net/tstun: diagnose /dev/net/tun fd leak, give better failure message

Updates #5029

Change-Id: Ibee5e0c9076fe764eb5d856d5ef8b09f4d0e2921
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>

cmd/tailscaled/tailscaled.go

@@ -515,7 +515,7 @@ func tryEngine(logf logger.Logf, linkMon *monitor.Mon, dialer *tsdial.Dialer, na
 	} else {
 		dev, devName, err := tstun.New(logf, name)
 		if err != nil {
-			tstun.Diagnose(logf, name)
+			tstun.Diagnose(logf, name, err)
 			return nil, false, fmt.Errorf("tstun.New(%q): %w", name, err)
 		}
 		conf.Tun = dev

net/tstun/tun.go

@@ -74,15 +74,18 @@ func New(logf logger.Logf, tunName string) (tun.Device, string, error) {
 
 // tunDiagnoseFailure, if non-nil, does OS-specific diagnostics of why
 // TUN failed to work.
-var tunDiagnoseFailure func(tunName string, logf logger.Logf)
+var tunDiagnoseFailure func(tunName string, logf logger.Logf, err error)
 
 // Diagnose tries to explain a tuntap device creation failure.
 // It pokes around the system and logs some diagnostic info that might
 // help debug why tun creation failed. Because device creation has
 // already failed and the program's about to end, log a lot.
-func Diagnose(logf logger.Logf, tunName string) {
+//
+// The tunName is the name of the tun device that was requested but failed.
+// The err error is how the tun creation failed.
+func Diagnose(logf logger.Logf, tunName string, err error) {
 	if tunDiagnoseFailure != nil {
-		tunDiagnoseFailure(tunName, logf)
+		tunDiagnoseFailure(tunName, logf, err)
 	} else {
 		logf("no TUN failure diagnostics for OS %q", runtime.GOOS)
 	}

net/tstun/tun_linux.go

@@ -6,6 +6,7 @@ package tstun
 
 import (
 	"bytes"
+	"errors"
 	"os"
 	"os/exec"
 	"strings"
@@ -19,7 +20,14 @@ func init() {
 	tunDiagnoseFailure = diagnoseLinuxTUNFailure
 }
 
-func diagnoseLinuxTUNFailure(tunName string, logf logger.Logf) {
+func diagnoseLinuxTUNFailure(tunName string, logf logger.Logf, createErr error) {
+	if errors.Is(createErr, syscall.EBUSY) {
+		logf("TUN device %s is busy; another process probably still has it open (from old version of Tailscale that had a bug)", tunName)
+		logf("To fix, kill the process that has it open. Find with:\n\n$ sudo lsof -n /dev/net/tun\n\n")
+		logf("... and then kill those PID(s)")
+		return
+	}
+
 	var un syscall.Utsname
 	err := syscall.Uname(&un)
 	if err != nil {

net/tstun/tun_macos.go

@@ -17,7 +17,7 @@ func init() {
 	tunDiagnoseFailure = diagnoseDarwinTUNFailure
 }
 
-func diagnoseDarwinTUNFailure(tunName string, logf logger.Logf) {
+func diagnoseDarwinTUNFailure(tunName string, logf logger.Logf, err error) {
 	if os.Getuid() != 0 {
 		logf("failed to create TUN device as non-root user; use 'sudo tailscaled', or run under launchd with 'sudo tailscaled install-system-daemon'")
 	}