tailscale/net/netmon/netmon.go
Brad Fitzpatrick b9adbe2002 net/{interfaces,netmon}, all: merge net/interfaces package into net/netmon
In prep for most of the package funcs in net/interfaces to become
methods in a long-lived netmon.Monitor that can cache things.  (Many
of the funcs are very heavy to call regularly, whereas the long-lived
netmon.Monitor can subscribe to things from the OS and remember
answers to questions it's asked regularly later)

Updates tailscale/corp#10910
Updates tailscale/corp#18960
Updates #7967
Updates #3299

Change-Id: Ie4e8dedb70136af2d611b990b865a822cd1797e5
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
2024-04-28 07:34:52 -07:00

627 lines
16 KiB
Go

// Copyright (c) Tailscale Inc & AUTHORS
// SPDX-License-Identifier: BSD-3-Clause
// Package monitor provides facilities for monitoring network
// interface and route changes. It primarily exists to know when
// portable devices move between different networks.
package netmon
import (
"encoding/json"
"errors"
"net/netip"
"runtime"
"sync"
"time"
"tailscale.com/types/logger"
"tailscale.com/util/clientmetric"
"tailscale.com/util/set"
)
// pollWallTimeInterval is how often we check the time to check
// for big jumps in wall (non-monotonic) time as a backup mechanism
// to get notified of a sleeping device waking back up.
// Usually there are also minor network change events on wake that let
// us check the wall time sooner than this.
const pollWallTimeInterval = 15 * time.Second
// message represents a message returned from an osMon.
type message interface {
// Ignore is whether we should ignore this message.
ignore() bool
}
// osMon is the interface that each operating system-specific
// implementation of the link monitor must implement.
type osMon interface {
Close() error
// Receive returns a new network interface change message. It
// should block until there's either something to return, or
// until the osMon is closed. After a Close, the returned
// error is ignored.
Receive() (message, error)
// IsInterestingInterface reports whether the provided interface should
// be considered for network change events.
IsInterestingInterface(iface string) bool
}
// Monitor represents a monitoring instance.
type Monitor struct {
logf logger.Logf
om osMon // nil means not supported on this platform
change chan bool // send false to wake poller, true to also force ChangeDeltas be sent
stop chan struct{} // closed on Stop
static bool // static Monitor that doesn't actually monitor
// Things that must be set early, before use,
// and not change at runtime.
tsIfName string // tailscale interface name, if known/set ("tailscale0", "utun3", ...)
mu sync.Mutex // guards all following fields
cbs set.HandleSet[ChangeFunc]
ruleDelCB set.HandleSet[RuleDeleteCallback]
ifState *State
gwValid bool // whether gw and gwSelfIP are valid
gw netip.Addr // our gateway's IP
gwSelfIP netip.Addr // our own IP address (that corresponds to gw)
started bool
closed bool
goroutines sync.WaitGroup
wallTimer *time.Timer // nil until Started; re-armed AfterFunc per tick
lastWall time.Time
timeJumped bool // whether we need to send a changed=true after a big time jump
}
// ChangeFunc is a callback function registered with Monitor that's called when the
// network changed.
type ChangeFunc func(*ChangeDelta)
// ChangeDelta describes the difference between two network states.
type ChangeDelta struct {
// Monitor is the network monitor that sent this delta.
Monitor *Monitor
// Old is the old interface state, if known.
// It's nil if the old state is unknown.
// Do not mutate it.
Old *State
// New is the new network state.
// It is always non-nil.
// Do not mutate it.
New *State
// Major is our legacy boolean of whether the network changed in some major
// way.
//
// Deprecated: do not remove. As of 2023-08-23 we're in a renewed effort to
// remove it and ask specific qustions of ChangeDelta instead. Look at Old
// and New (or add methods to ChangeDelta) instead of using Major.
Major bool
// TimeJumped is whether there was a big jump in wall time since the last
// time we checked. This is a hint that a mobile sleeping device might have
// come out of sleep.
TimeJumped bool
// TODO(bradfitz): add some lazy cached fields here as needed with methods
// on *ChangeDelta to let callers ask specific questions
}
// New instantiates and starts a monitoring instance.
// The returned monitor is inactive until it's started by the Start method.
