You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
tailscale/health/health.go

1094 lines
31 KiB
Go

// Copyright (c) Tailscale Inc & AUTHORS
// SPDX-License-Identifier: BSD-3-Clause
// Package health is a registry for other packages to report & check
// overall health status of the node.
package health
import (
"errors"
"fmt"
"maps"
"net/http"
"os"
"runtime"
"sort"
"sync"
"sync/atomic"
"time"
"tailscale.com/envknob"
"tailscale.com/tailcfg"
"tailscale.com/types/opt"
"tailscale.com/util/cibuild"
"tailscale.com/util/mak"
"tailscale.com/util/multierr"
"tailscale.com/util/set"
"tailscale.com/version"
)
var (
mu sync.Mutex
debugHandler map[string]http.Handler
)
// ReceiveFunc is one of the three magicsock Receive funcs (IPv4, IPv6, or
// DERP).
type ReceiveFunc int
// ReceiveFunc indices for Tracker.MagicSockReceiveFuncs.
const (
ReceiveIPv4 ReceiveFunc = 0
ReceiveIPv6 ReceiveFunc = 1
ReceiveDERP ReceiveFunc = 2
)
func (f ReceiveFunc) String() string {
if f < 0 || int(f) >= len(receiveNames) {
return fmt.Sprintf("ReceiveFunc(%d)", f)
}
return receiveNames[f]
}
var receiveNames = []string{
ReceiveIPv4: "ReceiveIPv4",
ReceiveIPv6: "ReceiveIPv6",
ReceiveDERP: "ReceiveDERP",
}
// Tracker tracks the health of various Tailscale subsystems,
// comparing each subsystems' state with each other to make sure
// they're consistent based on the user's intended state.
type Tracker struct {
// MagicSockReceiveFuncs tracks the state of the three
// magicsock receive functions: IPv4, IPv6, and DERP.
MagicSockReceiveFuncs [3]ReceiveFuncStats // indexed by ReceiveFunc values
// mu guards everything that follows.
mu sync.Mutex
warnables []*Warnable // keys ever set
warnableVal map[*Warnable]*warningState
// sysErr maps subsystems to their current error (or nil if the subsystem is healthy)
// Deprecated: using Warnables should be preferred
sysErr map[Subsystem]error
watchers set.HandleSet[func(*Warnable, *UnhealthyState)] // opt func to run if error state changes
timer *time.Timer
latestVersion *tailcfg.ClientVersion // or nil
checkForUpdates bool
inMapPoll bool
inMapPollSince time.Time
lastMapPollEndedAt time.Time
lastStreamedMapResponse time.Time
derpHomeRegion int
derpHomeless bool
derpRegionConnected map[int]bool
derpRegionHealthProblem map[int]string
derpRegionLastFrame map[int]time.Time
derpMap *tailcfg.DERPMap // last DERP map from control, could be nil if never received one
lastMapRequestHeard time.Time // time we got a 200 from control for a MapRequest
ipnState string
ipnWantRunning bool
anyInterfaceUp opt.Bool // empty means unknown (assume true)
udp4Unbound bool
controlHealth []string
lastLoginErr error
localLogConfigErr error
tlsConnectionErrors map[string]error // map[ServerName]error
}
// Subsystem is the name of a subsystem whose health can be monitored.
//
// Deprecated: Registering a Warnable using Register() and updating its health state
// with SetUnhealthy() and SetHealthy() should be preferred.
type Subsystem string
const (
// SysRouter is the name of the wgengine/router subsystem.
SysRouter = Subsystem("router")
// SysDNS is the name of the net/dns subsystem.
SysDNS = Subsystem("dns")
// SysDNSOS is the name of the net/dns OSConfigurator subsystem.
SysDNSOS = Subsystem("dns-os")
// SysDNSManager is the name of the net/dns manager subsystem.
SysDNSManager = Subsystem("dns-manager")
// SysTKA is the name of the tailnet key authority subsystem.
