tailscale/ipn/ipnlocal/local.go

// Copyright (c) Tailscale Inc & AUTHORS
// SPDX-License-Identifier: BSD-3-Clause

package ipnlocal

import (
	"context"
	"encoding/base64"
	"encoding/json"
	"errors"
	"fmt"
	"io"
	"log"
	"net"
	"net/http"
	"net/http/httputil"
	"net/netip"
	"net/url"
	"os"
	"os/user"
	"path/filepath"
	"runtime"
	"sort"
	"strconv"
	"strings"
	"sync"
	"sync/atomic"
	"time"

	"go4.org/mem"
	"go4.org/netipx"
	"golang.org/x/exp/slices"
	"gvisor.dev/gvisor/pkg/tcpip"
	"tailscale.com/client/tailscale/apitype"
	"tailscale.com/control/controlclient"
	"tailscale.com/doctor"
	"tailscale.com/doctor/permissions"
	"tailscale.com/doctor/routetable"
	"tailscale.com/envknob"
	"tailscale.com/health"
	"tailscale.com/health/healthmsg"
	"tailscale.com/hostinfo"
	"tailscale.com/ipn"
	"tailscale.com/ipn/ipnauth"
	"tailscale.com/ipn/ipnstate"
	"tailscale.com/ipn/policy"
	"tailscale.com/log/sockstatlog"
	"tailscale.com/logpolicy"
	"tailscale.com/net/dns"
	"tailscale.com/net/dnscache"
	"tailscale.com/net/dnsfallback"
	"tailscale.com/net/interfaces"
	"tailscale.com/net/netns"
	"tailscale.com/net/netutil"
	"tailscale.com/net/tsaddr"
	"tailscale.com/net/tsdial"
	"tailscale.com/paths"
	"tailscale.com/portlist"
	"tailscale.com/syncs"
	"tailscale.com/tailcfg"
	"tailscale.com/tka"
	"tailscale.com/tsd"
	"tailscale.com/types/dnstype"
	"tailscale.com/types/empty"
	"tailscale.com/types/key"
	"tailscale.com/types/logger"
	"tailscale.com/types/logid"
	"tailscale.com/types/netmap"
	"tailscale.com/types/persist"
	"tailscale.com/types/preftype"
	"tailscale.com/types/ptr"
	"tailscale.com/types/views"
	"tailscale.com/util/cmpx"
	"tailscale.com/util/deephash"
	"tailscale.com/util/dnsname"
	"tailscale.com/util/mak"
	"tailscale.com/util/multierr"
	"tailscale.com/util/osshare"
	"tailscale.com/util/set"
	"tailscale.com/util/systemd"
	"tailscale.com/util/uniq"
	"tailscale.com/version"
	"tailscale.com/version/distro"
	"tailscale.com/wgengine"
	"tailscale.com/wgengine/capture"
	"tailscale.com/wgengine/filter"
	"tailscale.com/wgengine/magicsock"
	"tailscale.com/wgengine/router"
	"tailscale.com/wgengine/wgcfg"
	"tailscale.com/wgengine/wgcfg/nmcfg"
)

var controlDebugFlags = getControlDebugFlags()

func getControlDebugFlags() []string {
	if e := envknob.String("TS_DEBUG_CONTROL_FLAGS"); e != "" {
		return strings.Split(e, ",")
	}
	return nil
}

// SSHServer is the interface of the conditionally linked ssh/tailssh.server.
type SSHServer interface {
	HandleSSHConn(net.Conn) error

	// OnPolicyChange is called when the SSH access policy changes,
	// so that existing sessions can be re-evaluated for validity
	// and closed if they'd no longer be accepted.
	OnPolicyChange()

	// Shutdown is called when tailscaled is shutting down.
	Shutdown()
}

type newSSHServerFunc func(logger.Logf, *LocalBackend) (SSHServer, error)

var newSSHServer newSSHServerFunc // or nil

// RegisterNewSSHServer lets the conditionally linked ssh/tailssh package register itself.
func RegisterNewSSHServer(fn newSSHServerFunc) {
	newSSHServer = fn
}

// LocalBackend is the glue between the major pieces of the Tailscale
// network software: the cloud control plane (via controlclient), the
// network data plane (via wgengine), and the user-facing UIs and CLIs
// (collectively called "frontends", via LocalBackend's implementation
// of the Backend interface).
//
// LocalBackend implements the overall state machine for the Tailscale
// application. Frontends, controlclient and wgengine can feed events
// into LocalBackend to advance the state machine, and advancing the
// state machine generates events back out to zero or more components.
type LocalBackend struct {
	// Elements that are thread-safe or constant after construction.
	ctx                   context.Context    // canceled by Close
	ctxCancel             context.CancelFunc // cancels ctx
	logf                  logger.Logf        // general logging
	keyLogf               logger.Logf        // for printing list of peers on change
	statsLogf             logger.Logf        // for printing peers stats on change
	sys                   *tsd.System
	e                     wgengine.Engine // non-nil; TODO(bradfitz): remove; use sys
	pm                    *profileManager
	store                 ipn.StateStore // non-nil; TODO(bradfitz): remove; use sys
	dialer                *tsdial.Dialer // non-nil; TODO(bradfitz): remove; use sys
	backendLogID          logid.PublicID
	unregisterNetMon      func()
	unregisterHealthWatch func()
	portpoll              *portlist.Poller // may be nil
	portpollOnce          sync.Once        // guards starting readPoller
	gotPortPollRes        chan struct{}    // closed upon first readPoller result
	newDecompressor       func() (controlclient.Decompressor, error)
	varRoot               string         // or empty if SetVarRoot never called
	logFlushFunc          func()         // or nil if SetLogFlusher wasn't called
	em                    *expiryManager // non-nil
	sshAtomicBool         atomic.Bool
	shutdownCalled        bool // if Shutdown has been called
	debugSink             *capture.Sink
	sockstatLogger        *sockstatlog.Logger

