2023-01-27 21:37:20 +00:00
// Copyright (c) Tailscale Inc & AUTHORS
// SPDX-License-Identifier: BSD-3-Clause
2020-02-05 22:16:58 +00:00
package controlclient
import (
2022-08-27 19:55:41 +00:00
"bufio"
2020-02-05 22:16:58 +00:00
"bytes"
"context"
"encoding/binary"
"encoding/json"
"errors"
"fmt"
"io"
"log"
2024-06-20 18:09:50 +00:00
"net"
2020-02-05 22:16:58 +00:00
"net/http"
2022-08-27 19:55:41 +00:00
"net/http/httptest"
all: convert more code to use net/netip directly
perl -i -npe 's,netaddr.IPPrefixFrom,netip.PrefixFrom,' $(git grep -l -F netaddr.)
perl -i -npe 's,netaddr.IPPortFrom,netip.AddrPortFrom,' $(git grep -l -F netaddr. )
perl -i -npe 's,netaddr.IPPrefix,netip.Prefix,g' $(git grep -l -F netaddr. )
perl -i -npe 's,netaddr.IPPort,netip.AddrPort,g' $(git grep -l -F netaddr. )
perl -i -npe 's,netaddr.IP\b,netip.Addr,g' $(git grep -l -F netaddr. )
perl -i -npe 's,netaddr.IPv6Raw\b,netip.AddrFrom16,g' $(git grep -l -F netaddr. )
goimports -w .
Then delete some stuff from the net/netaddr shim package which is no
longer neeed.
Updates #5162
Change-Id: Ia7a86893fe21c7e3ee1ec823e8aba288d4566cd8
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
2022-07-26 04:14:09 +00:00
"net/netip"
2020-04-27 15:18:35 +00:00
"net/url"
2020-02-05 22:16:58 +00:00
"os"
2020-04-02 00:18:39 +00:00
"reflect"
2020-07-28 04:14:28 +00:00
"runtime"
2023-08-30 20:34:22 +00:00
"slices"
2020-02-05 22:16:58 +00:00
"strings"
"sync"
2024-06-20 18:09:50 +00:00
"sync/atomic"
2020-02-05 22:16:58 +00:00
"time"
2021-09-03 20:17:46 +00:00
"go4.org/mem"
2021-06-22 22:29:01 +00:00
"tailscale.com/control/controlknobs"
2022-01-24 18:52:57 +00:00
"tailscale.com/envknob"
2021-02-18 16:58:13 +00:00
"tailscale.com/health"
2021-08-20 17:34:13 +00:00
"tailscale.com/hostinfo"
2021-06-08 15:02:05 +00:00
"tailscale.com/ipn/ipnstate"
2022-04-18 20:43:03 +00:00
"tailscale.com/logtail"
2020-11-11 20:37:53 +00:00
"tailscale.com/net/dnscache"
2021-02-26 20:49:54 +00:00
"tailscale.com/net/dnsfallback"
2023-04-18 21:26:58 +00:00
"tailscale.com/net/netmon"
2022-03-28 17:24:11 +00:00
"tailscale.com/net/netutil"
2020-04-25 20:24:53 +00:00
"tailscale.com/net/tlsdial"
2022-04-27 18:57:59 +00:00
"tailscale.com/net/tsdial"
2020-08-13 22:25:54 +00:00
"tailscale.com/net/tshttpproxy"
2020-02-05 22:16:58 +00:00
"tailscale.com/tailcfg"
2022-10-13 17:43:00 +00:00
"tailscale.com/tka"
2023-08-04 23:29:44 +00:00
"tailscale.com/tstime"
2021-09-03 20:17:46 +00:00
"tailscale.com/types/key"
2020-02-15 03:23:16 +00:00
"tailscale.com/types/logger"
2021-02-05 23:44:46 +00:00
"tailscale.com/types/netmap"
2021-02-05 23:23:01 +00:00
"tailscale.com/types/persist"
2023-08-10 02:56:43 +00:00
"tailscale.com/types/ptr"
2022-10-13 17:43:00 +00:00
"tailscale.com/types/tkatype"
2021-11-16 16:34:25 +00:00
"tailscale.com/util/clientmetric"
2022-03-07 18:55:02 +00:00
"tailscale.com/util/multierr"
2022-06-17 17:09:23 +00:00
"tailscale.com/util/singleflight"
2023-10-13 18:13:41 +00:00
"tailscale.com/util/syspolicy"
2020-11-24 23:35:04 +00:00
"tailscale.com/util/systemd"
2024-04-27 05:06:20 +00:00
"tailscale.com/util/testenv"
2024-03-21 19:20:38 +00:00
"tailscale.com/util/zstdframe"
2020-02-05 22:16:58 +00:00
)
// Direct is the client that connects to a tailcontrol server for a node.
type Direct struct {
2023-12-18 22:57:03 +00:00
httpc * http . Client // HTTP client used to talk to tailcontrol
2024-06-20 18:09:50 +00:00
interceptedDial * atomic . Bool // if non-nil, pointer to bool whether ScreenTime intercepted our dial
2023-12-18 22:57:03 +00:00
dialer * tsdial . Dialer
dnsCache * dnscache . Resolver
controlKnobs * controlknobs . Knobs // always non-nil
serverURL string // URL of the tailcontrol server
clock tstime . Clock
logf logger . Logf
2024-04-27 05:06:20 +00:00
netMon * netmon . Monitor // non-nil
2024-04-26 16:49:30 +00:00
health * health . Tracker
2023-12-18 22:57:03 +00:00
discoPubKey key . DiscoPublic
getMachinePrivKey func ( ) ( key . MachinePrivate , error )
debugFlags [ ] string
skipIPForwardingCheck bool
pinger Pinger
popBrowser func ( url string ) // or nil
c2nHandler http . Handler // or nil
onClientVersion func ( * tailcfg . ClientVersion ) // or nil
onControlTime func ( time . Time ) // or nil
onTailnetDefaultAutoUpdate func ( bool ) // or nil
2024-05-07 04:33:37 +00:00
panicOnUse bool // if true, panic if client is used (for testing)
2024-10-02 15:16:43 +00:00
closedCtx context . Context // alive until Direct.Close is called
closeCtx context . CancelFunc // cancels closedCtx
2020-02-05 22:16:58 +00:00
2022-09-23 17:06:55 +00:00
dialPlan ControlDialPlanner // can be nil
2024-04-02 04:23:27 +00:00
mu sync . Mutex // mutex guards the following fields
serverLegacyKey key . MachinePublic // original ("legacy") nacl crypto_box-based public key; only used for signRegisterRequest on Windows now
serverNoiseKey key . MachinePublic
2022-03-07 18:55:02 +00:00
2022-11-03 07:17:16 +00:00
sfGroup singleflight . Group [ struct { } , * NoiseClient ] // protects noiseClient creation.
noiseClient * NoiseClient
2022-03-07 23:32:53 +00:00
2023-03-09 01:15:47 +00:00
persist persist . PersistView
authKey string
tryingNewKey key . NodePrivate
2023-08-20 21:55:07 +00:00
expiry time . Time // or zero value if none/unknown
2023-03-09 01:15:47 +00:00
hostinfo * tailcfg . Hostinfo // always non-nil
netinfo * tailcfg . NetInfo
endpoints [ ] tailcfg . Endpoint
tkaHead string
lastPingURL string // last PingRequest.URL received, for dup suppression
2020-02-05 22:16:58 +00:00
}
2023-08-28 22:27:39 +00:00
// Observer is implemented by users of the control client (such as LocalBackend)
// to get notified of changes in the control client's status.
type Observer interface {
2023-09-02 19:04:03 +00:00
// SetControlClientStatus is called when the client has a new status to
// report. The Client is provided to allow the Observer to track which
// Client is reporting the status, allowing it to ignore stale status
// reports from previous Clients.
SetControlClientStatus ( Client , Status )
2023-08-28 22:27:39 +00:00
}
2020-02-05 22:16:58 +00:00
type Options struct {
2023-12-18 22:57:03 +00:00
Persist persist . Persist // initial persistent data
GetMachinePrivateKey func ( ) ( key . MachinePrivate , error ) // returns the machine key to use
ServerURL string // URL of the tailcontrol server
AuthKey string // optional node auth key for auto registration
Clock tstime . Clock
Hostinfo * tailcfg . Hostinfo // non-nil passes ownership, nil means to use default using os.Hostname, etc
DiscoPublicKey key . DiscoPublic
Logf logger . Logf
2024-04-27 05:06:20 +00:00
HTTPTestClient * http . Client // optional HTTP client to use (for tests only)
NoiseTestClient * http . Client // optional HTTP client to use for noise RPCs (tests only)
DebugFlags [ ] string // debug settings to send to control
2024-04-26 16:49:30 +00:00
HealthTracker * health . Tracker
2023-12-18 22:57:03 +00:00
PopBrowserURL func ( url string ) // optional func to open browser
OnClientVersion func ( * tailcfg . ClientVersion ) // optional func to inform GUI of client version status
OnControlTime func ( time . Time ) // optional func to notify callers of new time from control
OnTailnetDefaultAutoUpdate func ( bool ) // optional func to inform GUI of default auto-update setting for the tailnet
Dialer * tsdial . Dialer // non-nil
C2NHandler http . Handler // or nil
ControlKnobs * controlknobs . Knobs // or nil to ignore
2022-08-03 23:31:41 +00:00
2023-08-28 22:27:39 +00:00
// Observer is called when there's a change in status to report
// from the control client.
Observer Observer
2022-06-20 01:14:45 +00:00
2021-03-31 18:55:21 +00:00
// SkipIPForwardingCheck declares that the host's IP
// forwarding works and should not be double-checked by the
// controlclient package.
SkipIPForwardingCheck bool
2021-06-08 15:02:05 +00:00
// Pinger optionally specifies the Pinger to use to satisfy
// MapResponse.PingRequest queries from the control plane.
// If nil, PingRequest queries are not answered.
Pinger Pinger
2022-09-23 17:06:55 +00:00
// DialPlan contains and stores a previous dial plan that we received
// from the control server; if nil, we fall back to using DNS.
//
// If we receive a new DialPlan from the server, this value will be
// updated.
DialPlan ControlDialPlanner
}
// ControlDialPlanner is the interface optionally supplied when creating a
// control client to control exactly how TCP connections to the control plane
// are dialed.
//
// It is usually implemented by an atomic.Pointer.
type ControlDialPlanner interface {
// Load returns the current plan for how to connect to control.
//
// The returned plan can be nil. If so, connections should be made by
// resolving the control URL using DNS.
Load ( ) * tailcfg . ControlDialPlan
// Store updates the dial plan with new directions from the control
// server.
//
// The dial plan can span multiple connections to the control server.
// That is, a dial plan received when connected over Wi-Fi is still
// valid for a subsequent connection over LTE after a network switch.
Store ( * tailcfg . ControlDialPlan )
2021-06-08 15:02:05 +00:00
}
2022-06-06 20:07:14 +00:00
// Pinger is the LocalBackend.Ping method.
2021-06-08 15:02:05 +00:00
type Pinger interface {
2022-06-06 20:07:14 +00:00
// Ping is a request to do a ping with the peer handling the given IP.
2023-08-08 12:11:28 +00:00
Ping ( ctx context . Context , ip netip . Addr , pingType tailcfg . PingType , size int ) ( * ipnstate . PingResult , error )
2020-02-05 22:16:58 +00:00
}
2023-08-12 15:18:10 +00:00
// NetmapUpdater is the interface needed by the controlclient to enact change in
// the world as a function of updates received from the network.
type NetmapUpdater interface {
UpdateFullNetmap ( * netmap . NetworkMap )
// TODO(bradfitz): add methods to do fine-grained updates, mutating just
// parts of peers, without implementations of NetmapUpdater needing to do
// the diff themselves between the previous full & next full network maps.
}
2023-09-02 02:28:00 +00:00
// NetmapDeltaUpdater is an optional interface that can be implemented by
// NetmapUpdater implementations to receive delta updates from the controlclient
// rather than just full updates.
type NetmapDeltaUpdater interface {
// UpdateNetmapDelta is called with discrete changes to the network map.
//
// The ok result is whether the implementation was able to apply the
// mutations. It might return false if its internal state doesn't
// support applying them or a NetmapUpdater it's wrapping doesn't
// implement the NetmapDeltaUpdater optional method.
UpdateNetmapDelta ( [ ] netmap . NodeMutation ) ( ok bool )
}
2020-02-05 22:16:58 +00:00
// NewDirect returns a new Direct client.
func NewDirect ( opts Options ) ( * Direct , error ) {
if opts . ServerURL == "" {
return nil , errors . New ( "controlclient.New: no server URL specified" )
}
2021-03-31 15:51:22 +00:00
if opts . GetMachinePrivateKey == nil {
return nil , errors . New ( "controlclient.New: no GetMachinePrivateKey specified" )
2020-09-28 22:28:26 +00:00
}
2024-04-27 05:06:20 +00:00
if opts . Dialer == nil {
if testenv . InTest ( ) {
panic ( "no Dialer" )
}
return nil , errors . New ( "controlclient.New: no Dialer specified" )
}
netMon := opts . Dialer . NetMon ( )
if netMon == nil {
if testenv . InTest ( ) {
panic ( "no NetMon in Dialer" )
}
return nil , errors . New ( "controlclient.New: Dialer has nil NetMon" )
}
2023-09-11 19:03:39 +00:00
if opts . ControlKnobs == nil {
opts . ControlKnobs = & controlknobs . Knobs { }
}
2020-02-05 22:16:58 +00:00
opts . ServerURL = strings . TrimRight ( opts . ServerURL , "/" )
2020-04-27 15:18:35 +00:00
serverURL , err := url . Parse ( opts . ServerURL )
if err != nil {
return nil , err
}
2023-08-04 23:29:44 +00:00
if opts . Clock == nil {
opts . Clock = tstime . StdClock { }
2020-02-05 22:16:58 +00:00
}
if opts . Logf == nil {
// TODO(apenwarr): remove this default and fail instead.
