)")
+	}
+}

internal/machine/machine.go

@@ -0,0 +1,413 @@
+// Package machine implements the machine-mode orchestrator that dynamically
+// discovers nodes from the panel's machine API and manages their lifecycles.
+package machine
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"sync"
+	"time"
+
+	"github.com/cedar2025/xboard-node/internal/config"
+	"github.com/cedar2025/xboard-node/internal/controlplane"
+	"github.com/cedar2025/xboard-node/internal/model"
+	"github.com/cedar2025/xboard-node/internal/monitor"
+	"github.com/cedar2025/xboard-node/internal/nlog"
+	"github.com/cedar2025/xboard-node/internal/panel"
+	"github.com/cedar2025/xboard-node/internal/service"
+)
+
+// nodeHandle tracks a running node service.
+type nodeHandle struct {
+	cancel  context.CancelFunc
+	done    chan struct{}
+	mailbox *controlplane.NodeMailbox
+}
+
+// Orchestrator manages all nodes bound to a panel machine. It:
+//   - discovers nodes via GET /machine/nodes
+//   - starts / stops Service instances as nodes are added / removed
+//   - maintains a shared WS connection that demuxes events by node_id
+//   - reports machine-level load via POST /machine/status
+type Orchestrator struct {
+	cfg    *config.Config
+	client *panel.Client // machine-level client (no node_id)
+
+	mu    sync.Mutex
+	nodes map[int]*nodeHandle // node_id → handle
+
+	// Per-node mailbox keyed by node_id. Shared WS events are aggregated here
+	// and each node service drains the latest state when ready.
+	eventsMu  sync.RWMutex
+	mailboxes map[int]*controlplane.NodeMailbox
+	statuses  map[int]chan<- controlplane.StatusChange
+
+	// Shared WS client (nil when WS is disabled).
+	ws       *panel.WSClient
+	wsCancel context.CancelFunc
+
+	// runCtx is stored from Run() so that onWSEvent can trigger rediscover
+	// for sync.nodes events without blocking the main loop.
+	runCtx context.Context
+
+	pullInterval time.Duration
+	pushInterval time.Duration
+}
+
+// New creates a machine orchestrator from the given config.
+func New(cfg *config.Config) *Orchestrator {
+	panelCfg := config.PanelConfig{
+		URL:       cfg.Panel.URL,
+		Token:     cfg.Machine.Token,
+		MachineID: cfg.Machine.MachineID,
+	}
+	return &Orchestrator{
+		cfg:       cfg,
+		client:    panel.NewClient(panelCfg),
+		nodes:     make(map[int]*nodeHandle),
+		mailboxes: make(map[int]*controlplane.NodeMailbox),
+		statuses:  make(map[int]chan<- controlplane.StatusChange),
+	}
+}
+
+// Run is the main loop. It blocks until ctx is cancelled.
+func (o *Orchestrator) Run(ctx context.Context) error {
+	o.runCtx = ctx
+	nodesResp, err := o.client.GetMachineNodes()
+	if err != nil {
+		return fmt.Errorf("initial node discovery: %w", err)
+	}
+
+	o.applyIntervals(nodesResp.BaseConfig)
+	nlog.Core().Info(fmt.Sprintf("machine %d: discovered %d nodes",
+		o.cfg.Machine.MachineID, len(nodesResp.Nodes)))
+
+	// Start machine-level WS as early as possible so sync.nodes can reach an
+	// empty machine before the first node is attached.
+	o.tryStartWS(ctx)
+
+	// Start initial nodes.
+	for _, n := range nodesResp.Nodes {
+		o.startNode(ctx, n)
+	}
+
+	discoveryTicker := time.NewTicker(o.pullInterval)
+	statusTicker := time.NewTicker(o.pushInterval)
+	defer discoveryTicker.Stop()
+	defer statusTicker.Stop()
+
+	for {
+		select {
+		case <-ctx.Done():
+			o.stopAll()
+			return nil
+
+		case <-discoveryTicker.C:
+			o.rediscover(ctx)
+
+		case <-statusTicker.C:
+			o.reportMachineStatus()
+		}
+	}
+}
+
+// ─── Node lifecycle ──────────────────────────────────────────────────────
+
+func (o *Orchestrator) startNode(ctx context.Context, mn panel.MachineNode) {
+	o.mu.Lock()
+	if _, exists := o.nodes[mn.ID]; exists {
+		o.mu.Unlock()
+		return
+	}
+
+	nodeCtx, cancel := context.WithCancel(ctx)
+	done := make(chan struct{})
+	mb := controlplane.NewNodeMailbox()
+	o.nodes[mn.ID] = &nodeHandle{cancel: cancel, done: done, mailbox: mb}
+	o.mu.Unlock()
+
+	o.eventsMu.Lock()
+	o.mailboxes[mn.ID] = mb
+	o.eventsMu.Unlock()
+
+	nodeCfg := o.cfg.ExpandMachineNode(mn.ID, mn.Type)
+
+	perNodeClient := o.client.ForNode(mn.ID)
+
+	// Pre-fetch node config to detect transport-based kernel requirements.
+	// If the transport (e.g. xhttp) is incompatible with the configured kernel
+	// (e.g. singbox), auto-switch to the required kernel for this node.
+	if cfgSnapshot, err := perNodeClient.GetConfig(); err == nil && cfgSnapshot != nil {
+		if resolved := model.ResolveKernelForTransport(cfgSnapshot.Network, nodeCfg.Kernel.Type); resolved != nodeCfg.Kernel.Type {
+			nlog.Core().Info(fmt.Sprintf("machine: auto-switching kernel for node %d (%s→%s, transport=%s)",
+				mn.ID, nodeCfg.Kernel.Type, resolved, cfgSnapshot.Network))
+			nodeCfg.Kernel.Type = resolved
+		}
+	}
+	// Reset cached ETag so the subsequent GetConfig in Initial() gets a full response.
+	perNodeClient.ResetConfigETag()
+
+	var push controlplane.PushClient
+	if o.ws != nil {
+		push = &machineNodePush{
+			nodeID: mn.ID,
+			ws:     o.ws,
+		}
+	}
+
+	// The registerFn is called by MachinePanelControlPlane.Initial() to expose
+	// the node mailbox + status channel to the Service.
+	nodeID := mn.ID
+	registerFn := func(st chan<- controlplane.StatusChange) *controlplane.NodeMailbox {
+		o.registerNode(nodeID, st)
+		return mb
+	}
+
+	cp := controlplane.NewMachinePanelControlPlane(perNodeClient, nodeCfg.Kernel, push, registerFn)
+	svc := service.NewWithControlPlane(nodeCfg, cp)
+
+	nlog.Core().Info(fmt.Sprintf("machine: starting node %d (%s/%s)",
+		mn.ID, mn.Type, mn.Name))
+
+	go func() {
+		defer close(done)
+		defer o.unregisterNode(mn.ID)
+		if err := svc.Run(nodeCtx); err != nil {
+			nlog.Core().Error("machine node exited with error",
+				"node_id", mn.ID, "error", err)
+		}
+	}()
+}
+
+func (o *Orchestrator) stopNode(nodeID int) {
+	o.mu.Lock()
+	h, ok := o.nodes[nodeID]
+	if !ok {
+		o.mu.Unlock()
+		return
+	}
+	delete(o.nodes, nodeID)
+	o.mu.Unlock()
+
+	o.eventsMu.Lock()
+	delete(o.mailboxes, nodeID)
+	o.eventsMu.Unlock()
+
+	nlog.Core().Info(fmt.Sprintf("machine: stopping node %d", nodeID))
+	h.cancel()
+	<-h.done
+}
+
+func (o *Orchestrator) stopAll() {
+	o.mu.Lock()
+	handles := make(map[int]*nodeHandle, len(o.nodes))
+	for id, h := range o.nodes {
+		handles[id] = h
+	}
+	o.mu.Unlock()
+
+	for id, h := range handles {
+		nlog.Core().Info(fmt.Sprintf("machine: stopping node %d", id))
+		h.cancel()
+	}
+	for _, h := range handles {
+		<-h.done
+	}
+
+	if o.wsCancel != nil {
+		o.wsCancel()
+	}
+}
+
+// ─── Node discovery ──────────────────────────────────────────────────────
+
+func (o *Orchestrator) rediscover(ctx context.Context) {
+	nodesResp, err := o.client.GetMachineNodes()
+	if err != nil {
+		nlog.Core().Warn("machine node discovery failed", "error", err)
+		return
+	}
+
+	wanted := make(map[int]panel.MachineNode, len(nodesResp.Nodes))
+	for _, n := range nodesResp.Nodes {
+		wanted[n.ID] = n
+	}
+
+	o.mu.Lock()
+	var toRemove []int
+	for id := range o.nodes {
+		if _, ok := wanted[id]; !ok {
+			toRemove = append(toRemove, id)
+		}
+	}
+	o.mu.Unlock()
+
+	for _, id := range toRemove {
+		o.stopNode(id)
+	}
+
+	for _, n := range nodesResp.Nodes {
+		o.startNode(ctx, n) // no-op if already running
+	}
+}
+
+// ─── Machine status reporting ────────────────────────────────────────────
+
+func (o *Orchestrator) reportMachineStatus() {
+	s := monitor.Collect()
+	if err := o.client.ReportMachineStatus(
+		s.CPU,
+		[2]uint64{s.MemTotal, s.MemUsed},
+		[2]uint64{s.SwapTotal, s.SwapUsed},
+		[2]uint64{s.DiskTotal, s.DiskUsed},
+		s.NetInSpeed, s.NetOutSpeed,
+	); err != nil {
+		nlog.Core().Warn("machine status report failed", "error", err)
+	}
+}
+
+// ─── WS mux ─────────────────────────────────────────────────────────────
+
+func (o *Orchestrator) tryStartWS(ctx context.Context) {
+	hs, err := o.client.Handshake()
+	if err != nil {
+		nlog.Core().Warn("machine ws handshake failed, REST only", "error", err)
+		return
+	}
+	if !hs.WebSocket.Enabled || hs.WebSocket.WSURL == "" {
+		nlog.Core().Info("machine: ws disabled by panel, REST only")
+		return
+	}
+
+	wsCfg := panel.WSClientConfig{
+		StatusInterval:   time.Duration(o.cfg.WS.StatusInterval) * time.Second,
+		HandshakeTimeout: time.Duration(o.cfg.WS.HandshakeTimeout) * time.Second,
+		BackoffInitial:   time.Duration(o.cfg.WS.BackoffInitial) * time.Second,
+		BackoffMax:       time.Duration(o.cfg.WS.BackoffMax) * time.Second,
+		MachineID:        o.cfg.Machine.MachineID,
+	}
+
+	o.ws = panel.NewWSClient(
+		hs.WebSocket.WSURL,
+		o.cfg.Machine.Token,
+		0, // no single node_id
+		wsCfg,
+		o.onWSEvent,
+		o.onWSStatus,
+		nil, // per-node status is sent via machineNodePush
+	)
+
+	wsCtx, wsCancel := context.WithCancel(ctx)
+	o.wsCancel = wsCancel
+	go o.ws.Run(wsCtx)
+
+	nlog.Core().Info("machine: ws mux started")
+}
+
+// onWSEvent routes a WS event to the correct node's channel.
+// sync.nodes is a machine-level event that triggers immediate rediscovery.
+func (o *Orchestrator) onWSEvent(event panel.WSEvent) {
+	// sync.nodes is a machine-level event, not per-node
+	if event.Type == panel.WSEventSyncNodes {
+		nlog.Core().Info("machine received sync.nodes, triggering immediate rediscovery")
+		go o.rediscover(o.runCtx)
+		return
+	}
+
+	nodeID := event.NodeID
+	if nodeID == 0 {
+		nlog.Core().Debug("machine ws event missing node_id, dropping", "type", event.Type)
+		return
+	}
+
+	translated, err := controlplane.TranslateWSEvent(event, o.cfg.Kernel)
+	if err != nil {
+		nlog.Core().Warn("machine ws event translation failed",
+			"type", event.Type, "node_id", nodeID, "error", err)
+		return
+	}
+
+	o.eventsMu.RLock()
+	mailbox, ok := o.mailboxes[nodeID]
+	o.eventsMu.RUnlock()
+	if !ok {
+		nlog.Core().Debug("machine ws event for unknown node", "node_id", nodeID, "type", event.Type)
+		return
+	}
+	mailbox.Apply(translated)
+}
+
+// onWSStatus broadcasts WS connectivity changes to all registered nodes.
+func (o *Orchestrator) onWSStatus(status panel.WSStatusChange) {
+	change := controlplane.StatusChange{Connected: status.Connected}
+	o.eventsMu.RLock()
+	defer o.eventsMu.RUnlock()
+	for _, ch := range o.statuses {
+		select {
+		case ch <- change:
+		default:
+		}
+	}
+}
+
+func (o *Orchestrator) registerNode(nodeID int, st chan<- controlplane.StatusChange) {
+	o.eventsMu.Lock()
+	o.statuses[nodeID] = st
+	o.eventsMu.Unlock()
+}
+
+func (o *Orchestrator) unregisterNode(nodeID int) {
+	o.eventsMu.Lock()
+	delete(o.mailboxes, nodeID)
+	delete(o.statuses, nodeID)
+	o.eventsMu.Unlock()
+}
+
+func (o *Orchestrator) applyIntervals(bc panel.MachineBaseConfig) {
+	o.pullInterval = time.Duration(bc.PullInterval) * time.Second
+	if o.pullInterval < 30*time.Second {
+		o.pullInterval = 60 * time.Second
+	}
+	o.pushInterval = time.Duration(bc.PushInterval) * time.Second
+	if o.pushInterval < 10*time.Second {
+		o.pushInterval = 60 * time.Second
+	}
+}
+
+// ─── Virtual PushClient ─────────────────────────────────────────────────
+
+// machineNodePush implements controlplane.PushClient for a single node
+// backed by the shared machine WS connection. Events are routed by the
+// WS mux directly to the Service's channels; this adapter only provides
+// connectivity status and send capabilities.
+type machineNodePush struct {
+	nodeID int
+	ws     *panel.WSClient
+}
+
+func (p *machineNodePush) Run(ctx context.Context) {
+	// The shared WS mux pushes events into our channels; we just wait.
+	<-ctx.Done()
+}
+
+func (p *machineNodePush) IsConnected() bool {
+	return p.ws != nil && p.ws.IsConnected()
+}
+
+func (p *machineNodePush) SendDeviceReport(devices map[int][]string) {
+	if p.ws == nil {
+		return
+	}
+	payload := map[string]interface{}{
+		"node_id": p.nodeID,
+	}
+	// Flatten into the standard format with node_id wrapper.
+	strDevices := make(map[string][]string, len(devices))
+	for uid, ips := range devices {
+		strDevices[fmt.Sprintf("%d", uid)] = ips
+	}
+	payload["devices"] = strDevices
+	data, _ := json.Marshal(payload)
+	p.ws.SendRaw(panel.WSEventReportDevices, data)
+}

internal/model/custom_config_tags.go

@@ -0,0 +1,155 @@
+package model
+
+import (
+	"encoding/json"
+	"fmt"
+	"os"
+	"strings"
+
+	"gopkg.in/yaml.v3"
+)
+
+type outboundTagSource struct {
+	Tag    string
+	Source string
+}
+
+func collectAdditionalOutboundTagSources(path string, rawOutbounds []map[string]any) ([]outboundTagSource, error) {
+	sources := collectRawOutboundTagSources(rawOutbounds)
+	customSources, err := collectCustomConfigOutboundTagSources(path)
+	if err != nil {
+		return nil, err
+	}
+	sources = append(sources, customSources...)
+	return sources, nil
+}
+
+func collectRawOutboundTagSources(outbounds []map[string]any) []outboundTagSource {
+	if len(outbounds) == 0 {
+		return nil
+	}
+	sources := make([]outboundTagSource, 0, len(outbounds))
+	for i, outbound := range outbounds {
+		tag, _ := outbound["tag"].(string)
+		tag = strings.TrimSpace(tag)
+		if tag == "" {
+			continue
+		}
+		sources = append(sources, outboundTagSource{Tag: tag, Source: fmt.Sprintf("kernel.custom_outbound[%d]", i)})
+	}
+	return sources
+}
+
+func collectCustomConfigOutboundTagSources(path string) ([]outboundTagSource, error) {
+	if strings.TrimSpace(path) == "" {
+		return nil, nil
+	}
+	data, err := os.ReadFile(path)
+	if err != nil {
+		if os.IsNotExist(err) {
+			return nil, nil
+		}
+		return nil, fmt.Errorf("read custom config %q: %w", path, err)
+	}
+	if len(strings.TrimSpace(string(data))) == 0 {
+		return nil, nil
+	}
+
+	decoded := make(map[string]any)
+	trimmed := strings.TrimSpace(string(data))
+	if strings.HasPrefix(trimmed, "{") {
+		if err := json.Unmarshal(data, &decoded); err != nil {
+			return nil, fmt.Errorf("parse custom config as JSON: %w", err)
+		}
+	} else {
+		if err := yaml.Unmarshal(data, &decoded); err != nil {
+			return nil, fmt.Errorf("parse custom config as YAML: %w", err)
+		}
+	}
+
+	items, ok := toSliceOfMaps(decoded["outbounds"])
+	if !ok {
+		return nil, nil
+	}
+	sources := make([]outboundTagSource, 0, len(items))
+	for i, item := range items {
+		tag, _ := item["tag"].(string)
+		tag = strings.TrimSpace(tag)
+		if tag == "" {
+			continue
+		}
+		sources = append(sources, outboundTagSource{Tag: tag, Source: fmt.Sprintf("kernel.custom_config.outbounds[%d]", i)})
+	}
+	return sources, nil
+}
+
+func toSliceOfMaps(v any) ([]map[string]any, bool) {
+	if v == nil {
+		return nil, false
+	}
+	switch items := v.(type) {
+	case []any:
+		result := make([]map[string]any, 0, len(items))
+		for _, item := range items {
+			mapped, ok := toMap(item)
+			if ok {
+				result = append(result, mapped)
+			}
+		}
+		return result, len(result) > 0
+	case []map[string]any:
+		return items, len(items) > 0
+	default:
+		return nil, false
+	}
+}
+
+func toMap(v any) (map[string]any, bool) {
+	switch mapped := v.(type) {
+	case map[string]any:
+		return mapped, true
+	case map[any]any:
+		result := make(map[string]any, len(mapped))
+		for k, val := range mapped {
+			result[fmt.Sprint(k)] = val
+		}
+		return result, true
+	default:
+		return nil, false
+	}
+}
+
+func validateOutboundTagCollisions(structured []OutboundConfig, additional []outboundTagSource) error {
+	seen := make(map[string]string, len(additional))
+	for _, source := range additional {
+		key := strings.ToLower(strings.TrimSpace(source.Tag))
+		if key == "" {
+			continue
+		}
+		if prev, exists := seen[key]; exists {
+			return fmt.Errorf("%s.tag duplicates %q from %s", source.Source, source.Tag, prev)
+		}
+		seen[key] = source.Source
+	}
+	for i, outbound := range structured {
+		key := strings.ToLower(strings.TrimSpace(outbound.Tag))
+		if key == "" {
+			continue
+		}
+		if prev, exists := seen[key]; exists {
+			return fmt.Errorf("custom_outbounds[%d].tag duplicates %q from %s", i, outbound.Tag, prev)
+		}
+	}
+	return nil
+}
+
+func additionalTagNames(sources []outboundTagSource) []string {
+	if len(sources) == 0 {
+		return nil
+	}
+	tags := make([]string, 0, len(sources))
+	for _, source := range sources {
+		tags = append(tags, source.Tag)
+	}
+	return tags
+}

