fix(healthcheck): prevent race condition on the healthchecker (#3400)

Co-authored-by: Matt Van Horn <455140+mvanhorn@users.noreply.github.com>
Co-authored-by: Quentin McGaw <quentin.mcgaw@gmail.com>

Thanks also to @arturhoo for also trying to resolve this issue in #3410

internal/healthcheck/checker.go

@@ -8,7 +8,6 @@ import (
 	"net"
 	"net/netip"
 	"strings"
-	"sync"
 	"time"
 
 	"github.com/qdm12/gluetun/internal/healthcheck/dns"
@@ -24,7 +23,6 @@ type Checker struct {
 	icmpTargetIPs  []netip.Addr
 	smallCheckType string
 	startupOnFail  bool
-	configMutex    sync.Mutex
 
 	icmpNotPermitted *bool
 
@@ -55,8 +53,6 @@ func NewChecker(logger Logger) *Checker {
 func (c *Checker) SetConfig(tlsDialAddrs []string, icmpTargets []netip.Addr,
 	smallCheckType string, startupOnFail bool,
 ) {
-	c.configMutex.Lock()
-	defer c.configMutex.Unlock()
 	c.tlsDialAddrs = tlsDialAddrs
 	c.icmpTargetIPs = icmpTargets
 	c.smallCheckType = smallCheckType
@@ -166,10 +162,8 @@ func (c *Checker) Stop() error {
 }
 
 func (c *Checker) smallPeriodicCheck(ctx context.Context) error {
-	c.configMutex.Lock()
 	icmpTargetIPs := make([]netip.Addr, len(c.icmpTargetIPs))
 	copy(icmpTargetIPs, c.icmpTargetIPs)
-	c.configMutex.Unlock()
 	tryTimeouts := []time.Duration{
 		5 * time.Second,
 		5 * time.Second,

internal/vpn/loop.go

@@ -45,6 +45,7 @@ type Loop struct {
 	start       <-chan struct{}
 	running     chan<- models.LoopStatus
 	userTrigger bool
+	healthDone  <-chan struct{}
 	// Internal constant values
 	backoffTime time.Duration
 }
@@ -70,6 +71,11 @@ func NewLoop(vpnSettings settings.VPN, ipv6SupportLevel netlink.IPv6SupportLevel
 	statusManager := loopstate.New(constants.Stopped, start, running, stop, stopped)
 	state := state.New(statusManager, vpnSettings)
 
+	// Initialize healthDone channel to a closed channel so that the first time
+	// we call <-l.healthDone it does not block
+	healthDone := make(chan struct{})
+	close(healthDone)
+
 	return &Loop{
 		statusManager:    statusManager,
 		state:            state,
@@ -98,6 +104,7 @@ func NewLoop(vpnSettings settings.VPN, ipv6SupportLevel netlink.IPv6SupportLevel
 		stop:             stop,
 		stopped:          stopped,
 		userTrigger:      true,
+		healthDone:       healthDone,
 		backoffTime:      defaultBackoffTime,
 	}
 }

internal/vpn/tunnelup.go

@@ -81,6 +81,7 @@ func (l *Loop) onTunnelUp(ctx, loopCtx context.Context, data tunnelUpData) {
 
 	_, _ = l.dnsLooper.ApplyStatus(ctx, constants.Running)
 
+	<-l.healthDone // make sure the health checker is stopped before restarting it
 	icmpTargetIPs := l.healthSettings.ICMPTargetIPs
 	if len(icmpTargetIPs) == 1 && icmpTargetIPs[0].IsUnspecified() {
 		icmpTargetIPs = []netip.Addr{data.serverIP}
@@ -104,7 +105,12 @@ func (l *Loop) onTunnelUp(ctx, loopCtx context.Context, data tunnelUpData) {
 	// Start collecting health errors asynchronously, since
 	// we should not wait for the code below to complete
 	// to start monitoring health and auto-healing.
-	go l.collectHealthErrors(ctx, loopCtx, healthErrCh)
+	// We keep track of when this goroutine is done with the healthDone
+	// channel to avoid a race condition where the health checker is stopped
+	// after being reconfigured and restarted, instead of before.
+	healthDone := make(chan struct{})
+	l.healthDone = healthDone
+	go l.collectHealthErrors(ctx, loopCtx, healthDone, healthErrCh)
 
 	err = l.publicip.RunOnce(ctx)
 	if err != nil {
@@ -140,7 +146,10 @@ func (l *Loop) onTunnelUp(ctx, loopCtx context.Context, data tunnelUpData) {
 	}
 }
 
-func (l *Loop) collectHealthErrors(ctx, loopCtx context.Context, healthErrCh <-chan error) {
+func (l *Loop) collectHealthErrors(ctx, loopCtx context.Context,
+	done chan<- struct{}, healthErrCh <-chan error,
+) {
+	defer close(done)
 	var previousHealthErr error
 	for {
 		select {