net/interfaces: redo how we get the default interface on macOS and iOS

With #6566 we added an external mechanism for getting the default
interface, and used it on macOS and iOS (see tailscale/corp#8201).
The goal was to be able to get the default physical interface even when
using an exit node (in which case the routing table would say that the
Tailscale utun* interface is the default).

However, the external mechanism turns out to be unreliable in some
cases, e.g. when multiple cellular interfaces are present/toggled (I
have occasionally gotten my phone into a state where it reports the pdp_ip1
interface as the default, even though it can't actually route traffic).

It was observed that `ifconfig -v` on macOS reports an "effective interface"
for the Tailscale utn* interface, which seems promising. By examining
the ifconfig source code, it turns out that this is done via a
SIOCGIFDELEGATE ioctl syscall. Though this is a private API, it appears
to have been around for a long time (e.g. it's in the 10.13 xnu release
at https://opensource.apple.com/source/xnu/xnu-4570.41.2/bsd/net/if_types.h.auto.html)
and thus is unlikely to go away.

We can thus use this ioctl if the routing table says that a utun*
interface is the default, and go back to the simpler mechanism that
we had before #6566.

Updates #7184
Updates #7188

Signed-off-by: Mihai Parparita <mihai@tailscale.com>

cmd/derper/depaware.txt

@@ -81,7 +81,7 @@ tailscale.com/cmd/derper dependencies: (generated by github.com/tailscale/depawa
         tailscale.com/util/dnsname                                   from tailscale.com/hostinfo+
         tailscale.com/util/httpm                                     from tailscale.com/client/tailscale
         tailscale.com/util/lineread                                  from tailscale.com/hostinfo+
-        tailscale.com/util/mak                                       from tailscale.com/syncs
+        tailscale.com/util/mak                                       from tailscale.com/syncs+
         tailscale.com/util/multierr                                  from tailscale.com/health
         tailscale.com/util/set                                       from tailscale.com/health
         tailscale.com/util/singleflight                              from tailscale.com/net/dnscache

ipn/ipnlocal/local.go

@@ -3840,7 +3840,6 @@ func (b *LocalBackend) setNetMapLocked(nm *netmap.NetworkMap) {
 
 	// See the netns package for documentation on what this capability does.
 	netns.SetBindToInterfaceByRoute(hasCapability(nm, tailcfg.CapabilityBindToInterfaceByRoute))
-	interfaces.SetDisableAlternateDefaultRouteInterface(hasCapability(nm, tailcfg.CapabilityDebugDisableAlternateDefaultRouteInterface))
 	netns.SetDisableBindConnToInterface(hasCapability(nm, tailcfg.CapabilityDebugDisableBindConnToInterface))
 
 	b.setTCPPortsInterceptedFromNetmapAndPrefsLocked(b.pm.CurrentPrefs())

net/interfaces/interfaces.go

@@ -13,7 +13,6 @@ import (
 	"runtime"
 	"sort"
 	"strings"
-	"sync/atomic"
 
 	"tailscale.com/hostinfo"
 	"tailscale.com/net/netaddr"
@@ -757,14 +756,3 @@ func HasCGNATInterface() (bool, error) {
 	}
 	return hasCGNATInterface, nil
 }
-
-var disableAlternateDefaultRouteInterface atomic.Bool
-
-// SetDisableAlternateDefaultRouteInterface disables the optional external/
-// alternate mechanism for getting the default route network interface.
-//
-// Currently, this only changes the behaviour on BSD-like sytems (FreeBSD and
-// Darwin).
-func SetDisableAlternateDefaultRouteInterface(v bool) {
-	disableAlternateDefaultRouteInterface.Store(v)
-}

net/interfaces/interfaces_bsd.go

@@ -19,7 +19,6 @@ import (
 	"golang.org/x/net/route"
 	"golang.org/x/sys/unix"
 	"tailscale.com/net/netaddr"
-	"tailscale.com/syncs"
 )
 
 func defaultRoute() (d DefaultRouteDetails, err error) {
@@ -40,19 +39,6 @@ func defaultRoute() (d DefaultRouteDetails, err error) {
 // owns the default route. It returns the first IPv4 or IPv6 default route it
 // finds (it does not prefer one or the other).
 func DefaultRouteInterfaceIndex() (int, error) {
-	disabledAlternateDefaultRouteInterface := false
-	if f := defaultRouteInterfaceIndexFunc.Load(); f != nil {
-		if ifIndex := f(); ifIndex != 0 {
-			if !disableAlternateDefaultRouteInterface.Load() {
-				return ifIndex, nil
-			} else {
-				disabledAlternateDefaultRouteInterface = true
-				log.Printf("interfaces_bsd: alternate default route interface function disabled, would have returned interface %d", ifIndex)
-			}
-		}
-		// Fallthrough if we can't use the alternate implementation.
-	}
-
 	// $ netstat -nr
 	// Routing tables
 	// Internet:
@@ -81,8 +67,10 @@ func DefaultRouteInterfaceIndex() (int, error) {
 			continue
 		}
 		if isDefaultGateway(rm) {
-			if disabledAlternateDefaultRouteInterface {
-				log.Printf("interfaces_bsd: alternate default route interface function disabled, default implementation returned %d", rm.Index)
+			if delegatedIndex, err := getDelegatedInterface(rm.Index); err == nil && delegatedIndex != 0 {
+				return delegatedIndex, nil
+			} else if err != nil {
+				log.Printf("interfaces_bsd: could not get delegated interface: %v", err)
 			}
 			return rm.Index, nil
 		}
@@ -90,16 +78,6 @@ func DefaultRouteInterfaceIndex() (int, error) {
 	return 0, errors.New("no gateway index found")
 }
 