internal/model/custom_config_tags_test.go

@@ -0,0 +1,81 @@
+package model
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"github.com/cedar2025/xboard-node/internal/config"
+	"github.com/cedar2025/xboard-node/internal/panel"
+)
+
+func TestNodeSpecFromPanelValidated_AllowsTargetsFromCustomConfigOutbounds(t *testing.T) {
+	dir := t.TempDir()
+	customConfigPath := filepath.Join(dir, "custom.json")
+	if err := os.WriteFile(customConfigPath, []byte(`{"outbounds":[{"tag":"wg-exit","protocol":"wireguard"}]}`), 0o644); err != nil {
+		t.Fatalf("write custom config: %v", err)
+	}
+
+	_, err := NodeSpecFromPanelValidated(&panel.NodeConfig{
+		Protocol:   "shadowsocks",
+		ServerPort: 8388,
+		CustomOutbounds: []panel.OutboundConfig{{
+			Tag:      "proxy",
+			Protocol: "socks",
+			ProxyTag: "wg-exit",
+			Settings: map[string]any{"server": "2.2.2.2", "server_port": 1080},
+		}},
+		CustomRouteRules: []panel.CustomRouteRule{{
+			Action: panel.RouteAction{Type: "route", Target: "wg-exit"},
+			Match:  panel.RouteMatch{DomainSuffixes: []string{"example.com"}},
+		}},
+	}, config.KernelConfig{Type: "singbox", CustomConfig: customConfigPath})
+	if err != nil {
+		t.Fatalf("expected custom_config outbound tags to be available, got %v", err)
+	}
+}
+
+func TestNodeSpecFromPanelValidated_RejectsDuplicateTagsAcrossSources(t *testing.T) {
+	dir := t.TempDir()
+	customConfigPath := filepath.Join(dir, "custom.yaml")
+	if err := os.WriteFile(customConfigPath, []byte("outbounds:\n  - tag: warp\n    protocol: wireguard\n"), 0o644); err != nil {
+		t.Fatalf("write custom config: %v", err)
+	}
+
+	_, err := NodeSpecFromPanelValidated(&panel.NodeConfig{
+		Protocol:   "shadowsocks",
+		ServerPort: 8388,
+		CustomOutbounds: []panel.OutboundConfig{{
+			Tag:      "warp",
+			Protocol: "socks",
+			Settings: map[string]any{"server": "2.2.2.2", "server_port": 1080},
+		}},
+	}, config.KernelConfig{
+		Type:         "singbox",
+		CustomConfig: customConfigPath,
+		CustomOutbound: []map[string]any{{
+			"tag":      "proxy",
+			"protocol": "socks",
+		}},
+	})
+	if err == nil {
+		t.Fatal("expected error, got nil")
+	}
+	if !strings.Contains(err.Error(), `custom_outbounds[0].tag duplicates "warp" from kernel.custom_config.outbounds[0]`) {
+		t.Fatalf("unexpected error: %v", err)
+	}
+}
+
+func TestValidateOutboundTagCollisions_RejectsDuplicateRawSources(t *testing.T) {
+	err := validateOutboundTagCollisions(nil, []outboundTagSource{
+		{Tag: "warp", Source: "kernel.custom_outbound[0]"},
+		{Tag: "warp", Source: "kernel.custom_config.outbounds[0]"},
+	})
+	if err == nil {
+		t.Fatal("expected error, got nil")
+	}
+	if !strings.Contains(err.Error(), `kernel.custom_config.outbounds[0].tag duplicates "warp" from kernel.custom_outbound[0]`) {
+		t.Fatalf("unexpected error: %v", err)
+	}
+}

internal/model/custom_outbound_support.go

@@ -0,0 +1,50 @@
+package model
+
+func OutboundSupportMatrix() map[string]KernelOutboundSupport {
+	return map[string]KernelOutboundSupport{
+		"xray": {
+			Protocols: []string{
+				"vmess",
+				"vless",
+				"trojan",
+				"shadowsocks",
+				"socks",
+				"http",
+				"wireguard",
+			},
+			Features: []string{
+				"tag",
+				"protocol",
+				"settings",
+				"proxy_tag",
+			},
+		},
+		"singbox": {
+			Protocols: []string{
+				"vmess",
+				"vless",
+				"trojan",
+				"shadowsocks",
+				"socks",
+				"http",
+				"wireguard",
+				"tuic",
+				"hysteria2",
+				"anytls",
+				"naive",
+				"mieru",
+			},
+			Features: []string{
+				"tag",
+				"protocol",
+				"settings",
+				"proxy_tag",
+			},
+		},
+	}
+}
+
+type KernelOutboundSupport struct {
+	Protocols []string
+	Features  []string
+}

internal/model/custom_outbound_validate.go

@@ -0,0 +1,139 @@
+package model
+
+import (
+	"fmt"
+	"strconv"
+	"strings"
+)
+
+func ValidateCustomOutbounds(outbounds []OutboundConfig) error {
+	return ValidateCustomOutboundsForKernel(outbounds, "", nil)
+}
+
+func ValidateCustomOutboundsForKernel(outbounds []OutboundConfig, kernelType string, additionalTags []string) error {
+	if len(outbounds) == 0 {
+		return nil
+	}
+
+	kernelType = strings.ToLower(strings.TrimSpace(kernelType))
+	allowedProtocols := map[string]struct{}{}
+	if kernelType != "" {
+		support, ok := OutboundSupportMatrix()[kernelType]
+		if !ok {
+			return fmt.Errorf("unsupported kernel type %q", kernelType)
+		}
+		for _, protocol := range support.Protocols {
+			allowedProtocols[strings.ToLower(protocol)] = struct{}{}
+		}
+	}
+
+	seen := make(map[string]struct{}, len(outbounds))
+	availableTags := make(map[string]struct{}, len(outbounds)+len(additionalTags)+2)
+	availableTags["direct"] = struct{}{}
+	availableTags["block"] = struct{}{}
+	for _, tag := range additionalTags {
+		tag = strings.ToLower(strings.TrimSpace(tag))
+		if tag != "" {
+			availableTags[tag] = struct{}{}
+		}
+	}
+
+	for i, outbound := range outbounds {
+		tag := strings.TrimSpace(outbound.Tag)
+		protocol := strings.ToLower(strings.TrimSpace(outbound.Protocol))
+		if tag == "" {
+			return fmt.Errorf("custom_outbounds[%d].tag is required", i)
+		}
+		if protocol == "" {
+			return fmt.Errorf("custom_outbounds[%d].protocol is required", i)
+		}
+		if kernelType != "" {
+			if _, ok := allowedProtocols[protocol]; !ok {
+				return fmt.Errorf("custom_outbounds[%d].protocol %q is not supported by kernel %q", i, outbound.Protocol, kernelType)
+			}
+		}
+		if err := validateOutboundSettings(i, outbound.Settings); err != nil {
+			return err
+		}
+		key := strings.ToLower(tag)
+		if _, exists := seen[key]; exists {
+			return fmt.Errorf("custom_outbounds[%d].tag duplicates %q", i, tag)
+		}
+		seen[key] = struct{}{}
+		availableTags[key] = struct{}{}
+	}
+
+	for i, outbound := range outbounds {
+		if outbound.ProxyTag == "" {
+			continue
+		}
+		proxyTag := strings.ToLower(strings.TrimSpace(outbound.ProxyTag))
+		if proxyTag == "" {
+			return fmt.Errorf("custom_outbounds[%d].proxy_tag must not be blank", i)
+		}
+		if proxyTag == strings.ToLower(strings.TrimSpace(outbound.Tag)) {
+			return fmt.Errorf("custom_outbounds[%d].proxy_tag must not reference itself", i)
+		}
+		if _, exists := availableTags[proxyTag]; !exists {
+			return fmt.Errorf("custom_outbounds[%d].proxy_tag references unknown outbound %q", i, outbound.ProxyTag)
+		}
+	}
+
+	return nil
+}
+
+func validateOutboundSettings(index int, settings map[string]any) error {
+	if len(settings) == 0 {
+		return fmt.Errorf("custom_outbounds[%d].settings is required", index)
+	}
+	for _, reservedKey := range []string{"tag", "protocol", "proxy_tag", "proxyTag"} {
+		if _, exists := settings[reservedKey]; exists {
+			return fmt.Errorf("custom_outbounds[%d].settings.%s is reserved", index, reservedKey)
+		}
+	}
+	for _, field := range []string{"server", "uuid", "password", "method", "cipher", "private_key", "secret_key", "privateKey", "secretKey"} {
+		if value, ok := settings[field]; ok {
+			if strings.TrimSpace(fmt.Sprintf("%v", value)) == "" {
+				return fmt.Errorf("custom_outbounds[%d].settings.%s must not be blank", index, field)
+			}
+		}
+	}
+	if value, ok := settings["server_port"]; ok {
+		if _, err := asPort(value); err != nil {
+			return fmt.Errorf("custom_outbounds[%d].settings.server_port %w", index, err)
+		}
+	}
+	if value, ok := settings["serverPort"]; ok {
+		if _, err := asPort(value); err != nil {
+			return fmt.Errorf("custom_outbounds[%d].settings.serverPort %w", index, err)
+		}
+	}
+	return nil
+}
+
+func asPort(value any) (int, error) {
+	switch v := value.(type) {
+	case int:
+		if v < 1 || v > 65535 {
+			return 0, fmt.Errorf("must be between 1 and 65535")
+		}
+		return v, nil
+	case int32:
+		return asPort(int(v))
+	case int64:
+		return asPort(int(v))
+	case float64:
+		if v != float64(int(v)) {
+			return 0, fmt.Errorf("must be an integer")
+		}
+		return asPort(int(v))
+	case string:
+		parsed, err := strconv.Atoi(strings.TrimSpace(v))
+		if err != nil {
+			return 0, fmt.Errorf("must be an integer")
+		}
+		return asPort(parsed)
+	default:
+		return 0, fmt.Errorf("must be an integer")
+	}
+}

internal/model/custom_outbound_validate_test.go

@@ -0,0 +1,109 @@
+package model
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestValidateCustomOutbounds(t *testing.T) {
+	tests := []struct {
+		name      string
+		outbounds []OutboundConfig
+		kernel    string
+		wantErr   string
+	}{
+		{
+			name: "valid chain",
+			outbounds: []OutboundConfig{
+				{Tag: "warp", Protocol: "wireguard", Settings: map[string]any{"server": "1.1.1.1", "server_port": 2408, "private_key": "pk"}},
+				{Tag: "proxy", Protocol: "socks", ProxyTag: "warp", Settings: map[string]any{"server": "2.2.2.2", "server_port": 1080}},
+			},
+			kernel: "singbox",
+		},
+		{
+			name:      "valid native vmess settings preserved",
+			outbounds: []OutboundConfig{{Tag: "vmess-native", Protocol: "vmess", Settings: map[string]any{"vnext": []any{map[string]any{"address": "1.1.1.1", "port": 443}}}}},
+			kernel:    "xray",
+		},
+		{
+			name:      "missing tag",
+			outbounds: []OutboundConfig{{Protocol: "socks", Settings: map[string]any{"server": "2.2.2.2", "server_port": 1080}}},
+			wantErr:   "custom_outbounds[0].tag is required",
+		},
+		{
+			name:      "missing protocol",
+			outbounds: []OutboundConfig{{Tag: "proxy", Settings: map[string]any{"server": "2.2.2.2", "server_port": 1080}}},
+			wantErr:   "custom_outbounds[0].protocol is required",
+		},
+		{
+			name: "duplicate tag",
+			outbounds: []OutboundConfig{
+				{Tag: "proxy", Protocol: "socks", Settings: map[string]any{"server": "2.2.2.2", "server_port": 1080}},
+				{Tag: "proxy", Protocol: "http", Settings: map[string]any{"server": "3.3.3.3", "server_port": 8080}},
+			},
+			wantErr: `custom_outbounds[1].tag duplicates "proxy"`,
+		},
+		{
+			name:      "self proxy tag",
+			outbounds: []OutboundConfig{{Tag: "proxy", Protocol: "socks", ProxyTag: "proxy", Settings: map[string]any{"server": "2.2.2.2", "server_port": 1080}}},
+			wantErr:   "custom_outbounds[0].proxy_tag must not reference itself",
+		},
+		{
+			name:      "unknown proxy tag",
+			outbounds: []OutboundConfig{{Tag: "proxy", Protocol: "socks", ProxyTag: "missing", Settings: map[string]any{"server": "2.2.2.2", "server_port": 1080}}},
+			wantErr:   `custom_outbounds[0].proxy_tag references unknown outbound "missing"`,
+		},
+		{
+			name:      "built-in proxy tag allowed",
+			outbounds: []OutboundConfig{{Tag: "proxy", Protocol: "socks", ProxyTag: "direct", Settings: map[string]any{"server": "2.2.2.2", "server_port": 1080}}},
+		},
+		{
+			name:      "unsupported protocol for kernel",
+			outbounds: []OutboundConfig{{Tag: "hy2", Protocol: "hysteria2", Settings: map[string]any{"server": "2.2.2.2", "server_port": 8443}}},
+			kernel:    "xray",
+			wantErr:   `custom_outbounds[0].protocol "hysteria2" is not supported by kernel "xray"`,
+		},
+		{
+			name:      "missing settings",
+			outbounds: []OutboundConfig{{Tag: "proxy", Protocol: "socks"}},
+			wantErr:   "custom_outbounds[0].settings is required",
+		},
+		{
+			name:      "reserved snake case key in settings",
+			outbounds: []OutboundConfig{{Tag: "proxy", Protocol: "socks", Settings: map[string]any{"proxy_tag": "bad", "server": "2.2.2.2", "server_port": 1080}}},
+			wantErr:   "custom_outbounds[0].settings.proxy_tag is reserved",
+		},
+		{
+			name:      "reserved camel case key in settings",
+			outbounds: []OutboundConfig{{Tag: "proxy", Protocol: "socks", Settings: map[string]any{"proxyTag": "bad", "server": "2.2.2.2", "server_port": 1080}}},
+			wantErr:   "custom_outbounds[0].settings.proxyTag is reserved",
+		},
+		{
+			name:      "invalid server_port",
+			outbounds: []OutboundConfig{{Tag: "proxy", Protocol: "socks", Settings: map[string]any{"server": "2.2.2.2", "server_port": 70000}}},
+			wantErr:   "custom_outbounds[0].settings.server_port must be between 1 and 65535",
+		},
+		{
+			name:      "invalid camel case serverPort",
+			outbounds: []OutboundConfig{{Tag: "proxy", Protocol: "socks", Settings: map[string]any{"server": "2.2.2.2", "serverPort": 70000}}},
+			wantErr:   "custom_outbounds[0].settings.serverPort must be between 1 and 65535",
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			err := ValidateCustomOutboundsForKernel(tc.outbounds, tc.kernel, nil)
+			if tc.wantErr == "" && err != nil {
+				t.Fatalf("expected no error, got %v", err)
+			}
+			if tc.wantErr != "" {
+				if err == nil {
+					t.Fatal("expected error, got nil")
+				}
+				if !strings.Contains(err.Error(), tc.wantErr) {
+					t.Fatalf("unexpected error: %v", err)
+				}
+			}
+		})
+	}
+}

internal/model/custom_route.go

@@ -0,0 +1,58 @@
+package model
+
+type CustomRouteRule struct {
+	Name     string
+	Disabled bool
+	Match    RouteMatch
+	Action   RouteAction
+}
+
+type RouteMatch struct {
+	Domains        []string
+	DomainSuffixes []string
+	IPCIDRs        []string
+	Ports          []string
+	Networks       []string
+	SourceCIDRs    []string
+	SourcePorts    []string
+}
+
+type RouteAction struct {
+	Type   string
+	Target string
+}
+
+func cloneCustomRouteRules(src []CustomRouteRule) []CustomRouteRule {
+	if len(src) == 0 {
+		return nil
+	}
+	out := make([]CustomRouteRule, 0, len(src))
+	for _, rule := range src {
+		out = append(out, CustomRouteRule{
+			Name:     rule.Name,
+			Disabled: rule.Disabled,
+			Match:    cloneRouteMatch(rule.Match),
+			Action:   cloneRouteAction(rule.Action),
+		})
+	}
+	return out
+}
+
+func cloneRouteMatch(src RouteMatch) RouteMatch {
+	return RouteMatch{
+		Domains:        cloneStringSlice(src.Domains),
+		DomainSuffixes: cloneStringSlice(src.DomainSuffixes),
+		IPCIDRs:        cloneStringSlice(src.IPCIDRs),
+		Ports:          cloneStringSlice(src.Ports),
+		Networks:       cloneStringSlice(src.Networks),
+		SourceCIDRs:    cloneStringSlice(src.SourceCIDRs),
+		SourcePorts:    cloneStringSlice(src.SourcePorts),
+	}
+}
+
+func cloneRouteAction(src RouteAction) RouteAction {
+	return RouteAction{
+		Type:   src.Type,
+		Target: src.Target,
+	}
+}

internal/model/custom_route_support.go

@@ -0,0 +1,35 @@
+package model
+
+func RouteSupportMatrix() map[string]KernelRouteSupport {
+	return map[string]KernelRouteSupport{
+		"xray": {
+			Matchers: []string{
+				"domains",
+				"domain_suffixes",
+				"ip_cidrs",
+				"ports",
+				"networks",
+				"source_cidrs",
+				"source_ports",
+			},
+			Actions: []string{"block", "direct", "route"},
+		},
+		"singbox": {
+			Matchers: []string{
+				"domains",
+				"domain_suffixes",
+				"ip_cidrs",
+				"ports",
+				"networks",
+				"source_cidrs",
+				"source_ports",
+			},
+			Actions: []string{"block", "direct", "route"},
+		},
+	}
+}
+
+type KernelRouteSupport struct {
+	Matchers []string
+	Actions  []string
+}

internal/model/custom_route_validate.go

@@ -0,0 +1,178 @@
+package model
+
+import (
+	"fmt"
+	"strconv"
+	"strings"
+)
+
+func ValidateCustomRouteRules(rules []CustomRouteRule, kernelType string, availableTags map[string]struct{}) error {
+	if len(rules) == 0 {
+		return nil
+	}
+
+	kernelType = strings.ToLower(strings.TrimSpace(kernelType))
+	matcherSupport := map[string]struct{}{}
+	actionSupport := map[string]struct{}{}
+	if kernelType != "" {
+		support, ok := RouteSupportMatrix()[kernelType]
+		if !ok {
+			return fmt.Errorf("unsupported kernel type %q", kernelType)
+		}
+		for _, matcher := range support.Matchers {
+			matcherSupport[strings.ToLower(matcher)] = struct{}{}
+		}
+		for _, action := range support.Actions {
+			actionSupport[strings.ToLower(action)] = struct{}{}
+		}
+	}
+
+	for i, rule := range rules {
+		if rule.Disabled {
+			continue
+		}
+		if !hasRouteMatch(rule.Match) {
+			return fmt.Errorf("custom_route_rules[%d].match is required", i)
+		}
+		if err := ensureRouteMatcherSupported(i, kernelType, matcherSupport, rule.Match); err != nil {
+			return err
+		}
+		if err := validatePortSpecs(rule.Match.Ports, fmt.Sprintf("custom_route_rules[%d].match.ports", i)); err != nil {
+			return err
+		}
+		if err := validatePortSpecs(rule.Match.SourcePorts, fmt.Sprintf("custom_route_rules[%d].match.source_ports", i)); err != nil {
+			return err
+		}
+		if err := validateNetworks(rule.Match.Networks, fmt.Sprintf("custom_route_rules[%d].match.networks", i)); err != nil {
+			return err
+		}
+
+		actionType := strings.ToLower(strings.TrimSpace(rule.Action.Type))
+		if actionType == "" {
+			return fmt.Errorf("custom_route_rules[%d].action.type is required", i)
+		}
+		if kernelType != "" {
+			if _, ok := actionSupport[actionType]; !ok {
+				return fmt.Errorf("custom_route_rules[%d].action.type %q is not supported by kernel %q", i, rule.Action.Type, kernelType)
+			}
+		}
+		target := strings.TrimSpace(rule.Action.Target)
+		switch actionType {
+		case "route":
+			if target == "" {
+				return fmt.Errorf("custom_route_rules[%d].action.target is required when action.type is route", i)
+			}
+			if availableTags != nil {
+				if _, ok := availableTags[strings.ToLower(target)]; !ok {
+					return fmt.Errorf("custom_route_rules[%d].action.target references unknown outbound %q", i, rule.Action.Target)
+				}
+			}
+		case "direct", "block":
+			if target != "" {
+				return fmt.Errorf("custom_route_rules[%d].action.target is only allowed when action.type is route", i)
+			}
+		default:
+			return fmt.Errorf("custom_route_rules[%d].action.type %q is not supported", i, rule.Action.Type)
+		}
+	}
+
+	return nil
+}
+
+func hasRouteMatch(match RouteMatch) bool {
+	for _, values := range [][]string{
+		match.Domains,
+		match.DomainSuffixes,
+		match.IPCIDRs,
+		match.Ports,
+		match.Networks,
+		match.SourceCIDRs,
+		match.SourcePorts,
+	} {
+		for _, value := range values {
+			if strings.TrimSpace(value) != "" {
+				return true
+			}
+		}
+	}
+	return false
+}
+
+func ensureRouteMatcherSupported(index int, kernelType string, matcherSupport map[string]struct{}, match RouteMatch) error {
+	if kernelType == "" {
+		return nil
+	}
+	checks := map[string][]string{
+		"domains":         match.Domains,
+		"domain_suffixes": match.DomainSuffixes,
+		"ip_cidrs":        match.IPCIDRs,
+		"ports":           match.Ports,
+		"networks":        match.Networks,
+		"source_cidrs":    match.SourceCIDRs,
+		"source_ports":    match.SourcePorts,
+	}
+	for matcher, values := range checks {
+		if !hasAnyNonBlank(values) {
+			continue
+		}
+		if _, ok := matcherSupport[matcher]; !ok {
+			return fmt.Errorf("custom_route_rules[%d].match.%s is not supported by kernel %q", index, matcher, kernelType)
+		}
+	}
+	return nil
+}
+
+func hasAnyNonBlank(values []string) bool {
+	for _, value := range values {
+		if strings.TrimSpace(value) != "" {
+			return true
+		}
+	}
+	return false
+}
+
+func validateNetworks(values []string, field string) error {
+	for _, value := range values {
+		value = strings.ToLower(strings.TrimSpace(value))
+		if value == "" {
+			continue
+		}
+		if value != "tcp" && value != "udp" {
+			return fmt.Errorf("%s contains unsupported network %q", field, value)
+		}
+	}
+	return nil
+}
+
+func validatePortSpecs(values []string, field string) error {
+	for _, value := range values {
+		value = strings.TrimSpace(value)
+		if value == "" {
+			continue
+		}
+		if strings.Contains(value, "-") || strings.Contains(value, ":") {
+			delimiter := "-"
+			if strings.Contains(value, ":") {
+				delimiter = ":"
+			}
+			parts := strings.Split(value, delimiter)
+			if len(parts) != 2 {
+				return fmt.Errorf("%s contains invalid port range %q", field, value)
+			}
+			start, err := strconv.Atoi(strings.TrimSpace(parts[0]))
+			if err != nil || start < 1 || start > 65535 {
+				return fmt.Errorf("%s contains invalid port range %q", field, value)
+			}
+			end, err := strconv.Atoi(strings.TrimSpace(parts[1]))
+			if err != nil || end < 1 || end > 65535 || end < start {
+				return fmt.Errorf("%s contains invalid port range %q", field, value)
+			}
+			continue
+		}
+		port, err := strconv.Atoi(value)
+		if err != nil || port < 1 || port > 65535 {
+			return fmt.Errorf("%s contains invalid port %q", field, value)
+		}
+	}
+	return nil
+}