SysTKA = Subsystem("tailnet-lock")
)
var subsystemsWarnables = map[Subsystem]*Warnable{}
func init() {
for _, s := range []Subsystem{SysRouter, SysDNS, SysDNSOS, SysDNSManager, SysTKA} {
w := Register(&Warnable{
Code: WarnableCode(s),
Severity: SeverityMedium,
Text: func(args Args) string {
return args[legacyErrorArgKey]
},
})
subsystemsWarnables[s] = w
}
}
const legacyErrorArgKey = "LegacyError"
// Warnable returns a Warnable representing a legacy Subsystem. This is used
// temporarily (2024-06-14) while we migrate the old health infrastructure based
// on Subsystems to the new Warnables architecture.
func (s Subsystem) Warnable() *Warnable {
w, ok := subsystemsWarnables[s]
if !ok {
panic(fmt.Sprintf("health: no Warnable for Subsystem %q", s))
}
return w
}
var registeredWarnables = map[WarnableCode]*Warnable{}
// Register registers a new Warnable with the health package and returns it.
// Register panics if the Warnable was already registered, because Warnables
// should be unique across the program.
func Register(w *Warnable) *Warnable {
if registeredWarnables[w.Code] != nil {
panic(fmt.Sprintf("health: a Warnable with code %q was already registered", w.Code))
}
mak.Set(&registeredWarnables, w.Code, w)
return w
}
// unregister removes a Warnable from the health package. It should only be used
// for testing purposes.
func unregister(w *Warnable) {
if registeredWarnables[w.Code] == nil {
panic(fmt.Sprintf("health: attempting to unregister Warnable %q that was not registered", w.Code))
}
delete(registeredWarnables, w.Code)
}
// WarnableCode is a string that distinguishes each Warnable from others. It is globally unique within
// the program.
type WarnableCode string
// A Warnable is something that we might want to warn the user about, or not. A Warnable is either
// in an healthy or unhealth state. A Warnable is unhealthy if the Tracker knows about a WarningState
// affecting the Warnable.
// In most cases, Warnables are components of the backend (for instance, "DNS" or "Magicsock").
// Warnables are similar to the Subsystem type previously used in this package, but they provide
// a unique identifying code for each Warnable, along with more metadata that makes it easier for
// a GUI to display the Warnable in a user-friendly way.
type Warnable struct {
// Code is a string that uniquely identifies this Warnable across the entire Tailscale backend,
// and can be mapped to a user-displayable localized string.
Code WarnableCode
// Title is a string that the GUI uses as title for any message involving this Warnable. The title
// should be short and fit in a single line.
Title string
// Text is a function that generates an extended string that the GUI will display to the user when
// this Warnable is in an unhealthy state. The function can use the Args map to provide dynamic
// information to the user.
Text func(args Args) string
// Severity is the severity of the Warnable, which the GUI can use to determine how to display it.
// For instance, a Warnable with SeverityHigh could trigger a modal view, while a Warnable with
// SeverityLow could be displayed in a less intrusive way.
// TODO(angott): turn this into a SeverityFunc, which allows the Warnable to change its severity based on
// the Args of the unhappy state, just like we do in the Text function.
Severity Severity
// DependsOn is a set of Warnables that this Warnable depends, on and need to be healthy
// before this Warnable can also be healthy again. The GUI can use this information to ignore
// this Warnable if one of its dependencies is unhealthy.
DependsOn []*Warnable
// MapDebugFlag is a MapRequest.DebugFlag that is sent to control when this Warnable is unhealthy
//
// Deprecated: this is only used in one case, and will be removed in a future PR
MapDebugFlag string
// If true, this warnable is related to configuration of networking stack
// on the machine that impacts connectivity.
ImpactsConnectivity bool
}
// StaticMessage returns a function that always returns the input string, to be used in
// simple Warnables that do not use the Args map to generate their Text.
func StaticMessage(s string) func(Args) string {
return func(Args) string { return s }
}
// nil reports whether t is nil.
// It exists to accept nil *Tracker receivers on all methods
// to at least not crash. But because a nil receiver indicates
// some lost Tracker plumbing, we want to capture stack trace
// samples when it occurs.
func (t *Tracker) nil() bool {
if t != nil {
return false
}
if cibuild.On() {
stack := make([]byte, 1<<10)
stack = stack[:runtime.Stack(stack, false)]
fmt.Fprintf(os.Stderr, "## WARNING: (non-fatal) nil health.Tracker (being strict in CI):\n%s\n", stack)
}
// TODO(bradfitz): open source our "unexpected" package
// and use it here to capture samples of stacks where
// t is nil.
return true
}
// Severity represents how serious an error is. Each GUI interprets this severity value in different ways,
// to surface the error in a more or less visible way. For instance, the macOS GUI could change its menubar
// icon to display an exclamation mark and present a modal notification for SeverityHigh warnings, but not
// for SeverityLow messages, which would only appear in the Settings window.
type Severity string
const (
SeverityHigh Severity = "high"
SeverityMedium Severity = "medium"
SeverityLow Severity = "low"
)
// Args is a map of Args to string values that can be used to provide parameters regarding
// the unhealthy state of a Warnable.