// Use RegisterChangeCallback to get notified of network changes.
func New(logf logger.Logf) (*Monitor, error) {
logf = logger.WithPrefix(logf, "monitor: ")
m := &Monitor{
logf: logf,
change: make(chan bool, 1),
stop: make(chan struct{}),
lastWall: wallTime(),
}
st, err := m.interfaceStateUncached()
if err != nil {
return nil, err
}
m.ifState = st
m.om, err = newOSMon(logf, m)
if err != nil {
return nil, err
}
if m.om == nil {
return nil, errors.New("newOSMon returned nil, nil")
}
return m, nil
}
// NewStatic returns a Monitor that's a one-time snapshot of the network state
// but doesn't actually monitor for changes. It should only be used in tests
// and situations like cleanups or short-lived CLI programs.
func NewStatic() *Monitor {
m := &Monitor{static: true}
if st, err := m.interfaceStateUncached(); err == nil {
m.ifState = st
}
return m
}
// InterfaceState returns the latest snapshot of the machine's network
// interfaces.
//
// The returned value is owned by Mon; it must not be modified.
func (m *Monitor) InterfaceState() *State {
m.mu.Lock()
defer m.mu.Unlock()
return m.ifState
}
func (m *Monitor) interfaceStateUncached() (*State, error) {
return GetState()
}
// SetTailscaleInterfaceName sets the name of the Tailscale interface. For
// example, "tailscale0", "tun0", "utun3", etc.
//
// This must be called only early in tailscaled startup before the monitor is
// used.
func (m *Monitor) SetTailscaleInterfaceName(ifName string) {
m.tsIfName = ifName
}
// GatewayAndSelfIP returns the current network's default gateway, and
// the machine's default IP for that gateway.
//
// It's the same as interfaces.LikelyHomeRouterIP, but it caches the
// result until the monitor detects a network change.
func (m *Monitor) GatewayAndSelfIP() (gw, myIP netip.Addr, ok bool) {
if m.static {
return
}
m.mu.Lock()
defer m.mu.Unlock()
if m.gwValid {
return m.gw, m.gwSelfIP, true
}
gw, myIP, ok = LikelyHomeRouterIP()
changed := false
if ok {
changed = m.gw != gw || m.gwSelfIP != myIP
m.gw, m.gwSelfIP = gw, myIP
m.gwValid = true
}
if changed {
m.logf("gateway and self IP changed: gw=%v self=%v", m.gw, m.gwSelfIP)
}
return gw, myIP, ok
}
// RegisterChangeCallback adds callback to the set of parties to be
// notified (in their own goroutine) when the network state changes.
// To remove this callback, call unregister (or close the monitor).
func (m *Monitor) RegisterChangeCallback(callback ChangeFunc) (unregister func()) {
if m.static {
return func() {}
}
m.mu.Lock()
defer m.mu.Unlock()
handle := m.cbs.Add(callback)
return func() {
m.mu.Lock()
defer m.mu.Unlock()
delete(m.cbs, handle)
}
}
// RuleDeleteCallback is a callback when a Linux IP policy routing
// rule is deleted. The table is the table number (52, 253, 354) and
// priority is the priority order number (for Tailscale rules
// currently: 5210, 5230, 5250, 5270)
type RuleDeleteCallback func(table uint8, priority uint32)
// RegisterRuleDeleteCallback adds callback to the set of parties to be
// notified (in their own goroutine) when a Linux ip rule is deleted.
// To remove this callback, call unregister (or close the monitor).
func (m *Monitor) RegisterRuleDeleteCallback(callback RuleDeleteCallback) (unregister func()) {
if m.static {
return func() {}
}
m.mu.Lock()
defer m.mu.Unlock()
handle := m.ruleDelCB.Add(callback)
return func() {
m.mu.Lock()
defer m.mu.Unlock()
delete(m.ruleDelCB, handle)
}
}
// Start starts the monitor.
// A monitor can only be started & closed once.
func (m *Monitor) Start() {
if m.static {
return
}
m.mu.Lock()
defer m.mu.Unlock()
if m.started || m.closed {
return
}
m.started = true
if shouldMonitorTimeJump {
m.wallTimer = time.AfterFunc(pollWallTimeInterval, m.pollWallTime)
}
if m.om == nil {
return
}
m.goroutines.Add(2)
go m.pump()
go m.debounce()
}
// Close closes the monitor.
func (m *Monitor) Close() error {
if m.static {
return nil
}
m.mu.Lock()
if m.closed {
m.mu.Unlock()
return nil
}
m.closed = true
close(m.stop)
if m.wallTimer != nil {
m.wallTimer.Stop()
}
var err error
if m.om != nil {
err = m.om.Close()
}
started := m.started
m.mu.Unlock()
if started {
m.goroutines.Wait()
}
return err
}
// InjectEvent forces the monitor to pretend there was a network
// change and re-check the state of the network. Any registered
// ChangeFunc callbacks will be called within the event coalescing
// period (under a fraction of a second).
func (m *Monitor) InjectEvent() {
if m.static {
return
}
select {
case m.change <- true:
default:
// Another change signal is already
// buffered. Debounce will wake up soon
// enough.
