[client] Fix hanging status command during relay dial (#6694)

* Add regression test for relay state lock
* Make connect not hold a lock in openConnVia

client/internal/peer/guard/guard.go

@@ -85,7 +85,11 @@ func (g *Guard) reconnectLoopWithRetry(ctx context.Context, callback func()) {
 	defer g.srWatcher.RemoveListener(srReconnectedChan)
 
 	ticker := g.initialTicker(ctx)
-	defer ticker.Stop()
+	defer func() {
+		// If backoff.Ticker.send is blocked, context.Done will not close the Ticker goroutine.
+		// We have to explicitly call Stop, even if we use backoff.WithContext.
+		ticker.Stop()
+	}()
 
 	tickerChannel := ticker.C
 

client/internal/peer/guard/guard_leak_test.go

@@ -0,0 +1,92 @@
+package guard
+
+import (
+	"context"
+	"runtime"
+	"strings"
+	"sync"
+	"testing"
+	"time"
+
+	log "github.com/sirupsen/logrus"
+
+	"github.com/netbirdio/netbird/client/internal/peer/ice"
+)
+
+func newTestGuard(status connStatusFunc) *Guard {
+	srw := NewSRWatcher(nil, nil, nil, ice.Config{})
+	return NewGuard(log.WithField("test", "guard"), status, 50*time.Millisecond, srw)
+}
+
+// countBackoffTickerGoroutines returns how many goroutines are currently sitting
+// in backoff/v4.(*Ticker).run (a ticker goroutine that has not exited).
+func countBackoffTickerGoroutines() int {
+	buf := make([]byte, 1<<25) // 32MB
+	n := runtime.Stack(buf, true)
+	return strings.Count(string(buf[:n]), "backoff/v4.(*Ticker).run")
+}
+
+// TestGuard_ReconnectTicker_NoGoroutineLeakOnShutdown reproduces a observed
+// leak: after a shutdown burst, ticker run/send goroutines stay parked
+// forever even though every reconnect loop has exited.
+func TestGuard_ReconnectTicker_NoGoroutineLeakOnShutdown(t *testing.T) {
+	before := countBackoffTickerGoroutines()
+
+	const peers = 6000
+	cancels := make([]context.CancelFunc, 0, peers)
+	var wg sync.WaitGroup
+
+	// A status check slower than the tick cadence. This models the real
+	// isConnectedOnAllWay/callback doing work: while the loop is busy in the
+	// handler, the ticker fires the next tick and parks in send(), because
+	// send() never selects on ctx.
+	slowStatus := func() ConnStatus {
+		time.Sleep(70 * time.Millisecond)
+		return ConnStatusConnected
+	}
+
+	for range peers {
+		g := newTestGuard(slowStatus)
+		ctx, cancel := context.WithCancel(context.Background())
+		cancels = append(cancels, cancel)
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			g.Start(ctx, func() {})
+		}()
+		// Force the live ticker to be a newReconnectTicker.
+		g.SetRelayedConnDisconnected()
+	}
+
+	// Let the replacement tickers get past their 800ms initial interval, so
+	// many are parked in send() waiting on the (slow) consumer when we tear
+	// everything down.
+	time.Sleep(1500 * time.Millisecond)
+
+	// Shutdown burst: cancel every peer at once, like engine teardown.
+	for _, c := range cancels {
+		c()
+	}
+
+	// Every reconnect loop must return
+	waitCh := make(chan struct{})
+	go func() { wg.Wait(); close(waitCh) }()
+	select {
+	case <-waitCh:
+	case <-time.After(30 * time.Second):
+		t.Fatal("not all reconnect loops returned after ctx cancel")
+	}
+
+	// Give any correctly-stopped ticker goroutines time to unwind.
+	for range 50 {
+		runtime.Gosched()
+		time.Sleep(10 * time.Millisecond)
+	}
+
+	leaked := countBackoffTickerGoroutines() - before
+	t.Logf("backoff Ticker.run goroutines still parked after teardown of %d peers: %d", peers, leaked)
+	if leaked > 0 {
+		t.Errorf("LEAK: %d backoff ticker goroutines parked after all reconnect loops exited "+
+			"(defer ticker.Stop() stops the initial ticker, not the live replacement)", leaked)
+	}
+}