internal/model/custom_route_validate_test.go

@@ -0,0 +1,84 @@
+package model
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestValidateCustomRouteRules(t *testing.T) {
+	available := map[string]struct{}{"direct": {}, "block": {}, "warp": {}, "proxy": {}}
+	tests := []struct {
+		name    string
+		rules   []CustomRouteRule
+		kernel  string
+		wantErr string
+	}{
+		{
+			name: "valid route target",
+			rules: []CustomRouteRule{{
+				Match:  RouteMatch{DomainSuffixes: []string{"example.com"}, Ports: []string{"443"}, Networks: []string{"tcp"}},
+				Action: RouteAction{Type: "route", Target: "warp"},
+			}},
+			kernel: "singbox",
+		},
+		{
+			name:    "missing match",
+			rules:   []CustomRouteRule{{Action: RouteAction{Type: "direct"}}},
+			kernel:  "singbox",
+			wantErr: "custom_route_rules[0].match is required",
+		},
+		{
+			name:    "unknown target",
+			rules:   []CustomRouteRule{{Match: RouteMatch{DomainSuffixes: []string{"example.com"}}, Action: RouteAction{Type: "route", Target: "missing"}}},
+			kernel:  "singbox",
+			wantErr: `custom_route_rules[0].action.target references unknown outbound "missing"`,
+		},
+		{
+			name:    "route target required",
+			rules:   []CustomRouteRule{{Match: RouteMatch{DomainSuffixes: []string{"example.com"}}, Action: RouteAction{Type: "route"}}},
+			kernel:  "singbox",
+			wantErr: "custom_route_rules[0].action.target is required when action.type is route",
+		},
+		{
+			name:    "direct target forbidden",
+			rules:   []CustomRouteRule{{Match: RouteMatch{DomainSuffixes: []string{"example.com"}}, Action: RouteAction{Type: "direct", Target: "warp"}}},
+			kernel:  "singbox",
+			wantErr: "custom_route_rules[0].action.target is only allowed when action.type is route",
+		},
+		{
+			name:    "unsupported action",
+			rules:   []CustomRouteRule{{Match: RouteMatch{DomainSuffixes: []string{"example.com"}}, Action: RouteAction{Type: "dns"}}},
+			kernel:  "singbox",
+			wantErr: `custom_route_rules[0].action.type "dns" is not supported by kernel "singbox"`,
+		},
+		{
+			name:    "invalid port range",
+			rules:   []CustomRouteRule{{Match: RouteMatch{Ports: []string{"2000-1000"}}, Action: RouteAction{Type: "direct"}}},
+			kernel:  "singbox",
+			wantErr: `custom_route_rules[0].match.ports contains invalid port range "2000-1000"`,
+		},
+		{
+			name:    "invalid network",
+			rules:   []CustomRouteRule{{Match: RouteMatch{Networks: []string{"icmp"}}, Action: RouteAction{Type: "direct"}}},
+			kernel:  "singbox",
+			wantErr: `custom_route_rules[0].match.networks contains unsupported network "icmp"`,
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			err := ValidateCustomRouteRules(tc.rules, tc.kernel, available)
+			if tc.wantErr == "" && err != nil {
+				t.Fatalf("expected no error, got %v", err)
+			}
+			if tc.wantErr != "" {
+				if err == nil {
+					t.Fatal("expected error, got nil")
+				}
+				if !strings.Contains(err.Error(), tc.wantErr) {
+					t.Fatalf("unexpected error: %v", err)
+				}
+			}
+		})
+	}
+}

internal/model/panel.go

@@ -0,0 +1,276 @@
+package model
+
+import (
+	"github.com/cedar2025/xboard-node/internal/config"
+	"github.com/cedar2025/xboard-node/internal/panel"
+)
+
+func NodeSpecFromPanel(nc *panel.NodeConfig) *NodeSpec {
+	if nc == nil {
+		return nil
+	}
+
+	var certCfg *config.CertConfig
+	if nc.CertConfig != nil {
+		certCfg = &config.CertConfig{
+			CertMode:    nc.CertConfig.CertMode,
+			Domain:      nc.CertConfig.Domain,
+			Email:       nc.CertConfig.Email,
+			DNSProvider: nc.CertConfig.DNSProvider,
+			DNSEnv:      nc.CertConfig.DNSEnv,
+			HTTPPort:    nc.CertConfig.HTTPPort,
+			CertFile:    nc.CertConfig.CertFile,
+			KeyFile:     nc.CertConfig.KeyFile,
+			CertContent: nc.CertConfig.CertContent,
+			KeyContent:  nc.CertConfig.KeyContent,
+		}
+	}
+
+	var multiplex *MultiplexConfig
+	if nc.Multiplex != nil {
+		multiplex = &MultiplexConfig{
+			Enabled:        nc.Multiplex.Enabled,
+			Protocol:       nc.Multiplex.Protocol,
+			MaxConnections: nc.Multiplex.MaxConnections,
+			MinStreams:     nc.Multiplex.MinStreams,
+			MaxStreams:     nc.Multiplex.MaxStreams,
+			Padding:        nc.Multiplex.Padding,
+		}
+		if nc.Multiplex.Brutal != nil {
+			multiplex.Brutal = &BrutalConfig{
+				Enabled:  nc.Multiplex.Brutal.Enabled,
+				UpMbps:   nc.Multiplex.Brutal.UpMbps,
+				DownMbps: nc.Multiplex.Brutal.DownMbps,
+			}
+		}
+	}
+
+	routes := make([]RouteRule, 0, len(nc.Routes))
+	for _, route := range nc.Routes {
+		routes = append(routes, RouteRule{
+			ID:          route.ID,
+			Match:       cloneStringSlice(route.Match),
+			Action:      route.Action,
+			ActionValue: route.ActionValue,
+		})
+	}
+
+	outbounds := make([]OutboundConfig, 0, len(nc.CustomOutbounds))
+	for _, outbound := range nc.CustomOutbounds {
+		outbounds = append(outbounds, OutboundConfig{
+			Tag:      outbound.Tag,
+			Protocol: outbound.Protocol,
+			Settings: cloneAnyMap(outbound.Settings),
+			ProxyTag: outbound.ProxyTag,
+		})
+	}
+
+	customRouteRules := make([]CustomRouteRule, 0, len(nc.CustomRouteRules))
+	for _, rule := range nc.CustomRouteRules {
+		customRouteRules = append(customRouteRules, CustomRouteRule{
+			Name:     rule.Name,
+			Disabled: rule.Disabled,
+			Match: RouteMatch{
+				Domains:        cloneStringSlice(rule.Match.Domains),
+				DomainSuffixes: cloneStringSlice(rule.Match.DomainSuffixes),
+				IPCIDRs:        cloneStringSlice(rule.Match.IPCIDRs),
+				Ports:          cloneStringSlice(rule.Match.Ports),
+				Networks:       cloneStringSlice(rule.Match.Networks),
+				SourceCIDRs:    cloneStringSlice(rule.Match.SourceCIDRs),
+				SourcePorts:    cloneStringSlice(rule.Match.SourcePorts),
+			},
+			Action: RouteAction{
+				Type:   rule.Action.Type,
+				Target: rule.Action.Target,
+			},
+		})
+	}
+
+	return &NodeSpec{
+		Protocol:            nc.Protocol,
+		ListenIP:            nc.ListenIP,
+		ServerPort:          nc.ServerPort,
+		Network:             nc.Network,
+		NetworkSettings:     cloneAnyMap(nc.NetworkSettings),
+		Routes:              routes,
+		KernelType:          nc.KernelType,
+		KernelLogLevel:      nc.KernelLogLevel,
+		CustomOutbounds:     outbounds,
+		CustomRoutes:        cloneMapSlice(nc.CustomRoutes),
+		CustomRouteRules:    customRouteRules,
+		CertConfig:          certCfg,
+		AutoTLS:             nc.AutoTLS,
+		Domain:              nc.Domain,
+		Cipher:              nc.Cipher,
+		Plugin:              nc.Plugin,
+		PluginOpt:           nc.PluginOpt,
+		ServerKey:           nc.ServerKey,
+		TLS:                 nc.TLS,
+		Flow:                nc.Flow,
+		Decryption:          nc.Decryption,
+		TLSSettings:         cloneAnyMap(nc.TLSSettings),
+		Host:                nc.Host,
+		ServerName:          nc.ServerName,
+		Version:             nc.Version,
+		UpMbps:              nc.UpMbps,
+		DownMbps:            nc.DownMbps,
+		Obfs:                nc.Obfs,
+		ObfsPassword:        nc.ObfsPassword,
+		CongestionControl:   nc.CongestionControl,
+		PaddingScheme:       string(nc.PaddingScheme),
+		Transport:           nc.Transport,
+		TrafficPattern:      nc.TrafficPattern,
+		Multiplex:           multiplex,
+		AcceptProxyProtocol: nc.AcceptProxyProtocol,
+	}
+}
+
+func NodeSpecFromPanelValidated(nc *panel.NodeConfig, kcfg config.KernelConfig) (*NodeSpec, error) {
+	spec := NodeSpecFromPanel(nc)
+	if err := ValidateNodeSpec(spec, kcfg); err != nil {
+		return nil, err
+	}
+	return spec, nil
+}
+
+func UserSpecsFromPanel(users []panel.User) []UserSpec {
+	if users == nil {
+		return nil
+	}
+	out := make([]UserSpec, 0, len(users))
+	for _, user := range users {
+		out = append(out, UserSpec{ID: user.ID, UUID: user.UUID, SpeedLimit: user.SpeedLimit, DeviceLimit: user.DeviceLimit})
+	}
+	return out
+}
+
+func (n *NodeSpec) ToPanel() *panel.NodeConfig {
+	if n == nil {
+		return nil
+	}
+
+	var certCfg *panel.CertConfig
+	if n.CertConfig != nil {
+		certCfg = &panel.CertConfig{
+			CertMode:    n.CertConfig.CertMode,
+			Domain:      n.CertConfig.Domain,
+			Email:       n.CertConfig.Email,
+			DNSProvider: n.CertConfig.DNSProvider,
+			DNSEnv:      n.CertConfig.DNSEnv,
+			HTTPPort:    n.CertConfig.HTTPPort,
+			CertFile:    n.CertConfig.CertFile,
+			KeyFile:     n.CertConfig.KeyFile,
+			CertContent: n.CertConfig.CertContent,
+			KeyContent:  n.CertConfig.KeyContent,
+		}
+	}
+
+	var multiplex *panel.MultiplexConfig
+	if n.Multiplex != nil {
+		multiplex = &panel.MultiplexConfig{
+			Enabled:        n.Multiplex.Enabled,
+			Protocol:       n.Multiplex.Protocol,
+			MaxConnections: n.Multiplex.MaxConnections,
+			MinStreams:     n.Multiplex.MinStreams,
+			MaxStreams:     n.Multiplex.MaxStreams,
+			Padding:        n.Multiplex.Padding,
+		}
+		if n.Multiplex.Brutal != nil {
+			multiplex.Brutal = &panel.BrutalConfig{
+				Enabled:  n.Multiplex.Brutal.Enabled,
+				UpMbps:   n.Multiplex.Brutal.UpMbps,
+				DownMbps: n.Multiplex.Brutal.DownMbps,
+			}
+		}
+	}
+
+	routes := make([]panel.RouteRule, 0, len(n.Routes))
+	for _, route := range n.Routes {
+		routes = append(routes, panel.RouteRule{
+			ID:          route.ID,
+			Match:       cloneStringSlice(route.Match),
+			Action:      route.Action,
+			ActionValue: route.ActionValue,
+		})
+	}
+
+	outbounds := make([]panel.OutboundConfig, 0, len(n.CustomOutbounds))
+	for _, outbound := range n.CustomOutbounds {
+		outbounds = append(outbounds, panel.OutboundConfig{
+			Tag:      outbound.Tag,
+			Protocol: outbound.Protocol,
+			Settings: cloneAnyMap(outbound.Settings),
+			ProxyTag: outbound.ProxyTag,
+		})
+	}
+
+	customRouteRules := make([]panel.CustomRouteRule, 0, len(n.CustomRouteRules))
+	for _, rule := range n.CustomRouteRules {
+		customRouteRules = append(customRouteRules, panel.CustomRouteRule{
+			Name:     rule.Name,
+			Disabled: rule.Disabled,
+			Match: panel.RouteMatch{
+				Domains:        cloneStringSlice(rule.Match.Domains),
+				DomainSuffixes: cloneStringSlice(rule.Match.DomainSuffixes),
+				IPCIDRs:        cloneStringSlice(rule.Match.IPCIDRs),
+				Ports:          cloneStringSlice(rule.Match.Ports),
+				Networks:       cloneStringSlice(rule.Match.Networks),
+				SourceCIDRs:    cloneStringSlice(rule.Match.SourceCIDRs),
+				SourcePorts:    cloneStringSlice(rule.Match.SourcePorts),
+			},
+			Action: panel.RouteAction{
+				Type:   rule.Action.Type,
+				Target: rule.Action.Target,
+			},
+		})
+	}
+
+	return &panel.NodeConfig{
+		Protocol:            n.Protocol,
+		ListenIP:            n.ListenIP,
+		ServerPort:          n.ServerPort,
+		Network:             n.Network,
+		NetworkSettings:     cloneAnyMap(n.NetworkSettings),
+		Routes:              routes,
+		KernelType:          n.KernelType,
+		KernelLogLevel:      n.KernelLogLevel,
+		CustomOutbounds:     outbounds,
+		CustomRoutes:        cloneMapSlice(n.CustomRoutes),
+		CustomRouteRules:    customRouteRules,
+		CertConfig:          certCfg,
+		AutoTLS:             n.AutoTLS,
+		Domain:              n.Domain,
+		Cipher:              n.Cipher,
+		Plugin:              n.Plugin,
+		PluginOpt:           n.PluginOpt,
+		ServerKey:           n.ServerKey,
+		TLS:                 n.TLS,
+		Flow:                n.Flow,
+		Decryption:          n.Decryption,
+		TLSSettings:         cloneAnyMap(n.TLSSettings),
+		Host:                n.Host,
+		ServerName:          n.ServerName,
+		Version:             n.Version,
+		UpMbps:              n.UpMbps,
+		DownMbps:            n.DownMbps,
+		Obfs:                n.Obfs,
+		ObfsPassword:        n.ObfsPassword,
+		CongestionControl:   n.CongestionControl,
+		PaddingScheme:       panel.StringOrArray(n.PaddingScheme),
+		Transport:           n.Transport,
+		TrafficPattern:      n.TrafficPattern,
+		Multiplex:           multiplex,
+		AcceptProxyProtocol: n.AcceptProxyProtocol,
+	}
+}
+
+func UserSpecsToPanel(users []UserSpec) []panel.User {
+	if users == nil {
+		return nil
+	}
+	out := make([]panel.User, 0, len(users))
+	for _, user := range users {
+		out = append(out, panel.User{ID: user.ID, UUID: user.UUID, SpeedLimit: user.SpeedLimit, DeviceLimit: user.DeviceLimit})
+	}
+	return out
+}

internal/model/panel_test.go

@@ -0,0 +1,84 @@
+package model
+
+import (
+	"strings"
+	"testing"
+
+	"github.com/cedar2025/xboard-node/internal/config"
+	"github.com/cedar2025/xboard-node/internal/panel"
+)
+
+func TestNodeSpecFromPanelValidated(t *testing.T) {
+	t.Run("valid custom outbounds and route targets", func(t *testing.T) {
+		node, err := NodeSpecFromPanelValidated(&panel.NodeConfig{
+			Protocol:   "shadowsocks",
+			ServerPort: 8388,
+			CustomOutbounds: []panel.OutboundConfig{
+				{Tag: "warp", Protocol: "wireguard", Settings: map[string]any{"server": "1.1.1.1", "server_port": 2408, "private_key": "pk"}},
+				{Tag: "proxy", Protocol: "socks", ProxyTag: "warp", Settings: map[string]any{"server": "2.2.2.2", "server_port": 1080}},
+			},
+			CustomRouteRules: []panel.CustomRouteRule{{
+				Action: panel.RouteAction{Type: "route", Target: "proxy"},
+				Match:  panel.RouteMatch{DomainSuffixes: []string{"example.com"}},
+			}},
+		}, config.KernelConfig{Type: "singbox"})
+		if err != nil {
+			t.Fatalf("NodeSpecFromPanelValidated: %v", err)
+		}
+		if node == nil || len(node.CustomOutbounds) != 2 {
+			t.Fatalf("unexpected node spec: %#v", node)
+		}
+	})
+
+	t.Run("invalid custom outbounds", func(t *testing.T) {
+		_, err := NodeSpecFromPanelValidated(&panel.NodeConfig{
+			Protocol:   "shadowsocks",
+			ServerPort: 8388,
+			CustomOutbounds: []panel.OutboundConfig{{
+				Tag: "proxy", Protocol: "socks", ProxyTag: "missing", Settings: map[string]any{"server": "2.2.2.2", "server_port": 1080},
+			}},
+		}, config.KernelConfig{Type: "singbox"})
+		if err == nil {
+			t.Fatal("expected error, got nil")
+		}
+		if !strings.Contains(err.Error(), `custom_outbounds[0].proxy_tag references unknown outbound "missing"`) {
+			t.Fatalf("unexpected error: %v", err)
+		}
+	})
+
+	t.Run("reject unsupported protocol for kernel", func(t *testing.T) {
+		_, err := NodeSpecFromPanelValidated(&panel.NodeConfig{
+			Protocol:   "shadowsocks",
+			ServerPort: 8388,
+			CustomOutbounds: []panel.OutboundConfig{{
+				Tag: "hy2", Protocol: "hysteria2", Settings: map[string]any{"server": "2.2.2.2", "server_port": 8443},
+			}},
+		}, config.KernelConfig{Type: "xray"})
+		if err == nil {
+			t.Fatal("expected error, got nil")
+		}
+		if !strings.Contains(err.Error(), `protocol "hysteria2" is not supported by kernel "xray"`) {
+			t.Fatalf("unexpected error: %v", err)
+		}
+	})
+
+	t.Run("reject unknown route target", func(t *testing.T) {
+		_, err := NodeSpecFromPanelValidated(&panel.NodeConfig{
+			Protocol:   "shadowsocks",
+			ServerPort: 8388,
+			CustomOutbounds: []panel.OutboundConfig{{
+				Tag: "warp", Protocol: "wireguard", Settings: map[string]any{"server": "1.1.1.1", "server_port": 2408, "private_key": "pk"},
+			}},
+			CustomRouteRules: []panel.CustomRouteRule{{
+				Action: panel.RouteAction{Type: "route", Target: "missing"},
+				Match:  panel.RouteMatch{DomainSuffixes: []string{"example.com"}},
+			}},
+		}, config.KernelConfig{Type: "singbox"})
+		if err == nil {
+			t.Fatal("expected error, got nil")
+		}
+		if !strings.Contains(err.Error(), `custom_route_rules[0].action.target references unknown outbound "missing"`) {
+			t.Fatalf("unexpected error: %v", err)
+		}
+	})
+}

