mirror of
https://github.com/netbirdio/netbird.git
synced 2026-05-16 05:39:56 +00:00
Fix sleep detector races flagged in review
This commit is contained in:
Viktor Liu
@@ -20,33 +20,6 @@ const (
|
||||
kIOMessageSystemHasPoweredOn uintptr = 0xe0000300
|
||||
)
|
||||
|
||||
// IOKit / CoreFoundation symbols, resolved once at init.
|
||||
type iokitFuncs struct {
|
||||
IORegisterForSystemPower func(refcon uintptr, portRef *uintptr, callback uintptr, notifier *uintptr) uintptr
|
||||
IODeregisterForSystemPower func(notifier *uintptr) int32
|
||||
IOAllowPowerChange func(kernelPort uintptr, notificationID uintptr) int32
|
||||
IOServiceClose func(connect uintptr) int32
|
||||
IONotificationPortGetRunLoopSource func(port uintptr) uintptr
|
||||
IONotificationPortDestroy func(port uintptr)
|
||||
}
|
||||
|
||||
type cfFuncs struct {
|
||||
CFRunLoopGetCurrent func() uintptr
|
||||
CFRunLoopRun func()
|
||||
CFRunLoopStop func(rl uintptr)
|
||||
CFRunLoopAddSource func(rl, source, mode uintptr)
|
||||
CFRunLoopRemoveSource func(rl, source, mode uintptr)
|
||||
}
|
||||
|
||||
// runLoopSession bundles the handles owned by one CFRunLoop lifetime. A nil
|
||||
// session means no runloop is active and the next Register must start one.
|
||||
type runLoopSession struct {
|
||||
rl uintptr
|
||||
port uintptr
|
||||
notifier uintptr
|
||||
rp uintptr
|
||||
}
|
||||
|
||||
var (
|
||||
ioKit iokitFuncs
|
||||
cf cfFuncs
|
||||
@@ -67,110 +40,49 @@ var (
|
||||
lifecycleMu sync.Mutex
|
||||
)
|
||||
|
||||
func initLibs() error {
|
||||
libInitOnce.Do(func() {
|
||||
iokit, err := purego.Dlopen("/System/Library/Frameworks/IOKit.framework/IOKit", purego.RTLD_NOW|purego.RTLD_GLOBAL)
|
||||
if err != nil {
|
||||
libInitErr = fmt.Errorf("dlopen IOKit: %w", err)
|
||||
return
|
||||
}
|
||||
cfLib, err := purego.Dlopen("/System/Library/Frameworks/CoreFoundation.framework/CoreFoundation", purego.RTLD_NOW|purego.RTLD_GLOBAL)
|
||||
if err != nil {
|
||||
libInitErr = fmt.Errorf("dlopen CoreFoundation: %w", err)
|
||||
return
|
||||
}
|
||||
|
||||
purego.RegisterLibFunc(&ioKit.IORegisterForSystemPower, iokit, "IORegisterForSystemPower")
|
||||
purego.RegisterLibFunc(&ioKit.IODeregisterForSystemPower, iokit, "IODeregisterForSystemPower")
|
||||
purego.RegisterLibFunc(&ioKit.IOAllowPowerChange, iokit, "IOAllowPowerChange")
|
||||
purego.RegisterLibFunc(&ioKit.IOServiceClose, iokit, "IOServiceClose")
|
||||
purego.RegisterLibFunc(&ioKit.IONotificationPortGetRunLoopSource, iokit, "IONotificationPortGetRunLoopSource")
|
||||
purego.RegisterLibFunc(&ioKit.IONotificationPortDestroy, iokit, "IONotificationPortDestroy")
|
||||
|
||||
purego.RegisterLibFunc(&cf.CFRunLoopGetCurrent, cfLib, "CFRunLoopGetCurrent")
|
||||
purego.RegisterLibFunc(&cf.CFRunLoopRun, cfLib, "CFRunLoopRun")
|
||||
purego.RegisterLibFunc(&cf.CFRunLoopStop, cfLib, "CFRunLoopStop")
|
||||
purego.RegisterLibFunc(&cf.CFRunLoopAddSource, cfLib, "CFRunLoopAddSource")
|
||||
purego.RegisterLibFunc(&cf.CFRunLoopRemoveSource, cfLib, "CFRunLoopRemoveSource")
|
||||
|
||||
modeAddr, err := purego.Dlsym(cfLib, "kCFRunLoopCommonModes")
|
||||
if err != nil {
|
||||
libInitErr = fmt.Errorf("dlsym kCFRunLoopCommonModes: %w", err)
|
||||
return
|
||||
}
|
||||
// kCFRunLoopCommonModes is a CFStringRef variable. Launder the
|
||||
// uintptr-to-pointer conversion through the address of our Go
|
||||
// variable so go vet's unsafeptr analyzer doesn't flag it; the
|
||||
// address is a stable system-library global, not a Go heap pointer.