-var defaultRouteInterfaceIndexFunc syncs.AtomicValue[func() int]
-
-// SetDefaultRouteInterfaceIndexFunc allows an alternate implementation of
-// DefaultRouteInterfaceIndex to be provided. If none is set, or if f() returns a 0
-// (indicating an unknown interface index), then the default implementation (that parses
-// the routing table) will be used.
-func SetDefaultRouteInterfaceIndexFunc(f func() int) {
-	defaultRouteInterfaceIndexFunc.Store(f)
-}
-
 func init() {
 	likelyHomeRouterIP = likelyHomeRouterIPBSDFetchRIB
 }

net/interfaces/interfaces_darwin.go

@@ -4,9 +4,15 @@
 package interfaces
 
 import (
+	"net"
+	"strings"
+	"sync"
 	"syscall"
+	"unsafe"
 
 	"golang.org/x/net/route"
+	"golang.org/x/sys/unix"
+	"tailscale.com/util/mak"
 )
 
 // fetchRoutingTable calls route.FetchRIB, fetching NET_RT_DUMP2.
@@ -17,3 +23,73 @@ func fetchRoutingTable() (rib []byte, err error) {
 func parseRoutingTable(rib []byte) ([]route.Message, error) {
 	return route.ParseRIB(syscall.NET_RT_IFLIST2, rib)
 }
+
+var ifNames struct {
+	sync.Mutex
+	m map[int]string // ifindex => name
+}
+
+// getDelegatedInterface returns the interface index of the underlying interface
+// for the given interface index. 0 is returned if the interface does not
+// delegate.
+func getDelegatedInterface(ifIndex int) (int, error) {
+	ifNames.Lock()
+	defer ifNames.Unlock()
+
+	// To get the delegated interface, we do what ifconfig does and use the
+	// SIOCGIFDELEGATE ioctl. It operates in term of a ifreq struct, which
+	// has to be populated with a interface name. To avoid having to do a
+	// interface index -> name lookup every time, we cache interface names
+	// (since indexes and names are stable after boot).
+	ifName, ok := ifNames.m[ifIndex]
+	if !ok {
+		iface, err := net.InterfaceByIndex(ifIndex)
+		if err != nil {
+			return 0, err
+		}
+		ifName = iface.Name
+		mak.Set(&ifNames.m, ifIndex, ifName)
+	}
+
+	// Only tunnels (like Tailscale itself) have a delegated interface, avoid
+	// the ioctl if we can.
+	if !strings.HasPrefix(ifName, "utun") {
+		return 0, nil
+	}
+
+	// We don't cache the result of the ioctl, since the delegated interface can
+	// change, e.g. if the user changes the preferred service order in the
+	// network preference pane.
+	fd, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM, 0)
+	if err != nil {
+		return 0, err
+	}
+	defer unix.Close(fd)
+
+	// Match the ifreq struct/union from the bsd/net/if.h header in the Darwin
+	// open source release.
+	var ifr struct {
+		ifr_name      [unix.IFNAMSIZ]byte
+		ifr_delegated uint32
+	}
+	copy(ifr.ifr_name[:], ifName)
+
+	// SIOCGIFDELEGATE is not in the Go x/sys package or in the public macOS
+	// <sys/sockio.h> headers. However, it is in the Darwin/xnu open source
+	// release (and is used by ifconfig, see
+	// https://github.com/apple-oss-distributions/network_cmds/blob/6ccdc225ad5aa0d23ea5e7d374956245d2462427/ifconfig.tproj/ifconfig.c#L2183-L2187).
+	// We generate its value by evaluating the `_IOWR('i', 157, struct ifreq)`
+	// macro, which is how it's defined in
+	// https://github.com/apple/darwin-xnu/blob/2ff845c2e033bd0ff64b5b6aa6063a1f8f65aa32/bsd/sys/sockio.h#L264
+	const SIOCGIFDELEGATE = 0xc020699d
+
+	_, _, errno := syscall.Syscall(
+		syscall.SYS_IOCTL,
+		uintptr(fd),
+		uintptr(SIOCGIFDELEGATE),
+		uintptr(unsafe.Pointer(&ifr)))
+	if errno != 0 {
+		return 0, errno
+	}
+	return int(ifr.ifr_delegated), nil
+}

net/interfaces/interfaces_freebsd.go

@@ -22,3 +22,7 @@ func fetchRoutingTable() (rib []byte, err error) {
 func parseRoutingTable(rib []byte) ([]route.Message, error) {
 	return route.ParseRIB(syscall.NET_RT_IFLIST, rib)
 }
+
+func getDelegatedInterface(ifIndex int) (int, error) {
+	return 0, nil
+}