internal/model/standalone.go

@@ -0,0 +1,109 @@
+package model
+
+import "github.com/cedar2025/xboard-node/internal/config"
+
+func NodeSpecFromStandalone(cfg *config.Config) *NodeSpec {
+	sc := cfg.Standalone
+	if sc == nil {
+		return nil
+	}
+
+	routes := make([]RouteRule, 0, len(sc.Node.Routes))
+	for _, route := range sc.Node.Routes {
+		routes = append(routes, RouteRule{
+			ID:          route.ID,
+			Match:       cloneStringSlice(route.Match),
+			Action:      route.Action,
+			ActionValue: route.ActionValue,
+		})
+	}
+
+	customRouteRules := make([]CustomRouteRule, 0, len(sc.Node.CustomRouteRules))
+	for _, rule := range sc.Node.CustomRouteRules {
+		customRouteRules = append(customRouteRules, CustomRouteRule{
+			Name:     rule.Name,
+			Disabled: rule.Disabled,
+			Match: RouteMatch{
+				Domains:        cloneStringSlice(rule.Match.Domains),
+				DomainSuffixes: cloneStringSlice(rule.Match.DomainSuffixes),
+				IPCIDRs:        cloneStringSlice(rule.Match.IPCIDRs),
+				Ports:          cloneStringSlice(rule.Match.Ports),
+				Networks:       cloneStringSlice(rule.Match.Networks),
+				SourceCIDRs:    cloneStringSlice(rule.Match.SourceCIDRs),
+				SourcePorts:    cloneStringSlice(rule.Match.SourcePorts),
+			},
+			Action: RouteAction{Type: rule.Action.Type, Target: rule.Action.Target},
+		})
+	}
+
+	var multiplex *MultiplexConfig
+	if sc.Node.Multiplex != nil {
+		multiplex = &MultiplexConfig{
+			Enabled:        sc.Node.Multiplex.Enabled,
+			Protocol:       sc.Node.Multiplex.Protocol,
+			MaxConnections: sc.Node.Multiplex.MaxConnections,
+			MinStreams:     sc.Node.Multiplex.MinStreams,
+			MaxStreams:     sc.Node.Multiplex.MaxStreams,
+			Padding:        sc.Node.Multiplex.Padding,
+		}
+		if sc.Node.Multiplex.Brutal != nil {
+			multiplex.Brutal = &BrutalConfig{
+				Enabled:  sc.Node.Multiplex.Brutal.Enabled,
+				UpMbps:   sc.Node.Multiplex.Brutal.UpMbps,
+				DownMbps: sc.Node.Multiplex.Brutal.DownMbps,
+			}
+		}
+	}
+
+	return &NodeSpec{
+		Protocol:            sc.Node.Protocol,
+		ListenIP:            sc.Node.ListenIP,
+		ServerPort:          sc.Node.ServerPort,
+		Network:             sc.Node.Network,
+		NetworkSettings:     cloneAnyMap(sc.Node.NetworkSettings),
+		Routes:              routes,
+		CustomRouteRules:    customRouteRules,
+		KernelType:          cfg.Kernel.Type,
+		KernelLogLevel:      cfg.Kernel.LogLevel,
+		Cipher:              sc.Node.Cipher,
+		Plugin:              sc.Node.Plugin,
+		PluginOpt:           sc.Node.PluginOpt,
+		ServerKey:           sc.Node.ServerKey,
+		TLS:                 sc.Node.TLS,
+		Flow:                sc.Node.Flow,
+		Decryption:          sc.Node.Decryption,
+		TLSSettings:         cloneAnyMap(sc.Node.TLSSettings),
+		Host:                sc.Node.Host,
+		ServerName:          sc.Node.ServerName,
+		Version:             sc.Node.Version,
+		UpMbps:              sc.Node.UpMbps,
+		DownMbps:            sc.Node.DownMbps,
+		Obfs:                sc.Node.Obfs,
+		ObfsPassword:        sc.Node.ObfsPassword,
+		CongestionControl:   sc.Node.CongestionControl,
+		PaddingScheme:       sc.Node.PaddingScheme,
+		Transport:           sc.Node.Transport,
+		TrafficPattern:      sc.Node.TrafficPattern,
+		Multiplex:           multiplex,
+		AcceptProxyProtocol: sc.Node.AcceptProxyProtocol,
+	}
+}
+
+func NodeSpecFromStandaloneValidated(cfg *config.Config) (*NodeSpec, error) {
+	spec := NodeSpecFromStandalone(cfg)
+	if err := ValidateNodeSpec(spec, cfg.Kernel); err != nil {
+		return nil, err
+	}
+	return spec, nil
+}
+
+func UserSpecsFromStandalone(cfg *config.Config) []UserSpec {
+	if cfg.Standalone == nil {
+		return nil
+	}
+	users := make([]UserSpec, 0, len(cfg.Standalone.Users))
+	for _, user := range cfg.Standalone.Users {
+		users = append(users, UserSpec{ID: user.ID, UUID: user.UUID, SpeedLimit: user.SpeedLimit, DeviceLimit: user.DeviceLimit})
+	}
+	return users
+}

internal/model/standalone_test.go

@@ -0,0 +1,49 @@
+package model
+
+import (
+	"testing"
+
+	"github.com/cedar2025/xboard-node/internal/config"
+)
+
+func TestNodeSpecFromStandalone(t *testing.T) {
+	cfg := &config.Config{
+		Kernel: config.KernelConfig{Type: "xray", LogLevel: "info"},
+		Standalone: &config.StandaloneConfig{
+			Enabled: true,
+			Node: config.StandaloneNodeConfig{
+				Protocol:            "vless",
+				ListenIP:            "::",
+				ServerPort:          8443,
+				Network:             "ws",
+				NetworkSettings:     map[string]any{"path": "/ws"},
+				TLS:                 1,
+				Host:                "node.example.com",
+				ServerName:          "node.example.com",
+				AcceptProxyProtocol: true,
+				PaddingScheme:       "stop=8\n0=30-30",
+				Routes:              []config.StandaloneRouteRule{{ID: 1, Match: []string{"example.com"}, Action: "direct"}},
+				Multiplex:           &config.StandaloneMultiplexConfig{Enabled: true, Protocol: "smux", MaxConnections: 4},
+			},
+			Users: []config.StandaloneUser{{ID: 7, UUID: "11111111-1111-1111-1111-111111111111", SpeedLimit: 50, DeviceLimit: 2}},
+		},
+	}
+
+	nc := NodeSpecFromStandalone(cfg)
+	users := UserSpecsFromStandalone(cfg)
+	if nc.Protocol != "vless" || nc.ServerPort != 8443 {
+		t.Fatalf("unexpected node spec: %#v", nc)
+	}
+	if nc.NetworkSettings["path"] != "/ws" {
+		t.Fatalf("network_settings.path: got %v", nc.NetworkSettings["path"])
+	}
+	if nc.PaddingScheme != "stop=8\n0=30-30" {
+		t.Fatalf("padding_scheme: got %q", nc.PaddingScheme)
+	}
+	if nc.Multiplex == nil || !nc.Multiplex.Enabled {
+		t.Fatal("expected multiplex config to be copied")
+	}
+	if len(users) != 1 || users[0].DeviceLimit != 2 {
+		t.Fatalf("users: %#v", users)
+	}
+}

internal/model/types.go

@@ -0,0 +1,133 @@
+package model
+
+import "github.com/cedar2025/xboard-node/internal/config"
+
+type NodeSpec struct {
+	Protocol        string
+	ListenIP        string
+	ServerPort      int
+	Network         string
+	NetworkSettings map[string]any
+	Routes          []RouteRule
+
+	KernelType       string
+	KernelLogLevel   string
+	CustomOutbounds  []OutboundConfig
+	CustomRoutes     []map[string]any
+	CustomRouteRules []CustomRouteRule
+	CertConfig       *config.CertConfig
+	AutoTLS          bool
+	Domain           string
+
+	Cipher    string
+	Plugin    string
+	PluginOpt string
+	ServerKey string
+
+	TLS         int
+	Flow        string
+	Decryption  string
+	TLSSettings map[string]any
+
+	Host       string
+	ServerName string
+
+	Version      int
+	UpMbps       int
+	DownMbps     int
+	Obfs         string
+	ObfsPassword string
+
+	CongestionControl string
+	PaddingScheme     string
+	Transport         string
+	TrafficPattern    string
+
+	Multiplex           *MultiplexConfig
+	AcceptProxyProtocol bool
+}
+
+type OutboundConfig struct {
+	Tag      string
+	Protocol string
+	Settings map[string]any
+	ProxyTag string
+}
+
+type RouteRule struct {
+	ID          int
+	Match       []string
+	Action      string
+	ActionValue string
+}
+
+type MultiplexConfig struct {
+	Enabled        bool
+	Protocol       string
+	MaxConnections int
+	MinStreams     int
+	MaxStreams     int
+	Padding        bool
+	Brutal         *BrutalConfig
+}
+
+type BrutalConfig struct {
+	Enabled  bool
+	UpMbps   int
+	DownMbps int
+}
+
+type UserSpec struct {
+	ID          int
+	UUID        string
+	SpeedLimit  int
+	DeviceLimit int
+}
+
+func (n *NodeSpec) GetProxyProtocol() bool {
+	if n == nil {
+		return false
+	}
+	if n.AcceptProxyProtocol {
+		return true
+	}
+	if n.NetworkSettings != nil {
+		if v, ok := n.NetworkSettings["acceptProxyProtocol"]; ok {
+			if b, ok := v.(bool); ok {
+				return b
+			}
+		}
+	}
+	return false
+}
+
+func cloneAnyMap(src map[string]any) map[string]any {
+	if len(src) == 0 {
+		return nil
+	}
+	out := make(map[string]any, len(src))
+	for k, v := range src {
+		out[k] = v
+	}
+	return out
+}
+
+func cloneMapSlice(src []map[string]any) []map[string]any {
+	if len(src) == 0 {
+		return nil
+	}
+	out := make([]map[string]any, 0, len(src))
+	for _, item := range src {
+		out = append(out, cloneAnyMap(item))
+	}
+	return out
+}
+
+func cloneStringSlice(src []string) []string {
+	if len(src) == 0 {
+		return nil
+	}
+	out := make([]string, len(src))
+	copy(out, src)
+	return out
+}

internal/model/validate.go

@@ -0,0 +1,100 @@
+package model
+
+import (
+	"fmt"
+	"strings"
+
+	"github.com/cedar2025/xboard-node/internal/config"
+)
+
+func ValidateNodeSpec(n *NodeSpec, kcfg config.KernelConfig) error {
+	if n == nil {
+		return nil
+	}
+
+	effectiveKernelType := strings.TrimSpace(kcfg.Type)
+	if effectiveKernelType == "" {
+		effectiveKernelType = strings.TrimSpace(n.KernelType)
+	}
+	kernelType, err := normalizeKernelType(effectiveKernelType)
+	if err != nil {
+		return fmt.Errorf("normalize kernel type: %w", err)
+	}
+
+	additionalOutboundSources, err := collectAdditionalOutboundTagSources(kcfg.CustomConfig, kcfg.CustomOutbound)
+	if err != nil {
+		return fmt.Errorf("collect additional outbound tags: %w", err)
+	}
+	if err := validateOutboundTagCollisions(n.CustomOutbounds, additionalOutboundSources); err != nil {
+		return fmt.Errorf("validate outbound tags: %w", err)
+	}
+	additionalTags := additionalTagNames(additionalOutboundSources)
+	availableTags := buildAvailableOutboundTags(n.CustomOutbounds, additionalTags)
+	if err := ValidateCustomOutboundsForKernel(n.CustomOutbounds, kernelType, additionalTags); err != nil {
+		return fmt.Errorf("validate custom outbounds: %w", err)
+	}
+	if err := ValidateCustomRouteRules(n.CustomRouteRules, kernelType, availableTags); err != nil {
+		return fmt.Errorf("validate custom route rules: %w", err)
+	}
+	if err := validateTransportKernel(n.Network, kernelType); err != nil {
+		return err
+	}
+	return nil
+}
+
+// singboxUnsupportedTransports lists transport types that sing-box does not support.
+var singboxUnsupportedTransports = map[string]bool{
+	"xhttp":     true,
+	"splithttp": true,
+}
+
+func validateTransportKernel(network, kernelType string) error {
+	net := strings.ToLower(strings.TrimSpace(network))
+	if kernelType == "singbox" && singboxUnsupportedTransports[net] {
+		return fmt.Errorf("transport %q is not supported by sing-box kernel; use xray kernel instead", net)
+	}
+	return nil
+}
+
+// ResolveKernelForTransport returns the kernel type required by the given
+// transport. If the configured kernel cannot handle the transport, it returns
+// the kernel that can. Otherwise it returns configuredKernel unchanged.
+// This is used in machine mode to auto-switch kernel per node.
+func ResolveKernelForTransport(network, configuredKernel string) string {
+	net := strings.ToLower(strings.TrimSpace(network))
+	if configuredKernel == "singbox" && singboxUnsupportedTransports[net] {
+		return "xray"
+	}
+	return configuredKernel
+}
+
+func normalizeKernelType(value string) (string, error) {
+	switch strings.ToLower(strings.TrimSpace(value)) {
+	case "singbox", "sing-box":
+		return "singbox", nil
+	case "xray":
+		return "xray", nil
+	default:
+		return "", fmt.Errorf("unsupported kernel type %q", value)
+	}
+}
+
+func buildAvailableOutboundTags(structured []OutboundConfig, rawTags []string) map[string]struct{} {
+	available := map[string]struct{}{
+		"direct": {},
+		"block":  {},
+	}
+	for _, outbound := range structured {
+		tag := strings.ToLower(strings.TrimSpace(outbound.Tag))
+		if tag != "" {
+			available[tag] = struct{}{}
+		}
+	}
+	for _, tag := range rawTags {
+		tag = strings.ToLower(strings.TrimSpace(tag))
+		if tag != "" {
+			available[tag] = struct{}{}
+		}
+	}
+	return available
+}

internal/monitor/monitor.go

@@ -1,14 +1,17 @@
 package monitor
 
 import (
-	"log/slog"
 	"runtime"
+	"strings"
+	"sync"
 	"time"
 
+	"github.com/cedar2025/xboard-node/internal/nlog"
 	"github.com/shirou/gopsutil/v4/cpu"
 	"github.com/shirou/gopsutil/v4/disk"
 	"github.com/shirou/gopsutil/v4/load"
 	"github.com/shirou/gopsutil/v4/mem"
+	"github.com/shirou/gopsutil/v4/net"
 )
 
 var startTime = time.Now()
@@ -18,6 +21,9 @@ func init() {
 	// always returns 0% because it has no prior sample. This throwaway call
 	// seeds the baseline so subsequent Collect() calls return real values.
 	cpu.Percent(500*time.Millisecond, false)
+
+	// Seed the network baseline so the first Collect() can compute rates.
+	collectNetSpeed()
 }
 
 // Status holds system resource metrics
@@ -36,11 +42,81 @@ type Status struct {
 	DiskUsed   uint64
 	Goroutines int
 
+	// Network speed (bytes/sec), -1 means unavailable (first sample).
+	NetInSpeed  float64
+	NetOutSpeed float64
+
 	// GC metrics (process-wide)
 	NumGC       uint32
 	LastPauseMS float64
 }
 
+// netBaseline tracks the previous network counters for rate calculation.
+var (
+	netMu       sync.Mutex
+	netPrevRecv uint64
+	netPrevSent uint64
+	netPrevTime time.Time
+	netHasBase  bool
+)
+
+// skipInterface returns true for loopback and common virtual interfaces.
+func skipInterface(name string) bool {
+	lower := strings.ToLower(name)
+	for _, prefix := range []string{"lo", "docker", "veth", "br-", "virbr", "vnet", "tun", "tap"} {
+		if strings.HasPrefix(lower, prefix) {
+			return true
+		}
+	}
+	return false
+}
+
+// collectNetSpeed calculates network in/out bytes per second since last call.
+// Returns -1, -1 on first call or if counters decreased (reboot).
+func collectNetSpeed() (inSpeed, outSpeed float64) {
+	counters, err := net.IOCounters(true) // per-interface
+	if err != nil {
+		nlog.Core().Debug("failed to get network counters", "error", err)
+		return -1, -1
+	}
+
+	var totalRecv, totalSent uint64
+	for _, c := range counters {
+		if skipInterface(c.Name) {
+			continue
+		}
+		totalRecv += c.BytesRecv
+		totalSent += c.BytesSent
+	}
+
+	now := time.Now()
+
+	netMu.Lock()
+	defer netMu.Unlock()
+
+	if !netHasBase {
+		netPrevRecv, netPrevSent, netPrevTime, netHasBase = totalRecv, totalSent, now, true
+		return -1, -1
+	}
+
+	elapsed := now.Sub(netPrevTime).Seconds()
+	if elapsed <= 0 {
+		return -1, -1
+	}
+
+	// Counter decreased → system reboot or interface reset; reset baseline.
+	if totalRecv < netPrevRecv || totalSent < netPrevSent {
+		netPrevRecv, netPrevSent, netPrevTime = totalRecv, totalSent, now
+		return -1, -1
+	}
+
+	inSpeed = float64(totalRecv-netPrevRecv) / elapsed
+	outSpeed = float64(totalSent-netPrevSent) / elapsed
+
+	netPrevRecv, netPrevSent, netPrevTime = totalRecv, totalSent, now
+	return inSpeed, outSpeed
+}
+
 // Collect gathers current system metrics
 func Collect() Status {
 	var s Status
@@ -50,7 +126,7 @@ func Collect() Status {
 	if cpuPercent, err := cpu.Percent(0, false); err == nil && len(cpuPercent) > 0 {
 		s.CPU = cpuPercent[0]
 	} else if err != nil {
-		slog.Debug("failed to get CPU usage", "error", err)
+		nlog.Core().Debug("failed to get CPU usage", "error", err)
 	}
 
 	// Per-core CPU usage (best-effort; safe if it fails).
@@ -79,6 +155,8 @@ func Collect() Status {
 		s.DiskUsed = diskStat.Used
 	}
 
+	s.NetInSpeed, s.NetOutSpeed = collectNetSpeed()
+
 	// GC metrics
 	var ms runtime.MemStats
 	runtime.ReadMemStats(&ms)