// For instance, if you have a Warnable to track the health of DNS lookups, here you can include
// the hostname that failed to resolve, or the IP address of the DNS server that has been failing
// to respond. You can then use these parameters in the Text function of the Warnable to provide a detailed
// error message to the user.
type Args map[Arg]string
// A warningState is a condition affecting a Warnable. For each Warnable known to the Tracker, a Warnable
// is in an unhappy state if there is a warningState associated with the Warnable.
type warningState struct {
BrokenSince time.Time // when the Warnable became unhealthy
Args Args // args can be used to provide parameters to the function that generates the Text in the Warnable
}
func (ws *warningState) Equal(other *warningState) bool {
if ws == nil && other == nil {
return true
}
if ws == nil || other == nil {
return false
}
return ws.BrokenSince.Equal(other.BrokenSince) && maps.Equal(ws.Args, other.Args)
}
// SetUnhealthy sets a warningState for the given Warnable with the provided Args, and should be
// called when a Warnable becomes unhealthy, or its unhealthy status needs to be updated.
// SetUnhealthy takes ownership of args. The args can be nil if no additional information is
// needed for the unhealthy state.
func (t *Tracker) SetUnhealthy(w *Warnable, args Args) {
if t.nil() {
return
}
t.mu.Lock()
defer t.mu.Unlock()
t.setUnhealthyLocked(w, args)
}
func (t *Tracker) setUnhealthyLocked(w *Warnable, args Args) {
if w == nil {
return
}
// If we already have a warningState for this Warnable with an earlier BrokenSince time, keep that
// BrokenSince time.
brokenSince := time.Now()
if existingWS := t.warnableVal[w]; existingWS != nil {
brokenSince = existingWS.BrokenSince
}
if t.warnableVal[w] == nil {
t.warnables = append(t.warnables, w)
}
ws := &warningState{
BrokenSince: brokenSince,
Args: args,
}
prevWs := t.warnableVal[w]
mak.Set(&t.warnableVal, w, ws)
if !ws.Equal(prevWs) {
for _, cb := range t.watchers {
go cb(w, w.unhealthyState(ws))
}
}
}
// SetHealthy removes any warningState for the given Warnable.
func (t *Tracker) SetHealthy(w *Warnable) {
if t.nil() {
return
}
t.mu.Lock()
defer t.mu.Unlock()
t.setHealthyLocked(w)
}
func (t *Tracker) setHealthyLocked(w *Warnable) {
if t.warnableVal[w] == nil {
// Nothing to remove
return
}
delete(t.warnableVal, w)
for _, cb := range t.watchers {
go cb(w, nil)
}
}
// AppendWarnableDebugFlags appends to base any health items that are currently in failed
// state and were created with MapDebugFlag.
func (t *Tracker) AppendWarnableDebugFlags(base []string) []string {
if t.nil() {
return base
}
ret := base
t.mu.Lock()
defer t.mu.Unlock()
for w, err := range t.warnableVal {
if w.MapDebugFlag == "" {
continue
}
if err != nil {
ret = append(ret, w.MapDebugFlag)
}
}
sort.Strings(ret[len(base):]) // sort the new ones
return ret
}
// RegisterWatcher adds a function that will be called whenever the health state of any Warnable changes.
// If a Warnable becomes unhealthy or its unhealthy state is updated, the callback will be called with its
// current Representation.
// If a Warnable becomes healthy, the callback will be called with ws set to nil.