	// getTCPHandlerForFunnelFlow returns a handler for an incoming TCP flow for
	// the provided srcAddr and dstPort if one exists.
	//
	// srcAddr is the source address of the flow, not the address of the Funnel
	// node relaying the flow.
	// dstPort is the destination port of the flow.
	//
	// It returns nil if there is no known handler for this flow.
	//
	// This is specifically used to handle TCP flows for Funnel connections to tsnet
	// servers.
	//
	// It is set once during initialization, and can be nil if SetTCPHandlerForFunnelFlow
	// is never called.
	getTCPHandlerForFunnelFlow func(srcAddr netip.AddrPort, dstPort uint16) (handler func(net.Conn))

	// lastProfileID tracks the last profile we've seen from the ProfileManager.
	// It's used to detect when the user has changed their profile.
	lastProfileID ipn.ProfileID

	filterAtomic                 atomic.Pointer[filter.Filter]
	containsViaIPFuncAtomic      syncs.AtomicValue[func(netip.Addr) bool]
	shouldInterceptTCPPortAtomic syncs.AtomicValue[func(uint16) bool]
	numClientStatusCalls         atomic.Uint32

	// The mutex protects the following elements.
	mu             sync.Mutex
	filterHash     deephash.Sum
	httpTestClient *http.Client // for controlclient. nil by default, used by tests.
	ccGen          clientGen    // function for producing controlclient; lazily populated
	sshServer      SSHServer    // or nil, initialized lazily.
	notify         func(ipn.Notify)
	cc             controlclient.Client
	ccAuto         *controlclient.Auto // if cc is of type *controlclient.Auto
	machinePrivKey key.MachinePrivate
	tka            *tkaState
	state          ipn.State
	capFileSharing bool // whether netMap contains the file sharing capability
	capTailnetLock bool // whether netMap contains the tailnet lock capability
	// hostinfo is mutated in-place while mu is held.
	hostinfo *tailcfg.Hostinfo
	// netMap is not mutated in-place once set.
	netMap           *netmap.NetworkMap
	nmExpiryTimer    *time.Timer // for updating netMap on node expiry; can be nil
	nodeByAddr       map[netip.Addr]*tailcfg.Node
	activeLogin      string // last logged LoginName from netMap
	engineStatus     ipn.EngineStatus
	endpoints        []tailcfg.Endpoint
	blocked          bool
	keyExpired       bool
	authURL          string // cleared on Notify
	authURLSticky    string // not cleared on Notify
	interact         bool
	egg              bool
	prevIfState      *interfaces.State
	peerAPIServer    *peerAPIServer // or nil
	peerAPIListeners []*peerAPIListener
	loginFlags       controlclient.LoginFlags
	incomingFiles    map[*incomingFile]bool
	fileWaiters      set.HandleSet[context.CancelFunc] // of wake-up funcs
	notifyWatchers   set.HandleSet[chan *ipn.Notify]
	lastStatusTime   time.Time // status.AsOf value of the last processed status update
	// directFileRoot, if non-empty, means to write received files
	// directly to this directory, without staging them in an
	// intermediate buffered directory for "pick-up" later. If
	// empty, the files are received in a daemon-owned location
	// and the localapi is used to enumerate, download, and delete
	// them. This is used on macOS where the GUI lifetime is the
	// same as the Network Extension lifetime and we can thus avoid
	// double-copying files by writing them to the right location
	// immediately.
	// It's also used on several NAS platforms (Synology, TrueNAS, etc)
	// but in that case DoFinalRename is also set true, which moves the
	// *.partial file to its final name on completion.
	directFileRoot          string
	directFileDoFinalRename bool // false on macOS, true on several NAS platforms
	componentLogUntil       map[string]componentLogState

	// ServeConfig fields. (also guarded by mu)
	lastServeConfJSON mem.RO              // last JSON that was parsed into serveConfig
	serveConfig       ipn.ServeConfigView // or !Valid if none

	serveListeners     map[netip.AddrPort]*serveListener // addrPort => serveListener
	serveProxyHandlers sync.Map                          // string (HTTPHandler.Proxy) => *httputil.ReverseProxy

	// statusLock must be held before calling statusChanged.Wait() or
	// statusChanged.Broadcast().
	statusLock    sync.Mutex
	statusChanged *sync.Cond

	// dialPlan is any dial plan that we've received from the control
	// server during a previous connection; it is cleared on logout.
	dialPlan atomic.Pointer[tailcfg.ControlDialPlan]

	// tkaSyncLock is used to make tkaSyncIfNeeded an exclusive
	// section. This is needed to stop two map-responses in quick succession
	// from racing each other through TKA sync logic / RPCs.
	//
	// tkaSyncLock MUST be taken before mu (or inversely, mu must not be held
	// at the moment that tkaSyncLock is taken).
	tkaSyncLock sync.Mutex
}

// clientGen is a func that creates a control plane client.
// It's the type used by LocalBackend.SetControlClientGetterForTesting.
type clientGen func(controlclient.Options) (controlclient.Client, error)