|
||||
cfCommonModes = **(**uintptr)(unsafe.Pointer(&modeAddr))
|
||||
|
||||
// Register the callback once for the lifetime of the process. NewCallback slots
|
||||
// are a finite, non-reclaimable resource, so a single thunk that dispatches
|
||||
// to the current Detector set is safer than registering per Register().
|
||||
callbackThunk = purego.NewCallback(powerCallback)
|
||||
})
|
||||
return libInitErr
|
||||
// iokitFuncs holds IOKit symbols resolved once at init.
|
||||
type iokitFuncs struct {
|
||||
IORegisterForSystemPower func(refcon uintptr, portRef *uintptr, callback uintptr, notifier *uintptr) uintptr
|
||||
IODeregisterForSystemPower func(notifier *uintptr) int32
|
||||
IOAllowPowerChange func(kernelPort uintptr, notificationID uintptr) int32
|
||||
IOServiceClose func(connect uintptr) int32
|
||||
IONotificationPortGetRunLoopSource func(port uintptr) uintptr
|
||||
IONotificationPortDestroy func(port uintptr)
|
||||
}
|
||||
|
||||
// powerCallback is the IOServiceInterestCallback trampoline, invoked on the
|
||||
// runloop thread. A Go panic crossing the purego boundary has undefined
|
||||
// behavior, so contain it here.
|
||||
func powerCallback(refcon, service, messageType, messageArgument uintptr) uintptr {
|
||||
defer func() {
|
||||
if r := recover(); r != nil {
|
||||
log.Errorf("panic in sleep powerCallback: %v", r)
|
||||
}
|
||||
}()
|
||||
switch messageType {
|
||||
case kIOMessageCanSystemSleep:
|
||||
// Not acknowledging forces a 30s IOKit timeout before idle sleep.
|
||||
allowPowerChange(messageArgument)
|
||||
case kIOMessageSystemWillSleep:
|
||||
dispatchEvent(EventTypeSleep)
|
||||
allowPowerChange(messageArgument)
|
||||
case kIOMessageSystemHasPoweredOn:
|
||||
dispatchEvent(EventTypeWakeUp)
|
||||
}
|
||||
return 0
|
||||
// cfFuncs holds CoreFoundation symbols resolved once at init.
|
||||
type cfFuncs struct {
|
||||
CFRunLoopGetCurrent func() uintptr
|
||||
CFRunLoopRun func()
|
||||
CFRunLoopStop func(rl uintptr)
|
||||
CFRunLoopAddSource func(rl, source, mode uintptr)
|
||||
CFRunLoopRemoveSource func(rl, source, mode uintptr)
|
||||
}
|
||||
|
||||
func allowPowerChange(messageArgument uintptr) {
|
||||
serviceRegistryMu.Lock()
|
||||
var port uintptr
|
||||
if session != nil {
|
||||
port = session.rp
|
||||
}
|
||||
serviceRegistryMu.Unlock()
|
||||
if port != 0 {
|
||||
ioKit.IOAllowPowerChange(port, messageArgument)
|
||||
}
|
||||
// runLoopSession bundles the handles owned by one CFRunLoop lifetime. A nil
|
||||
// session means no runloop is active and the next Register must start one.
|
||||
type runLoopSession struct {
|
||||
rl uintptr
|
||||
port uintptr
|
||||
notifier uintptr
|
||||
rp uintptr
|
||||
}
|
||||
|
||||
func dispatchEvent(event EventType) {
|
||||
serviceRegistryMu.Lock()
|
||||
detectors := make([]*Detector, 0, len(serviceRegistry))
|
||||
for d := range serviceRegistry {
|
||||
detectors = append(detectors, d)
|
||||
}
|
||||
serviceRegistryMu.Unlock()
|
||||
|
||||
for _, d := range detectors {
|
||||
d.triggerCallback(event)
|
||||
}
|
||||
// detectorSnapshot pins a detector's callback and done channel so dispatch
|
||||
// runs with values valid at snapshot time, even if a concurrent
|
||||
// Deregister/Register rewrites the detector's fields.
|
||||
type detectorSnapshot struct {
|
||||
detector *Detector
|
||||
callback func(event EventType)
|
||||
done <-chan struct{}
|
||||
}
|
||||
|
||||
// Detector delivers sleep and wake events to a registered callback.