2020-02-25 18:04:20 +00:00
// TODO(bradfitz): ... but then it shouldn't be in Options.
2020-02-05 22:16:58 +00:00
opts . Logf = log . Printf
}
2020-04-25 20:24:53 +00:00
2023-04-30 03:25:37 +00:00
dnsCache := & dnscache . Resolver {
Forward : dnscache . Get ( ) . Forward , // use default cache's forwarder
UseLastGood : true ,
2024-04-27 05:06:20 +00:00
LookupIPFallback : dnsfallback . MakeLookupFunc ( opts . Logf , netMon ) ,
2023-04-30 03:25:37 +00:00
Logf : opts . Logf ,
}
2020-04-26 14:45:42 +00:00
httpc := opts . HTTPTestClient
2021-11-07 21:49:55 +00:00
if httpc == nil && runtime . GOOS == "js" {
// In js/wasm, net/http.Transport (as of Go 1.18) will
// only use the browser's Fetch API if you're using
// the DefaultClient (or a client without dial hooks
// etc set).
httpc = http . DefaultClient
}
2024-06-20 18:09:50 +00:00
var interceptedDial * atomic . Bool
2020-04-26 14:45:42 +00:00
if httpc == nil {
tr := http . DefaultTransport . ( * http . Transport ) . Clone ( )
2020-08-13 22:25:54 +00:00
tr . Proxy = tshttpproxy . ProxyFromEnvironment
2020-08-27 03:02:16 +00:00
tshttpproxy . SetTransportGetProxyConnectHeader ( tr )
2024-04-26 16:49:30 +00:00
tr . TLSClientConfig = tlsdial . Config ( serverURL . Hostname ( ) , opts . HealthTracker , tr . TLSClientConfig )
2024-06-20 18:09:50 +00:00
var dialFunc dialFunc
dialFunc , interceptedDial = makeScreenTimeDetectingDialFunc ( opts . Dialer . SystemDial )
tr . DialContext = dnscache . Dialer ( dialFunc , dnsCache )
tr . DialTLSContext = dnscache . TLSDialer ( dialFunc , dnsCache , tr . TLSClientConfig )
2020-04-26 14:45:42 +00:00
tr . ForceAttemptHTTP2 = true
2022-02-07 21:42:28 +00:00
// Disable implicit gzip compression; the various
// handlers (register, map, set-dns, etc) do their own
// zstd compression per naclbox.
tr . DisableCompression = true
2024-06-20 18:09:50 +00:00
2020-04-26 14:45:42 +00:00
httpc = & http . Client { Transport : tr }
}
2020-04-25 20:24:53 +00:00
2020-02-05 22:16:58 +00:00
c := & Direct {
2023-12-18 22:57:03 +00:00
httpc : httpc ,
2024-06-20 18:09:50 +00:00
interceptedDial : interceptedDial ,
2023-12-18 22:57:03 +00:00
controlKnobs : opts . ControlKnobs ,
getMachinePrivKey : opts . GetMachinePrivateKey ,
serverURL : opts . ServerURL ,
clock : opts . Clock ,
logf : opts . Logf ,
persist : opts . Persist . View ( ) ,
authKey : opts . AuthKey ,
discoPubKey : opts . DiscoPublicKey ,
debugFlags : opts . DebugFlags ,
2024-04-27 05:06:20 +00:00
netMon : netMon ,
2024-04-26 16:49:30 +00:00
health : opts . HealthTracker ,
2023-12-18 22:57:03 +00:00
skipIPForwardingCheck : opts . SkipIPForwardingCheck ,
pinger : opts . Pinger ,
popBrowser : opts . PopBrowserURL ,
onClientVersion : opts . OnClientVersion ,
onTailnetDefaultAutoUpdate : opts . OnTailnetDefaultAutoUpdate ,
onControlTime : opts . OnControlTime ,
c2nHandler : opts . C2NHandler ,
dialer : opts . Dialer ,
dnsCache : dnsCache ,
dialPlan : opts . DialPlan ,
2020-02-05 22:16:58 +00:00
}
2024-10-02 15:16:43 +00:00
c . closedCtx , c . closeCtx = context . WithCancel ( context . Background ( ) )
2020-02-05 22:16:58 +00:00
if opts . Hostinfo == nil {
2021-08-20 17:34:13 +00:00
c . SetHostinfo ( hostinfo . New ( ) )
2020-02-05 22:16:58 +00:00
} else {
2020-02-25 18:04:20 +00:00
c . SetHostinfo ( opts . Hostinfo )
2023-08-10 02:56:43 +00:00
if ni := opts . Hostinfo . NetInfo ; ni != nil {
2022-05-03 22:07:30 +00:00
c . SetNetInfo ( ni )
}
2020-02-05 22:16:58 +00:00
}
2022-09-06 23:34:16 +00:00
if opts . NoiseTestClient != nil {
2022-11-03 07:17:16 +00:00
c . noiseClient = & NoiseClient {
2022-09-06 23:34:16 +00:00
Client : opts . NoiseTestClient ,
}
c . serverNoiseKey = key . NewMachine ( ) . Public ( ) // prevent early error before hitting test client
}
2024-05-07 04:33:37 +00:00
if strings . Contains ( opts . ServerURL , "controlplane.tailscale.com" ) && envknob . Bool ( "TS_PANIC_IF_HIT_MAIN_CONTROL" ) {
c . panicOnUse = true
}
2020-02-05 22:16:58 +00:00
return c , nil
}
2022-03-07 23:32:53 +00:00
// Close closes the underlying Noise connection(s).
func ( c * Direct ) Close ( ) error {
2024-10-02 15:16:43 +00:00
c . closeCtx ( )
2022-03-07 23:32:53 +00:00
c . mu . Lock ( )
defer c . mu . Unlock ( )
if c . noiseClient != nil {
if err := c . noiseClient . Close ( ) ; err != nil {
return err
}
}
c . noiseClient = nil
2024-08-03 12:44:26 +00:00
if tr , ok := c . httpc . Transport . ( * http . Transport ) ; ok {
tr . CloseIdleConnections ( )
}
2022-03-07 23:32:53 +00:00
return nil
}
2020-02-25 18:04:20 +00:00
// SetHostinfo clones the provided Hostinfo and remembers it for the
2020-04-02 00:18:39 +00:00
// next update. It reports whether the Hostinfo has changed.
func ( c * Direct ) SetHostinfo ( hi * tailcfg . Hostinfo ) bool {
2020-02-25 18:04:20 +00:00
if hi == nil {
panic ( "nil Hostinfo" )
}
2023-08-10 02:56:43 +00:00
hi = ptr . To ( * hi )
hi . NetInfo = nil
2020-02-05 22:16:58 +00:00
c . mu . Lock ( )
defer c . mu . Unlock ( )
2020-04-02 00:18:39 +00:00
if hi . Equal ( c . hostinfo ) {
return false
}
2020-02-27 20:20:29 +00:00
c . hostinfo = hi . Clone ( )
2020-10-19 15:30:36 +00:00
j , _ := json . Marshal ( c . hostinfo )
2022-02-12 16:05:24 +00:00
c . logf ( "[v1] HostInfo: %s" , j )
2020-04-02 00:18:39 +00:00
return true
2020-02-05 22:16:58 +00:00
}
2020-03-04 06:21:56 +00:00
// SetNetInfo clones the provided NetInfo and remembers it for the
2020-04-02 00:18:39 +00:00
// next update. It reports whether the NetInfo has changed.
func ( c * Direct ) SetNetInfo ( ni * tailcfg . NetInfo ) bool {
2020-03-04 06:21:56 +00:00
if ni == nil {
panic ( "nil NetInfo" )
}
c . mu . Lock ( )
defer c . mu . Unlock ( )
2022-05-03 22:07:30 +00:00
if reflect . DeepEqual ( ni , c . netinfo ) {
2020-04-02 00:18:39 +00:00
return false
2020-03-04 06:21:56 +00:00
}
2022-05-03 22:07:30 +00:00
c . netinfo = ni . Clone ( )
c . logf ( "NetInfo: %v" , ni )
2020-04-02 00:18:39 +00:00
return true
2020-03-04 06:21:56 +00:00
}
2022-10-27 20:40:31 +00:00
// SetNetInfo stores a new TKA head value for next update.
// It reports whether the TKA head changed.
func ( c * Direct ) SetTKAHead ( tkaHead string ) bool {
c . mu . Lock ( )
defer c . mu . Unlock ( )
if tkaHead == c . tkaHead {
return false
}
c . tkaHead = tkaHead
c . logf ( "tkaHead: %v" , tkaHead )
return true
}
2022-11-09 05:58:10 +00:00
func ( c * Direct ) GetPersist ( ) persist . PersistView {
2020-02-05 22:16:58 +00:00
c . mu . Lock ( )
defer c . mu . Unlock ( )
2022-11-10 14:43:59 +00:00
return c . persist
2020-02-05 22:16:58 +00:00
}
func ( c * Direct ) TryLogout ( ctx context . Context ) error {
2022-02-12 16:05:24 +00:00
c . logf ( "[v1] direct.TryLogout()" )
2020-02-05 22:16:58 +00:00
2022-10-13 17:43:00 +00:00
mustRegen , newURL , _ , err := c . doLogin ( ctx , loginOpt { Logout : true } )
2022-02-12 16:05:24 +00:00
c . logf ( "[v1] TryLogout control response: mustRegen=%v, newURL=%v, err=%v" , mustRegen , newURL , err )
2020-02-05 22:16:58 +00:00
2021-04-08 04:06:31 +00:00
c . mu . Lock ( )
2022-11-10 14:43:59 +00:00
c . persist = new ( persist . Persist ) . View ( )
2021-04-08 04:06:31 +00:00
c . mu . Unlock ( )
return err
2020-02-05 22:16:58 +00:00
}
2024-05-17 21:54:41 +00:00
func ( c * Direct ) TryLogin ( ctx context . Context , flags LoginFlags ) ( url string , err error ) {
2024-05-06 21:08:28 +00:00
if strings . Contains ( c . serverURL , "controlplane.tailscale.com" ) && envknob . Bool ( "TS_PANIC_IF_HIT_MAIN_CONTROL" ) {
2024-05-07 04:33:37 +00:00
panic ( fmt . Sprintf ( "[unexpected] controlclient: TryLogin called on %s; tainted=%v" , c . serverURL , c . panicOnUse ) )
2024-05-06 21:08:28 +00:00
}
2024-05-17 21:54:41 +00:00
c . logf ( "[v1] direct.TryLogin(flags=%v)" , flags )
return c . doLoginOrRegen ( ctx , loginOpt { Flags : flags } )
2020-02-05 22:16:58 +00:00
}
2021-03-31 15:25:39 +00:00
// WaitLoginURL sits in a long poll waiting for the user to authenticate at url.
//
// On success, newURL and err will both be nil.
func ( c * Direct ) WaitLoginURL ( ctx context . Context , url string ) ( newURL string , err error ) {
2022-02-12 16:05:24 +00:00
c . logf ( "[v1] direct.WaitLoginURL" )
2021-04-08 04:06:31 +00:00
return c . doLoginOrRegen ( ctx , loginOpt { URL : url } )
2020-02-05 22:16:58 +00:00
}
2021-04-08 04:06:31 +00:00
func ( c * Direct ) doLoginOrRegen ( ctx context . Context , opt loginOpt ) ( newURL string , err error ) {
2022-10-13 17:43:00 +00:00
mustRegen , url , oldNodeKeySignature , err := c . doLogin ( ctx , opt )
2020-02-05 22:16:58 +00:00
if err != nil {
return url , err
}
2021-04-08 04:06:31 +00:00
if mustRegen {
opt . Regen = true
2022-10-13 17:43:00 +00:00
opt . OldNodeKeySignature = oldNodeKeySignature
_ , url , _ , err = c . doLogin ( ctx , opt )
2020-02-05 22:16:58 +00:00
}
return url , err
}
2022-03-09 22:42:42 +00:00
// SetExpirySooner attempts to shorten the expiry to the specified time.
func ( c * Direct ) SetExpirySooner ( ctx context . Context , expiry time . Time ) error {
c . logf ( "[v1] direct.SetExpirySooner()" )
newURL , err := c . doLoginOrRegen ( ctx , loginOpt { Expiry : & expiry } )
c . logf ( "[v1] SetExpirySooner control response: newURL=%v, err=%v" , newURL , err )
return err
}
2021-04-08 04:06:31 +00:00
type loginOpt struct {
2021-04-08 04:06:31 +00:00
Flags LoginFlags
2022-03-09 22:42:42 +00:00
Regen bool // generate a new nodekey, can be overridden in doLogin
2021-04-08 04:06:31 +00:00
URL string
2022-03-09 22:42:42 +00:00
Logout bool // set the expiry to the far past, expiring the node
// Expiry, if non-nil, attempts to set the node expiry to the
// specified time and cannot be used to extend the expiry.
// It is ignored if Logout is set since Logout works by setting a
// expiry time in the far past.
Expiry * time . Time
2022-10-13 17:43:00 +00:00
// OldNodeKeySignature indicates the former NodeKeySignature
// that must be resigned for the new node-key.