// NewLocalBackend returns a new LocalBackend that is ready to run,
// but is not actually running.
//
// If dialer is nil, a new one is made.
func NewLocalBackend(logf logger.Logf, logID logid.PublicID, sys *tsd.System, loginFlags controlclient.LoginFlags) (*LocalBackend, error) {
	e := sys.Engine.Get()
	store := sys.StateStore.Get()
	dialer := sys.Dialer.Get()

	pm, err := newProfileManager(store, logf)
	if err != nil {
		return nil, err
	}
	if sds, ok := store.(ipn.StateStoreDialerSetter); ok {
		sds.SetDialer(dialer.SystemDial)
	}

	hi := hostinfo.New()
	logf.JSON(1, "Hostinfo", hi)
	envknob.LogCurrent(logf)
	if dialer == nil {
		dialer = &tsdial.Dialer{Logf: logf}
	}

	osshare.SetFileSharingEnabled(false, logf)

	ctx, cancel := context.WithCancel(context.Background())
	portpoll := new(portlist.Poller)

	b := &LocalBackend{
		ctx:            ctx,
		ctxCancel:      cancel,
		logf:           logf,
		keyLogf:        logger.LogOnChange(logf, 5*time.Minute, time.Now),
		statsLogf:      logger.LogOnChange(logf, 5*time.Minute, time.Now),
		sys:            sys,
		e:              e,
		dialer:         dialer,
		store:          store,
		pm:             pm,
		backendLogID:   logID,
		state:          ipn.NoState,
		portpoll:       portpoll,
		em:             newExpiryManager(logf),
		gotPortPollRes: make(chan struct{}),
		loginFlags:     loginFlags,
	}

	netMon := sys.NetMon.Get()
	b.sockstatLogger, err = sockstatlog.NewLogger(logpolicy.LogsDir(logf), logf, logID, netMon)
	if err != nil {
		log.Printf("error setting up sockstat logger: %v", err)
	}
	// Enable sockstats logs only on unstable builds
	if version.IsUnstableBuild() && b.sockstatLogger != nil {
		b.sockstatLogger.SetLoggingEnabled(true)
	}

	// Default filter blocks everything and logs nothing, until Start() is called.
	b.setFilter(filter.NewAllowNone(logf, &netipx.IPSet{}))

	b.setTCPPortsIntercepted(nil)

	b.statusChanged = sync.NewCond(&b.statusLock)
	b.e.SetStatusCallback(b.setWgengineStatus)

	b.prevIfState = netMon.InterfaceState()
	// Call our linkChange code once with the current state, and
	// then also whenever it changes:
	b.linkChange(false, netMon.InterfaceState())
	b.unregisterNetMon = netMon.RegisterChangeCallback(b.linkChange)

	b.unregisterHealthWatch = health.RegisterWatcher(b.onHealthChange)

	if tunWrap, ok := b.sys.Tun.GetOK(); ok {
		tunWrap.PeerAPIPort = b.GetPeerAPIPort
	} else {
		b.logf("[unexpected] failed to wire up PeerAPI port for engine %T", e)
	}

	for _, component := range debuggableComponents {
		key := componentStateKey(component)
		if ut, err := ipn.ReadStoreInt(pm.Store(), key); err == nil {
			if until := time.Unix(ut, 0); until.After(time.Now()) {
				// conditional to avoid log spam at start when off
				b.SetComponentDebugLogging(component, until)
			}
		}
	}

	return b, nil
}

type componentLogState struct {
	until time.Time
	timer *time.Timer // if non-nil, the AfterFunc to disable it
}

var debuggableComponents = []string{
	"magicsock",
	"sockstats",
}

func componentStateKey(component string) ipn.StateKey {
	return ipn.StateKey("_debug_" + component + "_until")
}

// SetComponentDebugLogging sets component's debug logging enabled until the until time.
// If until is in the past, the component's debug logging is disabled.
//
// The following components are recognized:
//
//   - magicsock
//   - sockstats
func (b *LocalBackend) SetComponentDebugLogging(component string, until time.Time) error {
	b.mu.Lock()
	defer b.mu.Unlock()

	var setEnabled func(bool)
	switch component {
	case "magicsock":
		mc, err := b.magicConn()
		if err != nil {
			return err
		}
		setEnabled = mc.SetDebugLoggingEnabled
	case "sockstats":
		if b.sockstatLogger != nil {
			setEnabled = func(v bool) {
				b.sockstatLogger.SetLoggingEnabled(v)
				// Flush (and thus upload) logs when the enabled period ends,
				// so that the logs are available for debugging.
				if !v {
					b.sockstatLogger.Flush()
				}
			}
		}
	}
	if setEnabled == nil || !slices.Contains(debuggableComponents, component) {
		return fmt.Errorf("unknown component %q", component)
	}
	timeUnixOrZero := func(t time.Time) int64 {
		if t.IsZero() {
			return 0
		}
		return t.Unix()
	}
	ipn.PutStoreInt(b.store, componentStateKey(component), timeUnixOrZero(until))
	now := time.Now()
	on := now.Before(until)
	setEnabled(on)
	var onFor time.Duration
	if on {
		onFor = until.Sub(now)
		b.logf("debugging logging for component %q enabled for %v (until %v)", component, onFor.Round(time.Second), until.UTC().Format(time.RFC3339))
	} else {
		b.logf("debugging logging for component %q disabled", component)
	}
	if oldSt, ok := b.componentLogUntil[component]; ok && oldSt.timer != nil {
		oldSt.timer.Stop()
	}
	newSt := componentLogState{until: until}
	if on {
		newSt.timer = time.AfterFunc(onFor, func() {
			// Turn off logging after the timer fires, as long as the state is
			// unchanged when the timer actually fires.
			b.mu.Lock()
			defer b.mu.Unlock()
			if ls := b.componentLogUntil[component]; ls.until.Equal(until) {
				setEnabled(false)
				b.logf("debugging logging for component %q disabled (by timer)", component)
			}
		})
	}
	mak.Set(&b.componentLogUntil, component, newSt)
	return nil
}