|
||||
type Detector struct {
|
||||
callback func(event EventType)
|
||||
done chan struct{}
|
||||
}
|
||||
|
||||
func NewDetector() (*Detector, error) {
|
||||
if err := initLibs(); err != nil {
|
||||
return nil, err
|
||||
}
|
||||
return &Detector{}, nil
|
||||
}
|
||||
|
||||
// Register installs callback for power events. The first registration starts
|
||||
// the CFRunLoop on a dedicated OS-locked thread and blocks until IOKit
|
||||
// registration succeeds or fails; subsequent registrations just add to the
|
||||
@@ -228,7 +140,6 @@ func (d *Detector) Deregister() error {
|
||||
return nil
|
||||
}
|
||||
sess := session
|
||||
session = nil
|
||||
serviceRegistryMu.Unlock()
|
||||
|
||||
log.Info("sleep detection service stopping (deregister)")
|
||||
@@ -237,7 +148,6 @@ func (d *Detector) Deregister() error {
|
||||
return nil
|
||||
}
|
||||
|
||||
// CFRunLoop and IOKit teardown calls are safe from a non-runloop thread.
|
||||
if sess.rl != 0 && sess.port != 0 {
|
||||
source := ioKit.IONotificationPortGetRunLoopSource(sess.port)
|
||||
cf.CFRunLoopRemoveSource(sess.rl, source, cfCommonModes)
|
||||
@@ -246,6 +156,13 @@ func (d *Detector) Deregister() error {
|
||||
n := sess.notifier
|
||||
ioKit.IODeregisterForSystemPower(&n)
|
||||
}
|
||||
|
||||
// Clear session only after IODeregisterForSystemPower returns so any
|
||||
// in-flight powerCallback can still look up session.rp to ack sleep.
|
||||
serviceRegistryMu.Lock()
|
||||
session = nil
|
||||
serviceRegistryMu.Unlock()
|
||||
|
||||
if sess.rp != 0 {
|
||||
ioKit.IOServiceClose(sess.rp)
|
||||
}
|
||||
@@ -259,12 +176,17 @@ func (d *Detector) Deregister() error {
|
||||
return nil
|
||||
}
|
||||
|
||||
func (d *Detector) triggerCallback(event EventType) {
|
||||
cb := d.callback
|
||||
if cb == nil {
|
||||
func (d *Detector) triggerCallback(event EventType, cb func(event EventType), done <-chan struct{}) {
|
||||
if cb == nil || done == nil {
|
||||
return
|
||||
}
|
||||
|
||||
select {
|
||||
case <-done:
|
||||
return
|
||||
default:
|
||||
}
|
||||
|
||||
doneChan := make(chan struct{})
|
||||
timeout := time.NewTimer(500 * time.Millisecond)
|
||||
defer timeout.Stop()
|
||||
@@ -282,12 +204,113 @@ func (d *Detector) triggerCallback(event EventType) {
|
||||
|
||||
select {
|
||||
case <-doneChan:
|
||||
case <-d.done:
|
||||
case <-done:
|
||||
case <-timeout.C:
|
||||
log.Warn("sleep callback timed out")
|
||||
}
|
||||
}
|
||||
|
||||
// NewDetector initializes IOKit/CoreFoundation bindings and returns a Detector.
|
||||
func NewDetector() (*Detector, error) {
|
||||
if err := initLibs(); err != nil {
|
||||
return nil, err
|
||||
}
|
||||
return &Detector{}, nil
|
||||
}
|
||||
|
||||
func initLibs() error {
|
||||
libInitOnce.Do(func() {
|
||||
iokit, err := purego.Dlopen("/System/Library/Frameworks/IOKit.framework/IOKit", purego.RTLD_NOW|purego.RTLD_GLOBAL)
|
||||
if err != nil {
|
||||
libInitErr = fmt.Errorf("dlopen IOKit: %w", err)
|
||||
return
|
||||
}
|
||||
cfLib, err := purego.Dlopen("/System/Library/Frameworks/CoreFoundation.framework/CoreFoundation", purego.RTLD_NOW|purego.RTLD_GLOBAL)
|
||||
if err != nil {
|
||||
libInitErr = fmt.Errorf("dlopen CoreFoundation: %w", err)
|
||||
return
|
||||
}
|
||||
|
||||
purego.RegisterLibFunc(&ioKit.IORegisterForSystemPower, iokit, "IORegisterForSystemPower")
|
||||
purego.RegisterLibFunc(&ioKit.IODeregisterForSystemPower, iokit, "IODeregisterForSystemPower")
|
||||
purego.RegisterLibFunc(&ioKit.IOAllowPowerChange, iokit, "IOAllowPowerChange")
|
||||
purego.RegisterLibFunc(&ioKit.IOServiceClose, iokit, "IOServiceClose")
|
||||
purego.RegisterLibFunc(&ioKit.IONotificationPortGetRunLoopSource, iokit, "IONotificationPortGetRunLoopSource")
|
||||
purego.RegisterLibFunc(&ioKit.IONotificationPortDestroy, iokit, "IONotificationPortDestroy")
|
||||
|
||||
purego.RegisterLibFunc(&cf.CFRunLoopGetCurrent, cfLib, "CFRunLoopGetCurrent")
|
||||
purego.RegisterLibFunc(&cf.CFRunLoopRun, cfLib, "CFRunLoopRun")
|
||||
purego.RegisterLibFunc(&cf.CFRunLoopStop, cfLib, "CFRunLoopStop")
|
||||
purego.RegisterLibFunc(&cf.CFRunLoopAddSource, cfLib, "CFRunLoopAddSource")
|
||||
purego.RegisterLibFunc(&cf.CFRunLoopRemoveSource, cfLib, "CFRunLoopRemoveSource")
|
||||
|
||||
modeAddr, err := purego.Dlsym(cfLib, "kCFRunLoopCommonModes")
|
||||
if err != nil {
|
||||
libInitErr = fmt.Errorf("dlsym kCFRunLoopCommonModes: %w", err)
|
||||
return
|
||||
}
|
||||
// Launder the uintptr-to-pointer conversion through a Go variable so
|
||||
// go vet's unsafeptr analyzer doesn't flag a system-library global.