internal/nlog/nlog.go

@@ -0,0 +1,323 @@
+package nlog
+
+import (
+	"context"
+	"fmt"
+	"io"
+	"log/slog"
+	"os"
+	"strings"
+	"sync"
+	"time"
+)
+
+// ANSI color codes
+const (
+	ColorReset  = "\033[0m"
+	ColorGray   = "\033[90m" // 时间戳
+	ColorWhite  = "\033[37m" // INFO
+	ColorYellow = "\033[33m" // WARN
+	ColorRed    = "\033[31m" // ERROR
+	ColorCyan   = "\033[36m" // 节点前缀
+	ColorGreen  = "\033[32m" // 调试信息
+	ColorBlue   = "\033[34m" // core 前缀
+)
+
+// NodeLog provides structured logging with node context.
+// Format: LEVEL [protocol:port] message
+type NodeLog struct {
+	prefix string // e.g., "shadowsocks:10005" or "trojan:10033"
+}
+
+// Global logger state
+var (
+	mu          sync.RWMutex
+	defaultNode string
+	nodeLoggers = make(map[string]*NodeLog)
+
+	logMu      sync.RWMutex
+	logWriter  io.Writer = os.Stdout
+	logMin     = slog.LevelInfo
+	logColor   = true
+)
+
+// Init configures process-wide log output (called from config.InitLogger).
+// w must be non-nil. minLevel drops messages below that severity.
+// color enables ANSI coloring when true (typically TTY + stdout/stderr).
+func Init(w io.Writer, minLevel slog.Level, color bool) {
+	if w == nil {
+		w = os.Stdout
+	}
+	logMu.Lock()
+	defer logMu.Unlock()
+	logWriter = w
+	logMin = minLevel
+	logColor = color
+}
+
+// formatMsg appends alternating key-value pairs like slog: ("k", v, "k2", v2).
+func formatMsg(msg string, args []any) string {
+	if len(args) == 0 {
+		return msg
+	}
+	var b strings.Builder
+	b.WriteString(msg)
+	for i := 0; i < len(args); i += 2 {
+		if i+1 < len(args) {
+			fmt.Fprintf(&b, " %v=%v", args[i], args[i+1])
+		} else {
+			fmt.Fprintf(&b, " %v", args[i])
+		}
+	}
+	return b.String()
+}
+
+// SetDefault sets the default node prefix for core-level logs.
+func SetDefault(prefix string) {
+	mu.Lock()
+	defaultNode = prefix
+	mu.Unlock()
+}
+
+// ForNode returns a NodeLog for the given protocol and port.
+func ForNode(protocol string, port int) *NodeLog {
+	key := fmt.Sprintf("%s:%d", normalizeProto(protocol), port)
+	mu.RLock()
+	if nl, ok := nodeLoggers[key]; ok {
+		mu.RUnlock()
+		return nl
+	}
+	mu.RUnlock()
+
+	mu.Lock()
+	defer mu.Unlock()
+	// Double check
+	if nl, ok := nodeLoggers[key]; ok {
+		return nl
+	}
+	nl := &NodeLog{prefix: key}
+	nodeLoggers[key] = nl
+	return nl
+}
+
+// Core returns a NodeLog for core/system messages.
+func Core() *NodeLog {
+	mu.RLock()
+	if nl, ok := nodeLoggers["core"]; ok {
+		mu.RUnlock()
+		return nl
+	}
+	mu.RUnlock()
+
+	mu.Lock()
+	defer mu.Unlock()
+	if nl, ok := nodeLoggers["core"]; ok {
+		return nl
+	}
+	nl := &NodeLog{prefix: "core"}
+	nodeLoggers["core"] = nl
+	return nl
+}
+
+// normalizeProto returns the full protocol name for display.
+func normalizeProto(p string) string {
+	switch strings.ToLower(p) {
+	case "shadowsocks", "ss":
+		return "shadowsocks"
+	case "vmess":
+		return "vmess"
+	case "vless":
+		return "vless"
+	case "trojan":
+		return "trojan"
+	case "hysteria2", "hysteria":
+		return "hysteria2"
+	case "tuic":
+		return "tuic"
+	case "anytls":
+		return "anytls"
+	case "naive":
+		return "naive"
+	case "http":
+		return "http"
+	case "socks":
+		return "socks"
+	default:
+		return p
+	}
+}
+
+// ─── Logging Methods ────────────────────────────────────────────────────────
+
+func (nl *NodeLog) Debug(msg string, args ...any) {
+	logWithColor(slog.LevelDebug, nl.prefix, msg, args...)
+}
+
+func (nl *NodeLog) Info(msg string, args ...any) {
+	logWithColor(slog.LevelInfo, nl.prefix, msg, args...)
+}
+
+func (nl *NodeLog) Warn(msg string, args ...any) {
+	logWithColor(slog.LevelWarn, nl.prefix, msg, args...)
+}
+
+func (nl *NodeLog) Error(msg string, args ...any) {
+	logWithColor(slog.LevelError, nl.prefix, msg, args...)
+}
+
+// logWithColor writes one line to the configured writer (see Init).
+// Format: HH:MM:SS.mmm LEVEL [prefix] message [key=value ...]
+func logWithColor(level slog.Level, prefix, msg string, args ...any) {
+	logMu.RLock()
+	out := logWriter
+	min := logMin
+	color := logColor
+	logMu.RUnlock()
+
+	if level < min {
+		return
+	}
+
+	now := time.Now().Format("15:04:05.000")
+	fullMsg := formatMsg(msg, args)
+
+	var levelStr string
+	switch level {
+	case slog.LevelDebug:
+		levelStr = "DEBUG"
+	case slog.LevelInfo:
+		levelStr = "INFO "
+	case slog.LevelWarn:
+		levelStr = "WARN "
+	case slog.LevelError:
+		levelStr = "ERROR"
+	default:
+		levelStr = "?????"
+	}
+
+	if !color {
+		fmt.Fprintf(out, "%s %s [%s] %s\n", now, strings.TrimSpace(levelStr), prefix, fullMsg)
+		return
+	}
+
+	var levelColor string
+	switch level {
+	case slog.LevelDebug:
+		levelColor = ColorGreen
+	case slog.LevelInfo:
+		levelColor = ColorWhite
+	case slog.LevelWarn:
+		levelColor = ColorYellow
+	case slog.LevelError:
+		levelColor = ColorRed
+	default:
+		levelColor = ColorWhite
+	}
+
+	prefixColor := ColorCyan
+	if prefix == "core" {
+		prefixColor = ColorBlue
+	}
+
+	fmt.Fprintf(out, "%s%s%s %s%s%s %s[%s]%s %s\n",
+		ColorGray, now, ColorReset,
+		levelColor, levelStr, ColorReset,
+		prefixColor, prefix, ColorReset,
+		fullMsg,
+	)
+}
+
+// ─── Startup Summary ────────────────────────────────────────────────────────
+
+// StartupSummary logs a condensed startup summary.
+type StartupSummary struct {
+	mu    sync.Mutex
+	nodes []nodeInfo
+}
+
+type nodeInfo struct {
+	Protocol string
+	Port     int
+	Users    int
+}
+
+func NewStartupSummary() *StartupSummary {
+	return &StartupSummary{}
+}
+
+func (s *StartupSummary) AddNode(protocol string, port, users int) {
+	s.mu.Lock()
+	s.nodes = append(s.nodes, nodeInfo{protocol, port, users})
+	s.mu.Unlock()
+}
+
+func (s *StartupSummary) Print() {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	if len(s.nodes) == 0 {
+		Core().Warn("no nodes configured")
+		return
+	}
+
+	// Group by protocol
+	byProto := make(map[string][]nodeInfo)
+	for _, n := range s.nodes {
+		proto := normalizeProto(n.Protocol)
+		byProto[proto] = append(byProto[proto], n)
+	}
+
+	// Build summary line
+	parts := make([]string, 0, len(byProto))
+	for proto, nodes := range byProto {
+		ports := make([]string, 0, len(nodes))
+		for _, n := range nodes {
+			ports = append(ports, fmt.Sprintf("%d", n.Port))
+		}
+		parts = append(parts, fmt.Sprintf("%s:%s", proto, strings.Join(ports, ",")))
+	}
+
+	Core().Info(fmt.Sprintf("started %d nodes: %s", len(s.nodes), strings.Join(parts, " | ")))
+}
+
+// ─── Helpers ────────────────────────────────────────────────────────────────
+
+// ConfigUpdated logs a config update event.
+func ConfigUpdated(nl *NodeLog, users int) {
+	nl.Info(fmt.Sprintf("config updated, %d users", users))
+}
+
+// HotReload logs a hot-reload event (no restart).
+func HotReload(nl *NodeLog, what string) {
+	nl.Debug(fmt.Sprintf("hot-reload: %s", what))
+}
+
+// FullRestart logs a full kernel restart.
+func FullRestart(nl *NodeLog, reason string) {
+	nl.Info(fmt.Sprintf("kernel restart: %s", reason))
+}
+
+// ReportPushed logs a report push event.
+func ReportPushed(users, online int) {
+	Core().Info(fmt.Sprintf("report pushed: %d users, %d online", users, online))
+}
+
+// TrackerStats logs tracker statistics.
+func TrackerStats(conns, users int) {
+	Core().Debug(fmt.Sprintf("tracker: %d conns, %d users online", conns, users))
+}
+
+// Context-aware logging
+
+type ctxKey struct{}
+
+func WithNode(ctx context.Context, nl *NodeLog) context.Context {
+	return context.WithValue(ctx, ctxKey{}, nl)
+}
+
+func FromContext(ctx context.Context) *NodeLog {
+	if nl, ok := ctx.Value(ctxKey{}).(*NodeLog); ok {
+		return nl
+	}
+	return Core()
+}

internal/panel/client.go

@@ -5,7 +5,6 @@ import (
 	"encoding/json"
 	"fmt"
 	"io"
-	"log/slog"
 	"net/http"
 	"net/url"
 	"reflect"
@@ -16,6 +15,7 @@ import (
 	"time"
 
 	"github.com/cedar2025/xboard-node/internal/config"
+	"github.com/cedar2025/xboard-node/internal/nlog"
 	"github.com/go-viper/mapstructure/v2"
 )
 
@@ -25,12 +25,15 @@ var (
 	onlineMapPool  = sync.Pool{New: func() interface{} { return make(map[string]int) }}
 )
 
-// Client communicates with the Xboard panel API
+// Client communicates with the Xboard panel API.
+// When machineID > 0 the client uses machine-level authentication
+// (machine_id + token) instead of the legacy (token + node_type) scheme.
 type Client struct {
 	baseURL    string
 	token      string
 	nodeID     int
 	nodeType   string
+	machineID  int
 	httpClient *http.Client
 
 	configETag string
@@ -40,13 +43,14 @@ type Client struct {
 	apiFailure atomic.Uint64
 }
 
-// NewClient creates a new panel API client
+// NewClient creates a new panel API client.
 func NewClient(cfg config.PanelConfig) *Client {
 	return &Client{
-		baseURL:  strings.TrimRight(cfg.URL, "/"),
-		token:    cfg.Token,
-		nodeID:   cfg.NodeID,
-		nodeType: cfg.NodeType,
+		baseURL:   strings.TrimRight(cfg.URL, "/"),
+		token:     cfg.Token,
+		nodeID:    cfg.NodeID,
+		nodeType:  cfg.NodeType,
+		machineID: cfg.MachineID,
 		httpClient: &http.Client{
 			Timeout: 30 * time.Second,
 			Transport: &http.Transport{
@@ -58,6 +62,26 @@ func NewClient(cfg config.PanelConfig) *Client {
 	}
 }
 
+// ForNode returns a new client bound to a specific node_id, sharing the
+// same base URL, auth and HTTP transport. ETags and metric counters start fresh.
+func (c *Client) ForNode(nodeID int) *Client {
+	return &Client{
+		baseURL:    c.baseURL,
+		token:      c.token,
+		nodeID:     nodeID,
+		nodeType:   c.nodeType,
+		machineID:  c.machineID,
+		httpClient: c.httpClient,
+	}
+}
+
+// ResetConfigETag clears the cached config ETag so the next GetConfig call
+// returns a full response instead of 304. Used by machine mode after a
+// pre-fetch to probe the transport type.
+func (c *Client) ResetConfigETag() {
+	c.configETag = ""
+}
+
 // Handshake calls the new v2 API to get WS config + initial data in one shot.
 func (c *Client) Handshake() (*HandshakeResponse, error) {
 	resp, err := c.doRequest("POST", "/api/v2/server/handshake", nil, "")
@@ -174,9 +198,30 @@ func decodeWeakRaw(input map[string]interface{}, output interface{}) error {
 	return decoder.Decode(input)
 }
 
+// configPath returns the API path for config fetching.
+// Machine mode uses the v2 endpoint; legacy mode uses the v1 UniProxy endpoint.
+// fetchMethod returns POST for all API calls (token stays in body, not URL).
+// v2 already used POST; v1 legacy endpoints need panel route update to match.
+func (c *Client) fetchMethod() string { return "POST" }
+
+func (c *Client) configPath() string {
+	if c.machineID > 0 {
+		return "/api/v2/server/config"
+	}
+	return "/api/v1/server/UniProxy/config"
+}
+
+// userPath returns the API path for user fetching.
+func (c *Client) userPath() string {
+	if c.machineID > 0 {
+		return "/api/v2/server/user"
+	}
+	return "/api/v1/server/UniProxy/user"
+}
+
 // GetConfig fetches node configuration. Returns nil if not modified (304).
 func (c *Client) GetConfig() (*NodeConfig, error) {
-	resp, err := c.doRequest("GET", "/api/v1/server/UniProxy/config", nil, c.configETag)
+	resp, err := c.doRequest(c.fetchMethod(), c.configPath(), nil, c.configETag)
 	if err != nil {
 		return nil, fmt.Errorf("get config: %w", err)
 	}
@@ -214,7 +259,7 @@ func (c *Client) GetConfig() (*NodeConfig, error) {
 
 // GetUsers fetches available users. Returns nil if not modified (304).
 func (c *Client) GetUsers() ([]User, error) {
-	resp, err := c.doRequest("GET", "/api/v1/server/UniProxy/user", nil, c.userETag)
+	resp, err := c.doRequest(c.fetchMethod(), c.userPath(), nil, c.userETag)
 	if err != nil {
 		return nil, fmt.Errorf("get users: %w", err)
 	}
@@ -280,14 +325,78 @@ func (c *Client) ResetETags() {
 	c.userETag = ""
 }
 
-// postJSON marshals a map payload (with auth fields injected) and POSTs it.
-// Auth fields are injected before marshalling — single marshal pass.
-func (c *Client) postJSON(path string, payload map[string]interface{}) error {
-	payload["token"] = c.token
-	payload["node_id"] = c.nodeID
-	if c.nodeType != "" {
-		payload["node_type"] = c.nodeType
+// GetMachineNodes fetches the list of active nodes bound to this machine.
+func (c *Client) GetMachineNodes() (*MachineNodesResponse, error) {
+	resp, err := c.doRequest("POST", "/api/v2/server/machine/nodes", nil, "")
+	if err != nil {
+		return nil, fmt.Errorf("get machine nodes: %w", err)
 	}
+	defer drainAndClose(resp.Body)
+
+	if resp.StatusCode != http.StatusOK {
+		body, _ := io.ReadAll(io.LimitReader(resp.Body, 4096))
+		return nil, fmt.Errorf("machine nodes status %d: %s", resp.StatusCode, body)
+	}
+
+	var out MachineNodesResponse
+	if err := json.NewDecoder(resp.Body).Decode(&out); err != nil {
+		return nil, fmt.Errorf("decode machine nodes: %w", err)
+	}
+	return &out, nil
+}
+
+// ReportMachineStatus sends machine-level load metrics to the panel.
+// netIn/netOut are bytes/sec; negative values mean "unavailable" (first sample).
+func (c *Client) ReportMachineStatus(cpu float64, mem, swap, disk [2]uint64, netIn, netOut float64) error {
+	payload := map[string]interface{}{
+		"cpu":  cpu,
+		"mem":  map[string]interface{}{"total": mem[0], "used": mem[1]},
+		"swap": map[string]interface{}{"total": swap[0], "used": swap[1]},
+		"disk": map[string]interface{}{"total": disk[0], "used": disk[1]},
+	}
+	if netIn >= 0 && netOut >= 0 {
+		payload["net"] = map[string]interface{}{"in_speed": netIn, "out_speed": netOut}
+	}
+	return c.postJSON("/api/v2/server/machine/status", payload)
+}
+
+// injectAuth writes authentication fields into a payload map.
+func (c *Client) injectAuth(m map[string]interface{}) {
+	m["token"] = c.token
+	if c.machineID > 0 {
+		m["machine_id"] = c.machineID
+		if c.nodeID > 0 {
+			m["node_id"] = c.nodeID
+		}
+	} else {
+		m["node_id"] = c.nodeID
+		if c.nodeType != "" {
+			m["node_type"] = c.nodeType
+		}
+	}
+}
+
+// authQuery builds URL query parameters for GET requests.
+func (c *Client) authQuery() url.Values {
+	q := url.Values{}
+	q.Set("token", c.token)
+	if c.machineID > 0 {
+		q.Set("machine_id", strconv.Itoa(c.machineID))
+		if c.nodeID > 0 {
+			q.Set("node_id", strconv.Itoa(c.nodeID))
+		}
+	} else {
+		q.Set("node_id", strconv.Itoa(c.nodeID))
+		if c.nodeType != "" {
+			q.Set("node_type", c.nodeType)
+		}
+	}
+	return q
+}
+
+// postJSON marshals a map payload (with auth fields injected) and POSTs it.
+func (c *Client) postJSON(path string, payload map[string]interface{}) error {
+	c.injectAuth(payload)
 
 	body, err := json.Marshal(payload)
 	if err != nil {
@@ -313,23 +422,12 @@ func (c *Client) doRequest(method, path string, body []byte, ifNoneMatch string)
 
 	var bodyReader io.Reader
 	if method == "GET" {
-		q := url.Values{}
-		q.Set("token", c.token)
-		q.Set("node_id", strconv.Itoa(c.nodeID))
-		if c.nodeType != "" {
-			q.Set("node_type", c.nodeType)
-		}
-		fullURL += "?" + q.Encode()
+		fullURL += "?" + c.authQuery().Encode()
 	} else if body != nil {
 		bodyReader = bytes.NewReader(body)
 	} else {
-		authOnly := map[string]interface{}{
-			"token":   c.token,
-			"node_id": c.nodeID,
-		}
-		if c.nodeType != "" {
-			authOnly["node_type"] = c.nodeType
-		}
+		authOnly := make(map[string]interface{})
+		c.injectAuth(authOnly)
 		merged, _ := json.Marshal(authOnly)
 		bodyReader = bytes.NewReader(merged)
 	}
@@ -345,7 +443,7 @@ func (c *Client) doRequest(method, path string, body []byte, ifNoneMatch string)
 		req.Header.Set("If-None-Match", ifNoneMatch)
 	}
 
-	slog.Debug("panel request", "method", method, "path", path)
+	nlog.Core().Debug("panel request", "method", method, "path", path)
 
 	resp, err := c.httpClient.Do(req)
 	if err != nil {
@@ -359,7 +457,7 @@ func (c *Client) doRequest(method, path string, body []byte, ifNoneMatch string)
 		c.apiFailure.Add(1)
 	}
 
-	slog.Debug("panel response", "method", method, "path", path, "status", resp.StatusCode)
+	nlog.Core().Debug("panel response", "method", method, "path", path, "status", resp.StatusCode)
 	return resp, nil
 }
 

internal/panel/types.go

@@ -62,8 +62,29 @@ type Settings struct {
 	PullInterval int `json:"pull_interval"`
 }
 
+// MachineNode is a single entry returned by GET /api/v2/server/machine/nodes.
+type MachineNode struct {
+	ID   int    `json:"id"`
+	Type string `json:"type"`
+	Name string `json:"name"`
+}
+
+// MachineNodesResponse is the response from GET /api/v2/server/machine/nodes.
+type MachineNodesResponse struct {
+	Nodes      []MachineNode      `json:"nodes"`
+	BaseConfig MachineBaseConfig  `json:"base_config"`
+}
+
+// MachineBaseConfig holds polling intervals for machine mode.
+type MachineBaseConfig struct {
+	PushInterval int `json:"push_interval"`
+	PullInterval int `json:"pull_interval"`
+}
+
 // NodeConfig is the response from GET /api/v1/server/UniProxy/config
 type NodeConfig struct {
+	// NodeID is populated in machine-mode WS events for routing.
+	NodeID          int                    `json:"node_id,omitempty"`
 	Protocol        string                 `json:"protocol"`
 	ListenIP        string                 `json:"listen_ip"`
 	ServerPort      int                    `json:"server_port"`
@@ -73,10 +94,11 @@ type NodeConfig struct {
 	Routes          []RouteRule            `json:"routes"`
 
 	// Kernel settings (Xboard extension)
-	KernelType      string           `json:"kernel_type,omitempty"`      // "singbox" or "xray"
-	KernelLogLevel  string           `json:"kernel_log_level,omitempty"` // "info", "warn", etc.
-	CustomOutbounds []OutboundConfig `json:"custom_outbounds,omitempty"`
-	CustomRoutes    []map[string]any `json:"custom_routes,omitempty"`
+	KernelType       string            `json:"kernel_type,omitempty"`      // "singbox" or "xray"
+	KernelLogLevel   string            `json:"kernel_log_level,omitempty"` // "info", "warn", etc.
+	CustomOutbounds  []OutboundConfig  `json:"custom_outbounds,omitempty"`
+	CustomRoutes     []map[string]any  `json:"custom_routes,omitempty"`
+	CustomRouteRules []CustomRouteRule `json:"custom_route_rules,omitempty"`
 
 	// Certificate settings (Xboard extension)
 	CertConfig *CertConfig `json:"cert_config,omitempty"`
@@ -92,6 +114,7 @@ type NodeConfig struct {
 	// VMess / VLESS
 	TLS         int                    `json:"tls,omitempty"`
 	Flow        string                 `json:"flow,omitempty"`
+	Decryption  string                 `json:"decryption,omitempty"`
 	TLSSettings map[string]interface{} `json:"tls_settings,omitempty"`
 
 	// Trojan
@@ -118,10 +141,26 @@ type NodeConfig struct {
 	// Multiplex
 	Multiplex *MultiplexConfig `json:"multiplex,omitempty"`
 
-	// Proxy Protocol
+	// Proxy Protocol (supports both top-level and networkSettings for compatibility)
 	AcceptProxyProtocol bool `json:"accept_proxy_protocol,omitempty"`
 }
 
+// GetProxyProtocol returns true if AcceptProxyProtocol is set either at node level
+// or in networkSettings (for panel compatibility).
+func (nc *NodeConfig) GetProxyProtocol() bool {
+	if nc.AcceptProxyProtocol {
+		return true
+	}
+	if nc.NetworkSettings != nil {
+		if v, ok := nc.NetworkSettings["acceptProxyProtocol"]; ok {
+			if b, ok := v.(bool); ok {
+				return b
+			}
+		}
+	}
+	return false
+}
+
 type MultiplexConfig struct {
 	Enabled        bool          `json:"enabled"`
 	Protocol       string        `json:"protocol"`
@@ -138,9 +177,11 @@ type BrutalConfig struct {
 	DownMbps int  `json:"down_mbps"`
 }
 
-// CertConfig holds certificate automation settings from the panel
+// CertConfig holds certificate automation settings from the panel.
+// The panel may send the mode field as either "cert_mode" or "mode";
+// a custom UnmarshalJSON handles both.
 type CertConfig struct {
-	CertMode    string            `json:"cert_mode"`    // none, dns, http, self, file
+	CertMode    string            `json:"cert_mode"`    // none, dns, http, self, file, content
 	Domain      string            `json:"domain"`       // Certificate domain
 	Email       string            `json:"email"`        // ACME email
 	DNSProvider string            `json:"dns_provider"` // dns mode: cloudflare, alidns, etc.
@@ -152,6 +193,22 @@ type CertConfig struct {
 	KeyContent  string            `json:"key_content"`  // content mode: private key raw string
 }
 
+func (c *CertConfig) UnmarshalJSON(data []byte) error {
+	type plain CertConfig
+	var raw struct {
+		plain
+		ModeFallback string `json:"mode"`
+	}
+	if err := json.Unmarshal(data, &raw); err != nil {
+		return err
+	}
+	*c = CertConfig(raw.plain)
+	if c.CertMode == "" && raw.ModeFallback != "" {
+		c.CertMode = raw.ModeFallback
+	}
+	return nil
+}
+
 // OutboundConfig defines a custom outbound for kernel
 type OutboundConfig struct {
 	Tag      string         `json:"tag"`                 // Unique tag for routing
@@ -172,6 +229,28 @@ type RouteRule struct {
 	ActionValue string   `json:"action_value,omitempty"`
 }
 
+type CustomRouteRule struct {
+	Name     string      `json:"name,omitempty"`
+	Disabled bool        `json:"disabled,omitempty"`
+	Match    RouteMatch  `json:"match,omitempty"`
+	Action   RouteAction `json:"action"`
+}
+
+type RouteMatch struct {
+	Domains        []string `json:"domains,omitempty"`
+	DomainSuffixes []string `json:"domain_suffixes,omitempty"`
+	IPCIDRs        []string `json:"ip_cidrs,omitempty"`
+	Ports          []string `json:"ports,omitempty"`
+	Networks       []string `json:"networks,omitempty"`
+	SourceCIDRs    []string `json:"source_cidrs,omitempty"`
+	SourcePorts    []string `json:"source_ports,omitempty"`
+}
+
+type RouteAction struct {
+	Type   string `json:"type"`
+	Target string `json:"target,omitempty"`
+}
+
 // User represents a user returned by the panel
 type User struct {
 	ID          int    `json:"id"`

internal/panel/ws.go

@@ -4,13 +4,13 @@ import (
 	"context"
 	"encoding/json"
 	"fmt"
-	"log/slog"
 	"math/rand"
 	"net/url"
 	"strconv"
 	"sync/atomic"
 	"time"
 
