experimental/windivert: P2.1+P2.2 with WinDivert NETWORK+SOCKET layers

WIP snapshot before pivot to sing-box+TUN. Reached: - TCP redirect via streamdump pattern (swap+Outbound=0+reinject) - SOCKET layer for SYN-stage flow detection (avoids FLOW Establish-too-late race) - Lazy PID→name resolution (catches Update.exe inside procscan tick) - UDP forward via SOCKS5 UDP ASSOCIATE relay + manual reinject - Result: chat works, voice times out (Discord IP discovery / RTC handshake fails) Reason for pivot: WinDivert NAT-reinject pattern has subtle layer-3 semantics issues that DLL-injection / TUN-based proxies sidestep entirely. Going with embedded sing-box + wintun as the engine — proven path for Discord voice through SOCKS5. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-01 22:27:54 +03:00
parent 8ceb7775d7
commit 4074e68715
19 changed files with 2666 additions and 62 deletions
@@ -0,0 +1,12 @@
+//go:build !windows
+
+package main
+
+import (
+	"context"
+	"fmt"
+)
+
+func runDebugFlow(_ context.Context) error {
+	return fmt.Errorf("debug-flow requires Windows")
+}
@@ -0,0 +1,64 @@
+//go:build windows
+
+package main
+
+import (
+	"context"
+	"log"
+	"time"
+
+	"git.okcu.io/root/drover-go/internal/divert"
+)
+
+// runDebugFlow opens a WinDivert FLOW handle with the broadest possible
+// filter ("tcp") and logs every flow-establish/delete event for up to
+// 30 seconds. This is the simplest possible test that the FLOW layer
+// is delivering events to our handle.
+//
+// If we see events here but our process-targeted handle in `proxy`
+// stays silent, the bug is in our processId filter clause. If we see
+// nothing here, the FLOW layer is broken on this machine.
+func runDebugFlow(parent context.Context) error {
+	if _, err := divert.InstallDriver(); err != nil {
+		return err
+	}
+
+	ctx, cancel := context.WithTimeout(parent, 30*time.Second)
+	defer cancel()
+
+	log.Printf("debug-flow: opening FLOW handle with filter \"true\" (capture all flows)")
+	h, err := divert.OpenFlow("true")
+	if err != nil {
+		log.Printf("debug-flow: OpenFlow failed: %v", err)
+		return err
+	}
+	defer h.Close()
+	log.Printf("debug-flow: handle open, listening for 30s")
+
+	go func() {
+		<-ctx.Done()
+		_ = h.Close() // unblock RecvFlow
+	}()
+
+	count := 0
+	for {
+		ev, err := h.RecvFlow()
+		if err != nil {
+			if ctx.Err() != nil {
+				log.Printf("debug-flow: done — captured %d events in 30s", count)
+				return nil
+			}
+			log.Printf("debug-flow: RecvFlow err: %v", err)
+			return err
+		}
+		count++
+		log.Printf("debug-flow: event #%d est=%v pid=%d proto=%d %v:%d → %v:%d  rawLocal=%x  rawRemote=%x",
+			count, ev.Established, ev.ProcessID, ev.Protocol,
+			ev.SrcAddr, ev.SrcPort, ev.DstAddr, ev.DstPort,
+			ev.LocalRaw, ev.RemoteRaw)
+		if count >= 20 {
+			log.Printf("debug-flow: hit 20-event cap, stopping")
+			return nil
+		}
+	}
+}
@@ -3,7 +3,11 @@ package main

 import (
 	"fmt"
+	"io"
+	"log"
 	"os"
+	"path/filepath"
+	"time"

 	"github.com/spf13/cobra"

@@ -29,16 +33,30 @@ func main() {
 	// AttachConsole(ATTACH_PARENT_PROCESS) wires that up. No-op elsewhere.
 	attachToParentConsole()