|
||||
cfCommonModes = **(**uintptr)(unsafe.Pointer(&modeAddr))
|
||||
|
||||
// NewCallback slots are a finite, non-reclaimable resource, so register
|
||||
// a single thunk that dispatches to the current Detector set.
|
||||
callbackThunk = purego.NewCallback(powerCallback)
|
||||
})
|
||||
return libInitErr
|
||||
}
|
||||
|
||||
// powerCallback is the IOServiceInterestCallback trampoline, invoked on the
|
||||
// runloop thread. A Go panic crossing the purego boundary has undefined
|
||||
// behavior, so contain it here.
|
||||
func powerCallback(refcon, service, messageType, messageArgument uintptr) uintptr {
|
||||
defer func() {
|
||||
if r := recover(); r != nil {
|
||||
log.Errorf("panic in sleep powerCallback: %v", r)
|
||||
}
|
||||
}()
|
||||
switch messageType {
|
||||
case kIOMessageCanSystemSleep:
|
||||
// Not acknowledging forces a 30s IOKit timeout before idle sleep.
|
||||
allowPowerChange(messageArgument)
|
||||
case kIOMessageSystemWillSleep:
|
||||
dispatchEvent(EventTypeSleep)
|
||||
allowPowerChange(messageArgument)
|
||||
case kIOMessageSystemHasPoweredOn:
|
||||
dispatchEvent(EventTypeWakeUp)
|
||||
}
|
||||
return 0
|
||||
}
|
||||
|
||||
func allowPowerChange(messageArgument uintptr) {
|
||||
serviceRegistryMu.Lock()
|
||||
var port uintptr
|
||||
if session != nil {
|
||||
port = session.rp
|
||||
}
|
||||
serviceRegistryMu.Unlock()
|
||||
if port != 0 {
|
||||
ioKit.IOAllowPowerChange(port, messageArgument)
|
||||
}
|
||||
}
|
||||
|
||||
func dispatchEvent(event EventType) {
|
||||
serviceRegistryMu.Lock()
|
||||
snaps := make([]detectorSnapshot, 0, len(serviceRegistry))
|
||||
for d := range serviceRegistry {
|
||||
snaps = append(snaps, detectorSnapshot{
|
||||
detector: d,
|
||||
callback: d.callback,
|
||||
done: d.done,
|
||||
})
|
||||
}
|
||||
serviceRegistryMu.Unlock()
|
||||
|
||||
for _, s := range snaps {
|
||||
s.detector.triggerCallback(event, s.callback, s.done)
|
||||
}
|
||||
}
|
||||
|
||||
// runRunLoop owns the OS-locked thread that CFRunLoop is pinned to. Setup
|
||||
// result is reported on errCh so Register can surface failures synchronously.
|
||||
func runRunLoop(errCh chan<- error) {
|
||||
@@ -313,8 +336,8 @@ func runRunLoop(errCh chan<- error) {
|
||||
cf.CFRunLoopRun()
|
||||
}
|
||||
|
||||
// setupSession performs the IOKit registration on the current thread.
|
||||
// Panics are converted to errors so runRunLoop never leaves errCh unsent.
|
||||
// setupSession performs the IOKit registration on the current thread. Panics
|
||||
// are converted to errors so runRunLoop never leaves errCh unsent.
|
||||
func setupSession() (s *runLoopSession, err error) {
|
||||
defer func() {
|
||||
if r := recover(); r != nil {
|
||||
|
||||
Reference in New Issue
Block a user