// The provided callback function will be executed in its own goroutine. The returned function can be used
// to unregister the callback.
func (t *Tracker) RegisterWatcher(cb func(w *Warnable, r *UnhealthyState)) (unregister func()) {
if t.nil() {
return func() {}
}
t.mu.Lock()
defer t.mu.Unlock()
if t.watchers == nil {
t.watchers = set.HandleSet[func(*Warnable, *UnhealthyState)]{}
}
handle := t.watchers.Add(cb)
if t.timer == nil {
t.timer = time.AfterFunc(time.Minute, t.timerSelfCheck)
}
return func() {
t.mu.Lock()
defer t.mu.Unlock()
delete(t.watchers, handle)
if len(t.watchers) == 0 && t.timer != nil {
t.timer.Stop()
t.timer = nil
}
}
}
// SetRouterHealth sets the state of the wgengine/router.Router.
//
// Deprecated: Warnables should be preferred over Subsystem errors.
func (t *Tracker) SetRouterHealth(err error) { t.setErr(SysRouter, err) }
// RouterHealth returns the wgengine/router.Router error state.
//
// Deprecated: Warnables should be preferred over Subsystem errors.
func (t *Tracker) RouterHealth() error { return t.get(SysRouter) }
// SetDNSHealth sets the state of the net/dns.Manager
//
// Deprecated: Warnables should be preferred over Subsystem errors.
func (t *Tracker) SetDNSHealth(err error) { t.setErr(SysDNS, err) }
// DNSHealth returns the net/dns.Manager error state.
//
// Deprecated: Warnables should be preferred over Subsystem errors.
func (t *Tracker) DNSHealth() error { return t.get(SysDNS) }
// SetDNSOSHealth sets the state of the net/dns.OSConfigurator
//
// Deprecated: Warnables should be preferred over Subsystem errors.
func (t *Tracker) SetDNSOSHealth(err error) { t.setErr(SysDNSOS, err) }
// SetDNSManagerHealth sets the state of the Linux net/dns manager's
// discovery of the /etc/resolv.conf situation.
//
// Deprecated: Warnables should be preferred over Subsystem errors.
func (t *Tracker) SetDNSManagerHealth(err error) { t.setErr(SysDNSManager, err) }
// DNSOSHealth returns the net/dns.OSConfigurator error state.
//
// Deprecated: Warnables should be preferred over Subsystem errors.
func (t *Tracker) DNSOSHealth() error { return t.get(SysDNSOS) }
// SetTKAHealth sets the health of the tailnet key authority.
//
// Deprecated: Warnables should be preferred over Subsystem errors.
func (t *Tracker) SetTKAHealth(err error) { t.setErr(SysTKA, err) }
// TKAHealth returns the tailnet key authority error state.
//
// Deprecated: Warnables should be preferred over Subsystem errors.
func (t *Tracker) TKAHealth() error { return t.get(SysTKA) }
// SetLocalLogConfigHealth sets the error state of this client's local log configuration.
func (t *Tracker) SetLocalLogConfigHealth(err error) {
if t.nil() {
return
}
t.mu.Lock()
defer t.mu.Unlock()
t.localLogConfigErr = err
}
// SetTLSConnectionError sets the error state for connections to a specific
// host. Setting the error to nil will clear any previously-set error.
func (t *Tracker) SetTLSConnectionError(host string, err error) {
if t.nil() {
return
}
t.mu.Lock()
defer t.mu.Unlock()
if err == nil {
delete(t.tlsConnectionErrors, host)
} else {
mak.Set(&t.tlsConnectionErrors, host, err)
}
}
func RegisterDebugHandler(typ string, h http.Handler) {
mu.Lock()
defer mu.Unlock()
mak.Set(&debugHandler, typ, h)
}
func DebugHandler(typ string) http.Handler {
mu.Lock()
defer mu.Unlock()
return debugHandler[typ]
}
func (t *Tracker) get(key Subsystem) error {
if t.nil() {
return nil
}
t.mu.Lock()
defer t.mu.Unlock()
return t.sysErr[key]
}
func (t *Tracker) setErr(key Subsystem, err error) {
if t.nil() {
return
}
t.mu.Lock()
defer t.mu.Unlock()
t.setLocked(key, err)
}
func (t *Tracker) setLocked(key Subsystem, err error) {
if t.sysErr == nil {
t.sysErr = map[Subsystem]error{}
}
old, ok := t.sysErr[key]
if !ok && err == nil {
// Initial happy path.
t.sysErr[key] = nil
t.selfCheckLocked()
return
}
if ok && (old == nil) == (err == nil) {
// No change in overall error status (nil-vs-not), so
// don't run callbacks, but exact error might've
// changed, so note it.
if err != nil {
t.sysErr[key] = err
}
return
}
t.sysErr[key] = err
t.selfCheckLocked()
}
// updateLegacyErrorWarnableLocked takes a legacy Subsystem and an optional error, and
// updates the WarningState for that legacy Subsystem, setting it to healthy or unhealthy.
