tailscale/net/netns/netns_darwin.go
Mihai Parparita 7330aa593e all: avoid repeated default interface lookups
On some platforms (notably macOS and iOS) we look up the default
interface to bind outgoing connections to. This is both duplicated
work and results in logspam when the default interface is not available
(i.e. when a phone has no connectivity, we log an error and thus cause
more things that we will try to upload and fail).

Fixed by passing around a netmon.Monitor to more places, so that we can
use its cached interface state.

Fixes #7850
Updates #7621

Signed-off-by: Mihai Parparita <mihai@tailscale.com>
2023-04-20 15:46:01 -07:00

264 lines
7.3 KiB
Go

// Copyright (c) Tailscale Inc & AUTHORS
// SPDX-License-Identifier: BSD-3-Clause
//go:build darwin
package netns
import (
"errors"
"fmt"
"log"
"net"
"net/netip"
"os"
"strings"
"syscall"
"golang.org/x/net/route"
"golang.org/x/sys/unix"
"tailscale.com/envknob"
"tailscale.com/net/interfaces"
"tailscale.com/net/netmon"
"tailscale.com/types/logger"
)
func control(logf logger.Logf, netMon *netmon.Monitor) func(network, address string, c syscall.RawConn) error {
return func(network, address string, c syscall.RawConn) error {
return controlLogf(logf, netMon, network, address, c)
}
}
var bindToInterfaceByRouteEnv = envknob.RegisterBool("TS_BIND_TO_INTERFACE_BY_ROUTE")
var errInterfaceStateInvalid = errors.New("interface state invalid")
// controlLogf marks c as necessary to dial in a separate network namespace.
//
// It's intentionally the same signature as net.Dialer.Control
// and net.ListenConfig.Control.
func controlLogf(logf logger.Logf, netMon *netmon.Monitor, network, address string, c syscall.RawConn) error {
if isLocalhost(address) {
// Don't bind to an interface for localhost connections.
return nil
}
if disableBindConnToInterface.Load() {
logf("netns_darwin: binding connection to interfaces disabled")
return nil
}
idx, err := getInterfaceIndex(logf, netMon, address)
if err != nil {
// callee logged
return nil
}
return bindConnToInterface(c, network, address, idx, logf)
}
func getInterfaceIndex(logf logger.Logf, netMon *netmon.Monitor, address string) (int, error) {
// Helper so we can log errors.
defaultIdx := func() (int, error) {
if netMon == nil {
idx, err := interfaces.DefaultRouteInterfaceIndex()
if err != nil {
// It's somewhat common for there to be no default gateway route
// (e.g. on a phone with no connectivity), don't log those errors
// since they are expected.
if !errors.Is(err, interfaces.ErrNoGatewayIndexFound) {
logf("[unexpected] netns: DefaultRouteInterfaceIndex: %v", err)
}
return -1, err
}
return idx, nil
}
state := netMon.InterfaceState()
if state == nil {
return -1, errInterfaceStateInvalid
}
if iface, ok := state.Interface[state.DefaultRouteInterface]; ok {
return iface.Index, nil
}
return -1, errInterfaceStateInvalid
}
useRoute := bindToInterfaceByRoute.Load() || bindToInterfaceByRouteEnv()
if !useRoute {
return defaultIdx()
}
host, _, err := net.SplitHostPort(address)
if err != nil {
// No port number; use the string directly.
host = address
}
// If the address doesn't parse, use the default index.
addr, err := netip.ParseAddr(host)
if err != nil {
logf("[unexpected] netns: error parsing address %q: %v", host, err)
return defaultIdx()
}
idx, err := interfaceIndexFor(addr, true /* canRecurse */)
if err != nil {
logf("netns: error in interfaceIndexFor: %v", err)
return defaultIdx()
}
// Verify that we didn't just choose the Tailscale interface;
// if so, we fall back to binding from the default.
_, tsif, err2 := interfaces.Tailscale()
if err2 == nil && tsif.Index == idx {
logf("[unexpected] netns: interfaceIndexFor returned Tailscale interface")
return defaultIdx()
}
return idx, err
}
// interfaceIndexFor returns the interface index that we should bind to in
// order to send traffic to the provided address.