// GetComponentDebugLogging gets the time that component's debug logging is
// enabled until, or the zero time if component's time is not currently
// enabled.
func (b *LocalBackend) GetComponentDebugLogging(component string) time.Time {
	b.mu.Lock()
	defer b.mu.Unlock()

	now := time.Now()
	ls := b.componentLogUntil[component]
	if ls.until.IsZero() || ls.until.Before(now) {
		return time.Time{}
	}
	return ls.until
}

// Dialer returns the backend's dialer.
// It is always non-nil.
func (b *LocalBackend) Dialer() *tsdial.Dialer {
	return b.dialer
}

// SetDirectFileRoot sets the directory to download files to directly,
// without buffering them through an intermediate daemon-owned
// tailcfg.UserID-specific directory.
//
// This must be called before the LocalBackend starts being used.
func (b *LocalBackend) SetDirectFileRoot(dir string) {
	b.mu.Lock()
	defer b.mu.Unlock()
	b.directFileRoot = dir
}

// SetDirectFileDoFinalRename sets whether the peerapi file server should rename
// a received "name.partial" file to "name" when the download is complete.
//
// This only applies when SetDirectFileRoot is non-empty.
// The default is false.
func (b *LocalBackend) SetDirectFileDoFinalRename(v bool) {
	b.mu.Lock()
	defer b.mu.Unlock()
	b.directFileDoFinalRename = v
}

// b.mu must be held.
func (b *LocalBackend) maybePauseControlClientLocked() {
	if b.cc == nil {
		return
	}
	networkUp := b.prevIfState.AnyInterfaceUp()
	b.cc.SetPaused((b.state == ipn.Stopped && b.netMap != nil) || !networkUp)
}

// linkChange is our network monitor callback, called whenever the network changes.
// major is whether ifst is different than earlier.
func (b *LocalBackend) linkChange(major bool, ifst *interfaces.State) {
	b.mu.Lock()
	defer b.mu.Unlock()

	hadPAC := b.prevIfState.HasPAC()
	b.prevIfState = ifst
	b.maybePauseControlClientLocked()

	// If the PAC-ness of the network changed, reconfig wireguard+route to
	// add/remove subnets.
	if hadPAC != ifst.HasPAC() {
		b.logf("linkChange: in state %v; PAC changed from %v->%v", b.state, hadPAC, ifst.HasPAC())
		switch b.state {
		case ipn.NoState, ipn.Stopped:
			// Do nothing.
		default:
			go b.authReconfig()
		}
	}

	// If the local network configuration has changed, our filter may
	// need updating to tweak default routes.
	b.updateFilterLocked(b.netMap, b.pm.CurrentPrefs())

	if peerAPIListenAsync && b.netMap != nil && b.state == ipn.Running {
		want := len(b.netMap.Addresses)
		if len(b.peerAPIListeners) < want {
			b.logf("linkChange: peerAPIListeners too low; trying again")
			go b.initPeerAPIListener()
		}
	}
}

func (b *LocalBackend) onHealthChange(sys health.Subsystem, err error) {
	if err == nil {
		b.logf("health(%q): ok", sys)
	} else {
		b.logf("health(%q): error: %v", sys, err)
	}
}

// Shutdown halts the backend and all its sub-components. The backend
// can no longer be used after Shutdown returns.
func (b *LocalBackend) Shutdown() {
	b.mu.Lock()
	if b.shutdownCalled {
		b.mu.Unlock()
		return
	}
	b.shutdownCalled = true

	if b.loginFlags&controlclient.LoginEphemeral != 0 {
		b.mu.Unlock()
		ctx, cancel := context.WithTimeout(b.ctx, 5*time.Second)
		defer cancel()
		b.LogoutSync(ctx) // best effort
		b.mu.Lock()
	}
	cc := b.cc
	if b.sshServer != nil {
		b.sshServer.Shutdown()
		b.sshServer = nil
	}
	b.closePeerAPIListenersLocked()
	if b.debugSink != nil {
		b.e.InstallCaptureHook(nil)
		b.debugSink.Close()
		b.debugSink = nil
	}
	b.mu.Unlock()

	if b.sockstatLogger != nil {
		b.sockstatLogger.Shutdown()
	}

	b.unregisterNetMon()
	b.unregisterHealthWatch()
	if cc != nil {
		cc.Shutdown()
	}
	b.ctxCancel()
	b.e.Close()
	b.e.Wait()
}

func stripKeysFromPrefs(p ipn.PrefsView) ipn.PrefsView {
	if !p.Valid() || !p.Persist().Valid() {
		return p
	}

	p2 := p.AsStruct()
	p2.Persist.LegacyFrontendPrivateMachineKey = key.MachinePrivate{}
	p2.Persist.PrivateNodeKey = key.NodePrivate{}
	p2.Persist.OldPrivateNodeKey = key.NodePrivate{}
	p2.Persist.NetworkLockKey = key.NLPrivate{}
	return p2.View()
}

// Prefs returns a copy of b's current prefs, with any private keys removed.
func (b *LocalBackend) Prefs() ipn.PrefsView {
	b.mu.Lock()
	defer b.mu.Unlock()
	return b.sanitizedPrefsLocked()
}

func (b *LocalBackend) sanitizedPrefsLocked() ipn.PrefsView {
	return stripKeysFromPrefs(b.pm.CurrentPrefs())
}