OldNodeKeySignature tkatype . MarshaledSignature
2021-04-08 04:06:31 +00:00
}
2022-05-03 22:07:30 +00:00
// hostInfoLocked returns a Clone of c.hostinfo and c.netinfo.
// It must only be called with c.mu held.
func ( c * Direct ) hostInfoLocked ( ) * tailcfg . Hostinfo {
hi := c . hostinfo . Clone ( )
hi . NetInfo = c . netinfo . Clone ( )
return hi
}
2024-06-20 18:09:50 +00:00
var macOSScreenTime = health . Register ( & health . Warnable {
Code : "macos-screen-time-controlclient" ,
Severity : health . SeverityHigh ,
Title : "Tailscale blocked by Screen Time" ,
Text : func ( args health . Args ) string {
return "macOS Screen Time seems to be blocking Tailscale. Try disabling Screen Time in System Settings > Screen Time > Content & Privacy > Access to Web Content."
} ,
ImpactsConnectivity : true ,
} )
2022-10-13 17:43:00 +00:00
func ( c * Direct ) doLogin ( ctx context . Context , opt loginOpt ) ( mustRegen bool , newURL string , nks tkatype . MarshaledSignature , err error ) {
2024-05-07 04:33:37 +00:00
if c . panicOnUse {
panic ( "tainted client" )
}
2020-02-05 22:16:58 +00:00
c . mu . Lock ( )
2022-11-10 14:43:59 +00:00
persist := c . persist . AsStruct ( )
2020-02-05 22:16:58 +00:00
tryingNewKey := c . tryingNewKey
2024-04-02 04:23:27 +00:00
serverKey := c . serverLegacyKey
2022-03-08 18:53:19 +00:00
serverNoiseKey := c . serverNoiseKey
2024-06-27 11:02:15 +00:00
authKey , isWrapped , wrappedSig , wrappedKey := tka . DecodeWrappedAuthkey ( c . authKey , c . logf )
2022-05-03 22:07:30 +00:00
hi := c . hostInfoLocked ( )
2021-11-10 16:11:14 +00:00
backendLogID := hi . BackendLogID
2023-08-20 21:55:07 +00:00
expired := ! c . expiry . IsZero ( ) && c . expiry . Before ( c . clock . Now ( ) )
2020-02-05 22:16:58 +00:00
c . mu . Unlock ( )
2021-03-31 15:51:22 +00:00
machinePrivKey , err := c . getMachinePrivKey ( )
if err != nil {
2022-10-13 17:43:00 +00:00
return false , "" , nil , fmt . Errorf ( "getMachinePrivKey: %w" , err )
2021-03-31 15:51:22 +00:00
}
if machinePrivKey . IsZero ( ) {
2022-10-13 17:43:00 +00:00
return false , "" , nil , errors . New ( "getMachinePrivKey returned zero key" )
2020-02-05 22:16:58 +00:00
}
2021-04-08 04:06:31 +00:00
regen := opt . Regen
2021-04-08 04:06:31 +00:00
if opt . Logout {
c . logf ( "logging out..." )
} else {
if expired {
c . logf ( "Old key expired -> regen=true" )
systemd . Status ( "key expired; run 'tailscale up' to authenticate" )
regen = true
}
if ( opt . Flags & LoginInteractive ) != 0 {
c . logf ( "LoginInteractive -> regen=true" )
regen = true
}
2020-02-05 22:16:58 +00:00
}
2021-04-08 04:06:31 +00:00
c . logf ( "doLogin(regen=%v, hasUrl=%v)" , regen , opt . URL != "" )
2020-12-30 01:22:56 +00:00
if serverKey . IsZero ( ) {
2022-03-07 18:55:02 +00:00
keys , err := loadServerPubKeys ( ctx , c . httpc , c . serverURL )
2024-06-20 18:09:50 +00:00
if err != nil && c . interceptedDial != nil && c . interceptedDial . Load ( ) {
c . health . SetUnhealthy ( macOSScreenTime , nil )
} else {
c . health . SetHealthy ( macOSScreenTime )
}
2020-02-05 22:16:58 +00:00
if err != nil {
2022-10-13 17:43:00 +00:00
return regen , opt . URL , nil , err
2020-02-05 22:16:58 +00:00
}
2022-04-27 02:11:55 +00:00
c . logf ( "control server key from %s: ts2021=%s, legacy=%v" , c . serverURL , keys . PublicKey . ShortString ( ) , keys . LegacyPublicKey . ShortString ( ) )
2020-02-05 22:16:58 +00:00
c . mu . Lock ( )
2024-04-02 04:23:27 +00:00
c . serverLegacyKey = keys . LegacyPublicKey
2022-03-07 18:55:02 +00:00
c . serverNoiseKey = keys . PublicKey
2020-02-05 22:16:58 +00:00
c . mu . Unlock ( )
2022-03-07 18:55:02 +00:00
serverKey = keys . LegacyPublicKey
2022-03-08 18:53:19 +00:00
serverNoiseKey = keys . PublicKey
2022-03-10 21:19:21 +00:00
2024-04-02 04:23:27 +00:00
// Proactively shut down our TLS TCP connection.
2022-03-10 21:19:21 +00:00
// We're not going to need it and it's nicer to the
// server.
2024-04-02 04:23:27 +00:00
c . httpc . CloseIdleConnections ( )
}
if serverNoiseKey . IsZero ( ) {
return false , "" , nil , errors . New ( "control server is too old; no noise key" )
2020-02-05 22:16:58 +00:00
}
2024-04-02 04:23:27 +00:00
2021-10-28 21:22:51 +00:00
var oldNodeKey key . NodePublic
2021-04-08 04:06:31 +00:00
switch {
case opt . Logout :
2021-10-28 21:22:51 +00:00
tryingNewKey = persist . PrivateNodeKey
2021-04-08 04:06:31 +00:00
case opt . URL != "" :
// Nothing.
case regen || persist . PrivateNodeKey . IsZero ( ) :
2020-04-11 15:35:34 +00:00
c . logf ( "Generating a new nodekey." )
2020-02-05 22:16:58 +00:00
persist . OldPrivateNodeKey = persist . PrivateNodeKey
2021-10-28 21:22:51 +00:00
tryingNewKey = key . NewNode ( )
2021-04-08 04:06:31 +00:00
default :
2020-02-05 22:16:58 +00:00
// Try refreshing the current key first
2021-10-28 21:22:51 +00:00
tryingNewKey = persist . PrivateNodeKey
2020-02-05 22:16:58 +00:00
}
2020-10-01 00:13:41 +00:00
if ! persist . OldPrivateNodeKey . IsZero ( ) {
2021-10-28 21:22:51 +00:00
oldNodeKey = persist . OldPrivateNodeKey . Public ( )
2020-02-05 22:16:58 +00:00
}
2022-11-14 12:29:49 +00:00
if persist . NetworkLockKey . IsZero ( ) {
persist . NetworkLockKey = key . NewNLPrivate ( )
}
nlPub := persist . NetworkLockKey . Public ( )
2020-02-05 22:16:58 +00:00
2022-10-13 17:43:00 +00:00
if tryingNewKey . IsZero ( ) {
if opt . Logout {
return false , "" , nil , errors . New ( "no nodekey to log out" )
}
log . Fatalf ( "tryingNewKey is empty, give up" )
}
var nodeKeySignature tkatype . MarshaledSignature
2022-11-14 12:29:49 +00:00
if ! oldNodeKey . IsZero ( ) && opt . OldNodeKeySignature != nil {
2024-05-09 06:23:03 +00:00
if nodeKeySignature , err = tka . ResignNKS ( persist . NetworkLockKey , tryingNewKey . Public ( ) , opt . OldNodeKeySignature ) ; err != nil {
2022-11-14 12:29:49 +00:00
c . logf ( "Failed re-signing node-key signature: %v" , err )
2021-04-08 04:06:31 +00:00
}
2023-03-02 21:24:26 +00:00
} else if isWrapped {
// We were given a wrapped pre-auth key, which means that in addition
// to being a regular pre-auth key there was a suffix with information to
// generate a tailnet-lock signature.
2024-06-27 11:02:15 +00:00
nodeKeySignature , err = tka . SignByCredential ( wrappedKey , wrappedSig , tryingNewKey . Public ( ) )
2023-03-02 21:24:26 +00:00
if err != nil {
2024-06-27 11:02:15 +00:00
return false , "" , nil , err
2023-03-02 21:24:26 +00:00
}
2020-02-05 22:16:58 +00:00
}
2022-10-13 17:43:00 +00:00
2020-06-15 23:04:12 +00:00
if backendLogID == "" {
2020-02-05 22:16:58 +00:00
err = errors . New ( "hostinfo: BackendLogID missing" )
2022-10-13 17:43:00 +00:00
return regen , opt . URL , nil , err
2020-02-05 22:16:58 +00:00
}
2023-10-13 18:13:41 +00:00
tailnet , err := syspolicy . GetString ( syspolicy . Tailnet , "" )
if err != nil {
c . logf ( "unable to provide Tailnet field in register request. err: %v" , err )
}
2023-08-04 23:29:44 +00:00
now := c . clock . Now ( ) . Round ( time . Second )
2020-02-05 22:16:58 +00:00
request := tailcfg . RegisterRequest {
2022-10-13 17:43:00 +00:00
Version : 1 ,
OldNodeKey : oldNodeKey ,
NodeKey : tryingNewKey . Public ( ) ,
NLKey : nlPub ,
Hostinfo : hi ,
Followup : opt . URL ,
Timestamp : & now ,
Ephemeral : ( opt . Flags & LoginEphemeral ) != 0 ,
NodeKeySignature : nodeKeySignature ,
2023-10-13 18:13:41 +00:00
Tailnet : tailnet ,
2020-02-05 22:16:58 +00:00
}
2021-04-08 04:06:31 +00:00
if opt . Logout {
request . Expiry = time . Unix ( 123 , 0 ) // far in the past
2022-03-09 22:42:42 +00:00
} else if opt . Expiry != nil {
request . Expiry = * opt . Expiry
2021-04-08 04:06:31 +00:00
}
2022-10-13 17:43:00 +00:00
c . logf ( "RegisterReq: onode=%v node=%v fup=%v nks=%v" ,
2020-03-18 22:10:46 +00:00
request . OldNodeKey . ShortString ( ) ,
2022-10-13 17:43:00 +00:00
request . NodeKey . ShortString ( ) , opt . URL != "" , len ( nodeKeySignature ) > 0 )
2024-05-17 21:54:41 +00:00
if authKey != "" {
2024-04-20 17:04:04 +00:00
request . Auth = & tailcfg . RegisterResponseAuth {
2024-05-17 21:54:41 +00:00
AuthKey : authKey ,
2024-04-20 17:04:04 +00:00
}
}
2024-04-02 04:23:27 +00:00
err = signRegisterRequest ( & request , c . serverURL , c . serverLegacyKey , machinePrivKey . Public ( ) )
2021-03-26 14:01:08 +00:00
if err != nil {
// If signing failed, clear all related fields
request . SignatureType = tailcfg . SignatureNone
request . Timestamp = nil
request . DeviceCert = nil
request . Signature = nil
// Don't log the common error types. Signatures are not usually enabled,
// so these are expected.
2021-04-22 18:59:03 +00:00
if ! errors . Is ( err , errCertificateNotConfigured ) && ! errors . Is ( err , errNoCertStore ) {
2021-03-26 14:01:08 +00:00
c . logf ( "RegisterReq sign error: %v" , err )
}
}
2022-09-14 19:49:39 +00:00
if debugRegister ( ) {
2021-04-08 04:06:31 +00:00
j , _ := json . MarshalIndent ( request , "" , "\t" )
c . logf ( "RegisterRequest: %s" , j )
}
2022-03-08 18:53:19 +00:00
// URL and httpc are protocol specific.