client/internal/routemanager/exit_node_selection_test.go

@@ -0,0 +1,191 @@
+package routemanager
+
+import (
+	"net/netip"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/client/internal/routeselector"
+	"github.com/netbirdio/netbird/route"
+)
+
+func newExitNodeTestManager() *DefaultManager {
+	return &DefaultManager{routeSelector: routeselector.NewRouteSelector()}
+}
+
+func exitRoute(netID, peer string, skipAutoApply bool) *route.Route {
+	return &route.Route{
+		NetID:         route.NetID(netID),
+		Network:       netip.MustParsePrefix("0.0.0.0/0"),
+		Peer:          peer,
+		SkipAutoApply: skipAutoApply,
+	}
+}
+
+func TestPickPreferredExitNode(t *testing.T) {
+	tests := []struct {
+		name string
+		info exitNodeInfo
+		want route.NetID
+	}{
+		{
+			name: "persisted user selection wins over management",
+			info: exitNodeInfo{
+				allIDs:               []route.NetID{"a", "b", "c"},
+				userSelected:         []route.NetID{"b"},
+				selectedByManagement: []route.NetID{"a"},
+			},
+			want: "b",
+		},
+		{
+			name: "multiple user-selected self-heal to deterministic min",
+			info: exitNodeInfo{
+				allIDs:       []route.NetID{"a", "b", "c"},
+				userSelected: []route.NetID{"c", "a"},
+			},
+			want: "a",
+		},
+		{
+			name: "explicit opt-out keeps none",
+			info: exitNodeInfo{
+				allIDs:         []route.NetID{"a", "b"},
+				userDeselected: []route.NetID{"a", "b"},
+			},
+			want: "",
+		},
+		{
+			name: "fresh defaults to management auto-apply pick",
+			info: exitNodeInfo{
+				allIDs:               []route.NetID{"a", "b", "c"},
+				selectedByManagement: []route.NetID{"b"},
+			},
+			want: "b",
+		},
+		{
+			name: "no user pick and no management auto-apply selects none",
+			info: exitNodeInfo{
+				allIDs: []route.NetID{"c", "a", "b"},
+			},
+			want: "",
+		},
+		{
+			name: "user-deselect does not block a management auto-apply sibling",
+			info: exitNodeInfo{
+				allIDs:               []route.NetID{"a", "b"},
+				userDeselected:       []route.NetID{"a"},
+				selectedByManagement: []route.NetID{"b"},
+			},
+			want: "b",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			assert.Equal(t, tt.want, pickPreferredExitNode(tt.info), "preferred exit node")
+		})
+	}
+}
+
+func TestEnforceSingleExitNode(t *testing.T) {
+	m := newExitNodeTestManager()
+	all := []route.NetID{"a", "b", "c"}
+
+	m.enforceSingleExitNode("b", all)
+	assert.False(t, m.routeSelector.IsSelected("a"), "a should be deselected")
+	assert.True(t, m.routeSelector.IsSelected("b"), "b should be the only selected exit node")
+	assert.False(t, m.routeSelector.IsSelected("c"), "c should be deselected")
+
+	// Switching the preferred node moves the single selection.
+	m.enforceSingleExitNode("c", all)
+	assert.False(t, m.routeSelector.IsSelected("a"), "a stays deselected")
+	assert.False(t, m.routeSelector.IsSelected("b"), "b should now be deselected")
+	assert.True(t, m.routeSelector.IsSelected("c"), "c should now be selected")
+
+	// Empty preferred turns every exit node off.
+	m.enforceSingleExitNode("", all)
+	for _, id := range all {
+		assert.False(t, m.routeSelector.IsSelected(id), "no exit node should be selected")
+	}
+}
+
+func TestEnforceSingleExitNode_RespectsDeselectAll(t *testing.T) {
+	m := newExitNodeTestManager()
+	m.routeSelector.DeselectAllRoutes()
+