}
}
// Poll forces the monitor to pretend there was a network
// change and re-check the state of the network.
//
// This is like InjectEvent but only fires ChangeFunc callbacks
// if the network state differed at all.
func (m *Monitor) Poll() {
if m.static {
return
}
select {
case m.change <- false:
default:
}
}
func (m *Monitor) stopped() bool {
select {
case <-m.stop:
return true
default:
return false
}
}
// pump continuously retrieves messages from the connection, notifying
// the change channel of changes, and stopping when a stop is issued.
func (m *Monitor) pump() {
defer m.goroutines.Done()
for !m.stopped() {
msg, err := m.om.Receive()
if err != nil {
if m.stopped() {
return
}
// Keep retrying while we're not closed.
m.logf("error from link monitor: %v", err)
time.Sleep(time.Second)
continue
}
if rdm, ok := msg.(ipRuleDeletedMessage); ok {
m.notifyRuleDeleted(rdm)
continue
}
if msg.ignore() {
continue
}
m.Poll()
}
}
func (m *Monitor) notifyRuleDeleted(rdm ipRuleDeletedMessage) {
m.mu.Lock()
defer m.mu.Unlock()
for _, cb := range m.ruleDelCB {
go cb(rdm.table, rdm.priority)
}
}
// isInterestingInterface reports whether the provided interface should be
// considered when checking for network state changes.
// The ips parameter should be the IPs of the provided interface.
func (m *Monitor) isInterestingInterface(i Interface, ips []netip.Prefix) bool {
if !m.om.IsInterestingInterface(i.Name) {
return false
}
return true
}
// debounce calls the callback function with a delay between events
// and exits when a stop is issued.
func (m *Monitor) debounce() {
defer m.goroutines.Done()
for {
var forceCallbacks bool
select {
case <-m.stop:
return
case forceCallbacks = <-m.change:
}
if newState, err := m.interfaceStateUncached(); err != nil {
m.logf("interfaces.State: %v", err)
} else {
m.handlePotentialChange(newState, forceCallbacks)
}
select {
case <-m.stop:
return
case <-time.After(250 * time.Millisecond):
}
}
}
var (
metricChangeEq = clientmetric.NewCounter("netmon_link_change_eq")
metricChange = clientmetric.NewCounter("netmon_link_change")
metricChangeTimeJump = clientmetric.NewCounter("netmon_link_change_timejump")
metricChangeMajor = clientmetric.NewCounter("netmon_link_change_major")
)
// handlePotentialChange considers whether newState is different enough to wake
// up callers and updates the monitor's state if so.
//
// If forceCallbacks is true, they're always notified.
func (m *Monitor) handlePotentialChange(newState *State, forceCallbacks bool) {
m.mu.Lock()
defer m.mu.Unlock()
oldState := m.ifState
timeJumped := shouldMonitorTimeJump && m.checkWallTimeAdvanceLocked()
if !timeJumped && !forceCallbacks && oldState.Equal(newState) {
// Exactly equal. Nothing to do.
metricChangeEq.Add(1)
return
}
delta := &ChangeDelta{
Monitor: m,
Old: oldState,
New: newState,
TimeJumped: timeJumped,
}
delta.Major = m.IsMajorChangeFrom(oldState, newState)
if delta.Major {
m.gwValid = false
m.ifState = newState
if s1, s2 := oldState.String(), delta.New.String(); s1 == s2 {
m.logf("[unexpected] network state changed, but stringification didn't: %v", s1)
m.logf("[unexpected] old: %s", jsonSummary(oldState))
m.logf("[unexpected] new: %s", jsonSummary(newState))
}
}
// See if we have a queued or new time jump signal.
if timeJumped {
m.resetTimeJumpedLocked()
if !delta.Major {
// Only log if it wasn't an interesting change.
m.logf("time jumped (probably wake from sleep); synthesizing major change event")
delta.Major = true
}
}
metricChange.Add(1)
if delta.Major {
metricChangeMajor.Add(1)
}
if delta.TimeJumped {
metricChangeTimeJump.Add(1)
}
for _, cb := range m.cbs {
go cb(delta)
}
}
// IsMajorChangeFrom reports whether the transition from s1 to s2 is
// a "major" change, where major roughly means it's worth tearing down
// a bunch of connections and rebinding.