2024-04-02 04:23:27 +00:00
request . Version = tailcfg . CurrentCapabilityVersion
httpc , err := c . getNoiseClient ( )
if err != nil {
return regen , opt . URL , nil , fmt . Errorf ( "getNoiseClient: %w" , err )
2022-03-08 18:53:19 +00:00
}
2024-04-02 04:23:27 +00:00
url := fmt . Sprintf ( "%s/machine/register" , c . serverURL )
url = strings . Replace ( url , "http:" , "https:" , 1 )
bodyData , err := encode ( request )
2020-02-05 22:16:58 +00:00
if err != nil {
2022-10-13 17:43:00 +00:00
return regen , opt . URL , nil , err
2020-02-05 22:16:58 +00:00
}
2022-03-08 18:53:19 +00:00
req , err := http . NewRequestWithContext ( ctx , "POST" , url , bytes . NewReader ( bodyData ) )
2020-02-05 22:16:58 +00:00
if err != nil {
2022-10-13 17:43:00 +00:00
return regen , opt . URL , nil , err
2020-02-05 22:16:58 +00:00
}
2024-03-20 13:41:56 +00:00
addLBHeader ( req , request . OldNodeKey )
addLBHeader ( req , request . NodeKey )
2022-03-08 18:53:19 +00:00
res , err := httpc . Do ( req )
2020-02-05 22:16:58 +00:00
if err != nil {
2022-10-13 17:43:00 +00:00
return regen , opt . URL , nil , fmt . Errorf ( "register request: %w" , err )
2020-02-05 22:16:58 +00:00
}
2020-10-06 02:57:14 +00:00
if res . StatusCode != 200 {
2022-09-15 12:06:59 +00:00
msg , _ := io . ReadAll ( res . Body )
2020-10-06 02:57:14 +00:00
res . Body . Close ( )
2022-10-13 17:43:00 +00:00
return regen , opt . URL , nil , fmt . Errorf ( "register request: http %d: %.200s" ,
2020-10-06 02:57:14 +00:00
res . StatusCode , strings . TrimSpace ( string ( msg ) ) )
}
2020-02-05 22:16:58 +00:00
resp := tailcfg . RegisterResponse { }
2024-04-02 04:23:27 +00:00
if err := decode ( res , & resp ) ; err != nil {
2021-03-31 15:51:22 +00:00
c . logf ( "error decoding RegisterResponse with server key %s and machine key %s: %v" , serverKey , machinePrivKey . Public ( ) , err )
2022-10-13 17:43:00 +00:00
return regen , opt . URL , nil , fmt . Errorf ( "register request: %v" , err )
2020-02-05 22:16:58 +00:00
}
2022-09-14 19:49:39 +00:00
if debugRegister ( ) {
2021-04-08 04:06:31 +00:00
j , _ := json . MarshalIndent ( resp , "" , "\t" )
c . logf ( "RegisterResponse: %s" , j )
}
2020-09-28 22:28:26 +00:00
// Log without PII:
c . logf ( "RegisterReq: got response; nodeKeyExpired=%v, machineAuthorized=%v; authURL=%v" ,
resp . NodeKeyExpired , resp . MachineAuthorized , resp . AuthURL != "" )
2020-02-05 22:16:58 +00:00
2021-10-26 17:19:35 +00:00
if resp . Error != "" {
2022-10-13 17:43:00 +00:00
return false , "" , nil , UserVisibleError ( resp . Error )
}
if len ( resp . NodeKeySignature ) > 0 {
return true , "" , resp . NodeKeySignature , nil
2021-10-26 17:19:35 +00:00
}
2022-10-13 17:43:00 +00:00
2020-02-05 22:16:58 +00:00
if resp . NodeKeyExpired {
if regen {
2022-10-13 17:43:00 +00:00
return true , "" , nil , fmt . Errorf ( "weird: regen=true but server says NodeKeyExpired: %v" , request . NodeKey )
2020-02-05 22:16:58 +00:00
}
c . logf ( "server reports new node key %v has expired" ,
2020-03-18 22:10:46 +00:00
request . NodeKey . ShortString ( ) )
2022-10-13 17:43:00 +00:00
return true , "" , nil , nil
2020-02-05 22:16:58 +00:00
}
2022-11-09 05:58:10 +00:00
persist . UserProfile = tailcfg . UserProfile {
ID : resp . User . ID ,
DisplayName : resp . Login . DisplayName ,
ProfilePicURL : resp . Login . ProfilePicURL ,
LoginName : resp . Login . LoginName ,
}
2020-02-05 22:16:58 +00:00
// TODO(crawshaw): RegisterResponse should be able to mechanically
// communicate some extra instructions from the server:
// - new node key required
// - machine key no longer supported
// - user is disabled
if resp . AuthURL != "" {
2020-07-02 16:45:08 +00:00
c . logf ( "AuthURL is %v" , resp . AuthURL )
2020-02-05 22:16:58 +00:00
} else {
2022-02-12 16:05:24 +00:00
c . logf ( "[v1] No AuthURL" )
2020-02-05 22:16:58 +00:00
}
c . mu . Lock ( )
if resp . AuthURL == "" {
// key rotation is complete
2021-10-28 21:22:51 +00:00
persist . PrivateNodeKey = tryingNewKey
2020-02-05 22:16:58 +00:00
} else {
// save it for the retry-with-URL
c . tryingNewKey = tryingNewKey
}
2022-11-10 14:43:59 +00:00
c . persist = persist . View ( )
2020-02-05 22:16:58 +00:00
c . mu . Unlock ( )
if ctx . Err ( ) != nil {
2022-10-13 17:43:00 +00:00
return regen , "" , nil , ctx . Err ( )
2020-02-05 22:16:58 +00:00
}
2022-10-13 17:43:00 +00:00
return false , resp . AuthURL , nil , nil
}
2020-02-14 17:28:29 +00:00
// newEndpoints acquires c.mu and sets the local port and endpoints and reports
// whether they've changed.
//
// It does not retain the provided slice.
2022-06-19 23:31:54 +00:00
func ( c * Direct ) newEndpoints ( endpoints [ ] tailcfg . Endpoint ) ( changed bool ) {
2020-02-05 22:16:58 +00:00
c . mu . Lock ( )
defer c . mu . Unlock ( )
// Nothing new?
2023-08-30 20:34:22 +00:00
if slices . Equal ( c . endpoints , endpoints ) {
2020-02-05 22:16:58 +00:00
return false // unchanged
}
2023-08-30 20:34:22 +00:00
c . logf ( "[v2] client.newEndpoints(%v)" , endpoints )
c . endpoints = slices . Clone ( endpoints )
2020-02-05 22:16:58 +00:00
return true // changed
}
// SetEndpoints updates the list of locally advertised endpoints.
// It won't be replicated to the server until a *fresh* call to PollNetMap().
// You don't need to restart PollNetMap if we return changed==false.
2022-06-19 23:31:54 +00:00
func ( c * Direct ) SetEndpoints ( endpoints [ ] tailcfg . Endpoint ) ( changed bool ) {
2020-02-05 22:16:58 +00:00
// (no log message on function entry, because it clutters the logs
// if endpoints haven't changed. newEndpoints() will log it.)
2022-06-19 23:31:54 +00:00
return c . newEndpoints ( endpoints )
2020-02-05 22:16:58 +00:00
}
2023-08-12 15:18:10 +00:00
// PollNetMap makes a /map request to download the network map, calling
// NetmapUpdater on each update from the control plane.
//
// It always returns a non-nil error describing the reason for the failure or
// why the request ended.
func ( c * Direct ) PollNetMap ( ctx context . Context , nu NetmapUpdater ) error {
return c . sendMapRequest ( ctx , true , nu )
}
type rememberLastNetmapUpdater struct {
last * netmap . NetworkMap
}
func ( nu * rememberLastNetmapUpdater ) UpdateFullNetmap ( nm * netmap . NetworkMap ) {
nu . last = nm
2022-06-19 22:06:33 +00:00
}
2023-08-10 02:56:43 +00:00
// FetchNetMapForTest fetches the netmap once.
func ( c * Direct ) FetchNetMapForTest ( ctx context . Context ) ( * netmap . NetworkMap , error ) {
2023-08-12 15:18:10 +00:00
var nu rememberLastNetmapUpdater
err := c . sendMapRequest ( ctx , false , & nu )
if err == nil && nu . last == nil {
2022-06-19 22:06:33 +00:00
return nil , errors . New ( "[unexpected] sendMapRequest success without callback" )
}
2023-08-12 15:18:10 +00:00
return nu . last , err
2020-12-23 21:03:16 +00:00
}
2023-08-10 02:56:43 +00:00
// SendUpdate makes a /map request to update the server of our latest state, but
// does not fetch anything. It returns an error if the server did not return a
2020-12-23 21:03:16 +00:00
// successful 200 OK response.
2023-08-10 02:56:43 +00:00
func ( c * Direct ) SendUpdate ( ctx context . Context ) error {
return c . sendMapRequest ( ctx , false , nil )
2020-12-23 21:03:16 +00:00
}
2023-08-21 02:52:52 +00:00
// If we go more than watchdogTimeout without hearing from the server,
2021-04-21 21:17:21 +00:00
// end the long poll. We should be receiving a keep alive ping
// every minute.
2023-08-21 02:52:52 +00:00
const watchdogTimeout = 120 * time . Second
2021-04-21 21:17:21 +00:00
2023-08-10 02:56:43 +00:00
// sendMapRequest makes a /map request to download the network map, calling cb
// with each new netmap. If isStreaming, it will poll forever and only returns
// if the context expires or the server returns an error/closes the connection
// and as such always returns a non-nil error.
2023-07-14 03:45:12 +00:00
//
2023-11-05 18:41:13 +00:00
// If nu is nil, OmitPeers will be set to true.
2023-08-12 15:18:10 +00:00
func ( c * Direct ) sendMapRequest ( ctx context . Context , isStreaming bool , nu NetmapUpdater ) error {
2024-05-07 04:33:37 +00:00
if c . panicOnUse {
panic ( "tainted client" )
}
2023-08-12 15:18:10 +00:00
if isStreaming && nu == nil {
2023-08-10 02:56:43 +00:00
panic ( "cb must be non-nil if isStreaming is true" )
}
2021-11-16 16:34:25 +00:00
metricMapRequests . Add ( 1 )
metricMapRequestsActive . Add ( 1 )
defer metricMapRequestsActive . Add ( - 1 )
2023-08-10 02:56:43 +00:00
if isStreaming {
2021-11-16 16:34:25 +00:00
metricMapRequestsPoll . Add ( 1 )
} else {
metricMapRequestsLite . Add ( 1 )
}
2020-02-05 22:16:58 +00:00
c . mu . Lock ( )
persist := c . persist
serverURL := c . serverURL
2022-03-09 18:58:44 +00:00
serverNoiseKey := c . serverNoiseKey
2022-05-03 22:07:30 +00:00
hi := c . hostInfoLocked ( )
2021-11-10 16:11:14 +00:00
backendLogID := hi . BackendLogID
tailcfg: add Endpoint, EndpointType, MapRequest.EndpointType
Track endpoints internally with a new tailcfg.Endpoint type that
includes a typed netaddr.IPPort (instead of just a string) and
includes a type for how that endpoint was discovered (STUN, local,
etc).
Use []tailcfg.Endpoint instead of []string internally.
At the last second, send it to the control server as the existing
[]string for endpoints, but also include a new parallel
MapRequest.EndpointType []tailcfg.EndpointType, so the control server
can start filtering out less-important endpoint changes from
new-enough clients. Notably, STUN-discovered endpoints can be filtered
out from 1.6+ clients, as they can discover them amongst each other
via CallMeMaybe disco exchanges started over DERP. And STUN endpoints
change a lot, causing a lot of MapResposne updates. But portmapped
endpoints are worth keeping for now, as they they work right away
without requiring the firewall traversal extra RTT dance.
End result will be less control->client bandwidth. (despite negligible
increase in client->control bandwidth)
Updates tailscale/corp#1543
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
2021-04-12 20:24:29 +00:00
var epStrs [ ] string
2023-10-01 04:05:02 +00:00
var eps [ ] netip . AddrPort
tailcfg: add Endpoint, EndpointType, MapRequest.EndpointType
Track endpoints internally with a new tailcfg.Endpoint type that
includes a typed netaddr.IPPort (instead of just a string) and
includes a type for how that endpoint was discovered (STUN, local,
etc).
Use []tailcfg.Endpoint instead of []string internally.
At the last second, send it to the control server as the existing
[]string for endpoints, but also include a new parallel
MapRequest.EndpointType []tailcfg.EndpointType, so the control server
can start filtering out less-important endpoint changes from
new-enough clients. Notably, STUN-discovered endpoints can be filtered
out from 1.6+ clients, as they can discover them amongst each other
via CallMeMaybe disco exchanges started over DERP. And STUN endpoints
change a lot, causing a lot of MapResposne updates. But portmapped
endpoints are worth keeping for now, as they they work right away
without requiring the firewall traversal extra RTT dance.
End result will be less control->client bandwidth. (despite negligible
increase in client->control bandwidth)
Updates tailscale/corp#1543
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
2021-04-12 20:24:29 +00:00
var epTypes [ ] tailcfg . EndpointType
for _ , ep := range c . endpoints {
2023-10-01 04:05:02 +00:00
eps = append ( eps , ep . Addr )
tailcfg: add Endpoint, EndpointType, MapRequest.EndpointType
Track endpoints internally with a new tailcfg.Endpoint type that
includes a typed netaddr.IPPort (instead of just a string) and
includes a type for how that endpoint was discovered (STUN, local,
etc).
Use []tailcfg.Endpoint instead of []string internally.
At the last second, send it to the control server as the existing
[]string for endpoints, but also include a new parallel
MapRequest.EndpointType []tailcfg.EndpointType, so the control server
can start filtering out less-important endpoint changes from
new-enough clients. Notably, STUN-discovered endpoints can be filtered
out from 1.6+ clients, as they can discover them amongst each other
via CallMeMaybe disco exchanges started over DERP. And STUN endpoints
change a lot, causing a lot of MapResposne updates. But portmapped
endpoints are worth keeping for now, as they they work right away
without requiring the firewall traversal extra RTT dance.
End result will be less control->client bandwidth. (despite negligible
increase in client->control bandwidth)
Updates tailscale/corp#1543
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
2021-04-12 20:24:29 +00:00
epStrs = append ( epStrs , ep . Addr . String ( ) )
epTypes = append ( epTypes , ep . Type )
}
2020-02-05 22:16:58 +00:00
c . mu . Unlock ( )
2024-04-02 04:23:27 +00:00
if serverNoiseKey . IsZero ( ) {
return errors . New ( "control server is too old; no noise key" )
}
2021-03-31 15:51:22 +00:00
machinePrivKey , err := c . getMachinePrivKey ( )
if err != nil {
return fmt . Errorf ( "getMachinePrivKey: %w" , err )
}
if machinePrivKey . IsZero ( ) {
return errors . New ( "getMachinePrivKey returned zero key" )
}
2022-11-10 14:43:59 +00:00
if persist . PrivateNodeKey ( ) . IsZero ( ) {
2021-02-05 00:23:16 +00:00
return errors . New ( "privateNodeKey is zero" )
}
2020-06-15 23:04:12 +00:00
if backendLogID == "" {
2020-02-05 22:16:58 +00:00
return errors . New ( "hostinfo: BackendLogID missing" )
}
2023-08-10 02:56:43 +00:00
c . logf ( "[v1] PollNetMap: stream=%v ep=%v" , isStreaming , epStrs )
2020-02-05 22:16:58 +00:00
2020-04-11 16:22:33 +00:00
vlogf := logger . Discard
2022-09-14 19:49:39 +00:00
if DevKnob . DumpNetMaps ( ) {
2020-12-21 18:58:06 +00:00
// TODO(bradfitz): update this to use "[v2]" prefix perhaps? but we don't
// want to upload it always.