+	m.enforceSingleExitNode("b", []route.NetID{"a", "b"})
+
+	assert.True(t, m.routeSelector.IsDeselectAll(), "global deselect-all must stay in effect")
+	assert.False(t, m.routeSelector.IsSelected("b"), "no exit node should be forced on while deselect-all is set")
+}
+
+func TestUpdateRouteSelectorFromManagement_FreshSelectsOne(t *testing.T) {
+	m := newExitNodeTestManager()
+	routes := route.HAMap{
+		"exitA|0.0.0.0/0":    {exitRoute("exitA", "p1", false)},
+		"exitB|0.0.0.0/0":    {exitRoute("exitB", "p2", false)},
+		"lan|192.168.1.0/24": {{NetID: "lan", Network: netip.MustParsePrefix("192.168.1.0/24"), Peer: "p3"}},
+		"exitC|0.0.0.0/0":    {exitRoute("exitC", "p4", false)},
+	}
+
+	m.updateRouteSelectorFromManagement(routes)
+
+	// Exactly one exit node (the deterministic first) is selected.
+	assert.True(t, m.routeSelector.IsSelected("exitA"), "exitA is the deterministic default")
+	assert.False(t, m.routeSelector.IsSelected("exitB"), "exitB must not also be selected")
+	assert.False(t, m.routeSelector.IsSelected("exitC"), "exitC must not also be selected")
+	// Non-exit routes are left at their default-on state.
+	assert.True(t, m.routeSelector.IsSelected("lan"), "non-exit route selection is untouched")
+}
+
+func TestUpdateRouteSelectorFromManagement_HonorsPersistedPick(t *testing.T) {
+	m := newExitNodeTestManager()
+	routes := route.HAMap{
+		"exitA|0.0.0.0/0": {exitRoute("exitA", "p1", false)},
+		"exitB|0.0.0.0/0": {exitRoute("exitB", "p2", false)},
+	}
+	all := []route.NetID{"exitA", "exitB"}
+
+	// Simulate the state the runtime select path leaves behind: exactly one
+	// exit node explicitly selected, its sibling deselected.
+	require.NoError(t, m.routeSelector.SelectRoutes([]route.NetID{"exitB"}, true, all))
+	require.NoError(t, m.routeSelector.DeselectRoutes([]route.NetID{"exitA"}, all))
+
+	m.updateRouteSelectorFromManagement(routes)
+
+	assert.True(t, m.routeSelector.IsSelected("exitB"), "persisted pick must stay selected")
+	assert.False(t, m.routeSelector.IsSelected("exitA"), "the other exit node stays deselected")
+}
+
+func TestUpdateRouteSelectorFromManagement_OptOutKeepsNone(t *testing.T) {
+	m := newExitNodeTestManager()
+	routes := route.HAMap{
+		"exitA|0.0.0.0/0": {exitRoute("exitA", "p1", false)},
+		"exitB|0.0.0.0/0": {exitRoute("exitB", "p2", false)},
+	}
+	all := []route.NetID{"exitA", "exitB"}
+
+	// User deselected exit nodes and selected none.
+	require.NoError(t, m.routeSelector.DeselectRoutes(all, all))
+
+	m.updateRouteSelectorFromManagement(routes)
+
+	assert.False(t, m.routeSelector.IsSelected("exitA"), "opt-out keeps exitA off")
+	assert.False(t, m.routeSelector.IsSelected("exitB"), "opt-out keeps exitB off")
+}
+
+func TestUpdateRouteSelectorFromManagement_NoAutoApplySelectsNone(t *testing.T) {
+	m := newExitNodeTestManager()
+	// SkipAutoApply=true: management offers the exit nodes but doesn't request
+	// auto-activation, so none should be selected until the user picks one.
+	routes := route.HAMap{
+		"exitA|0.0.0.0/0": {exitRoute("exitA", "p1", true)},
+		"exitB|0.0.0.0/0": {exitRoute("exitB", "p2", true)},
+	}
+
+	m.updateRouteSelectorFromManagement(routes)
+
+	assert.False(t, m.routeSelector.IsSelected("exitA"), "no auto-apply keeps exitA off")
+	assert.False(t, m.routeSelector.IsSelected("exitB"), "no auto-apply keeps exitB off")
+}