//
// TODO(bradiftz): tigten this definition.
func (m *Monitor) IsMajorChangeFrom(s1, s2 *State) bool {
if s1 == nil && s2 == nil {
return false
}
if s1 == nil || s2 == nil {
return true
}
if s1.HaveV6 != s2.HaveV6 ||
s1.HaveV4 != s2.HaveV4 ||
s1.IsExpensive != s2.IsExpensive ||
s1.DefaultRouteInterface != s2.DefaultRouteInterface ||
s1.HTTPProxy != s2.HTTPProxy ||
s1.PAC != s2.PAC {
return true
}
for iname, i := range s1.Interface {
if iname == m.tsIfName {
// Ignore changes in the Tailscale interface itself.
continue
}
ips := s1.InterfaceIPs[iname]
if !m.isInterestingInterface(i, ips) {
continue
}
i2, ok := s2.Interface[iname]
if !ok {
return true
}
ips2, ok := s2.InterfaceIPs[iname]
if !ok {
return true
}
if !i.Equal(i2) || !prefixesMajorEqual(ips, ips2) {
return true
}
}
// Iterate over s2 in case there is a field in s2 that doesn't exist in s1
for iname, i := range s2.Interface {
if iname == m.tsIfName {
// Ignore changes in the Tailscale interface itself.
continue
}
ips := s2.InterfaceIPs[iname]
if !m.isInterestingInterface(i, ips) {
continue
}
i1, ok := s1.Interface[iname]
if !ok {
return true
}
ips1, ok := s1.InterfaceIPs[iname]
if !ok {
return true
}
if !i.Equal(i1) || !prefixesMajorEqual(ips, ips1) {
return true
}
}
return false
}
// prefixesMajorEqual reports whether a and b are equal after ignoring
// boring things like link-local, loopback, and multicast addresses.
func prefixesMajorEqual(a, b []netip.Prefix) bool {
// trim returns a subslice of p with link local unicast,
// loopback, and multicast prefixes removed from the front.
trim := func(p []netip.Prefix) []netip.Prefix {
for len(p) > 0 {
a := p[0].Addr()
if a.IsLinkLocalUnicast() || a.IsLoopback() || a.IsMulticast() {
p = p[1:]
continue
}
break
}
return p
}
for {
a = trim(a)
b = trim(b)
if len(a) == 0 || len(b) == 0 {
return len(a) == 0 && len(b) == 0
}
if a[0] != b[0] {
return false
}
a, b = a[1:], b[1:]
}
}
func jsonSummary(x any) any {
j, err := json.Marshal(x)
if err != nil {
return err
}
return j
}
func wallTime() time.Time {
// From time package's docs: "The canonical way to strip a
// monotonic clock reading is to use t = t.Round(0)."
return time.Now().Round(0)
}
func (m *Monitor) pollWallTime() {
m.mu.Lock()
defer m.mu.Unlock()
if m.closed {
return
}
if m.checkWallTimeAdvanceLocked() {
m.InjectEvent()
}
m.wallTimer.Reset(pollWallTimeInterval)
}
// shouldMonitorTimeJump is whether we keep a regular periodic timer running in
// the background watching for jumps in wall time.
//
// We don't do this on mobile platforms for battery reasons, and because these
// platforms don't really sleep in the same way.
const shouldMonitorTimeJump = runtime.GOOS != "android" && runtime.GOOS != "ios"
// checkWallTimeAdvanceLocked reports whether wall time jumped more than 150% of
// pollWallTimeInterval, indicating we probably just came out of sleep. Once a
// time jump is detected it must be reset by calling resetTimeJumpedLocked.
func (m *Monitor) checkWallTimeAdvanceLocked() bool {
if !shouldMonitorTimeJump {
panic("unreachable") // if callers are correct
}
now := wallTime()
if now.Sub(m.lastWall) > pollWallTimeInterval*3/2 {
m.timeJumped = true // it is reset by debounce.
}
m.lastWall = now
return m.timeJumped
}
// resetTimeJumpedLocked consumes the signal set by checkWallTimeAdvanceLocked.
func (m *Monitor) resetTimeJumpedLocked() {
m.timeJumped = false
}
type ipRuleDeletedMessage struct {
table uint8
priority uint32
}
func (ipRuleDeletedMessage) ignore() bool { return true }