// Status returns the latest status of the backend and its
// sub-components.
func (b *LocalBackend) Status() *ipnstate.Status {
	sb := &ipnstate.StatusBuilder{WantPeers: true}
	b.UpdateStatus(sb)
	return sb.Status()
}

// StatusWithoutPeers is like Status but omits any details
// of peers.
func (b *LocalBackend) StatusWithoutPeers() *ipnstate.Status {
	sb := &ipnstate.StatusBuilder{WantPeers: false}
	b.UpdateStatus(sb)
	return sb.Status()
}

// UpdateStatus implements ipnstate.StatusUpdater.
func (b *LocalBackend) UpdateStatus(sb *ipnstate.StatusBuilder) {
	b.e.UpdateStatus(sb)
	var extraLocked func(*ipnstate.StatusBuilder)
	if sb.WantPeers {
		extraLocked = b.populatePeerStatusLocked
	}
	b.updateStatus(sb, extraLocked)
}

// updateStatus populates sb with status.
//
// extraLocked, if non-nil, is called while b.mu is still held.
func (b *LocalBackend) updateStatus(sb *ipnstate.StatusBuilder, extraLocked func(*ipnstate.StatusBuilder)) {
	b.mu.Lock()
	defer b.mu.Unlock()
	sb.MutateStatus(func(s *ipnstate.Status) {
		s.Version = version.Long()
		s.TUN = !b.sys.IsNetstack()
		s.BackendState = b.state.String()
		s.AuthURL = b.authURLSticky
		if err := health.OverallError(); err != nil {
			switch e := err.(type) {
			case multierr.Error:
				for _, err := range e.Errors() {
					s.Health = append(s.Health, err.Error())
				}
			default:
				s.Health = append(s.Health, err.Error())
			}
		}
		if m := b.sshOnButUnusableHealthCheckMessageLocked(); m != "" {
			s.Health = append(s.Health, m)
		}
		if version.IsUnstableBuild() {
			s.Health = append(s.Health, "This is an unstable (development) version of Tailscale; frequent updates and bugs are likely")
		}
		if b.netMap != nil {
			s.CertDomains = append([]string(nil), b.netMap.DNS.CertDomains...)
			s.MagicDNSSuffix = b.netMap.MagicDNSSuffix()
			if s.CurrentTailnet == nil {
				s.CurrentTailnet = &ipnstate.TailnetStatus{}
			}
			s.CurrentTailnet.MagicDNSSuffix = b.netMap.MagicDNSSuffix()
			s.CurrentTailnet.MagicDNSEnabled = b.netMap.DNS.Proxied
			s.CurrentTailnet.Name = b.netMap.Domain
			if prefs := b.pm.CurrentPrefs(); prefs.Valid() {
				if !prefs.RouteAll() && b.netMap.AnyPeersAdvertiseRoutes() {
					s.Health = append(s.Health, healthmsg.WarnAcceptRoutesOff)
				}
				if !prefs.ExitNodeID().IsZero() {
					if exitPeer, ok := b.netMap.PeerWithStableID(prefs.ExitNodeID()); ok {
						var online = false
						if exitPeer.Online != nil {
							online = *exitPeer.Online
						}
						s.ExitNodeStatus = &ipnstate.ExitNodeStatus{
							ID:           prefs.ExitNodeID(),
							Online:       online,
							TailscaleIPs: exitPeer.Addresses,
						}
					}
				}
			}
		}
	})
	sb.MutateSelfStatus(func(ss *ipnstate.PeerStatus) {
		ss.Online = health.GetInPollNetMap()
		if b.netMap != nil {
			ss.InNetworkMap = true
			ss.HostName = b.netMap.Hostinfo.Hostname
			ss.DNSName = b.netMap.Name
			ss.UserID = b.netMap.User
			if sn := b.netMap.SelfNode; sn != nil {
				peerStatusFromNode(ss, sn)
				if c := sn.Capabilities; len(c) > 0 {
					ss.Capabilities = append([]string(nil), c...)
				}
			}
		} else {
			ss.HostName, _ = os.Hostname()
		}
		for _, pln := range b.peerAPIListeners {
			ss.PeerAPIURL = append(ss.PeerAPIURL, pln.urlStr)
		}
	})
	// TODO: hostinfo, and its networkinfo
	// TODO: EngineStatus copy (and deprecate it?)

	if extraLocked != nil {
		extraLocked(sb)
	}
}

func (b *LocalBackend) populatePeerStatusLocked(sb *ipnstate.StatusBuilder) {
	if b.netMap == nil {
		return
	}
	for id, up := range b.netMap.UserProfiles {
		sb.AddUser(id, up)
	}
	exitNodeID := b.pm.CurrentPrefs().ExitNodeID()
	for _, p := range b.netMap.Peers {
		var lastSeen time.Time
		if p.LastSeen != nil {
			lastSeen = *p.LastSeen
		}
		var tailscaleIPs = make([]netip.Addr, 0, len(p.Addresses))
		for _, addr := range p.Addresses {
			if addr.IsSingleIP() && tsaddr.IsTailscaleIP(addr.Addr()) {
				tailscaleIPs = append(tailscaleIPs, addr.Addr())
			}
		}
		ps := &ipnstate.PeerStatus{
			InNetworkMap: true,
			UserID:       p.User,
			TailscaleIPs: tailscaleIPs,
			HostName:     p.Hostinfo.Hostname(),
			DNSName:      p.Name,
			OS:           p.Hostinfo.OS(),
			LastSeen:     lastSeen,
			Online:       p.Online != nil && *p.Online,
			ShareeNode:   p.Hostinfo.ShareeNode(),
			ExitNode:     p.StableID != "" && p.StableID == exitNodeID,
			SSH_HostKeys: p.Hostinfo.SSH_HostKeys().AsSlice(),
			Location:     p.Hostinfo.Location(),
		}
		peerStatusFromNode(ps, p)

		p4, p6 := peerAPIPorts(p)
		if u := peerAPIURL(nodeIP(p, netip.Addr.Is4), p4); u != "" {
			ps.PeerAPIURL = append(ps.PeerAPIURL, u)
		}
		if u := peerAPIURL(nodeIP(p, netip.Addr.Is6), p6); u != "" {
			ps.PeerAPIURL = append(ps.PeerAPIURL, u)
		}
		sb.AddPeer(p.Key, ps)
	}
}