2020-04-11 16:22:33 +00:00
vlogf = c . logf
}
2024-03-20 13:41:56 +00:00
nodeKey := persist . PublicNodeKey ( )
2021-02-25 05:29:51 +00:00
request := & tailcfg . MapRequest {
2022-03-06 17:32:52 +00:00
Version : tailcfg . CurrentCapabilityVersion ,
2023-09-01 05:28:47 +00:00
KeepAlive : true ,
2024-03-20 13:41:56 +00:00
NodeKey : nodeKey ,
tailcfg: add Endpoint, EndpointType, MapRequest.EndpointType
Track endpoints internally with a new tailcfg.Endpoint type that
includes a typed netaddr.IPPort (instead of just a string) and
includes a type for how that endpoint was discovered (STUN, local,
etc).
Use []tailcfg.Endpoint instead of []string internally.
At the last second, send it to the control server as the existing
[]string for endpoints, but also include a new parallel
MapRequest.EndpointType []tailcfg.EndpointType, so the control server
can start filtering out less-important endpoint changes from
new-enough clients. Notably, STUN-discovered endpoints can be filtered
out from 1.6+ clients, as they can discover them amongst each other
via CallMeMaybe disco exchanges started over DERP. And STUN endpoints
change a lot, causing a lot of MapResposne updates. But portmapped
endpoints are worth keeping for now, as they they work right away
without requiring the firewall traversal extra RTT dance.
End result will be less control->client bandwidth. (despite negligible
increase in client->control bandwidth)
Updates tailscale/corp#1543
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
2021-04-12 20:24:29 +00:00
DiscoKey : c . discoPubKey ,
2023-10-01 04:05:02 +00:00
Endpoints : eps ,
tailcfg: add Endpoint, EndpointType, MapRequest.EndpointType
Track endpoints internally with a new tailcfg.Endpoint type that
includes a typed netaddr.IPPort (instead of just a string) and
includes a type for how that endpoint was discovered (STUN, local,
etc).
Use []tailcfg.Endpoint instead of []string internally.
At the last second, send it to the control server as the existing
[]string for endpoints, but also include a new parallel
MapRequest.EndpointType []tailcfg.EndpointType, so the control server
can start filtering out less-important endpoint changes from
new-enough clients. Notably, STUN-discovered endpoints can be filtered
out from 1.6+ clients, as they can discover them amongst each other
via CallMeMaybe disco exchanges started over DERP. And STUN endpoints
change a lot, causing a lot of MapResposne updates. But portmapped
endpoints are worth keeping for now, as they they work right away
without requiring the firewall traversal extra RTT dance.
End result will be less control->client bandwidth. (despite negligible
increase in client->control bandwidth)
Updates tailscale/corp#1543
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
2021-04-12 20:24:29 +00:00
EndpointTypes : epTypes ,
2023-08-10 02:56:43 +00:00
Stream : isStreaming ,
2021-11-10 16:11:14 +00:00
Hostinfo : hi ,
tailcfg: add Endpoint, EndpointType, MapRequest.EndpointType
Track endpoints internally with a new tailcfg.Endpoint type that
includes a typed netaddr.IPPort (instead of just a string) and
includes a type for how that endpoint was discovered (STUN, local,
etc).
Use []tailcfg.Endpoint instead of []string internally.
At the last second, send it to the control server as the existing
[]string for endpoints, but also include a new parallel
MapRequest.EndpointType []tailcfg.EndpointType, so the control server
can start filtering out less-important endpoint changes from
new-enough clients. Notably, STUN-discovered endpoints can be filtered
out from 1.6+ clients, as they can discover them amongst each other
via CallMeMaybe disco exchanges started over DERP. And STUN endpoints
change a lot, causing a lot of MapResposne updates. But portmapped
endpoints are worth keeping for now, as they they work right away
without requiring the firewall traversal extra RTT dance.
End result will be less control->client bandwidth. (despite negligible
increase in client->control bandwidth)
Updates tailscale/corp#1543
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
2021-04-12 20:24:29 +00:00
DebugFlags : c . debugFlags ,
2023-08-12 15:18:10 +00:00
OmitPeers : nu == nil ,
2022-10-27 20:40:31 +00:00
TKAHead : c . tkaHead ,
2020-02-05 22:16:58 +00:00
}
2021-02-18 16:58:13 +00:00
var extraDebugFlags [ ] string
2023-04-18 21:26:58 +00:00
if hi != nil && c . netMon != nil && ! c . skipIPForwardingCheck &&
ipForwardingBroken ( hi . RoutableIPs , c . netMon . InterfaceState ( ) ) {
2021-02-18 16:58:13 +00:00
extraDebugFlags = append ( extraDebugFlags , "warn-ip-forwarding-off" )
}
2024-04-26 16:49:30 +00:00
if c . health . RouterHealth ( ) != nil {
2021-02-18 16:58:13 +00:00
extraDebugFlags = append ( extraDebugFlags , "warn-router-unhealthy" )
}
2024-04-26 16:49:30 +00:00
extraDebugFlags = c . health . AppendWarnableDebugFlags ( extraDebugFlags )
2021-11-10 16:09:29 +00:00
if hostinfo . DisabledEtcAptSource ( ) {
extraDebugFlags = append ( extraDebugFlags , "warn-etc-apt-source-disabled" )
}
2021-02-18 16:58:13 +00:00
if len ( extraDebugFlags ) > 0 {
2020-11-04 21:48:50 +00:00
old := request . DebugFlags
2021-02-18 16:58:13 +00:00
request . DebugFlags = append ( old [ : len ( old ) : len ( old ) ] , extraDebugFlags ... )
2020-11-04 21:48:50 +00:00
}
2023-09-10 14:24:51 +00:00
request . Compress = "zstd"
2020-02-05 22:16:58 +00:00
2024-04-02 04:23:27 +00:00
bodyData , err := encode ( request )
2020-02-05 22:16:58 +00:00
if err != nil {
2020-04-11 16:22:33 +00:00
vlogf ( "netmap: encode: %v" , err )
2020-02-05 22:16:58 +00:00
return err
}
2020-10-14 21:01:33 +00:00
ctx , cancel := context . WithCancel ( ctx )
defer cancel ( )
2021-09-03 20:17:46 +00:00
machinePubKey := machinePrivKey . Public ( )
2023-08-04 23:29:44 +00:00
t0 := c . clock . Now ( )
2020-10-14 21:01:33 +00:00
2024-04-02 04:23:27 +00:00
httpc , err := c . getNoiseClient ( )
if err != nil {
return fmt . Errorf ( "getNoiseClient: %w" , err )
2022-03-09 18:58:44 +00:00
}
2024-04-02 04:23:27 +00:00
url := fmt . Sprintf ( "%s/machine/map" , serverURL )
url = strings . Replace ( url , "http:" , "https:" , 1 )
2022-03-09 18:58:44 +00:00
2024-03-27 22:04:38 +00:00
// Create a watchdog timer that breaks the connection if we don't receive a
// MapResponse from the network at least once every two minutes. The
// watchdog timer is stopped every time we receive a MapResponse (so it
// doesn't run when we're processing a MapResponse message, including any
// long-running requested operations like Debug.Sleep) and is reset whenever
// we go back to blocking on network reads.
// The watchdog timer also covers the initial request (effectively the
// pre-body and initial-body read timeouts) as we do not have any other
// keep-alive mechanism for the initial request.
watchdogTimer , watchdogTimedOut := c . clock . NewTimer ( watchdogTimeout )
defer watchdogTimer . Stop ( )
go func ( ) {
select {
case <- ctx . Done ( ) :
vlogf ( "netmap: ending timeout goroutine" )
return
case <- watchdogTimedOut :
c . logf ( "map response long-poll timed out!" )
cancel ( )
return
}
} ( )
2022-03-09 18:58:44 +00:00
req , err := http . NewRequestWithContext ( ctx , "POST" , url , bytes . NewReader ( bodyData ) )
2020-02-05 22:16:58 +00:00
if err != nil {
return err
}
2024-03-20 13:41:56 +00:00
addLBHeader ( req , nodeKey )
2020-02-05 22:16:58 +00:00
2022-03-09 18:58:44 +00:00
res , err := httpc . Do ( req )
2020-02-05 22:16:58 +00:00
if err != nil {
2020-04-11 16:22:33 +00:00
vlogf ( "netmap: Do: %v" , err )
2020-02-05 22:16:58 +00:00
return err
}
2020-04-11 16:22:33 +00:00
vlogf ( "netmap: Do = %v after %v" , res . StatusCode , time . Since ( t0 ) . Round ( time . Millisecond ) )
2020-02-05 22:16:58 +00:00
if res . StatusCode != 200 {
2022-09-15 12:06:59 +00:00
msg , _ := io . ReadAll ( res . Body )
2020-02-05 22:16:58 +00:00
res . Body . Close ( )
2020-10-06 02:57:14 +00:00
return fmt . Errorf ( "initial fetch failed %d: %.200s" ,
2020-02-05 22:16:58 +00:00
res . StatusCode , strings . TrimSpace ( string ( msg ) ) )
}
defer res . Body . Close ( )
2024-04-26 16:49:30 +00:00
c . health . NoteMapRequestHeard ( request )
2024-03-27 22:04:38 +00:00
watchdogTimer . Reset ( watchdogTimeout )
2021-02-25 05:29:51 +00:00
2023-08-12 15:18:10 +00:00
if nu == nil {
2022-09-15 12:06:59 +00:00
io . Copy ( io . Discard , res . Body )
2020-12-23 21:03:16 +00:00
return nil
}
2023-09-11 19:03:39 +00:00
sess := newMapSession ( persist . PrivateNodeKey ( ) , nu , c . controlKnobs )
2023-08-21 02:52:52 +00:00
defer sess . Close ( )
sess . cancel = cancel
2021-04-19 02:49:25 +00:00
sess . logf = c . logf
sess . vlogf = vlogf
2023-08-21 02:52:52 +00:00
sess . altClock = c . clock
2021-04-19 02:49:25 +00:00
sess . machinePubKey = machinePubKey
2023-08-21 02:52:52 +00:00
sess . onDebug = c . handleDebugMessage
sess . onSelfNodeChanged = func ( nm * netmap . NetworkMap ) {
c . mu . Lock ( )
defer c . mu . Unlock ( )
// If we are the ones who last updated persist, then we can update it
// again. Otherwise, we should not touch it. Also, it's only worth
// change it if the Node info changed.
if persist == c . persist {
newPersist := persist . AsStruct ( )
2023-08-21 17:53:57 +00:00
newPersist . NodeID = nm . SelfNode . StableID ( )
2023-08-21 02:52:52 +00:00
newPersist . UserProfile = nm . UserProfiles [ nm . User ( ) ]
c . persist = newPersist . View ( )
persist = c . persist
}
c . expiry = nm . Expiry
}
2023-11-05 18:41:13 +00:00
2023-08-21 02:52:52 +00:00
// gotNonKeepAliveMessage is whether we've yet received a MapResponse message without
// KeepAlive set.
var gotNonKeepAliveMessage bool
2020-05-17 16:51:38 +00:00
2020-02-05 22:16:58 +00:00
// If allowStream, then the server will use an HTTP long poll to
// return incremental results. There is always one response right
// away, followed by a delay, and eventually others.
// If !allowStream, it'll still send the first result in exactly
// the same format before just closing the connection.