// It is used temporarily while we migrate from Subsystems to Warnables.
//
// Deprecated: this function will be removed after migrating all subsystem errors to use
// Warnables instead.
func (t *Tracker) updateLegacyErrorWarnableLocked(key Subsystem, err error) {
w := key.Warnable()
if err != nil {
t.setUnhealthyLocked(key.Warnable(), Args{legacyErrorArgKey: err.Error()})
} else {
t.setHealthyLocked(w)
}
}
func (t *Tracker) SetControlHealth(problems []string) {
if t.nil() {
return
}
t.mu.Lock()
defer t.mu.Unlock()
t.controlHealth = problems
t.selfCheckLocked()
}
// GotStreamedMapResponse notes that we got a tailcfg.MapResponse
// message in streaming mode, even if it's just a keep-alive message.
//
// This also notes that a map poll is in progress. To unset that, call
// SetOutOfPollNetMap().
func (t *Tracker) GotStreamedMapResponse() {
if t.nil() {
return
}
t.mu.Lock()
defer t.mu.Unlock()
t.lastStreamedMapResponse = time.Now()
if !t.inMapPoll {
t.inMapPoll = true
t.inMapPollSince = time.Now()
}
t.selfCheckLocked()
}
// SetOutOfPollNetMap records that the client is no longer in
// an HTTP map request long poll to the control plane.
func (t *Tracker) SetOutOfPollNetMap() {
if t.nil() {
return
}
t.mu.Lock()
defer t.mu.Unlock()
if !t.inMapPoll {
return
}
t.inMapPoll = false
t.lastMapPollEndedAt = time.Now()
t.selfCheckLocked()
}
// GetInPollNetMap reports whether the client has an open
// HTTP long poll open to the control plane.
func (t *Tracker) GetInPollNetMap() bool {
if t.nil() {
return false
}
t.mu.Lock()
defer t.mu.Unlock()
return t.inMapPoll
}
// SetMagicSockDERPHome notes what magicsock's view of its home DERP is.
//
// The homeless parameter is whether magicsock is running in DERP-disconnected
// mode, without discovering and maintaining a connection to its home DERP.
func (t *Tracker) SetMagicSockDERPHome(region int, homeless bool) {
if t.nil() {
return
}
t.mu.Lock()
defer t.mu.Unlock()
t.derpHomeRegion = region
t.derpHomeless = homeless
t.selfCheckLocked()
}
// NoteMapRequestHeard notes whenever we successfully sent a map request
// to control for which we received a 200 response.
func (t *Tracker) NoteMapRequestHeard(mr *tailcfg.MapRequest) {
if t.nil() {
return
}
t.mu.Lock()
defer t.mu.Unlock()
// TODO: extract mr.HostInfo.NetInfo.PreferredDERP, compare
// against SetMagicSockDERPHome and
// SetDERPRegionConnectedState
t.lastMapRequestHeard = time.Now()
t.selfCheckLocked()
}
func (t *Tracker) SetDERPRegionConnectedState(region int, connected bool) {
if t.nil() {
return
}
t.mu.Lock()
defer t.mu.Unlock()
mak.Set(&t.derpRegionConnected, region, connected)
t.selfCheckLocked()
}
// SetDERPRegionHealth sets or clears any problem associated with the
// provided DERP region.
func (t *Tracker) SetDERPRegionHealth(region int, problem string) {
if t.nil() {
return
}
t.mu.Lock()
defer t.mu.Unlock()
if problem == "" {
delete(t.derpRegionHealthProblem, region)
} else {
mak.Set(&t.derpRegionHealthProblem, region, problem)
}
t.selfCheckLocked()
}
// NoteDERPRegionReceivedFrame is called to note that a frame was received from
// the given DERP region at the current time.
func (t *Tracker) NoteDERPRegionReceivedFrame(region int) {
if t.nil() {
return
}
t.mu.Lock()
defer t.mu.Unlock()
mak.Set(&t.derpRegionLastFrame, region, time.Now())
t.selfCheckLocked()
}
// GetDERPRegionReceivedTime returns the last time that a frame was received
// from the given DERP region, or the zero time if no communication with that
// region has occurred.