func interfaceIndexFor(addr netip.Addr, canRecurse bool) (int, error) {
fd, err := unix.Socket(unix.AF_ROUTE, unix.SOCK_RAW, unix.AF_UNSPEC)
if err != nil {
return 0, fmt.Errorf("creating AF_ROUTE socket: %w", err)
}
defer unix.Close(fd)
var routeAddr route.Addr
if addr.Is4() {
routeAddr = &route.Inet4Addr{IP: addr.As4()}
} else {
routeAddr = &route.Inet6Addr{IP: addr.As16()}
}
rm := route.RouteMessage{
// NOTE: This is unix.RTM_VERSION, but we want to pin this to a
// particular constant so that it doesn't change under us if
// the x/sys/unix package changes down the road. Currently this
// is 0x5 on both Darwin x86 and ARM64.
Version: 0x5,
Type: unix.RTM_GET,
Flags: unix.RTF_UP,
ID: uintptr(os.Getpid()),
Seq: 1,
Addrs: []route.Addr{
unix.RTAX_DST: routeAddr,
},
}
b, err := rm.Marshal()
if err != nil {
return 0, fmt.Errorf("marshaling RouteMessage: %w", err)
}
_, err = unix.Write(fd, b)
if err != nil {
return 0, fmt.Errorf("writing message: %w", err)
}
// On macOS, the RTM_GET call should return exactly one route message.
// Given the following sizes and constants:
// - sizeof(struct rt_msghdr) = 92
// - RTAX_MAX = 8
// - sizeof(struct sockaddr_in6) = 28
// - sizeof(struct sockaddr_in) = 16
// - sizeof(struct sockaddr_dl) = 20
//
// The maximum buffer size should be:
// sizeof(struct rt_msghdr) + RTAX_MAX*sizeof(struct sockaddr_in6)
// = 92 + 8*28
// = 316
//
// During my testing, responses are typically ~120 bytes.
//
// We provide a much larger buffer just in case we're off by a bit, or
// the kernel decides to return more than one message; 2048 bytes
// should be plenty here. This also means we can do a single Read.
var buf [2048]byte
n, err := unix.Read(fd, buf[:])
if err != nil {
return 0, fmt.Errorf("reading message: %w", err)
}
msgs, err := route.ParseRIB(route.RIBTypeRoute, buf[:n])
if err != nil {
return 0, fmt.Errorf("route.ParseRIB: %w", err)
}
if len(msgs) == 0 {
return 0, fmt.Errorf("no messages")
}
for _, msg := range msgs {
rm, ok := msg.(*route.RouteMessage)
if !ok {
continue
}
if rm.Version < 3 || rm.Version > 5 || rm.Type != unix.RTM_GET {
continue
}
if len(rm.Addrs) < unix.RTAX_GATEWAY {
continue
}
switch addr := rm.Addrs[unix.RTAX_GATEWAY].(type) {
case *route.LinkAddr:
return addr.Index, nil
case *route.Inet4Addr:
// We can get a gateway IP; recursively call ourselves
// (exactly once) to get the link (and thus index) for
// the gateway IP.
if canRecurse {
return interfaceIndexFor(netip.AddrFrom4(addr.IP), false)
}
case *route.Inet6Addr:
// As above.
if canRecurse {
return interfaceIndexFor(netip.AddrFrom16(addr.IP), false)
}
default:
// Unknown type; skip it
continue
}
}
return 0, fmt.Errorf("no valid address found")
}
// SetListenConfigInterfaceIndex sets lc.Control such that sockets are bound
// to the provided interface index.
func SetListenConfigInterfaceIndex(lc *net.ListenConfig, ifIndex int) error {
if lc == nil {
return errors.New("nil ListenConfig")
}
if lc.Control != nil {
return errors.New("ListenConfig.Control already set")
}
lc.Control = func(network, address string, c syscall.RawConn) error {
return bindConnToInterface(c, network, address, ifIndex, log.Printf)
}
return nil
}
func bindConnToInterface(c syscall.RawConn, network, address string, ifIndex int, logf logger.Logf) error {
v6 := strings.Contains(address, "]:") || strings.HasSuffix(network, "6") // hacky test for v6
proto := unix.IPPROTO_IP
opt := unix.IP_BOUND_IF
if v6 {
proto = unix.IPPROTO_IPV6
opt = unix.IPV6_BOUND_IF
}
var sockErr error
err := c.Control(func(fd uintptr) {
sockErr = unix.SetsockoptInt(int(fd), proto, opt, ifIndex)
})
if sockErr != nil {
logf("[unexpected] netns: bindConnToInterface(%q, %q), v6=%v, index=%v: %v", network, address, v6, ifIndex, sockErr)
}
if err != nil {
return fmt.Errorf("RawConn.Control on %T: %w", c, err)
}
return sockErr
}