client/internal/routemanager/manager.go

@@ -701,7 +701,13 @@ func resolveURLsToIPs(urls []string) []net.IP {
 	return ips
 }
 
-// updateRouteSelectorFromManagement updates the route selector based on the isSelected status from the management server
+// updateRouteSelectorFromManagement reconciles exit-node selection on every
+// network map: it keeps at most one exit node selected — the user's persisted
+// pick, else whatever management marks for auto-apply (SkipAutoApply=false),
+// else none. We never auto-activate an exit node the map doesn't request; it
+// stays off until the user picks it. Exit nodes are mutually exclusive, but the
+// RouteSelector stores routes with default-on semantics, so without this every
+// available exit node would report selected at once.
 func (m *DefaultManager) updateRouteSelectorFromManagement(clientRoutes route.HAMap) {
 	m.mirrorV6ExitPairSelections(clientRoutes)
 
@@ -712,13 +718,14 @@ func (m *DefaultManager) updateRouteSelectorFromManagement(clientRoutes route.HA
 		return
 	}
 
-	exitNodeInfo := m.collectExitNodeInfo(clientRoutes)
-	if len(exitNodeInfo.allIDs) == 0 {
+	info := m.collectExitNodeInfo(clientRoutes)
+	if len(info.allIDs) == 0 {
 		return
 	}
 
-	m.updateExitNodeSelections(exitNodeInfo)
-	m.logExitNodeUpdate(exitNodeInfo)
+	preferred := pickPreferredExitNode(info)
+	m.enforceSingleExitNode(preferred, info.allIDs)
+	m.logExitNodeUpdate(info, preferred)
 }
 
 // mirrorV6ExitPairSelections keeps every synthesized "-v6" exit route's selection
@@ -746,6 +753,10 @@ type exitNodeInfo struct {
 	userDeselected       []route.NetID
 }
 
+// collectExitNodeInfo categorises the available exit nodes by their persisted
+// selection state. It keys on the base (v4) NetID and skips the synthesized
+// "-v6" partner, which inherits its base's selection through the RouteSelector
+// — counting it separately would double-count the pair.
 func (m *DefaultManager) collectExitNodeInfo(clientRoutes route.HAMap) exitNodeInfo {
 	var info exitNodeInfo
 
@@ -755,6 +766,9 @@ func (m *DefaultManager) collectExitNodeInfo(clientRoutes route.HAMap) exitNodeI
 		}
 
 		netID := haID.NetID()
+		if strings.HasSuffix(string(netID), route.V6ExitSuffix) {
+			continue
+		}
 		info.allIDs = append(info.allIDs, netID)
 
 		if m.routeSelector.HasUserSelectionForRoute(netID) {
@@ -791,45 +805,52 @@ func (m *DefaultManager) checkManagementSelection(routes []*route.Route, netID r
 	}
 }
 