func (t *Tracker) GetDERPRegionReceivedTime(region int) time.Time {
if t.nil() {
return time.Time{}
}
t.mu.Lock()
defer t.mu.Unlock()
return t.derpRegionLastFrame[region]
}
// SetDERPMap sets the last fetched DERP map in the Tracker. The DERP map is used
// to provide a region name in user-facing DERP-related warnings.
func (t *Tracker) SetDERPMap(dm *tailcfg.DERPMap) {
if t.nil() {
return
}
t.mu.Lock()
defer t.mu.Unlock()
t.derpMap = dm
t.selfCheckLocked()
}
// state is an ipn.State.String() value: "Running", "Stopped", "NeedsLogin", etc.
func (t *Tracker) SetIPNState(state string, wantRunning bool) {
if t.nil() {
return
}
t.mu.Lock()
defer t.mu.Unlock()
t.ipnState = state
t.ipnWantRunning = wantRunning
t.selfCheckLocked()
}
// SetAnyInterfaceUp sets whether any network interface is up.
func (t *Tracker) SetAnyInterfaceUp(up bool) {
if t.nil() {
return
}
t.mu.Lock()
defer t.mu.Unlock()
t.anyInterfaceUp.Set(up)
t.selfCheckLocked()
}
// SetUDP4Unbound sets whether the udp4 bind failed completely.
func (t *Tracker) SetUDP4Unbound(unbound bool) {
if t.nil() {
return
}
t.mu.Lock()
defer t.mu.Unlock()
t.udp4Unbound = unbound
t.selfCheckLocked()
}
// SetAuthRoutineInError records the latest error encountered as a result of a
// login attempt. Providing a nil error indicates successful login, or that
// being logged in w/coordination is not currently desired.
func (t *Tracker) SetAuthRoutineInError(err error) {
if t.nil() {
return
}
t.mu.Lock()
defer t.mu.Unlock()
if err == nil && t.lastLoginErr == nil {
return
}
t.lastLoginErr = err
t.selfCheckLocked()
}
// SetLatestVersion records the latest version of the Tailscale client.
// v can be nil if unknown.
func (t *Tracker) SetLatestVersion(v *tailcfg.ClientVersion) {
if t.nil() {
return
}
t.mu.Lock()
defer t.mu.Unlock()
t.latestVersion = v
t.selfCheckLocked()
}
// SetCheckForUpdates sets whether the client wants to check for updates.
func (t *Tracker) SetCheckForUpdates(v bool) {
if t.nil() {
return
}
t.mu.Lock()
defer t.mu.Unlock()
if t.checkForUpdates == v {
return
}
t.checkForUpdates = v
t.selfCheckLocked()
}
func (t *Tracker) timerSelfCheck() {
if t.nil() {
return
}
t.mu.Lock()
defer t.mu.Unlock()
t.checkReceiveFuncsLocked()
t.selfCheckLocked()
if t.timer != nil {
t.timer.Reset(time.Minute)
}
}
func (t *Tracker) selfCheckLocked() {
if t.ipnState == "" {
// Don't check yet.
return
}
t.updateBuiltinWarnablesLocked()
}
// OverallError returns a summary of the health state.
//
// If there are multiple problems, the error will be of type
// multierr.Error.
func (t *Tracker) OverallError() error {
if t.nil() {
return nil
}
t.mu.Lock()
defer t.mu.Unlock()
t.updateBuiltinWarnablesLocked()
return t.multiErrLocked()
}
// Strings() returns a string array containing the Text of all Warnings
// currently known to the Tracker. These strings can be presented to the
// user, although ideally you would use the Code property on each Warning
// to show a localized version of them instead.
// This function is here for legacy compatibility purposes and is deprecated.
func (t *Tracker) Strings() []string {
if t.nil() {
return nil
}
t.mu.Lock()
defer t.mu.Unlock()
return t.stringsLocked()
}
func (t *Tracker) stringsLocked() []string {
result := []string{}
for w, ws := range t.warnableVal {
if ws.Args == nil {
result = append(result, w.Text(Args{}))
} else {
result = append(result, w.Text(ws.Args))
}
}
return result
}
// errorsLocked returns an array of errors where each error is the Text
// of a Warning known to the Tracker.