+	"github.com/cedar2025/xboard-node/internal/nlog"
 	"github.com/gorilla/websocket"
 )
 
@@ -19,6 +19,9 @@ const (
 	WSEventSyncConfig    = "sync.config"
 	WSEventSyncUsers     = "sync.users"
 	WSEventSyncUserDelta = "sync.user.delta"
+	WSEventSyncDevices   = "sync.devices"   // panel → node: global device state
+	WSEventSyncNodes     = "sync.nodes"     // panel → machine: node list changed
+	WSEventReportDevices = "report.devices" // node → panel: report device snapshot
 )
 
 // WSEvent is a parsed data event delivered to the service layer.
@@ -28,6 +31,13 @@ type WSEvent struct {
 	Users       []User
 	DeltaAction string // "add" or "remove" (only for sync.user.delta)
 	DeltaUsers  []User // users affected by the delta
+
+	// Device sync fields
+	DeviceUsers map[int][]string // userID -> IPs (for sync.devices)
+	NodeID      int
+
+	// Machine node discovery fields (for sync.nodes)
+	Nodes []MachineNode
 }
 
 // WSStatusChange notifies the service when WS connectivity changes.
@@ -42,21 +52,45 @@ type wsMessage struct {
 	Timestamp int64           `json:"timestamp,omitempty"`
 }
 
-// Payload structures for data events
+// Payload structures for data events.
 type syncConfigPayload struct {
 	Config    NodeConfig `json:"config"`
 	Timestamp int64      `json:"timestamp"`
+	NodeID    int        `json:"node_id"`
 }
 
 type syncUsersPayload struct {
 	Users     []User `json:"users"`
 	Timestamp int64  `json:"timestamp"`
+	NodeID    int    `json:"node_id"`
 }
 
 type syncUserDeltaPayload struct {
 	Action    string `json:"action"`
 	Users     []User `json:"users"`
 	Timestamp int64  `json:"timestamp"`
+	NodeID    int    `json:"node_id"`
+}
+
+// syncDevicesPayload carries global device state from panel.
+type syncDevicesPayload struct {
+	Users     map[int][]string `json:"users"`
+	Timestamp int64            `json:"timestamp"`
+	NodeID    int              `json:"node_id"`
+}
+
+// syncNodesPayload carries the updated node list for a machine.
+type syncNodesPayload struct {
+	Nodes []MachineNode `json:"nodes"`
+}
+
+// WSClientConfig holds WebSocket client tuning options.
+type WSClientConfig struct {
+	StatusInterval   time.Duration
+	HandshakeTimeout time.Duration
+	BackoffInitial   time.Duration
+	BackoffMax       time.Duration
+	MachineID        int
 }
 
 // WSClient connects to the panel's Workerman WS server using native WebSocket.
@@ -70,17 +104,37 @@ type WSClient struct {
 	onStatus func(WSStatusChange)
 	onPing   func() map[string]interface{}
 
+	cfg WSClientConfig
+
 	connected atomic.Bool
+
+	// writeCh allows sending messages from outside the connect loop.
+	// It is set in connect() and cleared on disconnect.
+	writeCh chan wsMessage
 }
 
 // NewWSClient creates a new WebSocket client.
 // wsURL is the base WebSocket URL (e.g. "ws://panel.example.com:8076").
 // token and nodeID are used for authentication via query parameters.
-func NewWSClient(wsURL string, token string, nodeID int, onEvent func(WSEvent), onStatus func(WSStatusChange), onPing func() map[string]interface{}) *WSClient {
+func NewWSClient(wsURL string, token string, nodeID int, cfg WSClientConfig, onEvent func(WSEvent), onStatus func(WSStatusChange), onPing func() map[string]interface{}) *WSClient {
+	// Apply defaults
+	if cfg.StatusInterval == 0 {
+		cfg.StatusInterval = 10 * time.Second
+	}
+	if cfg.HandshakeTimeout == 0 {
+		cfg.HandshakeTimeout = 15 * time.Second
+	}
+	if cfg.BackoffInitial == 0 {
+		cfg.BackoffInitial = time.Second
+	}
+	if cfg.BackoffMax == 0 {
+		cfg.BackoffMax = 60 * time.Second
+	}
 	return &WSClient{
 		wsURL:    wsURL,
 		token:    token,
 		nodeID:   nodeID,
+		cfg:      cfg,
 		onEvent:  onEvent,
 		onStatus: onStatus,
 		onPing:   onPing,
@@ -97,7 +151,7 @@ func (w *WSClient) notifyStatus(connected bool) {
 
 // Run connects and reconnects until ctx is cancelled.
 func (w *WSClient) Run(ctx context.Context) {
-	backoff := time.Second
+	backoff := w.cfg.BackoffInitial
 	for {
 		start := time.Now()
 		err := w.connect(ctx)
@@ -108,7 +162,7 @@ func (w *WSClient) Run(ctx context.Context) {
 		}
 
 		if err != nil {
-			slog.Warn("ws disconnected", "error", err)
+			nlog.Core().Warn("ws disconnected", "error", err)
 			if !wasConnected {
 				w.notifyStatus(false)
 			}
@@ -122,7 +176,7 @@ func (w *WSClient) Run(ctx context.Context) {
 
 		// Reset backoff if the connection was up for a meaningful duration.
 		if time.Since(start) > 2*time.Minute {
-			backoff = time.Second
+			backoff = w.cfg.BackoffInitial
 		}
 
 		// Apply exponential backoff with jitter to prevent thundering herd.
@@ -136,27 +190,30 @@ func (w *WSClient) Run(ctx context.Context) {
 			return
 		case <-timer.C:
 		}
-		if backoff < 60*time.Second {
-			backoff = min(backoff*2, 60*time.Second)
+		if backoff < w.cfg.BackoffMax {
+			backoff = min(backoff*2, w.cfg.BackoffMax)
 		}
 	}
 }
 
 func (w *WSClient) connect(ctx context.Context) error {
-	// Build WS URL with auth query params
 	u, err := url.Parse(w.wsURL)
 	if err != nil {
 		return fmt.Errorf("parse ws url: %w", err)
 	}
 	q := u.Query()
 	q.Set("token", w.token)
-	q.Set("node_id", strconv.Itoa(w.nodeID))
+	if w.cfg.MachineID > 0 {
+		q.Set("machine_id", strconv.Itoa(w.cfg.MachineID))
+	} else {
+		q.Set("node_id", strconv.Itoa(w.nodeID))
+	}
 	u.RawQuery = q.Encode()
 
-	slog.Debug("ws connecting", "url", u.Redacted())
+	nlog.Core().Debug("ws connecting", "url", u.String())
 
 	dialer := websocket.Dialer{
-		HandshakeTimeout: 15 * time.Second,
+		HandshakeTimeout: w.cfg.HandshakeTimeout,
 	}
 	conn, _, err := dialer.DialContext(ctx, u.String(), nil)
 	if err != nil {
@@ -171,12 +228,15 @@ func (w *WSClient) connect(ctx context.Context) error {
 	if err := conn.ReadJSON(&firstMsg); err != nil {
 		return fmt.Errorf("read auth response: %w", err)
 	}
+	nlog.Core().Debug("ws recv", "event", firstMsg.Event, "data", string(firstMsg.Data))
 
 	if firstMsg.Event == "error" {
 		var errData struct {
 			Message string `json:"message"`
 		}
-		json.Unmarshal(firstMsg.Data, &errData)
+		if err := json.Unmarshal(firstMsg.Data, &errData); err != nil {
+			return fmt.Errorf("auth failed (unable to parse error: %v)", err)
+		}
 		return fmt.Errorf("auth failed: %s", errData.Message)
 	}
 
@@ -185,19 +245,22 @@ func (w *WSClient) connect(ctx context.Context) error {
 		// Process it and continue
 		w.connected.Store(true)
 		w.notifyStatus(true)
-		slog.Info("ws connected (implicit auth)")
 		w.handleMessage(firstMsg)
 	} else {
 		w.connected.Store(true)
 		w.notifyStatus(true)
-		slog.Info("ws connected and authenticated")
 	}
 
 	// Ping interval: send pong responses to server pings.
-	// We also use this timer to trigger periodic status pushes (10s interval).
-	const reportInterval = 10 * time.Second
+	// We also use this timer to trigger periodic status pushes.
+	reportTicker := time.NewTicker(w.cfg.StatusInterval)
+	defer reportTicker.Stop()
+
+	// writeCh decouples data collection from network I/O.
+	writeCh := make(chan wsMessage, 16)
+	w.writeCh = writeCh
+	defer func() { w.writeCh = nil }()
 
-	msgCh := make(chan wsMessage, 16)
 	errCh := make(chan error, 1)
 	done := make(chan struct{})
 	go func() {
@@ -211,20 +274,18 @@ func (w *WSClient) connect(ctx context.Context) error {
 				}
 				return
 			}
-			select {
-			case msgCh <- msg:
-			case <-ctx.Done():
-				return
+			nlog.Core().Debug("ws recv", "event", msg.Event, "data", string(msg.Data))
+			w.handleMessage(msg)
+			if msg.Event == "ping" {
+				select {
+				case writeCh <- wsMessage{Event: "pong"}:
+				default:
+					nlog.Core().Warn("ws write channel full, skipping pong")
+				}
 			}
 		}
 	}()
 
-	reportTicker := time.NewTicker(reportInterval)
-	defer reportTicker.Stop()
-
-	// writeCh decouples data collection from network I/O.
-	writeCh := make(chan wsMessage, 8)
-
 	for {
 		select {
 		case <-ctx.Done():
@@ -237,17 +298,6 @@ func (w *WSClient) connect(ctx context.Context) error {
 		case err := <-errCh:
 			return fmt.Errorf("read: %w", err)
 
-		case msg := <-msgCh:
-			w.handleMessage(msg)
-			if msg.Event == "ping" {
-				// Queue an immediate pong
-				select {
-				case writeCh <- wsMessage{Event: "pong"}:
-				default:
-					slog.Warn("ws write channel full, skipping pong")
-				}
-			}
-
 		case <-reportTicker.C:
 			// Send periodic node.status via WebSocket
 			if w.onPing != nil {
@@ -260,12 +310,13 @@ func (w *WSClient) connect(ctx context.Context) error {
 				select {
 				case writeCh <- msg:
 				default:
-					slog.Warn("ws write channel full, skipping status push (network slow?)")
+					nlog.Core().Warn("ws write channel full, skipping status push (network slow?)")
 				}
 			}
 
 		case msg := <-writeCh:
 			// Perform the actual network write asynchronously in this loop.
+			nlog.Core().Debug("ws send", "event", msg.Event, "data", string(msg.Data))
 			conn.SetWriteDeadline(time.Now().Add(10 * time.Second))
 			if err := conn.WriteJSON(msg); err != nil {
 				return fmt.Errorf("write: %w", err)
@@ -278,10 +329,10 @@ func (w *WSClient) handleMessage(msg wsMessage) {
 	switch msg.Event {
 	case "ping":
 		// Server ping — handled by the pong timer reset above
-		slog.Debug("ws received ping")
+		nlog.Core().Debug("ws received ping")
 
 	case "auth.success":
-		slog.Debug("ws auth confirmed")
+		nlog.Core().Debug("ws auth confirmed")
 
 	case WSEventSyncConfig:
 		w.handleDataEvent(msg)
@@ -292,8 +343,14 @@ func (w *WSClient) handleMessage(msg wsMessage) {
 	case WSEventSyncUserDelta:
 		w.handleDataEvent(msg)
 
+	case WSEventSyncDevices:
+		w.handleDataEvent(msg)
+
+	case WSEventSyncNodes:
+		w.handleDataEvent(msg)
+
 	default:
-		slog.Debug("ws unknown event", "event", msg.Event)
+		nlog.Core().Debug("ws unknown event", "event", msg.Event)
 	}
 }
 
@@ -312,33 +369,151 @@ func (w *WSClient) handleDataEvent(msg wsMessage) {
 
 	switch msg.Event {
 	case WSEventSyncConfig:
-		slog.Debug("ws sync config event received")
+		nlog.Core().Debug("ws sync config event received")
 		var p syncConfigPayload
-		if err := decodeData(msg.Data, &p); err != nil && p.Config.Protocol != "" {
-			slog.Warn("ws: cannot decode config payload", "error", err)
+		if err := decodeData(msg.Data, &p); err != nil {
+			nlog.Core().Warn("ws: cannot decode config payload", "error", err)
+			return
+		}
+		if p.Config.Protocol == "" {
+			nlog.Core().Warn("ws: config payload missing protocol")
 			return
 		}
 		event.Config = &p.Config
+		if p.NodeID > 0 {
+			event.NodeID = p.NodeID
+		} else if p.Config.NodeID > 0 {
+			event.NodeID = p.Config.NodeID
+		}
 
 	case WSEventSyncUsers:
-		slog.Debug("ws sync users event received")
+		nlog.Core().Debug("ws sync users event received")
 		var p syncUsersPayload
-		if err := decodeData(msg.Data, &p); err != nil && len(p.Users) > 0 {
-			slog.Warn("ws: cannot decode users payload", "error", err)
+		if err := decodeData(msg.Data, &p); err != nil {
+			nlog.Core().Warn("ws: cannot decode users payload", "error", err)
+			return
+		}
+		if len(p.Users) == 0 {
+			nlog.Core().Warn("ws: users payload empty")
 			return
 		}
 		event.Users = p.Users
+		event.NodeID = p.NodeID
 
 	case WSEventSyncUserDelta:
-		slog.Debug("ws sync user delta event received")
+		nlog.Core().Debug("ws sync user delta event received")
 		var p syncUserDeltaPayload
 		if err := decodeData(msg.Data, &p); err != nil {
-			slog.Warn("ws: cannot decode user delta payload", "error", err)
+			nlog.Core().Warn("ws: cannot decode user delta payload", "error", err)
+			return
+		}
+		if p.Action == "" {
+			nlog.Core().Warn("ws: user delta payload missing action")
+			return
+		}
+		if len(p.Users) == 0 {
+			nlog.Core().Warn("ws: user delta payload has no users")
 			return
 		}
 		event.DeltaAction = p.Action
 		event.DeltaUsers = p.Users
+		event.NodeID = p.NodeID
+
+	case WSEventSyncDevices:
+		nlog.Core().Debug("ws sync devices event received")
+		var p syncDevicesPayload
+		if err := decodeData(msg.Data, &p); err != nil {
+			nlog.Core().Warn("ws: cannot decode devices payload", "error", err)
+			return
+		}
+		event.DeviceUsers = p.Users
+		event.NodeID = p.NodeID
+
+	case WSEventSyncNodes:
+		nlog.Core().Info("ws sync nodes event received (machine node list changed)")
+		var p syncNodesPayload
+		if err := decodeData(msg.Data, &p); err != nil {
+			nlog.Core().Warn("ws: cannot decode nodes payload", "error", err)
+			return
+		}
+		event.Nodes = p.Nodes
 	}
 
 	w.onEvent(event)
 }
+
+// SendDeviceReport sends local device snapshot to panel via WS.
+func (w *WSClient) SendDeviceReport(devices map[int][]string) {
+	w.SendDeviceReportForNode(0, devices)
+}
+
+// SendDeviceReportForNode sends a device report tagged with a specific node_id.
+// nodeID == 0 omits the field (legacy single-node mode).
+func (w *WSClient) SendDeviceReportForNode(nodeID int, devices map[int][]string) {
+	if !w.connected.Load() {
+		return
+	}
+
+	var payload interface{}
+	if nodeID > 0 {
+		payload = map[string]interface{}{"node_id": nodeID, "devices": devices}
+	} else {
+		payload = devices
+	}
+
+	d, err := json.Marshal(payload)
+	if err != nil {
+		return
+	}
+	msg := wsMessage{
+		Event:     WSEventReportDevices,
+		Data:      d,
+		Timestamp: time.Now().Unix(),
+	}
+
+	select {
+	case w.writeCh <- msg:
+	default:
+		nlog.Core().Warn("ws write channel full, skipping device report")
+	}
+}
+
+// SendNodeStatus sends a node.status event with the given node_id (machine mode).
+func (w *WSClient) SendNodeStatus(nodeID int, stats map[string]interface{}) {
+	if !w.connected.Load() || stats == nil {
+		return
+	}
+	if nodeID > 0 {
+		stats["node_id"] = nodeID
+	}
+	data, _ := json.Marshal(stats)
+	msg := wsMessage{
+		Event:     "node.status",
+		Data:      data,
+		Timestamp: time.Now().Unix(),
+	}
+	select {
+	case w.writeCh <- msg:
+	default:
+		nlog.Core().Warn("ws write channel full, skipping node status")
+	}
+}
+
+// SendRaw sends a raw message to the panel via WebSocket.
+func (w *WSClient) SendRaw(event string, data json.RawMessage) {
+	if !w.connected.Load() {
+		return
+	}
+
+	msg := wsMessage{
+		Event:     event,
+		Data:      data,
+		Timestamp: time.Now().Unix(),
+	}
+
+	select {
+	case w.writeCh <- msg:
+	default:
+		nlog.Core().Warn("ws write channel full, skipping raw message", "event", event)
+	}
+}

internal/panel/ws_test.go

@@ -86,7 +86,7 @@ func TestWSClient_ConnectAndReceiveDataEvents(t *testing.T) {
 	var mu sync.Mutex
 	var received []WSEvent
 
-	ws := NewWSClient("ws://"+host, "test-token", 1, func(event WSEvent) {
+	ws := NewWSClient("ws://"+host, "test-token", 1, WSClientConfig{}, func(event WSEvent) {
 		mu.Lock()
 		received = append(received, event)
 		mu.Unlock()
@@ -187,7 +187,7 @@ func TestWSClient_ReconnectOnDisconnect(t *testing.T) {
 	defer server.Close()
 
 	host := strings.TrimPrefix(server.URL, "http://")
-	ws := NewWSClient("ws://"+host, "reconnect-token", 1, func(WSEvent) {}, nil, func() map[string]interface{} { return nil })
+	ws := NewWSClient("ws://"+host, "reconnect-token", 1, WSClientConfig{}, func(WSEvent) {}, nil, func() map[string]interface{} { return nil })
 
 	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
 	defer cancel()
@@ -218,7 +218,7 @@ func TestWSClient_ReconnectOnDisconnect(t *testing.T) {
 }
 
 func TestWSClient_FallbackWhenNoServer(t *testing.T) {
-	ws := NewWSClient("ws://127.0.0.1:19999", "fallback-token", 1, func(WSEvent) {}, nil, func() map[string]interface{} { return nil })
+	ws := NewWSClient("ws://127.0.0.1:19999", "fallback-token", 1, WSClientConfig{}, func(WSEvent) {}, nil, func() map[string]interface{} { return nil })
 
 	ctx, cancel := context.WithTimeout(context.Background(), 3*time.Second)
 	defer cancel()
@@ -250,7 +250,7 @@ func TestWSClient_UserDeltaEvent(t *testing.T) {
 	var mu sync.Mutex
 	var received []WSEvent
 