-func (m *DefaultManager) updateExitNodeSelections(info exitNodeInfo) {
-	routesToDeselect := m.getRoutesToDeselect(info.allIDs)
-	m.deselectExitNodes(routesToDeselect)
-	m.selectExitNodesByManagement(info.selectedByManagement, info.allIDs)
+// pickPreferredExitNode chooses the single exit node to keep selected. In order:
+//   - a persisted user selection wins (deterministic if several survive from
+//     legacy state, so the set self-heals down to one);
+//   - otherwise activate only what management marks for auto-apply
+//     (SkipAutoApply=false); the lexicographically first if it marks several.
+//
+// Returns "" when neither holds — we never force an arbitrary exit node on. A
+// route the map doesn't auto-apply stays off until the user selects it.
+// info.userDeselected is informational only: an explicit deselect simply keeps
+// that route out of both lists above, so it can't be picked.
+func pickPreferredExitNode(info exitNodeInfo) route.NetID {
+	if len(info.userSelected) > 0 {
+		return minNetID(info.userSelected)
+	}
+	if len(info.selectedByManagement) > 0 {
+		return minNetID(info.selectedByManagement)
+	}
+	return ""
 }
 
-func (m *DefaultManager) getRoutesToDeselect(allIDs []route.NetID) []route.NetID {
-	var routesToDeselect []route.NetID
-	for _, netID := range allIDs {
-		if !m.routeSelector.HasUserSelectionForRoute(netID) {
-			routesToDeselect = append(routesToDeselect, netID)
+// enforceSingleExitNode makes preferred the only selected exit node: every other
+// available exit node is deselected and preferred (if any) is selected, without
+// disturbing non-exit route selections. The whole reconciliation runs under a
+// single RouteSelector lock (SetExclusiveExitNode) so a concurrent deselect-all
+// cannot interleave and get undone; a global deselect-all is left untouched so
+// the user's "all off" stays in effect.
+func (m *DefaultManager) enforceSingleExitNode(preferred route.NetID, allIDs []route.NetID) {
+	m.routeSelector.SetExclusiveExitNode(preferred, allIDs)
+}
+
+func (m *DefaultManager) logExitNodeUpdate(info exitNodeInfo, preferred route.NetID) {
+	log.Debugf("Exit node selection: %d available, preferred=%q (%d user-selected, %d user-deselected, %d management-selected)",
+		len(info.allIDs), preferred, len(info.userSelected), len(info.userDeselected), len(info.selectedByManagement))
+}
+
+// minNetID returns the lexicographically smallest NetID, for a deterministic
+// default pick that stays stable across restarts.
+func minNetID(ids []route.NetID) route.NetID {
+	if len(ids) == 0 {
+		return ""
+	}
+	best := ids[0]
+	for _, id := range ids[1:] {
+		if id < best {
+			best = id
 		}
 	}
-	return routesToDeselect
-}
-
-func (m *DefaultManager) deselectExitNodes(routesToDeselect []route.NetID) {
-	if len(routesToDeselect) == 0 {
-		return
-	}
-
-	err := m.routeSelector.DeselectRoutes(routesToDeselect, routesToDeselect)
-	if err != nil {
-		log.Warnf("Failed to deselect exit nodes: %v", err)
-	}
-}
-
-func (m *DefaultManager) selectExitNodesByManagement(selectedByManagement []route.NetID, allIDs []route.NetID) {
-	if len(selectedByManagement) == 0 {
-		return
-	}
-
-	err := m.routeSelector.SelectRoutes(selectedByManagement, true, allIDs)
-	if err != nil {
-		log.Warnf("Failed to select exit nodes: %v", err)
-	}
-}
-
-func (m *DefaultManager) logExitNodeUpdate(info exitNodeInfo) {
-	log.Debugf("Updated route selector: %d exit nodes available, %d selected by management, %d user-selected, %d user-deselected",
-		len(info.allIDs), len(info.selectedByManagement), len(info.userSelected), len(info.userDeselected))
+	return best
 }

client/internal/routeselector/routeselector.go

@@ -115,7 +115,38 @@ func (rs *RouteSelector) DeselectAllRoutes() {
 	clear(rs.selectedRoutes)
 }
 