// We can use this same read loop either way.
var msg [ ] byte
2023-11-05 15:31:06 +00:00
for mapResIdx := 0 ; mapResIdx == 0 || isStreaming ; mapResIdx ++ {
2023-11-05 18:41:13 +00:00
watchdogTimer . Reset ( watchdogTimeout )
2023-08-21 02:52:52 +00:00
vlogf ( "netmap: starting size read after %v (poll %v)" , time . Since ( t0 ) . Round ( time . Millisecond ) , mapResIdx )
2020-02-05 22:16:58 +00:00
var siz [ 4 ] byte
if _ , err := io . ReadFull ( res . Body , siz [ : ] ) ; err != nil {
2020-04-11 16:22:33 +00:00
vlogf ( "netmap: size read error after %v: %v" , time . Since ( t0 ) . Round ( time . Millisecond ) , err )
2020-02-05 22:16:58 +00:00
return err
}
size := binary . LittleEndian . Uint32 ( siz [ : ] )
2020-04-11 16:22:33 +00:00
vlogf ( "netmap: read size %v after %v" , size , time . Since ( t0 ) . Round ( time . Millisecond ) )
2020-02-05 22:16:58 +00:00
msg = append ( msg [ : 0 ] , make ( [ ] byte , size ) ... )
if _ , err := io . ReadFull ( res . Body , msg ) ; err != nil {
2020-04-11 16:22:33 +00:00
vlogf ( "netmap: body read error: %v" , err )
2020-02-05 22:16:58 +00:00
return err
}
2020-04-11 16:22:33 +00:00
vlogf ( "netmap: read body after %v" , time . Since ( t0 ) . Round ( time . Millisecond ) )
2020-02-05 22:16:58 +00:00
var resp tailcfg . MapResponse
2024-04-02 04:23:27 +00:00
if err := c . decodeMsg ( msg , & resp ) ; err != nil {
2024-01-05 12:06:12 +00:00
vlogf ( "netmap: decode error: %v" , err )
2020-02-05 22:16:58 +00:00
return err
}
2023-11-05 18:41:13 +00:00
watchdogTimer . Stop ( )
2020-08-08 03:44:04 +00:00
2021-11-16 16:34:25 +00:00
metricMapResponseMessages . Add ( 1 )
2023-08-10 02:56:43 +00:00
if isStreaming {
2024-04-26 16:49:30 +00:00
c . health . GotStreamedMapResponse ( )
2021-02-25 05:29:51 +00:00
}
2021-06-15 19:28:34 +00:00
if pr := resp . PingRequest ; pr != nil && c . isUniquePingRequest ( pr ) {
2021-11-16 16:34:25 +00:00
metricMapResponsePings . Add ( 1 )
2022-08-10 21:52:02 +00:00
go c . answerPing ( pr )
2021-03-05 04:54:44 +00:00
}
2022-03-21 21:10:25 +00:00
if u := resp . PopBrowserURL ; u != "" && u != sess . lastPopBrowserURL {
sess . lastPopBrowserURL = u
if c . popBrowser != nil {
c . logf ( "netmap: control says to open URL %v; opening..." , u )
c . popBrowser ( u )
} else {
c . logf ( "netmap: control says to open URL %v; no popBrowser func" , u )
}
}
2022-11-24 03:13:41 +00:00
if resp . ClientVersion != nil && c . onClientVersion != nil {
c . onClientVersion ( resp . ClientVersion )
}
2022-02-18 18:14:14 +00:00
if resp . ControlTime != nil && ! resp . ControlTime . IsZero ( ) {
2022-02-18 21:00:08 +00:00
c . logf . JSON ( 1 , "controltime" , resp . ControlTime . UTC ( ) )
2023-01-11 20:59:08 +00:00
if c . onControlTime != nil {
c . onControlTime ( * resp . ControlTime )
}
2022-02-17 02:36:04 +00:00
}
2020-02-05 22:16:58 +00:00
if resp . KeepAlive {
2020-04-11 16:22:33 +00:00
vlogf ( "netmap: got keep-alive" )
2020-08-07 04:24:31 +00:00
} else {
vlogf ( "netmap: got new map" )
}
2022-09-23 17:06:55 +00:00
if resp . ControlDialPlan != nil {
if c . dialPlan != nil {
c . logf ( "netmap: got new dial plan from control" )
c . dialPlan . Store ( resp . ControlDialPlan )
} else {
c . logf ( "netmap: [unexpected] new dial plan; nowhere to store it" )
}
}
2020-08-07 04:24:31 +00:00
if resp . KeepAlive {
2021-11-16 16:34:25 +00:00
metricMapResponseKeepAlives . Add ( 1 )
2020-02-05 22:16:58 +00:00
continue
}
2023-12-18 22:57:03 +00:00
if au , ok := resp . DefaultAutoUpdate . Get ( ) ; ok {
if c . onTailnetDefaultAutoUpdate != nil {
c . onTailnetDefaultAutoUpdate ( au )
}
}
2020-08-08 03:44:04 +00:00
2021-11-16 16:34:25 +00:00
metricMapResponseMap . Add ( 1 )
2023-08-21 02:52:52 +00:00
if gotNonKeepAliveMessage {
// If we've already seen a non-keep-alive message, this is a delta update.
2021-11-16 16:34:25 +00:00
metricMapResponseMapDelta . Add ( 1 )
2023-08-21 02:52:52 +00:00
} else if resp . Node == nil {
// The very first non-keep-alive message should have Node populated.
c . logf ( "initial MapResponse lacked Node" )
return errors . New ( "initial MapResponse lacked node" )
2021-11-16 16:34:25 +00:00
}
2023-08-21 02:52:52 +00:00
gotNonKeepAliveMessage = true
2021-11-16 16:34:25 +00:00
2023-08-21 02:52:52 +00:00
if err := sess . HandleNonKeepAliveMapResponse ( ctx , & resp ) ; err != nil {
return err
2023-08-04 05:29:27 +00:00
}
2020-02-05 22:16:58 +00:00
}
if ctx . Err ( ) != nil {
return ctx . Err ( )
}
return nil
}
2023-11-05 18:41:13 +00:00
func ( c * Direct ) handleDebugMessage ( ctx context . Context , debug * tailcfg . Debug ) error {
2023-08-21 02:52:52 +00:00
if code := debug . Exit ; code != nil {
c . logf ( "exiting process with status %v per controlplane" , * code )
os . Exit ( * code )
}
if debug . DisableLogTail {
logtail . Disable ( )
envknob . SetNoLogsNoSupport ( )
}
if sleep := time . Duration ( debug . SleepSeconds * float64 ( time . Second ) ) ; sleep > 0 {
2023-11-05 18:41:13 +00:00
if err := sleepAsRequested ( ctx , c . logf , sleep , c . clock ) ; err != nil {
2023-08-21 02:52:52 +00:00
return err
}
}
return nil
}
2023-08-20 22:17:25 +00:00
// initDisplayNames mutates any tailcfg.Nodes in resp to populate their display names,
// calling InitDisplayNames on each.
//
// The magicDNSSuffix used is based on selfNode.
func initDisplayNames ( selfNode tailcfg . NodeView , resp * tailcfg . MapResponse ) {
if resp . Node == nil && len ( resp . Peers ) == 0 && len ( resp . PeersChanged ) == 0 {
// Fast path for a common case (delta updates). No need to compute
// magicDNSSuffix.
return
}
magicDNSSuffix := netmap . MagicDNSSuffixOfNodeName ( selfNode . Name ( ) )
if resp . Node != nil {
resp . Node . InitDisplayNames ( magicDNSSuffix )
}
for _ , n := range resp . Peers {
n . InitDisplayNames ( magicDNSSuffix )
}
for _ , n := range resp . PeersChanged {
n . InitDisplayNames ( magicDNSSuffix )
}
}
2024-04-02 04:23:27 +00:00
// decode JSON decodes the res.Body into v.
func decode ( res * http . Response , v any ) error {
2020-02-05 22:16:58 +00:00
defer res . Body . Close ( )
2022-09-15 12:06:59 +00:00
msg , err := io . ReadAll ( io . LimitReader ( res . Body , 1 << 20 ) )
2020-02-05 22:16:58 +00:00
if err != nil {
return err
}
if res . StatusCode != 200 {
return fmt . Errorf ( "%d: %v" , res . StatusCode , string ( msg ) )
}
2024-04-02 04:23:27 +00:00
return json . Unmarshal ( msg , v )
2020-02-05 22:16:58 +00:00
}
2021-04-08 04:06:31 +00:00
var (
2022-09-14 19:49:39 +00:00
debugMap = envknob . RegisterBool ( "TS_DEBUG_MAP" )
debugRegister = envknob . RegisterBool ( "TS_DEBUG_REGISTER" )
2021-04-08 04:06:31 +00:00
)
2020-09-15 16:54:52 +00:00
2020-11-12 21:31:29 +00:00
var jsonEscapedZero = [ ] byte ( ` \u0000 ` )
2022-03-09 18:58:44 +00:00
// decodeMsg is responsible for uncompressing msg and unmarshaling into v.
2024-04-02 04:23:27 +00:00
func ( c * Direct ) decodeMsg ( compressedMsg [ ] byte , v any ) error {
b , err := zstdframe . AppendDecode ( nil , compressedMsg )
2023-09-10 14:24:51 +00:00
if err != nil {
return err
2020-02-05 22:16:58 +00:00
}
2022-09-14 19:49:39 +00:00
if debugMap ( ) {
2020-09-15 16:54:52 +00:00
var buf bytes . Buffer
json . Indent ( & buf , b , "" , " " )
log . Printf ( "MapResponse: %s" , buf . Bytes ( ) )
}
2020-11-12 21:31:29 +00:00
if bytes . Contains ( b , jsonEscapedZero ) {
log . Printf ( "[unexpected] zero byte in controlclient.Direct.decodeMsg into %T: %q" , v , b )
}
2020-02-05 22:16:58 +00:00
if err := json . Unmarshal ( b , v ) ; err != nil {
return fmt . Errorf ( "response: %v" , err )
}
return nil
}
2024-04-02 04:23:27 +00:00
// encode JSON encodes v as JSON, logging tailcfg.MapRequest values if
// debugMap is set.
func encode ( v any ) ( [ ] byte , error ) {
2020-02-05 22:16:58 +00:00
b , err := json . Marshal ( v )
if err != nil {
return nil , err
}
2022-09-14 19:49:39 +00:00
if debugMap ( ) {
2021-03-29 19:42:43 +00:00
if _ , ok := v . ( * tailcfg . MapRequest ) ; ok {
2020-03-04 06:21:56 +00:00
log . Printf ( "MapRequest: %s" , b )
}
}
2024-04-02 04:23:27 +00:00
return b , nil
2020-02-05 22:16:58 +00:00
}
2022-03-07 18:55:02 +00:00
func loadServerPubKeys ( ctx context . Context , httpc * http . Client , serverURL string ) ( * tailcfg . OverTLSPublicKeyResponse , error ) {
keyURL := fmt . Sprintf ( "%v/key?v=%d" , serverURL , tailcfg . CurrentCapabilityVersion )
req , err := http . NewRequestWithContext ( ctx , "GET" , keyURL , nil )
2020-02-05 22:16:58 +00:00
if err != nil {
2022-03-07 18:55:02 +00:00
return nil , fmt . Errorf ( "create control key request: %v" , err )
2020-02-05 22:16:58 +00:00
}
res , err := httpc . Do ( req )
if err != nil {
2022-03-07 18:55:02 +00:00
return nil , fmt . Errorf ( "fetch control key: %v" , err )
2020-02-05 22:16:58 +00:00
}
defer res . Body . Close ( )
2022-09-15 12:06:59 +00:00
b , err := io . ReadAll ( io . LimitReader ( res . Body , 64 << 10 ) )
2020-02-05 22:16:58 +00:00
if err != nil {
2022-03-07 18:55:02 +00:00
return nil , fmt . Errorf ( "fetch control key response: %v" , err )
2020-02-05 22:16:58 +00:00
}
if res . StatusCode != 200 {
2022-03-07 18:55:02 +00:00
return nil , fmt . Errorf ( "fetch control key: %d" , res . StatusCode )
2020-02-05 22:16:58 +00:00
}
2022-03-07 18:55:02 +00:00
var out tailcfg . OverTLSPublicKeyResponse
jsonErr := json . Unmarshal ( b , & out )
if jsonErr == nil {
return & out , nil
}
// Some old control servers might not be updated to send the new format.
// Accept the old pre-JSON format too.
out = tailcfg . OverTLSPublicKeyResponse { }
2021-09-03 20:17:46 +00:00
k , err := key . ParseMachinePublicUntyped ( mem . B ( b ) )
2020-02-05 22:16:58 +00:00
if err != nil {
2022-03-07 18:55:02 +00:00
return nil , multierr . New ( jsonErr , err )
2020-02-05 22:16:58 +00:00
}
2022-03-07 18:55:02 +00:00
out . LegacyPublicKey = k
return & out , nil
2020-02-05 22:16:58 +00:00
}
2020-06-28 18:53:37 +00:00
2022-08-09 17:16:34 +00:00
// DevKnob contains temporary internal-only debug knobs.
2020-06-28 18:53:37 +00:00
// They're unexported to not draw attention to them.
2022-08-09 17:16:34 +00:00
var DevKnob = initDevKnob ( )
2020-06-28 18:53:37 +00:00
2022-08-09 17:16:34 +00:00
type devKnobs struct {
2022-09-14 19:49:39 +00:00
DumpNetMaps func ( ) bool
ForceProxyDNS func ( ) bool
StripEndpoints func ( ) bool // strip endpoints from control (only use disco messages)
StripCaps func ( ) bool // strip all local node's control-provided capabilities
2020-06-28 18:53:37 +00:00
}
2022-08-09 17:16:34 +00:00
func initDevKnob ( ) devKnobs {
return devKnobs {
2022-09-14 19:49:39 +00:00
DumpNetMaps : envknob . RegisterBool ( "TS_DEBUG_NETMAP" ) ,
ForceProxyDNS : envknob . RegisterBool ( "TS_DEBUG_PROXY_DNS" ) ,
StripEndpoints : envknob . RegisterBool ( "TS_DEBUG_STRIP_ENDPOINTS" ) ,
StripCaps : envknob . RegisterBool ( "TS_DEBUG_STRIP_CAPS" ) ,
2020-06-28 18:53:37 +00:00
}
}
2023-08-04 23:29:44 +00:00
var clock tstime . Clock = tstime . StdClock { }
2021-04-15 23:03:59 +00:00
2020-11-04 21:48:50 +00:00
// ipForwardingBroken reports whether the system's IP forwarding is disabled
// and will definitely not work for the routes provided.
//
// It should not return false positives.