-	ws := NewWSClient("ws://"+host, "delta-token", 1, func(event WSEvent) {
+	ws := NewWSClient("ws://"+host, "delta-token", 1, WSClientConfig{}, func(event WSEvent) {
 		mu.Lock()
 		received = append(received, event)
 		mu.Unlock()

internal/service/service_test.go

@@ -0,0 +1,298 @@
+package service
+
+import (
+	"context"
+	"errors"
+	"strings"
+	"testing"
+
+	"github.com/cedar2025/xboard-node/internal/cert"
+	"github.com/cedar2025/xboard-node/internal/config"
+	"github.com/cedar2025/xboard-node/internal/kernel"
+	"github.com/cedar2025/xboard-node/internal/limiter"
+	"github.com/cedar2025/xboard-node/internal/model"
+	"golang.org/x/time/rate"
+)
+
+type fakeKernel struct {
+	running bool
+
+	startErr  error
+	updateErr error
+	addErr    error
+
+	startCalls  int
+	updateCalls int
+	addCalls    int
+	removeCalls int
+
+	onUpdateUsers func([]model.UserSpec)
+	onAddUsers    func([]model.UserSpec)
+	onRemoveUsers func([]model.UserSpec)
+
+	speedLimitFunc  func(string) *rate.Limiter
+	deviceLimitFunc func(string) (int, bool)
+}
+
+func (f *fakeKernel) Name() string { return "fake" }
+func (f *fakeKernel) Protocols() []string { return []string{"vless"} }
+func (f *fakeKernel) Capabilities() kernel.Capabilities { return kernel.Capabilities{} }
+func (f *fakeKernel) Start(nodeConfig *model.NodeSpec, users []model.UserSpec, tls kernel.TLSCert) error {
+	_, _, _ = nodeConfig, users, tls
+	f.startCalls++
+	if f.startErr != nil {
+		return f.startErr
+	}
+	f.running = true
+	return nil
+}
+func (f *fakeKernel) Stop() { f.running = false }
+func (f *fakeKernel) IsRunning() bool { return f.running }
+func (f *fakeKernel) Reload(nodeConfig *model.NodeSpec, users []model.UserSpec, tls kernel.TLSCert) error {
+	_, _, _ = nodeConfig, users, tls
+	return nil
+}
+func (f *fakeKernel) AddUsers(users []model.UserSpec) (int, error) {
+	f.addCalls++
+	if f.onAddUsers != nil {
+		f.onAddUsers(users)
+	}
+	if f.addErr != nil {
+		return 0, f.addErr
+	}
+	return len(users), nil
+}
+func (f *fakeKernel) RemoveUsers(users []model.UserSpec) (int, error) {
+	f.removeCalls++
+	if f.onRemoveUsers != nil {
+		f.onRemoveUsers(users)
+	}
+	return len(users), nil
+}
+func (f *fakeKernel) UpdateUsers(users []model.UserSpec) (int, int, error) {
+	f.updateCalls++
+	if f.onUpdateUsers != nil {
+		f.onUpdateUsers(users)
+	}
+	if f.updateErr != nil {
+		return 0, 0, f.updateErr
+	}
+	return len(users), 0, nil
+}
+func (f *fakeKernel) GetUserTraffic(ctx context.Context) (map[int][2]int64, map[int]map[string]bool, int, error) {
+	_ = ctx
+	return nil, nil, 0, nil
+}
+func (f *fakeKernel) CloseConnection(ctx context.Context, connID string) error {
+	_, _ = ctx, connID
+	return nil
+}
+func (f *fakeKernel) CloseUserConnections(ctx context.Context, uuid string) error {
+	_, _ = ctx, uuid
+	return nil
+}
+func (f *fakeKernel) SetSpeedLimitFunc(fn func(uuid string) *rate.Limiter) { f.speedLimitFunc = fn }
+func (f *fakeKernel) SetDeviceLimitFunc(fn func(uuid string) (int, bool)) { f.deviceLimitFunc = fn }
+func (f *fakeKernel) UpdateGlobalDevices(users map[int][]string) { _ = users }
+func (f *fakeKernel) ClearGlobalDevices() {}
+
+func newTestService(k *fakeKernel) *Service {
+	sharedLimiter := limiter.New()
+	s := &Service{
+		kernel:       k,
+		limiter:      sharedLimiter,
+		speedTracker: limiter.NewSpeedTracker(sharedLimiter),
+		cert:         cert.NewManager(config.CertConfig{}),
+	}
+	k.SetSpeedLimitFunc(s.speedTracker.GetLimiter)
+	k.SetDeviceLimitFunc(s.limiter.GetDeviceLimitByUUID)
+	return s
+}
+
+func TestApplyUserUpdatePreparesLimiterBeforeKernelUpdate(t *testing.T) {
+	k := &fakeKernel{running: true}
+	s := newTestService(k)
+	s.lastConfig = &model.NodeSpec{Protocol: "vless"}
+	oldUsers := []model.UserSpec{{ID: 1, UUID: "uuid-old", SpeedLimit: 4}}
+	s.updateUserState(oldUsers)
+
+	newUsers := []model.UserSpec{{ID: 2, UUID: "uuid-new", SpeedLimit: 8}}
+	k.onUpdateUsers = func(users []model.UserSpec) {
+		if len(users) != 1 || users[0].UUID != "uuid-new" {
+			t.Fatalf("unexpected users passed to UpdateUsers: %#v", users)
+		}
+		if got := k.speedLimitFunc("uuid-new"); got == nil {
+			t.Fatal("expected new user's limiter to be visible before kernel UpdateUsers")
+		}
+	}
+
+	s.applyUserUpdate(context.Background(), newUsers, computeUserHash(newUsers))
+
+	if got := k.updateCalls; got != 1 {
+		t.Fatalf("UpdateUsers call count = %d, want 1", got)
+	}
+	if len(s.lastUsers) != 1 || s.lastUsers[0].UUID != "uuid-new" {
+		t.Fatalf("lastUsers = %#v, want new users", s.lastUsers)
+	}
+	if s.speedTracker.GetLimiter("uuid-new") == nil {
+		t.Fatal("expected limiter for new user after successful update")
+	}
+}
+
+func TestApplyUserUpdateRestoresStateWhenKernelAndRestartFail(t *testing.T) {
+	k := &fakeKernel{
+		running:   true,
+		updateErr: errors.New("update failed"),
+		startErr:  errors.New("restart failed"),
+	}
+	s := newTestService(k)
+	s.lastConfig = &model.NodeSpec{Protocol: "vless"}
+	oldUsers := []model.UserSpec{{ID: 1, UUID: "uuid-old", SpeedLimit: 4}}
+	s.updateUserState(oldUsers)
+	oldHash := s.lastUserHash
+
+	newUsers := []model.UserSpec{{ID: 2, UUID: "uuid-new", SpeedLimit: 8}}
+	s.applyUserUpdate(context.Background(), newUsers, computeUserHash(newUsers))
+
+	if got := k.startCalls; got != 1 {
+		t.Fatalf("Start call count = %d, want 1", got)
+	}
+	if len(s.lastUsers) != 1 || s.lastUsers[0].UUID != "uuid-old" {
+		t.Fatalf("lastUsers = %#v, want restored old users", s.lastUsers)
+	}
+	if s.lastUserHash != oldHash {
+		t.Fatalf("lastUserHash = %q, want %q", s.lastUserHash, oldHash)
+	}
+	if s.speedTracker.GetLimiter("uuid-old") == nil {
+		t.Fatal("expected old limiter to be restored after rollback")
+	}
+	if s.speedTracker.GetLimiter("uuid-new") != nil {
+		t.Fatal("expected new limiter to be removed after rollback")
+	}
+}
+
+func TestApplyUserDeltaAddPreparesLimiterBeforeKernelUpdate(t *testing.T) {
+	k := &fakeKernel{running: true}
+	s := newTestService(k)
+	s.lastConfig = &model.NodeSpec{Protocol: "vless"}
+	oldUsers := []model.UserSpec{{ID: 1, UUID: "uuid-old", SpeedLimit: 4}}
+	s.updateUserState(oldUsers)
+
+	delta := []model.UserSpec{{ID: 2, UUID: "uuid-new", SpeedLimit: 8}}
+	k.onAddUsers = func(users []model.UserSpec) {
+		if len(users) != 1 || users[0].UUID != "uuid-new" {
+			t.Fatalf("unexpected users passed to AddUsers: %#v", users)
+		}
+		if got := k.speedLimitFunc("uuid-new"); got == nil {
+			t.Fatal("expected delta user's limiter to be visible before kernel AddUsers")
+		}
+	}
+
+	s.applyUserDelta(context.Background(), "add", delta)
+
+	if got := k.addCalls; got != 1 {
+		t.Fatalf("AddUsers call count = %d, want 1", got)
+	}
+	if s.speedTracker.GetLimiter("uuid-new") == nil {
+		t.Fatal("expected limiter for delta-added user after successful update")
+	}
+}
+
+
+func TestValidateNodeRuntimeRejectsUnsupportedDNSProvider(t *testing.T) {
+	cfg := &config.Config{Kernel: config.KernelConfig{Type: "singbox"}}
+	err := validateNodeRuntime(cfg, []string{"http"}, &model.NodeSpec{
+		Protocol: "http",
+		CertConfig: &config.CertConfig{
+			CertMode:    "dns",
+			DNSProvider: "3123123",
+			Domain:      "example.com",
+		},
+	}, kernel.TLSCert{CertPEM: []byte("CERT"), KeyPEM: []byte("KEY")})
+	if err == nil {
+		t.Fatal("expected error, got nil")
+	}
+	if err.Error() == "" {
+		t.Fatal("expected non-empty error")
+	}
+	if got := err.Error(); !strings.HasPrefix(got, `unsupported cert_config.dns_provider "3123123" (supported: `) {
+		t.Fatalf("unexpected error: %v", got)
+	}
+}
+
+func TestValidateNodeRuntimeAllowsSelfManagedTLSBeforeFilesExist(t *testing.T) {
+	cfg := &config.Config{Kernel: config.KernelConfig{Type: "singbox"}}
+	err := validateNodeRuntime(cfg, []string{"anytls", "hysteria"}, &model.NodeSpec{
+		Protocol: "anytls",
+		CertConfig: &config.CertConfig{
+			CertMode: "self",
+			Domain:   "example.com",
+		},
+	}, kernel.TLSCert{})
+	if err != nil {
+		t.Fatalf("expected self-managed TLS config to pass validation, got %v", err)
+	}
+}
+
+func TestValidateNodeRuntimeAllowsSingboxRealityWithRequiredFields(t *testing.T) {
+	cfg := &config.Config{Kernel: config.KernelConfig{Type: "singbox"}}
+	err := validateNodeRuntime(cfg, []string{"vless"}, &model.NodeSpec{
+		Protocol: "vless",
+		TLS:      2,
+		TLSSettings: map[string]any{
+			"private_key": "test-key",
+			"server_name": "example.com",
+		},
+	}, kernel.TLSCert{CertPEM: []byte("CERT"), KeyPEM: []byte("KEY")})
+	if err != nil {
+		t.Fatalf("expected sing-box reality validation to pass, got %v", err)
+	}
+}
+
+func TestValidateNodeRuntimeRejectsRealityWithoutTLSSettings(t *testing.T) {
+	cfg := &config.Config{Kernel: config.KernelConfig{Type: "singbox"}}
+	err := validateNodeRuntime(cfg, []string{"vless"}, &model.NodeSpec{
+		Protocol: "vless",
+		TLS:      2,
+	}, kernel.TLSCert{CertPEM: []byte("CERT"), KeyPEM: []byte("KEY")})
+	if err == nil {
+		t.Fatal("expected error, got nil")
+	}
+	if got := err.Error(); got != "reality tls requires tls_settings" {
+		t.Fatalf("unexpected error: %v", got)
+	}
+}
+
+func TestValidateNodeRuntimeRejectsRealityWithoutPrivateKey(t *testing.T) {
+	cfg := &config.Config{Kernel: config.KernelConfig{Type: "singbox"}}
+	err := validateNodeRuntime(cfg, []string{"vless"}, &model.NodeSpec{
+		Protocol: "vless",
+		TLS:      2,
+		TLSSettings: map[string]any{
+			"server_name": "example.com",
+		},
+	}, kernel.TLSCert{CertPEM: []byte("CERT"), KeyPEM: []byte("KEY")})
+	if err == nil {
+		t.Fatal("expected error, got nil")
+	}
+	if got := err.Error(); got != "reality tls requires tls_settings.private_key" {
+		t.Fatalf("unexpected error: %v", got)
+	}
+}
+
+func TestValidateNodeRuntimeRejectsRealityWithoutServerNameOrDest(t *testing.T) {
+	cfg := &config.Config{Kernel: config.KernelConfig{Type: "singbox"}}
+	err := validateNodeRuntime(cfg, []string{"vless"}, &model.NodeSpec{
+		Protocol: "vless",
+		TLS:      2,
+		TLSSettings: map[string]any{
+			"private_key": "test-key",
+		},
+	}, kernel.TLSCert{CertPEM: []byte("CERT"), KeyPEM: []byte("KEY")})
+	if err == nil {
+		t.Fatal("expected error, got nil")
+	}
+	if got := err.Error(); got != "reality tls requires tls_settings.server_name or tls_settings.dest" {
+		t.Fatalf("unexpected error: %v", got)
+	}
+}

internal/tracker/tracker.go

@@ -1,275 +1,309 @@
 package tracker
 
 import (
-	"log/slog"
+	"crypto/sha256"
+	"encoding/hex"
+	"sort"
+	"sync"
+	"sync/atomic"
 
-	"github.com/cedar2025/xboard-node/internal/kernel"
+	"github.com/cedar2025/xboard-node/internal/nlog"
 )
 
-const (
-	// DefaultMaxConnState caps how many connection-state entries we maintain.
-	// Prevents unbounded memory growth from rogue / DDoS connection floods.
-	DefaultMaxConnState = 100_000
+// snapshot is an immutable point-in-time view of tracker state.
+// It is swapped atomically so readers never block writers.
+type snapshot struct {
+	traffic   map[int][2]int64        // userID → [upload, download] delta
+	aliveIPs  map[int]map[string]bool // userID → set of source IPs
+	online    map[int]int             // userID → distinct IP count
+	connCount int
+	inSpeed   int64
+	outSpeed  int64
+}
 
-	// DefaultMaxAliveIPsPerUser caps the per-user IP set tracked per push interval.
-	// 200 IPs per user is generous for any legitimate use case.
-	DefaultMaxAliveIPsPerUser = 200
-)
-
-// Tracker tracks per-connection traffic deltas and accumulates per-user totals.
+// Tracker computes per-user traffic deltas from cumulative counters
+// provided by the kernel, and accumulates totals for panel reporting.
 //
-// Thread safety model: Tracker is designed for SINGLE-GOROUTINE use only.
-// All methods (Process, Flush*, Restore*, CurrentOnline, speed accessors) are
-// called exclusively from the service goroutine's select loop. No internal
-// locking is provided. Do NOT call Tracker methods from multiple goroutines.
+// Architecture: the kernel maintains per-user atomic counters and IP sets.
+// Each tick, the service calls Process() with cumulative per-user traffic.
+// Tracker computes deltas against the previous cycle's values — O(users),
+// not O(connections).
 //
-// Hot-path optimisation (②): activeConns is allocated once and cleared in-place
-// every Process() call instead of being reallocated, saving GC pressure at high
-// connection counts.
-//
-// OOM protection (④): connState and aliveIPs are capped at configurable maximums.
+// Thread safety: Process() acquires mu to update internal state, then
+// atomically publishes a new snapshot. All read methods (Flush*,
+// CurrentOnline, LogStats, *Speed) read the snapshot lock-free.
+// This eliminates contention between the 10s Process tick and the 60s
+// flush/push tick.
 type Tracker struct {
-	connState map[string][2]int64     // conn_id → last [upload, download]
-	traffic   map[int][2]int64        // user_id → accumulated [upload, download]
-	aliveIPs  map[int]map[string]bool // user_id → set of source IPs
-	online    map[int]int             // user_id → current active connection count
+	mu sync.Mutex
 
-	// ② Reused buffer: cleared each Process() call, not reallocated.
-	activeConns map[string]struct{}
+	// lastSeen stores the cumulative traffic from the previous Process() call.
+	// Protected by mu — only written by Process.
+	lastSeen map[int][2]int64 // userID → [upload, download] cumulative
 
-	// ④ Limits (set at construction, never mutated).
-	maxConnState       int
-	maxAliveIPsPerUser int
+	// pending accumulates deltas between flushes.
+	// Protected by mu — written by Process, drained by FlushTraffic.
+	pendingTraffic map[int][2]int64
 
-	// lastActiveConns stores the last observed number of active connections.
-	// This is updated on each Process() call and read by the service when
-	// building node metrics for panel reporting.
-	lastActiveConns int
-	totalConns      int64 // cumulative number of connections seen since startup
+	// live holds the current snapshot, swapped atomically.
+	// Readers load this pointer without any lock.
+	live atomic.Pointer[snapshot]
 
-	// lastUserCount is used to pre-size intervalTraffic each cycle, avoiding
-	// repeated map rehashing under heavy user counts.
-	lastUserCount int
+	// aliveIPsBuf is a reusable buffer for FlushAliveIPs output.
+	// Avoids allocating a new map+slice every 60s.
+	aliveIPsBuf map[int][]string
 
-	inSpeed  int64
-	outSpeed int64
+	// lastAliveIPsHash detects changes to avoid duplicate reports.
+	lastAliveIPsHash string
 }
 
 func New() *Tracker {
-	return NewWithLimits(DefaultMaxConnState, DefaultMaxAliveIPsPerUser)
+	t := &Tracker{
+		lastSeen:       make(map[int][2]int64),
+		pendingTraffic: make(map[int][2]int64),
+		aliveIPsBuf:    make(map[int][]string),
+	}
+	// Publish initial empty snapshot.
+	t.live.Store(&snapshot{
+		traffic:  make(map[int][2]int64),
+		aliveIPs: make(map[int]map[string]bool),
+		online:   make(map[int]int),
+	})
+	return t
 }
 
-func NewWithLimits(maxConnState, maxAliveIPsPerUser int) *Tracker {
-	return &Tracker{
-		connState:          make(map[string][2]int64),
-		traffic:            make(map[int][2]int64),
-		aliveIPs:           make(map[int]map[string]bool),
-		online:             make(map[int]int),
-		activeConns:        make(map[string]struct{}),
-		maxConnState:       maxConnState,
-		maxAliveIPsPerUser: maxAliveIPsPerUser,
-	}
-}
+// Process computes per-user traffic deltas from cumulative kernel counters.
+// Also stores alive IPs and connection count. O(users).
+//
+// This is the only writer to lastSeen and pendingTraffic.
+// After computing deltas, it publishes a new snapshot for lock-free reads.
+func (t *Tracker) Process(
+	cumTraffic map[int][2]int64,
+	kernelAliveIPs map[int]map[string]bool,
+	connCount int,
+) {
+	t.mu.Lock()
+	defer t.mu.Unlock()
 
-// Process calculates per-user traffic deltas for this interval.
-// Internally accumulates totals for panel push and tracks alive IPs.
-// Returns per-user interval traffic [upload, download] for the limiter.
-func (t *Tracker) Process(conns []kernel.Connection) map[int][2]int64 {
-	// Pre-size using last cycle's user count to avoid rehashing under heavy load.
-	intervalTraffic := make(map[int][2]int64, t.lastUserCount)
+	var cycleIn, cycleOut int64
 
-	// Clear activeConns in-place (no alloc, backing array retained).
-	for k := range t.activeConns {
-		delete(t.activeConns, k)
-	}
+	for uid, cum := range cumTraffic {
+		prev := t.lastSeen[uid]
+		deltaUp := cum[0] - prev[0]
+		deltaDown := cum[1] - prev[1]
 
-	// Rebuild aliveIPs each cycle so only currently-active IPs are reported.
-	// Previously, IPs accumulated across cycles leading to stale entries.
-	for uid := range t.aliveIPs {
-		for ip := range t.aliveIPs[uid] {
-			delete(t.aliveIPs[uid], ip)
+		// Guard against counter reset (kernel restart).
+		if deltaUp < 0 {
+			deltaUp = cum[0]
 		}
-	}
-
-	// Clear online counts
-	for uid := range t.online {
-		delete(t.online, uid)
-	}
-
-	for _, conn := range conns {
-		t.activeConns[conn.ID] = struct{}{}
-
-		if conn.UserID == 0 {
-			continue
+		if deltaDown < 0 {
+			deltaDown = cum[1]
 		}
 
-		t.online[conn.UserID]++
-
-		var deltaUp, deltaDown int64
-		prev, exists := t.connState[conn.ID]
-		if exists {
-			deltaUp = conn.Upload - prev[0]
-			deltaDown = conn.Download - prev[1]
-			if deltaUp < 0 {
-				deltaUp = conn.Upload
-			}
-			if deltaDown < 0 {
-				deltaDown = conn.Download
-			}
-		} else {
-			// ④ New connection: only track if below the state cap.
-			if len(t.connState) >= t.maxConnState {
-				slog.Debug("connState cap reached, skipping new connection",
-					"cap", t.maxConnState, "conn_id", conn.ID)
-				continue
-			}
-			deltaUp = conn.Upload
-			deltaDown = conn.Download
-			t.totalConns++
-		}
-		t.connState[conn.ID] = [2]int64{conn.Upload, conn.Download}
+		t.lastSeen[uid] = cum
 
 		if deltaUp > 0 || deltaDown > 0 {
-			cur := t.traffic[conn.UserID]
+			cur := t.pendingTraffic[uid]
 			cur[0] += deltaUp
 			cur[1] += deltaDown
-			t.traffic[conn.UserID] = cur
+			t.pendingTraffic[uid] = cur
 
-			it := intervalTraffic[conn.UserID]
-			it[0] += deltaUp
-			it[1] += deltaDown
-			intervalTraffic[conn.UserID] = it
-		}
-
-		// ④ Track alive IPs, capped per user.
-		if conn.SourceIP != "" {
-			ips := t.aliveIPs[conn.UserID]
-			if ips == nil {
-				ips = make(map[string]bool)
-				t.aliveIPs[conn.UserID] = ips
-			}
-			if len(ips) < t.maxAliveIPsPerUser {
-				ips[conn.SourceIP] = true
-			}
+			cycleOut += deltaUp
+			cycleIn += deltaDown
 		}
 	}
 
-	// Cleanup stale connState entries for connections that are no longer active.
-	for id := range t.connState {
-		if _, active := t.activeConns[id]; !active {
-			delete(t.connState, id)
-		}
+	// Compute online from alive IPs.
+	online := make(map[int]int, len(kernelAliveIPs))
+	for uid, ips := range kernelAliveIPs {
+		online[uid] = len(ips)
 	}
 
-	// Snapshot current active connection count for metrics reporting.
-	t.lastActiveConns = len(t.connState)
-
-	// Derive speed directly from intervalTraffic deltas accumulated this cycle.
-	// This avoids the negative-diff bug that occurs when connState entries for
-	// closed connections are removed, which made cumulative totals drop and
-	// clamped speed to 0 for one cycle after a mass disconnect.
-	var cycleIn, cycleOut int64
-	for _, it := range intervalTraffic {
-		cycleOut += it[0]
-		cycleIn += it[1]
-	}
-	t.inSpeed = cycleIn
-	t.outSpeed = cycleOut
-
-	// Update hint for next cycle's map pre-sizing.
-	if n := len(intervalTraffic); n > 0 {
-		t.lastUserCount = n
-	}
-
-	return intervalTraffic
+	// Publish new snapshot (readers will see this atomically).
+	t.live.Store(&snapshot{
+		traffic:   copyTrafficMap(t.pendingTraffic),
+		aliveIPs:  kernelAliveIPs, // kernel provides fresh copy each tick
+		online:    online,
+		connCount: connCount,
+		inSpeed:   cycleIn,
+		outSpeed:  cycleOut,
+	})
 }
 