+	// Open a debug log file at %LOCALAPPDATA%\Drover\debug.log so we have
+	// post-mortem visibility into engine startup failures even when the
+	// process was launched via UAC re-elevation (which detaches stderr
+	// from the parent terminal).
+	setupDebugLog()
+
 	// Detect if we need admin for the command in os.Args[1:]. If we do and
 	// we're not admin, re-launch via ShellExecute("runas", ...) and exit.
 	// CLI subcommands like "check", "version", "update" don't need admin
 	// and will run without UAC prompt.
-	if CmdNeedsAdmin(os.Args[1:]) && !IsAdmin() {
+	needsAdm := CmdNeedsAdmin(os.Args[1:])
+	isAdm := IsAdmin()
+	log.Printf("main: post-console admin=%v needsAdmin=%v args=%v", isAdm, needsAdm, os.Args[1:])
+	if needsAdm && !isAdm {
+		log.Printf("main: invoking ReElevate")
 		if err := ReElevate(os.Args[1:]); err != nil {
+			log.Printf("main: ReElevate returned err: %v", err)
 			fmt.Fprintf(os.Stderr, "failed to re-elevate: %v\n", err)
+		} else {
+			log.Printf("main: ReElevate returned ok, exiting parent")
 		}
 		os.Exit(0)
 	}
+	log.Printf("main: continuing in current process (no re-elevation needed)")

 	// Inject our build version so the updater package can stamp it on the
 	// User-Agent header it sends to git.okcu.io.
@@ -50,6 +68,39 @@ func main() {
 	}
 }

+// setupDebugLog wires the standard `log` package to write to both stderr
+// and %LOCALAPPDATA%\Drover\debug.log. Survives UAC re-launch (each
+// process opens its own append-mode handle).
+func setupDebugLog() {
+	dir := os.Getenv("LOCALAPPDATA")
+	if dir == "" {
+		dir = os.Getenv("TEMP")
+	}
+	if dir == "" {
+		return
+	}
+	dir = filepath.Join(dir, "Drover")
+	_ = os.MkdirAll(dir, 0755)
+	// Truncate on each startup — keeps the log focused on the current
+	// run instead of accumulating past sessions. If you need history,
+	// rotate before launch.
+	f, err := os.OpenFile(filepath.Join(dir, "debug.log"), os.O_TRUNC|os.O_CREATE|os.O_WRONLY, 0644)
+	if err != nil {
+		return
+	}
+	// On a UAC-elevated launch (Start-Process -Verb RunAs) we have no
+	// parent console — os.Stderr points at an invalid handle. Writing
+	// to it via MultiWriter fails the *entire* write, so logs silently
+	// drop. Just write to the file; CLI subcommands launched from a
+	// real console can grep the file.
+	log.SetOutput(f)
+	_ = io.Discard // keep io import used
+
+	log.SetFlags(log.LstdFlags | log.Lmicroseconds)
+	log.Printf("=== drover %s start pid=%d args=%v admin=%v at %s ===",
+		Version, os.Getpid(), os.Args[1:], IsAdmin(), time.Now().Format(time.RFC3339))
+}
+
 func newRootCmd() *cobra.Command {
 	root := &cobra.Command{
 		Use:           "drover",
@@ -76,10 +127,52 @@ func newRootCmd() *cobra.Command {
 	root.AddCommand(newUpdateCmd())
 	root.AddCommand(newServiceCmd())
 	root.AddCommand(newGUICmd())
+	root.AddCommand(newProxyCmd())
+	root.AddCommand(newDebugFlowCmd())