-// IsDeselectAll reports whether the user has explicitly deselected all routes.
+// SetExclusiveExitNode atomically makes preferred the only selected exit node
+// among exitIDs: every other ID in exitIDs is deselected and preferred (when
+// non-empty) is selected, all under a single lock. Holding the lock across the
+// whole reconciliation prevents a concurrent DeselectAllRoutes from interleaving
+// between the deselect and select steps and being silently undone. A global
+// deselect-all is left untouched so the user's "all off" stays in effect;
+// non-exit routes are never referenced, so their selection is preserved.
+func (rs *RouteSelector) SetExclusiveExitNode(preferred route.NetID, exitIDs []route.NetID) {
+	rs.mu.Lock()
+	defer rs.mu.Unlock()
+
+	if rs.deselectAll {
+		return
+	}
+
+	for _, id := range exitIDs {
+		if id == preferred {
+			continue
+		}
+		rs.deselectedRoutes[id] = struct{}{}
+		delete(rs.selectedRoutes, id)
+	}
+
+	if preferred != "" {
+		delete(rs.deselectedRoutes, preferred)
+		rs.selectedRoutes[preferred] = struct{}{}
+	}
+}
+
+// IsDeselectAll reports whether the global "deselect all" flag is set, i.e. the
+// user explicitly disabled every route. Callers enforcing per-route invariants
+// (e.g. single exit node) should leave the selection untouched when it is.
 func (rs *RouteSelector) IsDeselectAll() bool {
 	rs.mu.RLock()
 	defer rs.mu.RUnlock()

client/server/mdm.go

@@ -3,6 +3,7 @@ package server
 import (
 	"context"
 	"fmt"
+	"net/url"
 	"time"
 
 	log "github.com/sirupsen/logrus"
@@ -181,6 +182,37 @@ func conflictBool(key string, p *bool) conflictCheck {
 	}
 }
 
+func canonicalURL(s string) string {
+	u, err := url.ParseRequestURI(s)
+	if err != nil {
+		return s
+	}
+	if u.Port() == "" {
+		switch u.Scheme {
+		case "https":
+			u.Host += ":443"
+		case "http":
+			u.Host += ":80"
+		}
+	}
+	return u.String()
+}
+
+// conflictURL is conflictString for URL-typed keys: both sides are
+// normalized via canonicalURL before comparison.
+func conflictURL(key, got string) conflictCheck {
+	return conflictCheck{
+		key: key,
+		check: func(pol *mdm.Policy) bool {
+			if got == "" {
+				return true
+			}
+			want, ok := pol.GetString(key)
+			return ok && canonicalURL(want) == canonicalURL(got)
+		},
+	}
+}
+
 // conflictString builds a conflictCheck for a string MDM key. An empty
 // `got` is treated as "field not set" (no override requested); otherwise
 // the check returns true only when the policy contains the key and its
@@ -256,7 +288,7 @@ func mdmManagedFieldConflicts(msg *proto.SetConfigRequest, policy *mdm.Policy) [
 	}
 
 	return resolveConflicts(policy, []conflictCheck{
-		conflictString(mdm.KeyManagementURL, msg.ManagementUrl),
+		conflictURL(mdm.KeyManagementURL, msg.ManagementUrl),
 		conflictString(mdm.KeyPreSharedKey, pskGot),
 		conflictBool(mdm.KeyRosenpassEnabled, msg.RosenpassEnabled),
 		conflictBool(mdm.KeyRosenpassPermissive, msg.RosenpassPermissive),
@@ -377,7 +409,7 @@ func loginRequestMDMConflicts(msg *proto.LoginRequest, policy *mdm.Policy) []str
 	}
 
 	return resolveConflicts(policy, []conflictCheck{
-		conflictString(mdm.KeyManagementURL, msg.ManagementUrl),
+		conflictURL(mdm.KeyManagementURL, msg.ManagementUrl),
 		conflictString(mdm.KeyPreSharedKey, pskGot),
 		conflictBool(mdm.KeyRosenpassEnabled, msg.RosenpassEnabled),
 		conflictBool(mdm.KeyRosenpassPermissive, msg.RosenpassPermissive),

client/server/setconfig_mdm_test.go

@@ -181,6 +181,43 @@ func TestSetConfig_MDMAllow_NonManagedFields(t *testing.T) {
 	require.NotNil(t, resp)
 }
 