-// FlushTraffic returns accumulated per-user traffic and resets the counter.
+// FlushTraffic returns accumulated per-user traffic and resets the pending buffer.
+// Lock-free: reads from live snapshot, then acquires mu only to drain.
 func (t *Tracker) FlushTraffic() map[int][2]int64 {
-	data := t.traffic
-	t.traffic = make(map[int][2]int64, len(data))
+	t.mu.Lock()
+	data := t.pendingTraffic
+	t.pendingTraffic = make(map[int][2]int64, len(data))
+	t.mu.Unlock()
 	return data
 }
 
 // RestoreTraffic adds traffic back (used when push to panel fails).
 func (t *Tracker) RestoreTraffic(data map[int][2]int64) {
+	t.mu.Lock()
 	for uid, d := range data {
-		cur := t.traffic[uid]
+		cur := t.pendingTraffic[uid]
 		cur[0] += d[0]
 		cur[1] += d[1]
-		t.traffic[uid] = cur
+		t.pendingTraffic[uid] = cur
 	}
+	t.mu.Unlock()
 }
 
 // HasTraffic returns true if there is accumulated traffic to report.
 func (t *Tracker) HasTraffic() bool {
-	return len(t.traffic) > 0
+	t.mu.Lock()
+	defer t.mu.Unlock()
+	return len(t.pendingTraffic) > 0
 }
 
-// FlushAliveIPs returns per-user alive IPs and resets the tracker.
+// FlushAliveIPs returns per-user alive IPs.
+// Reuses internal buffer. Returns nil if unchanged.
 func (t *Tracker) FlushAliveIPs() map[int][]string {
-	data := make(map[int][]string, len(t.aliveIPs))
-	for uid, ips := range t.aliveIPs {
+	s := t.live.Load()
+
+	t.mu.Lock()
+	defer t.mu.Unlock()
+
+	// Calculate hash of current aliveIPs
+	currentHash := calcAliveIPsHash(s.aliveIPs)
+
+	// If no changes, return nil to avoid duplicate reporting
+	if currentHash == t.lastAliveIPsHash {
+		return nil
+	}
+
+	t.lastAliveIPsHash = currentHash
+
+	// Clear old buffer entries.
+	for k := range t.aliveIPsBuf {
+		delete(t.aliveIPsBuf, k)
+	}
+
+	// Fill buffer from snapshot.
+	for uid, ips := range s.aliveIPs {
+		buf := t.aliveIPsBuf[uid]
+		if buf == nil {
+			buf = make([]string, 0, len(ips))
+		}
+		buf = buf[:0]
+		for ip := range ips {
+			buf = append(buf, ip)
+		}
+		t.aliveIPsBuf[uid] = buf
+	}
+
+	return t.aliveIPsBuf
+}
+
+// calcAliveIPsHash computes a deterministic hash for change detection.
+func calcAliveIPsHash(aliveIPs map[int]map[string]bool) string {
+	if len(aliveIPs) == 0 {
+		return ""
+	}
+
+	h := sha256.New()
+
+	// Sort user IDs for consistent hashing
+	userIDs := make([]int, 0, len(aliveIPs))
+	for uid := range aliveIPs {
+		userIDs = append(userIDs, uid)
+	}
+	sort.Ints(userIDs)
+
+	for _, uid := range userIDs {
+		ips := aliveIPs[uid]
+		// Sort IPs for consistent hashing
 		ipList := make([]string, 0, len(ips))
 		for ip := range ips {
 			ipList = append(ipList, ip)
 		}
-		data[uid] = ipList
+		sort.Strings(ipList)
+
+		// Write user ID and IPs to hash
+		h.Write([]byte{byte(uid >> 24), byte(uid >> 16), byte(uid >> 8), byte(uid)})
+		for _, ip := range ipList {
+			h.Write([]byte(ip))
+		}
 	}
-	t.aliveIPs = make(map[int]map[string]bool, len(t.aliveIPs))
-	return data
+
+	return hex.EncodeToString(h.Sum(nil))
 }
 
-// CurrentOnline returns a map of user_id to active connection count.
+// CurrentOnline returns a snapshot copy of user_id → device count (distinct IPs).
+// Lock-free: reads from live snapshot.
 func (t *Tracker) CurrentOnline() map[int]int {
-	return t.online
+	s := t.live.Load()
+	cp := make(map[int]int, len(s.online))
+	for k, v := range s.online {
+		cp[k] = v
+	}
+	return cp
 }
 
 // RestoreAliveIPs merges alive IPs back in (used when push to panel fails).
+// Note: this operates on the buffer, which will be overwritten next Process().
 func (t *Tracker) RestoreAliveIPs(data map[int][]string) {
+	t.mu.Lock()
+	defer t.mu.Unlock()
 	for uid, ipList := range data {
-		ips := t.aliveIPs[uid]
+		ips := t.aliveIPsBuf[uid]
 		if ips == nil {
-			ips = make(map[string]bool, len(ipList))
-			t.aliveIPs[uid] = ips
+			ips = make([]string, 0, len(ipList))
+		}
+		// Use map for O(n) dedup instead of O(n²) linear search
+		existMap := make(map[string]struct{}, len(ips)+len(ipList))
+		for _, existing := range ips {
+			existMap[existing] = struct{}{}
 		}
 		for _, ip := range ipList {
-			if len(ips) < t.maxAliveIPsPerUser {
-				ips[ip] = true
+			if _, exists := existMap[ip]; !exists {
+				ips = append(ips, ip)
+				existMap[ip] = struct{}{}
 			}
 		}
+		t.aliveIPsBuf[uid] = ips
 	}
 }
 
 // LogStats logs current tracking statistics.
+// Lock-free: reads from live snapshot.
 func (t *Tracker) LogStats() {
-	slog.Debug("tracker stats",
-		"active_connections", len(t.connState),
-		"users_with_traffic", len(t.traffic),
-		"users_with_ips", len(t.aliveIPs),
-	)
+	s := t.live.Load()
+	nlog.TrackerStats(s.connCount, len(s.online))
 }
 
 // ActiveConnections returns the last observed active connection count.
-// This is cheap and safe because Tracker is only mutated from the service
-// goroutine; reads happen from the same goroutine when pushing reports.
+// Lock-free: reads from live snapshot.
 func (t *Tracker) ActiveConnections() int {
-	return t.lastActiveConns
+	return t.live.Load().connCount
 }
 
-// TotalConnections returns the cumulative connection count since startup.
+// TotalConnections is deprecated — no longer tracked per-connection.
+// Returns 0 for backward compatibility.
 func (t *Tracker) TotalConnections() int64 {
-	return t.totalConns
+	return 0
 }
 
 // InboundSpeed returns the last observed inbound (download) speed in bytes/second.
-// Process() is called every 10 seconds, so we divide the 10-second total by 10.
+// Lock-free: reads from live snapshot.
 func (t *Tracker) InboundSpeed() int64 {
-	return t.inSpeed / 10
+	return t.live.Load().inSpeed / 10
 }
 
 // OutboundSpeed returns the last observed outbound (upload) speed in bytes/second.
-// Process() is called every 10 seconds, so we divide the 10-second total by 10.
+// Lock-free: reads from live snapshot.
 func (t *Tracker) OutboundSpeed() int64 {
-	return t.outSpeed / 10
+	return t.live.Load().outSpeed / 10
+}
+
+// copyTrafficMap creates a shallow copy of the traffic map.
+func copyTrafficMap(src map[int][2]int64) map[int][2]int64 {
+	dst := make(map[int][2]int64, len(src))
+	for k, v := range src {
+		dst[k] = v
+	}
+	return dst
 }

internal/tracker/tracker_test.go

@@ -2,24 +2,26 @@ package tracker
 
 import (
 	"testing"
-
-	"github.com/cedar2025/xboard-node/internal/kernel"
 )
 
 func TestProcess_InitialTraffic(t *testing.T) {
 	tr := New()
-	conns := []kernel.Connection{
-		{ID: "c1", UserID: 1, Upload: 100, Download: 200, SourceIP: "1.1.1.1"},
-		{ID: "c2", UserID: 2, Upload: 50, Download: 80, SourceIP: "2.2.2.2"},
+	cumTraffic := map[int][2]int64{
+		1: {100, 200},
+		2: {50, 80},
 	}
-
-	interval := tr.Process(conns)
-
-	if interval[1] != [2]int64{100, 200} {
-		t.Errorf("user 1 interval: got %v", interval[1])
+	aliveIPs := map[int]map[string]bool{
+		1: {"1.1.1.1": true},
+		2: {"2.2.2.2": true},
 	}
-	if interval[2] != [2]int64{50, 80} {
-		t.Errorf("user 2 interval: got %v", interval[2])
+	tr.Process(cumTraffic, aliveIPs, 2)
+
+	flushed := tr.FlushTraffic()
+	if flushed[1] != [2]int64{100, 200} {
+		t.Errorf("user 1 traffic: got %v", flushed[1])
+	}
+	if flushed[2] != [2]int64{50, 80} {
+		t.Errorf("user 2 traffic: got %v", flushed[2])
 	}
 }
 
@@ -27,18 +29,15 @@ func TestProcess_DeltaCalculation(t *testing.T) {
 	tr := New()
 
 	// First tick
-	conns := []kernel.Connection{
-		{ID: "c1", UserID: 1, Upload: 100, Download: 200},
-	}
-	tr.Process(conns)
+	tr.Process(map[int][2]int64{1: {100, 200}}, nil, 1)
 
 	// Second tick — counters advanced
-	conns[0].Upload = 300
-	conns[0].Download = 500
-	interval := tr.Process(conns)
+	tr.Process(map[int][2]int64{1: {300, 500}}, nil, 1)
 
-	if interval[1] != [2]int64{200, 300} {
-		t.Errorf("delta: got %v, want [200,300]", interval[1])
+	flushed := tr.FlushTraffic()
+	// Total: first tick 100/200 + delta 200/300 = 300/500
+	if flushed[1] != [2]int64{300, 500} {
+		t.Errorf("accumulated: got %v, want [300,500]", flushed[1])
 	}
 }
 
@@ -46,73 +45,52 @@ func TestProcess_CounterReset(t *testing.T) {
 	tr := New()
 
 	// First tick with high counters
-	conns := []kernel.Connection{
-		{ID: "c1", UserID: 1, Upload: 1000, Download: 2000},
-	}
-	tr.Process(conns)
+	tr.Process(map[int][2]int64{1: {1000, 2000}}, nil, 1)
+	tr.FlushTraffic() // clear accumulator
 
 	// Counter reset (new value < old value) — treat as fresh
-	conns[0].Upload = 50
-	conns[0].Download = 80
-	interval := tr.Process(conns)
+	tr.Process(map[int][2]int64{1: {50, 80}}, nil, 1)
 
-	if interval[1] != [2]int64{50, 80} {
-		t.Errorf("counter reset: got %v, want [50,80]", interval[1])
+	flushed := tr.FlushTraffic()
+	if flushed[1] != [2]int64{50, 80} {
+		t.Errorf("counter reset: got %v, want [50,80]", flushed[1])
 	}
 }
 
-func TestProcess_ConnectionCleanup(t *testing.T) {
+func TestProcess_UserDisappears(t *testing.T) {
 	tr := New()
 
-	// Tick 1: connection exists
-	tr.Process([]kernel.Connection{
-		{ID: "c1", UserID: 1, Upload: 100, Download: 200},
-	})
+	// Tick 1: user present
+	tr.Process(map[int][2]int64{1: {100, 200}}, nil, 1)
 
-	// Tick 2: connection gone — should clean up state
-	tr.Process([]kernel.Connection{})
+	// Tick 2: user gone
+	tr.Process(map[int][2]int64{}, nil, 0)
 
-	// Tick 3: same conn ID reappears — should count full bytes (not delta)
-	interval := tr.Process([]kernel.Connection{
-		{ID: "c1", UserID: 1, Upload: 50, Download: 30},
-	})
+	// Tick 3: user reappears — should count full bytes (treated as new)
+	tr.Process(map[int][2]int64{1: {50, 30}}, nil, 1)
 
-	if interval[1] != [2]int64{50, 30} {
-		t.Errorf("after cleanup: got %v, want [50,30]", interval[1])
+	flushed := tr.FlushTraffic()
+	// First tick: 100/200, tick 2: 0, tick 3: 50/30 = total 150/230
+	if flushed[1] != [2]int64{150, 230} {
+		t.Errorf("after disappear: got %v, want [150,230]", flushed[1])
 	}
 }
 
-func TestProcess_ZeroUserID(t *testing.T) {
+func TestProcess_ZeroTraffic(t *testing.T) {
 	tr := New()
-	interval := tr.Process([]kernel.Connection{
-		{ID: "c1", UserID: 0, Upload: 100, Download: 200},
-	})
+	tr.Process(map[int][2]int64{}, nil, 0)
 
-	if len(interval) != 0 {
-		t.Errorf("expected no interval traffic for userID=0, got %v", interval)
-	}
-}
-
-func TestProcess_MultipleConnectionsSameUser(t *testing.T) {
-	tr := New()
-	conns := []kernel.Connection{
-		{ID: "c1", UserID: 1, Upload: 100, Download: 200},
-		{ID: "c2", UserID: 1, Upload: 50, Download: 80},
-	}
-
-	interval := tr.Process(conns)
-
-	if interval[1] != [2]int64{150, 280} {
-		t.Errorf("aggregated: got %v, want [150,280]", interval[1])
+	if tr.HasTraffic() {
+		t.Error("expected no traffic for empty input")
 	}
 }
 
 func TestFlushTraffic(t *testing.T) {
 	tr := New()
-	tr.Process([]kernel.Connection{
-		{ID: "c1", UserID: 1, Upload: 100, Download: 200},
-		{ID: "c2", UserID: 2, Upload: 50, Download: 80},
-	})
+	tr.Process(map[int][2]int64{
+		1: {100, 200},
+		2: {50, 80},
+	}, nil, 2)
 
 	flushed := tr.FlushTraffic()
 	if flushed[1] != [2]int64{100, 200} {
@@ -134,15 +112,9 @@ func TestFlushTraffic(t *testing.T) {
 
 func TestRestoreTraffic(t *testing.T) {
 	tr := New()
-
-	// Generate some traffic
-	tr.Process([]kernel.Connection{
-		{ID: "c1", UserID: 1, Upload: 100, Download: 200},
-	})
+	tr.Process(map[int][2]int64{1: {100, 200}}, nil, 1)
 
 	flushed := tr.FlushTraffic()
-
-	// Restore (simulating failed push)
 	tr.RestoreTraffic(flushed)
 
 	if !tr.HasTraffic() {
@@ -158,27 +130,19 @@ func TestRestoreTraffic(t *testing.T) {
 func TestRestoreTraffic_Additive(t *testing.T) {
 	tr := New()
 
-	// Generate traffic
-	tr.Process([]kernel.Connection{
-		{ID: "c1", UserID: 1, Upload: 100, Download: 200},
-	})
-
-	// Generate more traffic
-	tr.Process([]kernel.Connection{
-		{ID: "c1", UserID: 1, Upload: 200, Download: 400},
-	})
+	// Two ticks accumulate
+	tr.Process(map[int][2]int64{1: {100, 200}}, nil, 1)
+	tr.Process(map[int][2]int64{1: {200, 400}}, nil, 1)
 
 	// Flush and restore
 	first := tr.FlushTraffic()
 	tr.RestoreTraffic(first)
 
 	// Generate additional traffic
-	tr.Process([]kernel.Connection{
-		{ID: "c1", UserID: 1, Upload: 250, Download: 500},
-	})
+	tr.Process(map[int][2]int64{1: {250, 500}}, nil, 1)
 
 	final := tr.FlushTraffic()
-	// Should be restored(200,400) + delta(50,100) = (250,500)
+	// Restored: 200/400 (initial 100/200 + delta 100/200) + new delta 50/100 = 250/500
 	if final[1][0] != 250 || final[1][1] != 500 {
 		t.Errorf("additive restore: got %v", final[1])
 	}
@@ -186,37 +150,37 @@ func TestRestoreTraffic_Additive(t *testing.T) {
 
 func TestFlushAliveIPs(t *testing.T) {
 	tr := New()
-	tr.Process([]kernel.Connection{
-		{ID: "c1", UserID: 1, Upload: 100, Download: 200, SourceIP: "1.1.1.1"},
-		{ID: "c2", UserID: 1, Upload: 50, Download: 80, SourceIP: "2.2.2.2"},
-		{ID: "c3", UserID: 2, Upload: 30, Download: 40, SourceIP: "3.3.3.3"},
-	})
-
-	aliveIPs := tr.FlushAliveIPs()
-	if len(aliveIPs[1]) != 2 {
-		t.Errorf("user 1 IPs: got %d, want 2", len(aliveIPs[1]))
+	aliveIPs := map[int]map[string]bool{
+		1: {"1.1.1.1": true, "2.2.2.2": true},
+		2: {"3.3.3.3": true},
 	}
-	if len(aliveIPs[2]) != 1 {
-		t.Errorf("user 2 IPs: got %d, want 1", len(aliveIPs[2]))
+	tr.Process(map[int][2]int64{1: {100, 200}, 2: {30, 40}}, aliveIPs, 3)
+
+	flushed := tr.FlushAliveIPs()
+	if len(flushed[1]) != 2 {
+		t.Errorf("user 1 IPs: got %d, want 2", len(flushed[1]))
+	}
+	if len(flushed[2]) != 1 {
+		t.Errorf("user 2 IPs: got %d, want 1", len(flushed[2]))
 	}
 
-	// After flush, should be empty
-	aliveIPs2 := tr.FlushAliveIPs()
-	if len(aliveIPs2) != 0 {
-		t.Errorf("expected empty after second flush, got %v", aliveIPs2)
+	// Without another Process call, hash is same → returns nil (skip duplicate)
+	flushed2 := tr.FlushAliveIPs()
+	if flushed2 != nil {
+		t.Errorf("expected nil on duplicate flush, got %v", flushed2)
 	}
 }
 
 func TestFlushAliveIPs_DedupSameIP(t *testing.T) {
 	tr := New()
-	tr.Process([]kernel.Connection{
-		{ID: "c1", UserID: 1, Upload: 100, Download: 200, SourceIP: "1.1.1.1"},
-		{ID: "c2", UserID: 1, Upload: 50, Download: 80, SourceIP: "1.1.1.1"},
-	})
+	aliveIPs := map[int]map[string]bool{
+		1: {"1.1.1.1": true},
+	}
+	tr.Process(map[int][2]int64{1: {100, 200}}, aliveIPs, 2)
 
-	aliveIPs := tr.FlushAliveIPs()
-	if len(aliveIPs[1]) != 1 {
-		t.Errorf("expected dedup to 1 IP, got %d", len(aliveIPs[1]))
+	flushed := tr.FlushAliveIPs()
+	if len(flushed[1]) != 1 {
+		t.Errorf("expected 1 IP, got %d", len(flushed[1]))
 	}
 }
 
@@ -226,9 +190,7 @@ func TestHasTraffic(t *testing.T) {
 		t.Error("new tracker should not have traffic")
 	}
 
-	tr.Process([]kernel.Connection{
-		{ID: "c1", UserID: 1, Upload: 100, Download: 200},
-	})
+	tr.Process(map[int][2]int64{1: {100, 200}}, nil, 1)
 	if !tr.HasTraffic() {
 		t.Error("should have traffic after process")
 	}
@@ -242,15 +204,13 @@ func TestHasTraffic(t *testing.T) {
 func TestProcess_NoTrafficDelta(t *testing.T) {
 	tr := New()
 
-	conns := []kernel.Connection{
-		{ID: "c1", UserID: 1, Upload: 100, Download: 200},
-	}
-	tr.Process(conns)
+	tr.Process(map[int][2]int64{1: {100, 200}}, nil, 1)
+	tr.FlushTraffic() // clear
 
 	// Same values — no delta
-	interval := tr.Process(conns)
-	if len(interval) != 0 {
-		t.Errorf("expected no interval traffic for zero delta, got %v", interval)
+	tr.Process(map[int][2]int64{1: {100, 200}}, nil, 1)
+	if tr.HasTraffic() {
+		t.Error("expected no traffic for zero delta")
 	}
 }
 
@@ -258,17 +218,11 @@ func TestTrafficAccumulation(t *testing.T) {
 	tr := New()
 
 	// Tick 1
-	tr.Process([]kernel.Connection{
-		{ID: "c1", UserID: 1, Upload: 100, Download: 200},
-	})
-	// Tick 2
-	tr.Process([]kernel.Connection{
-		{ID: "c1", UserID: 1, Upload: 300, Download: 500},
-	})
-	// Tick 3
-	tr.Process([]kernel.Connection{
-		{ID: "c1", UserID: 1, Upload: 350, Download: 600},
-	})
+	tr.Process(map[int][2]int64{1: {100, 200}}, nil, 1)
+	// Tick 2 (cumulative 300, 500 → delta 200, 300)
+	tr.Process(map[int][2]int64{1: {300, 500}}, nil, 1)
+	// Tick 3 (cumulative 350, 600 → delta 50, 100)
+	tr.Process(map[int][2]int64{1: {350, 600}}, nil, 1)
 
 	flushed := tr.FlushTraffic()
 	// Total: 100+200+50 = 350 upload, 200+300+100 = 600 download