// peerStatusFromNode copies fields that exist in the Node struct for
// current node and peers into the provided PeerStatus.
func peerStatusFromNode(ps *ipnstate.PeerStatus, n *tailcfg.Node) {
	ps.ID = n.StableID
	ps.Created = n.Created
	ps.ExitNodeOption = tsaddr.ContainsExitRoutes(n.AllowedIPs)
	if n.Tags != nil {
		v := views.SliceOf(n.Tags)
		ps.Tags = &v
	}
	if n.PrimaryRoutes != nil {
		v := views.IPPrefixSliceOf(n.PrimaryRoutes)
		ps.PrimaryRoutes = &v
	}

	if n.Expired {
		ps.Expired = true
	}
	if t := n.KeyExpiry; !t.IsZero() {
		t = t.Round(time.Second)
		ps.KeyExpiry = &t
	}
}

// WhoIs reports the node and user who owns the node with the given IP:port.
// If the IP address is a Tailscale IP, the provided port may be 0.
// If ok == true, n and u are valid.
func (b *LocalBackend) WhoIs(ipp netip.AddrPort) (n *tailcfg.Node, u tailcfg.UserProfile, ok bool) {
	b.mu.Lock()
	defer b.mu.Unlock()
	n, ok = b.nodeByAddr[ipp.Addr()]
	if !ok {
		var ip netip.Addr
		if ipp.Port() != 0 {
			ip, ok = b.e.WhoIsIPPort(ipp)
		}
		if !ok {
			return nil, u, false
		}
		n, ok = b.nodeByAddr[ip]
		if !ok {
			return nil, u, false
		}
	}
	u, ok = b.netMap.UserProfiles[n.User]
	if !ok {
		return nil, u, false
	}
	return n, u, true
}

// PeerCaps returns the capabilities that remote src IP has to
// ths current node.
func (b *LocalBackend) PeerCaps(src netip.Addr) tailcfg.PeerCapMap {
	b.mu.Lock()
	defer b.mu.Unlock()
	return b.peerCapsLocked(src)
}

func (b *LocalBackend) peerCapsLocked(src netip.Addr) tailcfg.PeerCapMap {
	if b.netMap == nil {
		return nil
	}
	filt := b.filterAtomic.Load()
	if filt == nil {
		return nil
	}
	for _, a := range b.netMap.Addresses {
		if !a.IsSingleIP() {
			continue
		}
		dst := a.Addr()
		if dst.BitLen() == src.BitLen() { // match on family
			return filt.CapsWithValues(src, dst)
		}
	}
	return nil
}

// SetDecompressor sets a decompression function, which must be a zstd
// reader.
//
// This exists because the iOS/Mac NetworkExtension is very resource
// constrained, and the zstd package is too heavy to fit in the
// constrained RSS limit.
func (b *LocalBackend) SetDecompressor(fn func() (controlclient.Decompressor, error)) {
	b.newDecompressor = fn
}

// setClientStatus is the callback invoked by the control client whenever it posts a new status.
// Among other things, this is where we update the netmap, packet filters, DNS and DERP maps.
func (b *LocalBackend) setClientStatus(st controlclient.Status) {
	// The following do not depend on any data for which we need to lock b.
	if st.Err != nil {
		// TODO(crawshaw): display in the UI.
		if errors.Is(st.Err, io.EOF) {
			b.logf("[v1] Received error: EOF")
			return
		}
		b.logf("Received error: %v", st.Err)
		var uerr controlclient.UserVisibleError
		if errors.As(st.Err, &uerr) {
			s := uerr.UserVisibleError()
			b.send(ipn.Notify{ErrMessage: &s})
		}
		return
	}

	// Track the number of calls
	currCall := b.numClientStatusCalls.Add(1)

	b.mu.Lock()

	// Handle node expiry in the netmap
	if st.NetMap != nil {
		now := time.Now()
		b.em.flagExpiredPeers(st.NetMap, now)

		// Always stop the existing netmap timer if we have a netmap;
		// it's possible that we have no nodes expiring, so we should
		// always cancel the timer and then possibly restart it below.
		if b.nmExpiryTimer != nil {
			// Ignore if we can't stop; the atomic check in the
			// AfterFunc (below) will skip running.
			b.nmExpiryTimer.Stop()

			// Nil so we don't attempt to stop on the next netmap
			b.nmExpiryTimer = nil
		}

		// Figure out when the next node in the netmap is expiring so we can
		// start a timer to reconfigure at that point.
		nextExpiry := b.em.nextPeerExpiry(st.NetMap, now)
		if !nextExpiry.IsZero() {
			tmrDuration := nextExpiry.Sub(now) + 10*time.Second
			b.nmExpiryTimer = time.AfterFunc(tmrDuration, func() {
				// Skip if the world has moved on past the
				// saved call (e.g. if we race stopping this
				// timer).
				if b.numClientStatusCalls.Load() != currCall {
					return
				}

				b.logf("setClientStatus: netmap expiry timer triggered after %v", tmrDuration)