2021-03-31 18:55:21 +00:00
//
2022-03-28 17:24:11 +00:00
// TODO(bradfitz): Change controlclient.Options.SkipIPForwardingCheck into a
all: convert more code to use net/netip directly
perl -i -npe 's,netaddr.IPPrefixFrom,netip.PrefixFrom,' $(git grep -l -F netaddr.)
perl -i -npe 's,netaddr.IPPortFrom,netip.AddrPortFrom,' $(git grep -l -F netaddr. )
perl -i -npe 's,netaddr.IPPrefix,netip.Prefix,g' $(git grep -l -F netaddr. )
perl -i -npe 's,netaddr.IPPort,netip.AddrPort,g' $(git grep -l -F netaddr. )
perl -i -npe 's,netaddr.IP\b,netip.Addr,g' $(git grep -l -F netaddr. )
perl -i -npe 's,netaddr.IPv6Raw\b,netip.AddrFrom16,g' $(git grep -l -F netaddr. )
goimports -w .
Then delete some stuff from the net/netaddr shim package which is no
longer neeed.
Updates #5162
Change-Id: Ia7a86893fe21c7e3ee1ec823e8aba288d4566cd8
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
2022-07-26 04:14:09 +00:00
// func([]netip.Prefix) error signature instead.
2024-04-28 04:01:54 +00:00
func ipForwardingBroken ( routes [ ] netip . Prefix , state * netmon . State ) bool {
2022-03-28 17:24:11 +00:00
warn , err := netutil . CheckIPForwarding ( routes , state )
if err != nil {
// Oh well, we tried. This is just for debugging.
// We don't want false positives.
// TODO: maybe we want a different warning for inability to check?
2020-11-04 21:48:50 +00:00
return false
}
2022-03-28 17:24:11 +00:00
return warn != nil
2020-11-04 21:48:50 +00:00
}
2021-03-05 04:54:44 +00:00
2021-06-15 19:28:34 +00:00
// isUniquePingRequest reports whether pr contains a new PingRequest.URL
// not already handled, noting its value when returning true.
func ( c * Direct ) isUniquePingRequest ( pr * tailcfg . PingRequest ) bool {
if pr == nil || pr . URL == "" {
// Bogus.
return false
}
c . mu . Lock ( )
defer c . mu . Unlock ( )
if pr . URL == c . lastPingURL {
return false
}
c . lastPingURL = pr . URL
return true
}
2022-08-10 21:52:02 +00:00
func ( c * Direct ) answerPing ( pr * tailcfg . PingRequest ) {
httpc := c . httpc
2024-04-02 04:23:27 +00:00
useNoise := pr . URLIsNoise || pr . Types == "c2n"
2022-08-27 19:55:41 +00:00
if useNoise {
2022-08-10 21:52:02 +00:00
nc , err := c . getNoiseClient ( )
if err != nil {
c . logf ( "failed to get noise client for ping request: %v" , err )
return
}
httpc = nc . Client
}
2021-03-05 04:54:44 +00:00
if pr . URL == "" {
2022-08-10 21:52:02 +00:00
c . logf ( "invalid PingRequest with no URL" )
2021-03-05 04:54:44 +00:00
return
}
2022-08-27 19:55:41 +00:00
switch pr . Types {
case "" :
2022-08-10 21:52:02 +00:00
answerHeadPing ( c . logf , httpc , pr )
2022-04-15 16:00:27 +00:00
return
2022-08-27 19:55:41 +00:00
case "c2n" :
if ! useNoise && ! envknob . Bool ( "TS_DEBUG_PERMIT_HTTP_C2N" ) {
c . logf ( "refusing to answer c2n ping without noise" )
return
}
answerC2NPing ( c . logf , c . c2nHandler , httpc , pr )
return
2022-04-15 16:00:27 +00:00
}
for _ , t := range strings . Split ( pr . Types , "," ) {
2022-04-22 01:49:01 +00:00
switch pt := tailcfg . PingType ( t ) ; pt {
2022-06-06 20:07:14 +00:00
case tailcfg . PingTSMP , tailcfg . PingDisco , tailcfg . PingICMP , tailcfg . PingPeerAPI :
2022-08-10 21:52:02 +00:00
go doPingerPing ( c . logf , httpc , pr , c . pinger , pt )
2022-04-15 16:00:27 +00:00
default :
2022-08-10 21:52:02 +00:00
c . logf ( "unsupported ping request type: %q" , t )
2022-04-15 16:00:27 +00:00
}
}
}
func answerHeadPing ( logf logger . Logf , c * http . Client , pr * tailcfg . PingRequest ) {
2021-03-05 04:54:44 +00:00
ctx , cancel := context . WithTimeout ( context . Background ( ) , 15 * time . Second )
defer cancel ( )
req , err := http . NewRequestWithContext ( ctx , "HEAD" , pr . URL , nil )
if err != nil {
2022-04-15 16:00:27 +00:00
logf ( "answerHeadPing: NewRequestWithContext: %v" , err )
2021-03-05 04:54:44 +00:00
return
}
if pr . Log {
2022-04-15 16:00:27 +00:00
logf ( "answerHeadPing: sending HEAD ping to %v ..." , pr . URL )
2021-03-05 04:54:44 +00:00
}
2023-08-04 23:29:44 +00:00
t0 := clock . Now ( )
2021-03-05 04:54:44 +00:00
_ , err = c . Do ( req )
2023-08-04 23:29:44 +00:00
d := clock . Since ( t0 ) . Round ( time . Millisecond )
2021-03-05 04:54:44 +00:00
if err != nil {
2022-04-15 16:00:27 +00:00
logf ( "answerHeadPing error: %v to %v (after %v)" , err , pr . URL , d )
2021-03-05 04:54:44 +00:00
} else if pr . Log {
2022-04-15 16:00:27 +00:00
logf ( "answerHeadPing complete to %v (after %v)" , pr . URL , d )
2021-03-05 04:54:44 +00:00
}
}
2021-04-21 21:17:21 +00:00
2022-08-27 19:55:41 +00:00
func answerC2NPing ( logf logger . Logf , c2nHandler http . Handler , c * http . Client , pr * tailcfg . PingRequest ) {
if c2nHandler == nil {
logf ( "answerC2NPing: c2nHandler not defined" )
return
}
hreq , err := http . ReadRequest ( bufio . NewReader ( bytes . NewReader ( pr . Payload ) ) )
if err != nil {
logf ( "answerC2NPing: ReadRequest: %v" , err )
return
}
if pr . Log {
logf ( "answerC2NPing: got c2n request for %v ..." , hreq . RequestURI )
}
handlerTimeout := time . Minute
if v := hreq . Header . Get ( "C2n-Handler-Timeout" ) ; v != "" {
handlerTimeout , _ = time . ParseDuration ( v )
}
handlerCtx , cancel := context . WithTimeout ( context . Background ( ) , handlerTimeout )
defer cancel ( )
hreq = hreq . WithContext ( handlerCtx )
rec := httptest . NewRecorder ( )
c2nHandler . ServeHTTP ( rec , hreq )
cancel ( )
c2nResBuf := new ( bytes . Buffer )
rec . Result ( ) . Write ( c2nResBuf )
replyCtx , cancel := context . WithTimeout ( context . Background ( ) , time . Minute )
defer cancel ( )
req , err := http . NewRequestWithContext ( replyCtx , "POST" , pr . URL , c2nResBuf )
if err != nil {
logf ( "answerC2NPing: NewRequestWithContext: %v" , err )
return
}
if pr . Log {
logf ( "answerC2NPing: sending POST ping to %v ..." , pr . URL )
}
2023-08-04 23:29:44 +00:00
t0 := clock . Now ( )
2022-08-27 19:55:41 +00:00
_ , err = c . Do ( req )
d := time . Since ( t0 ) . Round ( time . Millisecond )
if err != nil {
logf ( "answerC2NPing error: %v to %v (after %v)" , err , pr . URL , d )
} else if pr . Log {
logf ( "answerC2NPing complete to %v (after %v)" , pr . URL , d )
}
}
2023-08-21 02:52:52 +00:00
// sleepAsRequest implements the sleep for a tailcfg.Debug message requesting
// that the client sleep. The complication is that while we're sleeping (if for
// a long time), we need to periodically reset the watchdog timer before it
// expires.
2023-11-05 18:41:13 +00:00
func sleepAsRequested ( ctx context . Context , logf logger . Logf , d time . Duration , clock tstime . Clock ) error {
2021-04-21 21:17:21 +00:00
const maxSleep = 5 * time . Minute
if d > maxSleep {
logf ( "sleeping for %v, capped from server-requested %v ..." , maxSleep , d )
d = maxSleep
} else {
logf ( "sleeping for server-requested %v ..." , d )
}
2023-08-04 23:29:44 +00:00
timer , timerChannel := clock . NewTimer ( d )
2021-04-21 21:17:21 +00:00
defer timer . Stop ( )
2023-11-05 18:41:13 +00:00
select {
case <- ctx . Done ( ) :
return ctx . Err ( )
case <- timerChannel :
return nil
2021-04-21 21:17:21 +00:00
}
}
2021-06-07 23:03:16 +00:00
2022-03-07 23:32:53 +00:00
// getNoiseClient returns the noise client, creating one if one doesn't exist.
2022-11-03 07:17:16 +00:00
func ( c * Direct ) getNoiseClient ( ) ( * NoiseClient , error ) {
2022-03-07 23:32:53 +00:00
c . mu . Lock ( )
serverNoiseKey := c . serverNoiseKey
nc := c . noiseClient
c . mu . Unlock ( )
if serverNoiseKey . IsZero ( ) {
return nil , errors . New ( "zero serverNoiseKey" )
}
if nc != nil {
return nc , nil
}
2022-09-23 17:06:55 +00:00
var dp func ( ) * tailcfg . ControlDialPlan
if c . dialPlan != nil {
dp = c . dialPlan . Load
}
2022-11-03 07:17:16 +00:00
nc , err , _ := c . sfGroup . Do ( struct { } { } , func ( ) ( * NoiseClient , error ) {
2022-03-07 23:32:53 +00:00
k , err := c . getMachinePrivKey ( )
if err != nil {
return nil , err
}
2023-09-23 18:21:47 +00:00
c . logf ( "[v1] creating new noise client" )
2023-04-30 03:25:37 +00:00
nc , err := NewNoiseClient ( NoiseOpts {
2024-04-26 17:12:46 +00:00
PrivKey : k ,
ServerPubKey : serverNoiseKey ,
ServerURL : c . serverURL ,
Dialer : c . dialer ,
DNSCache : c . dnsCache ,
Logf : c . logf ,
NetMon : c . netMon ,
HealthTracker : c . health ,
DialPlan : dp ,
2023-04-30 03:25:37 +00:00
} )
2022-03-07 23:32:53 +00:00
if err != nil {
return nil , err
}
c . mu . Lock ( )
defer c . mu . Unlock ( )
c . noiseClient = nc
return nc , nil
} )
if err != nil {
return nil , err
}
2022-06-17 17:09:23 +00:00
return nc , nil
2022-03-07 23:32:53 +00:00
}
2022-03-08 00:16:15 +00:00
// setDNSNoise sends the SetDNSRequest request to the control plane server over Noise,
// requesting a DNS record be created or updated.
func ( c * Direct ) setDNSNoise ( ctx context . Context , req * tailcfg . SetDNSRequest ) error {
newReq := * req
newReq . Version = tailcfg . CurrentCapabilityVersion
2022-09-23 20:09:33 +00:00
nc , err := c . getNoiseClient ( )
2022-03-08 00:16:15 +00:00
if err != nil {
return err
}
2024-03-20 13:41:56 +00:00
res , err := nc . post ( ctx , "/machine/set-dns" , newReq . NodeKey , & newReq )
2022-03-08 00:16:15 +00:00
if err != nil {
return err
}
defer res . Body . Close ( )
if res . StatusCode != 200 {
2022-09-15 12:06:59 +00:00
msg , _ := io . ReadAll ( res . Body )
2022-03-08 00:16:15 +00:00
return fmt . Errorf ( "set-dns response: %v, %.200s" , res . Status , strings . TrimSpace ( string ( msg ) ) )
}
var setDNSRes tailcfg . SetDNSResponse
if err := json . NewDecoder ( res . Body ) . Decode ( & setDNSRes ) ; err != nil {
c . logf ( "error decoding SetDNSResponse: %v" , err )
return fmt . Errorf ( "set-dns-response: %w" , err )
}
return nil
}
2021-06-07 23:03:16 +00:00
// SetDNS sends the SetDNSRequest request to the control plane server,
// requesting a DNS record be created or updated.
2021-11-16 16:34:25 +00:00
func ( c * Direct ) SetDNS ( ctx context . Context , req * tailcfg . SetDNSRequest ) ( err error ) {
metricSetDNS . Add ( 1 )
defer func ( ) {
if err != nil {
metricSetDNSError . Add ( 1 )
}
} ( )
2024-04-02 04:23:27 +00:00
return c . setDNSNoise ( ctx , req )
2021-06-07 23:03:16 +00:00
}
2021-07-16 14:51:54 +00:00
2022-03-10 18:28:42 +00:00
func ( c * Direct ) DoNoiseRequest ( req * http . Request ) ( * http . Response , error ) {
2024-05-07 04:33:37 +00:00
if c . panicOnUse {
panic ( "tainted client" )
}
2022-03-10 18:28:42 +00:00
nc , err := c . getNoiseClient ( )
if err != nil {
return nil , err
}
return nc . Do ( req )
}
2022-11-02 03:37:13 +00:00
// GetSingleUseNoiseRoundTripper returns a RoundTripper that can be only be used
// once (and must be used once) to make a single HTTP request over the noise
// channel to the coordination server.