// This function is here for legacy compatibility purposes and is deprecated.
func (t *Tracker) errorsLocked() []error {
strs := t.stringsLocked()
errs := []error{}
for _, str := range strs {
errs = append(errs, errors.New(str))
}
return errs
}
// multiErrLocked returns an error listing all errors known to the Tracker.
// This function is here for legacy compatibility purposes and is deprecated.
func (t *Tracker) multiErrLocked() error {
errs := t.errorsLocked()
return multierr.New(errs...)
}
var fakeErrForTesting = envknob.RegisterString("TS_DEBUG_FAKE_HEALTH_ERROR")
// updateBuiltinWarnablesLocked performs a number of checks on the state of the backend,
// and adds/removes Warnings from the Tracker as needed.
func (t *Tracker) updateBuiltinWarnablesLocked() {
if t.checkForUpdates {
if cv := t.latestVersion; cv != nil && !cv.RunningLatest && cv.LatestVersion != "" {
if cv.UrgentSecurityUpdate {
t.setUnhealthyLocked(securityUpdateAvailableWarnable, Args{
ArgCurrentVersion: version.Short(),
ArgAvailableVersion: cv.LatestVersion,
})
} else {
t.setUnhealthyLocked(updateAvailableWarnable, Args{
ArgCurrentVersion: version.Short(),
ArgAvailableVersion: cv.LatestVersion,
})
}
}
}
if version.IsUnstableBuild() {
t.setUnhealthyLocked(unstableWarnable, Args{
ArgCurrentVersion: version.Short(),
})
}
if v, ok := t.anyInterfaceUp.Get(); ok && !v {
t.setUnhealthyLocked(NetworkStatusWarnable, nil)
return
} else {
t.setHealthyLocked(NetworkStatusWarnable)
}
if t.localLogConfigErr != nil {
t.setUnhealthyLocked(localLogWarnable, Args{
ArgError: t.localLogConfigErr.Error(),
})
return
} else {
t.setHealthyLocked(localLogWarnable)
}
if !t.ipnWantRunning {
t.setUnhealthyLocked(IPNStateWarnable, Args{
"State": t.ipnState,
})
return
} else {
t.setHealthyLocked(IPNStateWarnable)
}
if t.lastLoginErr != nil {
t.setUnhealthyLocked(LoginStateWarnable, Args{
ArgError: t.lastLoginErr.Error(),
})
return
} else {
t.setHealthyLocked(LoginStateWarnable)
}
now := time.Now()
if !t.inMapPoll && (t.lastMapPollEndedAt.IsZero() || now.Sub(t.lastMapPollEndedAt) > 10*time.Second) {
t.setUnhealthyLocked(notInMapPollWarnable, nil)
return
} else {
t.setHealthyLocked(notInMapPollWarnable)
}
const tooIdle = 2*time.Minute + 5*time.Second
if d := now.Sub(t.lastStreamedMapResponse).Round(time.Second); d > tooIdle {
t.setUnhealthyLocked(mapResponseTimeoutWarnable, Args{
ArgDuration: d.String(),
})
return
} else {
t.setHealthyLocked(mapResponseTimeoutWarnable)
}
if !t.derpHomeless {
rid := t.derpHomeRegion
if rid == 0 {
t.setUnhealthyLocked(noDERPHomeWarnable, nil)
return
} else if !t.derpRegionConnected[rid] {
t.setUnhealthyLocked(noDERPConnectionWarnable, Args{
ArgDERPRegionID: fmt.Sprint(rid),
ArgDERPRegionName: t.derpMap.Regions[rid].RegionName,
})
return
} else if d := now.Sub(t.derpRegionLastFrame[rid]).Round(time.Second); d > tooIdle {
t.setUnhealthyLocked(derpTimeoutWarnable, Args{
ArgDERPRegionID: fmt.Sprint(rid),
ArgDERPRegionName: t.derpMap.Regions[rid].RegionName,
ArgDuration: d.String(),
})
return
}
}
t.setHealthyLocked(noDERPHomeWarnable)
t.setHealthyLocked(noDERPConnectionWarnable)
t.setHealthyLocked(derpTimeoutWarnable)
if t.udp4Unbound {
t.setUnhealthyLocked(noUDP4BindWarnable, nil)
return
} else {
t.setHealthyLocked(noUDP4BindWarnable)
}
// TODO: use
_ = t.inMapPollSince
_ = t.lastMapPollEndedAt
_ = t.lastStreamedMapResponse
_ = t.lastMapRequestHeard
shouldClearMagicsockWarnings := false