				// Call ourselves with the current status again; the logic in
				// setClientStatus will take care of updating the expired field
				// of peers in the netmap.
				b.setClientStatus(st)
			})
		}
	}

	wasBlocked := b.blocked
	keyExpiryExtended := false
	if st.NetMap != nil {
		wasExpired := b.keyExpired
		isExpired := !st.NetMap.Expiry.IsZero() && st.NetMap.Expiry.Before(time.Now())
		if wasExpired && !isExpired {
			keyExpiryExtended = true
		}
		b.keyExpired = isExpired
	}
	b.mu.Unlock()

	if keyExpiryExtended && wasBlocked {
		// Key extended, unblock the engine
		b.blockEngineUpdates(false)
	}

	if st.LoginFinished != nil && wasBlocked {
		// Auth completed, unblock the engine
		b.blockEngineUpdates(false)
		b.authReconfig()
		b.send(ipn.Notify{LoginFinished: &empty.Message{}})
	}

	// Lock b once and do only the things that require locking.
	b.mu.Lock()

	if st.LogoutFinished != nil {
		if p := b.pm.CurrentPrefs(); !p.Persist().Valid() || p.Persist().LoginName() == "" {
			b.mu.Unlock()
			return
		}
		if err := b.pm.DeleteProfile(b.pm.CurrentProfile().ID); err != nil {
			b.logf("error deleting profile: %v", err)
		}
		if err := b.resetForProfileChangeLockedOnEntry(); err != nil {
			b.logf("resetForProfileChangeLockedOnEntry err: %v", err)
		}
		return
	}

	prefsChanged := false
	prefs := b.pm.CurrentPrefs().AsStruct()
	netMap := b.netMap
	interact := b.interact

	if prefs.ControlURL == "" {
		// Once we get a message from the control plane, set
		// our ControlURL pref explicitly. This causes a
		// future "tailscale up" to start checking for
		// implicit setting reverts, which it doesn't do when
		// ControlURL is blank.
		prefs.ControlURL = prefs.ControlURLOrDefault()
		prefsChanged = true
	}
	if st.Persist != nil && st.Persist.Valid() {
		if !prefs.Persist.View().Equals(*st.Persist) {
			prefsChanged = true
			prefs.Persist = st.Persist.AsStruct()
		}
	}
	if st.URL != "" {
		b.authURL = st.URL
		b.authURLSticky = st.URL
	}
	if wasBlocked && st.LoginFinished != nil {
		// Interactive login finished successfully (URL visited).
		// After an interactive login, the user always wants
		// WantRunning.
		if !prefs.WantRunning || prefs.LoggedOut {
			prefsChanged = true
		}
		prefs.WantRunning = true
		prefs.LoggedOut = false
	}
	if findExitNodeIDLocked(prefs, st.NetMap) {
		prefsChanged = true
	}

	// Perform all mutations of prefs based on the netmap here.
	if st.NetMap != nil {
		if b.updatePersistFromNetMapLocked(st.NetMap, prefs) {
			prefsChanged = true
		}
	}
	// Prefs will be written out if stale; this is not safe unless locked or cloned.
	if prefsChanged {
		if err := b.pm.SetPrefs(prefs.View()); err != nil {
			b.logf("Failed to save new controlclient state: %v", err)
		}
	}
	// initTKALocked is dependent on CurrentProfile.ID, which is initialized
	// (for new profiles) on the first call to b.pm.SetPrefs.
	if err := b.initTKALocked(); err != nil {
		b.logf("initTKALocked: %v", err)
	}

	// Perform all reconfiguration based on the netmap here.
	if st.NetMap != nil {
		b.capTailnetLock = hasCapability(st.NetMap, tailcfg.CapabilityTailnetLock)

		b.mu.Unlock() // respect locking rules for tkaSyncIfNeeded
		if err := b.tkaSyncIfNeeded(st.NetMap, prefs.View()); err != nil {
			b.logf("[v1] TKA sync error: %v", err)
		}
		b.mu.Lock()
		if b.tka != nil {
			head, err := b.tka.authority.Head().MarshalText()
			if err != nil {
				b.logf("[v1] error marshalling tka head: %v", err)
			} else {
				b.cc.SetTKAHead(string(head))
			}
		} else {
			b.cc.SetTKAHead("")
		}

		if !envknob.TKASkipSignatureCheck() {
			b.tkaFilterNetmapLocked(st.NetMap)
		}
		b.setNetMapLocked(st.NetMap)
		b.updateFilterLocked(st.NetMap, prefs.View())
	}
	b.mu.Unlock()

	// Now complete the lock-free parts of what we started while locked.
	if prefsChanged {
		b.send(ipn.Notify{Prefs: ptr.To(prefs.View())})
	}

	if st.NetMap != nil {
		if envknob.NoLogsNoSupport() && hasCapability(st.NetMap, tailcfg.CapabilityDataPlaneAuditLogs) {
			msg := "tailnet requires logging to be enabled. Remove --no-logs-no-support from tailscaled command line."
			health.SetLocalLogConfigHealth(errors.New(msg))
			// Connecting to this tailnet without logging is forbidden; boot us outta here.
			b.mu.Lock()
			prefs.WantRunning = false
			p := prefs.View()
			if err := b.pm.SetPrefs(p); err != nil {
				b.logf("Failed to save new controlclient state: %v", err)
			}
			b.mu.Unlock()
			b.send(ipn.Notify{ErrMessage: &msg, Prefs: &p})
			return
		}
		if netMap != nil {
			diff := st.NetMap.ConciseDiffFrom(netMap)
			if strings.TrimSpace(diff) == "" {
				b.logf("[v1] netmap diff: (none)")
			} else {
				b.logf("[v1] netmap diff:\n%v", diff)
			}
		}

		b.e.SetNetworkMap(st.NetMap)
		b.e.SetDERPMap(st.NetMap.DERPMap)