 	return root
 }

+// newDebugFlowCmd opens a WinDivert FLOW handle with filter "tcp"
+// (capture all TCP flow events from any process) and logs every event
+// for 30 seconds. Useful to verify the FLOW layer is working at all
+// without process-targeting interference.
+func newDebugFlowCmd() *cobra.Command {
+	return &cobra.Command{
+		Use:    "debug-flow",
+		Short:  "[debug] open broad FLOW handle, log events for 30s",
+		Hidden: true,
+		RunE: func(cmd *cobra.Command, args []string) error {
+			return runDebugFlow(cmd.Context())
+		},
+	}
+}
+
+// newProxyCmd is the headless engine-only mode: no Wails, no tray —
+// just spin up the WinDivert + SOCKS5 pipeline against the configured
+// upstream and block on Ctrl+C. Useful for debugging without the GUI
+// stack in the way; everything still goes to %LOCALAPPDATA%\Drover\debug.log.
+func newProxyCmd() *cobra.Command {
+	var host, login, password string
+	var port int
+	var auth bool
+	cmd := &cobra.Command{
+		Use:   "proxy",
+		Short: "Run the WinDivert+SOCKS5 engine in headless mode (no GUI, blocks until Ctrl+C)",
+		RunE: func(cmd *cobra.Command, args []string) error {
+			return runProxy(cmd.Context(), host, port, auth, login, password)
+		},
+	}
+	cmd.Flags().StringVar(&host, "host", "", "upstream SOCKS5 host (required)")
+	cmd.Flags().IntVar(&port, "port", 0, "upstream SOCKS5 port (required)")
+	cmd.Flags().BoolVar(&auth, "auth", false, "enable user/pass auth")
+	cmd.Flags().StringVar(&login, "login", "", "SOCKS5 login (when --auth)")
+	cmd.Flags().StringVar(&password, "password", "", "SOCKS5 password (when --auth)")
+	_ = cmd.MarkFlagRequired("host")
+	_ = cmd.MarkFlagRequired("port")
+	return cmd
+}
+
 func newGUICmd() *cobra.Command {
 	return &cobra.Command{
 		Use:   "gui",
@@ -0,0 +1,14 @@
+//go:build !windows
+
+package main
+
+import (
+	"context"
+	"fmt"
+)
+
+// runProxy stub for non-Windows builds (drover only ships for Windows;
+// this stub keeps `go build ./...` clean on Linux dev/CI machines).
+func runProxy(_ context.Context, _ string, _ int, _ bool, _, _ string) error {
+	return fmt.Errorf("the proxy subcommand requires Windows (WinDivert is Windows-only)")
+}
@@ -0,0 +1,80 @@
+//go:build windows
+
+package main
+
+import (
+	"context"
+	"fmt"
+	"log"
+	"os"
+	"os/signal"
+	"syscall"
+	"time"
+
+	"git.okcu.io/root/drover-go/internal/engine"
+)
+
+// runProxy is the body of the `drover proxy` subcommand. It builds an
+// engine.Engine from the supplied flags, calls Start, and blocks until
+// the process receives SIGINT (Ctrl+C) or SIGTERM. On signal, it
+// gracefully Stops the engine and exits.
+//
+// All output is mirrored to stderr (visible when launched from a
+// console session) AND %LOCALAPPDATA%\Drover\debug.log. setupDebugLog
+// in main.go has already wired the log package to write to both.
+func runProxy(parent context.Context, host string, port int, auth bool, login, password string) error {
+	if host == "" || port == 0 {
+		return fmt.Errorf("--host and --port are required")
+	}
+
+	ctx, cancel := signal.NotifyContext(parent, os.Interrupt, syscall.SIGTERM)
+	defer cancel()
+
+	cfg := engine.Config{
+		ProxyAddr: fmt.Sprintf("%s:%d", host, port),
+		UseAuth:   auth,
+		Login:     login,
+		Password:  password,
+		Targets:   []string{"Discord.exe", "DiscordCanary.exe", "DiscordPTB.exe", "Update.exe"},
+	}
+
+	log.Printf("proxy: building engine (proxy=%s auth=%v targets=%v)", cfg.ProxyAddr, cfg.UseAuth, cfg.Targets)
+	e, err := engine.New(cfg)
+	if err != nil {
+		return fmt.Errorf("engine.New: %w", err)
+	}
+	startCtx, startCancel := context.WithTimeout(ctx, 15*time.Second)
+	defer startCancel()
+	if err := e.Start(startCtx); err != nil {
+		log.Printf("proxy: Start failed: %v", err)
+		return fmt.Errorf("engine.Start: %w", err)
+	}
+	log.Printf("proxy: engine status=%s — press Ctrl+C to stop", e.Status())
+
+	// Periodic status ping so the user sees the engine is alive.
+	statusTk := time.NewTicker(10 * time.Second)
+	defer statusTk.Stop()
+
+	for {
+		select {
+		case <-ctx.Done():
+			log.Printf("proxy: signal received, shutting down")
+			if err := e.Stop(); err != nil {
+				log.Printf("proxy: Stop returned: %v", err)
+			}
+			log.Printf("proxy: bye")
+			return nil
+		case <-statusTk.C:
+			if le := e.LastError(); le != nil {
+				log.Printf("proxy: heartbeat status=%s lastErr=%v", e.Status(), le)
+			} else {
+				log.Printf("proxy: heartbeat status=%s", e.Status())
+			}
+			if e.Status() == engine.StatusFailed {
+				log.Printf("proxy: engine entered Failed state, exiting")
+				_ = e.Stop()
+				return fmt.Errorf("engine failed: %v", e.LastError())
+			}
+		}
+	}
+}