for i := range t.MagicSockReceiveFuncs {
f := &t.MagicSockReceiveFuncs[i]
if f.missing {
t.setUnhealthyLocked(magicsockReceiveFuncWarnable, Args{
ArgMagicsockFunctionName: f.name,
})
shouldClearMagicsockWarnings = false
}
}
if shouldClearMagicsockWarnings {
t.setHealthyLocked(magicsockReceiveFuncWarnable)
}
// Iterates over the legacy subsystems and their error, and turns them into structured errors
for sys, err := range t.sysErr {
t.updateLegacyErrorWarnableLocked(sys, err)
}
if len(t.derpRegionHealthProblem) > 0 {
for regionID, problem := range t.derpRegionHealthProblem {
t.setUnhealthyLocked(derpRegionErrorWarnable, Args{
ArgDERPRegionID: fmt.Sprint(regionID),
ArgError: problem,
})
}
} else {
t.setHealthyLocked(derpRegionErrorWarnable)
}
if len(t.controlHealth) > 0 {
for _, s := range t.controlHealth {
t.setUnhealthyLocked(controlHealthWarnable, Args{
ArgError: s,
})
}
} else {
t.setHealthyLocked(controlHealthWarnable)
}
if err := envknob.ApplyDiskConfigError(); err != nil {
t.setUnhealthyLocked(applyDiskConfigWarnable, Args{
ArgError: err.Error(),
})
} else {
t.setHealthyLocked(applyDiskConfigWarnable)
}
if len(t.tlsConnectionErrors) > 0 {
for serverName, err := range t.tlsConnectionErrors {
t.setUnhealthyLocked(tlsConnectionFailedWarnable, Args{
ArgServerName: serverName,
ArgError: err.Error(),
})
}
} else {
t.setHealthyLocked(tlsConnectionFailedWarnable)
}
if e := fakeErrForTesting(); len(t.warnables) == 0 && e != "" {
t.setUnhealthyLocked(testWarnable, Args{
ArgError: e,
})
} else {
t.setHealthyLocked(testWarnable)
}
}
// ReceiveFuncStats tracks the calls made to a wireguard-go receive func.
type ReceiveFuncStats struct {
// name is the name of the receive func.
// It's lazily populated.
name string
// numCalls is the number of times the receive func has ever been called.
// It is required because it is possible for a receive func's wireguard-go goroutine
// to be active even though the receive func isn't.
// The wireguard-go goroutine alternates between calling the receive func and
// processing what the func returned.
numCalls atomic.Uint64
// prevNumCalls is the value of numCalls last time the health check examined it.
prevNumCalls uint64
// inCall indicates whether the receive func is currently running.
inCall atomic.Bool
// missing indicates whether the receive func is not running.
missing bool
}
func (s *ReceiveFuncStats) Enter() {
s.numCalls.Add(1)
s.inCall.Store(true)
}
func (s *ReceiveFuncStats) Exit() {
s.inCall.Store(false)
}
// ReceiveFuncStats returns the ReceiveFuncStats tracker for the given func
// type.
//
// If t is nil, it returns nil.
func (t *Tracker) ReceiveFuncStats(which ReceiveFunc) *ReceiveFuncStats {
if t == nil {
return nil
}
return &t.MagicSockReceiveFuncs[which]
}
func (t *Tracker) checkReceiveFuncsLocked() {
for i := range t.MagicSockReceiveFuncs {
f := &t.MagicSockReceiveFuncs[i]
if f.name == "" {
f.name = (ReceiveFunc(i)).String()
}
if runtime.GOOS == "js" && i < 2 {
// Skip IPv4 and IPv6 on js.
continue
}
f.missing = false
prev := f.prevNumCalls
numCalls := f.numCalls.Load()
f.prevNumCalls = numCalls
if numCalls > prev {
// OK: the function has gotten called since last we checked
continue
}
if f.inCall.Load() {
// OK: the function is active, probably blocked due to inactivity
continue
}
// Not OK: The function is not active, and not accumulating new calls.
// It is probably MIA.
f.missing = true
}
}