		// Update our cached DERP map
		dnsfallback.UpdateCache(st.NetMap.DERPMap, b.logf)

		b.send(ipn.Notify{NetMap: st.NetMap})
	}
	if st.URL != "" {
		b.logf("Received auth URL: %.20v...", st.URL)
		if interact {
			b.popBrowserAuthNow()
		}
	}
	b.stateMachine()
	// This is currently (2020-07-28) necessary; conditionally disabling it is fragile!
	// This is where netmap information gets propagated to router and magicsock.
	b.authReconfig()
}

// findExitNodeIDLocked updates prefs to reference an exit node by ID, rather
// than by IP. It returns whether prefs was mutated.
func findExitNodeIDLocked(prefs *ipn.Prefs, nm *netmap.NetworkMap) (prefsChanged bool) {
	if nm == nil {
		// No netmap, can't resolve anything.
		return false
	}

	// If we have a desired IP on file, try to find the corresponding
	// node.
	if !prefs.ExitNodeIP.IsValid() {
		return false
	}

	// IP takes precedence over ID, so if both are set, clear ID.
	if prefs.ExitNodeID != "" {
		prefs.ExitNodeID = ""
		prefsChanged = true
	}

	for _, peer := range nm.Peers {
		for _, addr := range peer.Addresses {
			if !addr.IsSingleIP() || addr.Addr() != prefs.ExitNodeIP {
				continue
			}
			// Found the node being referenced, upgrade prefs to
			// reference it directly for next time.
			prefs.ExitNodeID = peer.StableID
			prefs.ExitNodeIP = netip.Addr{}
			return true
		}
	}

	return prefsChanged
}

// setWgengineStatus is the callback by the wireguard engine whenever it posts a new status.
// This updates the endpoints both in the backend and in the control client.
func (b *LocalBackend) setWgengineStatus(s *wgengine.Status, err error) {
	if err != nil {
		b.logf("wgengine status error: %v", err)
		b.broadcastStatusChanged()
		return
	}
	if s == nil {
		b.logf("[unexpected] non-error wgengine update with status=nil: %v", s)
		b.broadcastStatusChanged()
		return
	}

	b.mu.Lock()
	if s.AsOf.Before(b.lastStatusTime) {
		// Don't process a status update that is older than the one we have
		// already processed. (corp#2579)
		b.mu.Unlock()
		return
	}
	b.lastStatusTime = s.AsOf
	es := b.parseWgStatusLocked(s)
	cc := b.cc
	b.engineStatus = es
	needUpdateEndpoints := !endpointsEqual(s.LocalAddrs, b.endpoints)
	if needUpdateEndpoints {
		b.endpoints = append([]tailcfg.Endpoint{}, s.LocalAddrs...)
	}
	b.mu.Unlock()

	if cc != nil {
		if needUpdateEndpoints {
			cc.UpdateEndpoints(s.LocalAddrs)
		}
		b.stateMachine()
	}
	b.broadcastStatusChanged()
	b.send(ipn.Notify{Engine: &es})
}

func (b *LocalBackend) broadcastStatusChanged() {
	// The sync.Cond docs say: "It is allowed but not required for the caller to hold c.L during the call."
	// In this particular case, we must acquire b.statusLock. Otherwise we might broadcast before
	// the waiter (in requestEngineStatusAndWait) starts to wait, in which case
	// the waiter can get stuck indefinitely. See PR 2865.
	b.statusLock.Lock()
	b.statusChanged.Broadcast()
	b.statusLock.Unlock()
}

func endpointsEqual(x, y []tailcfg.Endpoint) bool {
	if len(x) != len(y) {
		return false
	}
	for i := range x {
		if x[i] != y[i] {
			return false
		}
	}
	return true
}

func (b *LocalBackend) SetNotifyCallback(notify func(ipn.Notify)) {
	b.mu.Lock()
	defer b.mu.Unlock()
	b.notify = notify
}

// SetHTTPTestClient sets an alternate HTTP client to use with
// connections to the coordination server. It exists for
// testing. Using nil means to use the default.
func (b *LocalBackend) SetHTTPTestClient(c *http.Client) {
	b.mu.Lock()
	defer b.mu.Unlock()
	b.httpTestClient = c
}

// SetControlClientGetterForTesting sets the func that creates a
// control plane client. It can be called at most once, before Start.
func (b *LocalBackend) SetControlClientGetterForTesting(newControlClient func(controlclient.Options) (controlclient.Client, error)) {
	b.mu.Lock()
	defer b.mu.Unlock()
	if b.ccGen != nil {
		panic("invalid use of SetControlClientGetterForTesting after Start")
	}
	b.ccGen = newControlClient
}

func (b *LocalBackend) getNewControlClientFunc() clientGen {
	b.mu.Lock()
	defer b.mu.Unlock()
	if b.ccGen == nil {
		// Initialize it rather than just returning the
		// default to make any future call to
		// SetControlClientGetterForTesting panic.
		b.ccGen = func(opts controlclient.Options) (controlclient.Client, error) {
			return controlclient.New(opts)
		}
	}
	return b.ccGen