//
// In addition to the RoundTripper, it returns the HTTP/2 channel's early noise
// payload, if any.
func ( c * Direct ) GetSingleUseNoiseRoundTripper ( ctx context . Context ) ( http . RoundTripper , * tailcfg . EarlyNoise , error ) {
nc , err := c . getNoiseClient ( )
if err != nil {
return nil , nil , err
}
2022-11-03 07:17:16 +00:00
return nc . GetSingleUseRoundTripper ( ctx )
2022-11-02 03:37:13 +00:00
}
2022-04-15 16:00:27 +00:00
// doPingerPing sends a Ping to pr.IP using pinger, and sends an http request back to
// pr.URL with ping response data.
2022-04-22 01:49:01 +00:00
func doPingerPing ( logf logger . Logf , c * http . Client , pr * tailcfg . PingRequest , pinger Pinger , pingType tailcfg . PingType ) {
2022-07-25 03:08:42 +00:00
if pr . URL == "" || ! pr . IP . IsValid ( ) || pinger == nil {
2022-04-15 16:00:27 +00:00
logf ( "invalid ping request: missing url, ip or pinger" )
return
2021-07-16 14:51:54 +00:00
}
2023-08-04 23:29:44 +00:00
start := clock . Now ( )
2022-06-06 20:07:14 +00:00
ctx , cancel := context . WithTimeout ( context . Background ( ) , 10 * time . Second )
defer cancel ( )
2023-08-08 12:11:28 +00:00
res , err := pinger . Ping ( ctx , pr . IP , pingType , 0 )
2022-06-06 20:07:14 +00:00
if err != nil {
d := time . Since ( start ) . Round ( time . Millisecond )
logf ( "doPingerPing: ping error of type %q to %v after %v: %v" , pingType , pr . IP , d , err )
return
}
postPingResult ( start , logf , c , pr , res . ToPingResponse ( pingType ) )
2021-07-16 14:51:54 +00:00
}
2022-04-15 16:00:27 +00:00
func postPingResult ( start time . Time , logf logger . Logf , c * http . Client , pr * tailcfg . PingRequest , res * tailcfg . PingResponse ) error {
duration := time . Since ( start )
2021-07-16 14:51:54 +00:00
if pr . Log {
2022-04-15 16:00:27 +00:00
if res . Err == "" {
logf ( "ping to %v completed in %v. pinger.Ping took %v seconds" , pr . IP , res . LatencySeconds , duration )
} else {
logf ( "ping to %v failed after %v: %v" , pr . IP , duration , res . Err )
}
2021-07-16 14:51:54 +00:00
}
ctx , cancel := context . WithTimeout ( context . Background ( ) , 15 * time . Second )
defer cancel ( )
jsonPingRes , err := json . Marshal ( res )
if err != nil {
return err
}
// Send the results of the Ping, back to control URL.
2022-04-15 16:00:27 +00:00
req , err := http . NewRequestWithContext ( ctx , "POST" , pr . URL , bytes . NewReader ( jsonPingRes ) )
2021-07-16 14:51:54 +00:00
if err != nil {
return fmt . Errorf ( "http.NewRequestWithContext(%q): %w" , pr . URL , err )
}
if pr . Log {
2022-04-15 16:00:27 +00:00
logf ( "postPingResult: sending ping results to %v ..." , pr . URL )
2021-07-16 14:51:54 +00:00
}
2023-08-04 23:29:44 +00:00
t0 := clock . Now ( )
2021-07-16 14:51:54 +00:00
_ , err = c . Do ( req )
d := time . Since ( t0 ) . Round ( time . Millisecond )
if err != nil {
2022-04-15 16:00:27 +00:00
return fmt . Errorf ( "postPingResult error: %w to %v (after %v)" , err , pr . URL , d )
2021-07-16 14:51:54 +00:00
} else if pr . Log {
2022-04-15 16:00:27 +00:00
logf ( "postPingResult complete to %v (after %v)" , pr . URL , d )
2021-07-16 14:51:54 +00:00
}
return nil
}
2021-11-16 16:34:25 +00:00
2022-06-21 14:44:11 +00:00
// ReportHealthChange reports to the control plane a change to this node's
health: begin work to use structured health warnings instead of strings, pipe changes into ipn.Notify (#12406)
Updates tailscale/tailscale#4136
This PR is the first round of work to move from encoding health warnings as strings and use structured data instead. The current health package revolves around the idea of Subsystems. Each subsystem can have (or not have) a Go error associated with it. The overall health of the backend is given by the concatenation of all these errors.
This PR polishes the concept of Warnable introduced by @bradfitz a few weeks ago. Each Warnable is a component of the backend (for instance, things like 'dns' or 'magicsock' are Warnables). Each Warnable has a unique identifying code. A Warnable is an entity we can warn the user about, by setting (or unsetting) a WarningState for it. Warnables have:
- an identifying Code, so that the GUI can track them as their WarningStates come and go
- a Title, which the GUIs can use to tell the user what component of the backend is broken
- a Text, which is a function that is called with a set of Args to generate a more detailed error message to explain the unhappy state
Additionally, this PR also begins to send Warnables and their WarningStates through LocalAPI to the clients, using ipn.Notify messages. An ipn.Notify is only issued when a warning is added or removed from the Tracker.
In a next PR, we'll get rid of subsystems entirely, and we'll start using structured warnings for all errors affecting the backend functionality.
Signed-off-by: Andrea Gottardo <andrea@gottardo.me>
2024-06-14 18:53:56 +00:00
// health. w must be non-nil. us can be nil to indicate a healthy state for w.
func ( c * Direct ) ReportHealthChange ( w * health . Warnable , us * health . UnhealthyState ) {
if w == health . NetworkStatusWarnable || w == health . IPNStateWarnable || w == health . LoginStateWarnable {
2022-06-21 14:44:11 +00:00
// We don't report these. These include things like the network is down
// (in which case we can't report anyway) or the user wanted things
// stopped, as opposed to the more unexpected failure types in the other
// subsystems.
return
}
np , err := c . getNoiseClient ( )
if err != nil {
// Don't report errors to control if the server doesn't support noise.
return
}
2024-03-21 19:19:57 +00:00
nodeKey , ok := c . GetPersist ( ) . PublicNodeKeyOK ( )
if ! ok {
return
}
2024-05-07 04:33:37 +00:00
if c . panicOnUse {
panic ( "tainted client" )
}
health: begin work to use structured health warnings instead of strings, pipe changes into ipn.Notify (#12406)
Updates tailscale/tailscale#4136
This PR is the first round of work to move from encoding health warnings as strings and use structured data instead. The current health package revolves around the idea of Subsystems. Each subsystem can have (or not have) a Go error associated with it. The overall health of the backend is given by the concatenation of all these errors.
This PR polishes the concept of Warnable introduced by @bradfitz a few weeks ago. Each Warnable is a component of the backend (for instance, things like 'dns' or 'magicsock' are Warnables). Each Warnable has a unique identifying code. A Warnable is an entity we can warn the user about, by setting (or unsetting) a WarningState for it. Warnables have:
- an identifying Code, so that the GUI can track them as their WarningStates come and go
- a Title, which the GUIs can use to tell the user what component of the backend is broken
- a Text, which is a function that is called with a set of Args to generate a more detailed error message to explain the unhappy state
Additionally, this PR also begins to send Warnables and their WarningStates through LocalAPI to the clients, using ipn.Notify messages. An ipn.Notify is only issued when a warning is added or removed from the Tracker.
In a next PR, we'll get rid of subsystems entirely, and we'll start using structured warnings for all errors affecting the backend functionality.
Signed-off-by: Andrea Gottardo <andrea@gottardo.me>
2024-06-14 18:53:56 +00:00
// TODO(angott): at some point, update `Subsys` in the request to be `Warnable`
2022-06-21 14:44:11 +00:00
req := & tailcfg . HealthChangeRequest {
health: begin work to use structured health warnings instead of strings, pipe changes into ipn.Notify (#12406)
Updates tailscale/tailscale#4136
This PR is the first round of work to move from encoding health warnings as strings and use structured data instead. The current health package revolves around the idea of Subsystems. Each subsystem can have (or not have) a Go error associated with it. The overall health of the backend is given by the concatenation of all these errors.
This PR polishes the concept of Warnable introduced by @bradfitz a few weeks ago. Each Warnable is a component of the backend (for instance, things like 'dns' or 'magicsock' are Warnables). Each Warnable has a unique identifying code. A Warnable is an entity we can warn the user about, by setting (or unsetting) a WarningState for it. Warnables have:
- an identifying Code, so that the GUI can track them as their WarningStates come and go
- a Title, which the GUIs can use to tell the user what component of the backend is broken
- a Text, which is a function that is called with a set of Args to generate a more detailed error message to explain the unhappy state
Additionally, this PR also begins to send Warnables and their WarningStates through LocalAPI to the clients, using ipn.Notify messages. An ipn.Notify is only issued when a warning is added or removed from the Tracker.
In a next PR, we'll get rid of subsystems entirely, and we'll start using structured warnings for all errors affecting the backend functionality.
Signed-off-by: Andrea Gottardo <andrea@gottardo.me>
2024-06-14 18:53:56 +00:00
Subsys : string ( w . Code ) ,
2024-03-20 13:41:56 +00:00
NodeKey : nodeKey ,
2022-06-21 14:44:11 +00:00
}
health: begin work to use structured health warnings instead of strings, pipe changes into ipn.Notify (#12406)
Updates tailscale/tailscale#4136
This PR is the first round of work to move from encoding health warnings as strings and use structured data instead. The current health package revolves around the idea of Subsystems. Each subsystem can have (or not have) a Go error associated with it. The overall health of the backend is given by the concatenation of all these errors.
This PR polishes the concept of Warnable introduced by @bradfitz a few weeks ago. Each Warnable is a component of the backend (for instance, things like 'dns' or 'magicsock' are Warnables). Each Warnable has a unique identifying code. A Warnable is an entity we can warn the user about, by setting (or unsetting) a WarningState for it. Warnables have:
- an identifying Code, so that the GUI can track them as their WarningStates come and go
- a Title, which the GUIs can use to tell the user what component of the backend is broken
- a Text, which is a function that is called with a set of Args to generate a more detailed error message to explain the unhappy state
Additionally, this PR also begins to send Warnables and their WarningStates through LocalAPI to the clients, using ipn.Notify messages. An ipn.Notify is only issued when a warning is added or removed from the Tracker.
In a next PR, we'll get rid of subsystems entirely, and we'll start using structured warnings for all errors affecting the backend functionality.
Signed-off-by: Andrea Gottardo <andrea@gottardo.me>
2024-06-14 18:53:56 +00:00
if us != nil {
req . Error = us . Text
2022-06-21 14:44:11 +00:00
}
// Best effort, no logging:
2024-10-02 15:16:43 +00:00
ctx , cancel := context . WithTimeout ( c . closedCtx , 5 * time . Second )
2022-06-21 14:44:11 +00:00
defer cancel ( )
2024-03-20 13:41:56 +00:00
res , err := np . post ( ctx , "/machine/update-health" , nodeKey , req )
2022-06-21 14:44:11 +00:00
if err != nil {
return
}
res . Body . Close ( )
}
2024-03-20 13:41:56 +00:00
func addLBHeader ( req * http . Request , nodeKey key . NodePublic ) {
if ! nodeKey . IsZero ( ) {
req . Header . Add ( tailcfg . LBHeader , nodeKey . String ( ) )
}
}
2024-06-20 18:09:50 +00:00
type dialFunc = func ( ctx context . Context , network , addr string ) ( net . Conn , error )
// makeScreenTimeDetectingDialFunc returns dialFunc, optionally wrapped (on
// Apple systems) with a func that sets the returned atomic.Bool for whether
// Screen Time seemed to intercept the connection.
//
// The returned *atomic.Bool is nil on non-Apple systems.
func makeScreenTimeDetectingDialFunc ( dial dialFunc ) ( dialFunc , * atomic . Bool ) {
switch runtime . GOOS {
case "darwin" , "ios" :
// Continue below.
default :
return dial , nil
}
ab := new ( atomic . Bool )
return func ( ctx context . Context , network , addr string ) ( net . Conn , error ) {
c , err := dial ( ctx , network , addr )
if err != nil {
return nil , err
}
ab . Store ( isTCPLoopback ( c . LocalAddr ( ) ) && isTCPLoopback ( c . RemoteAddr ( ) ) )
return c , nil
} , ab
}
func isTCPLoopback ( a net . Addr ) bool {
if ta , ok := a . ( * net . TCPAddr ) ; ok {
return ta . IP . IsLoopback ( )
}
return false
}
2021-11-16 16:34:25 +00:00
var (
metricMapRequestsActive = clientmetric . NewGauge ( "controlclient_map_requests_active" )
metricMapRequests = clientmetric . NewCounter ( "controlclient_map_requests" )
metricMapRequestsLite = clientmetric . NewCounter ( "controlclient_map_requests_lite" )
metricMapRequestsPoll = clientmetric . NewCounter ( "controlclient_map_requests_poll" )
metricMapResponseMessages = clientmetric . NewCounter ( "controlclient_map_response_message" ) // any message type
metricMapResponsePings = clientmetric . NewCounter ( "controlclient_map_response_ping" )
metricMapResponseKeepAlives = clientmetric . NewCounter ( "controlclient_map_response_keepalive" )
metricMapResponseMap = clientmetric . NewCounter ( "controlclient_map_response_map" ) // any non-keepalive map response
metricMapResponseMapDelta = clientmetric . NewCounter ( "controlclient_map_response_map_delta" ) // 2nd+ non-keepalive map response
metricSetDNS = clientmetric . NewCounter ( "controlclient_setdns" )
metricSetDNSError = clientmetric . NewCounter ( "controlclient_setdns_error" )
)