+// TestSetConfig_MDMAllow_ManagementURLPortNormalized covers the
+// regression from discussion #6483: MDM URL without explicit port vs
+// UI echo with the parseURL-appended default port must be treated as
+// a no-op echo, not a conflict.
+func TestSetConfig_MDMAllow_ManagementURLPortNormalized(t *testing.T) {
+	tests := []struct {
+		name      string
+		mdmURL    string
+		submitURL string
+	}{
+		{"policy_no_port_submit_with_443", "https://netbird.corp.example", "https://netbird.corp.example:443"},
+		{"policy_with_443_submit_no_port", "https://netbird.corp.example:443", "https://netbird.corp.example"},
+		{"http_policy_no_port_submit_with_80", "http://netbird.corp.example", "http://netbird.corp.example:80"},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			withMDMPolicy(t, mdm.NewPolicy(map[string]any{
+				mdm.KeyManagementURL: tc.mdmURL,
+			}))
+
+			s, ctx, profName, username, _ := setupServerWithProfile(t)
+
+			rosenpassEnabled := true
+			resp, err := s.SetConfig(ctx, &proto.SetConfigRequest{
+				ProfileName:      profName,
+				Username:         username,
+				ManagementUrl:    tc.submitURL,
+				RosenpassEnabled: &rosenpassEnabled,
+			})
+
+			require.NoError(t, err, "port-normalized URL echo must not trip MDM conflict gate")
+			require.NotNil(t, resp)
+		})
+	}
+}
+
 func TestSetConfig_MDMEmpty_NoEnforcement(t *testing.T) {
 	// No MDM policy active: any field can be written.
 	withMDMPolicy(t, mdm.NewPolicy(nil))

client/system/process.go

@@ -7,7 +7,7 @@ import (
 	"os"
 	"slices"
 
-	"github.com/shirou/gopsutil/v3/process"
+	"github.com/shirou/gopsutil/v4/process"
 )
 
 // getRunningProcesses returns a list of running process paths. The context bounds the work:

client/system/process_test.go

@@ -4,7 +4,7 @@ import (
 	"context"
 	"testing"
 
-	"github.com/shirou/gopsutil/v3/process"
+	"github.com/shirou/gopsutil/v4/process"
 )
 
 func Benchmark_getRunningProcesses(b *testing.B) {

go.mod

@@ -104,6 +104,7 @@ require (
 	github.com/redis/go-redis/v9 v9.7.3
 	github.com/rs/xid v1.3.0
 	github.com/shirou/gopsutil/v3 v3.24.4
+	github.com/shirou/gopsutil/v4 v4.25.8
 	github.com/skratchdot/open-golang v0.0.0-20200116055534-eef842397966
 	github.com/songgao/water v0.0.0-20200317203138-2b4b6d7c09d8
 	github.com/stretchr/testify v1.11.1
@@ -308,7 +309,6 @@ require (
 	github.com/russellhaering/goxmldsig v1.6.0 // indirect
 	github.com/ryanuber/go-glob v1.0.0 // indirect
 	github.com/rymdport/portal v0.4.2 // indirect
-	github.com/shirou/gopsutil/v4 v4.25.8 // indirect
 	github.com/shoenig/go-m1cpu v0.2.1 // indirect
 	github.com/shopspring/decimal v1.4.0 // indirect
 	github.com/spf13/cast v1.7.0 // indirect

shared/relay/client/manager.go

@@ -30,11 +30,16 @@ type RelayTrack struct {
 	relayClient *Client
 	err         error
 	created     time.Time
+	// ready is closed once the dial started by openConnVia finishes (relayClient
+	// or err is set). Callers reusing a track wait on this instead of the track
+	// lock, so the dial never runs under rt.Lock.
+	ready chan struct{}
 }
 
 func NewRelayTrack() *RelayTrack {
 	return &RelayTrack{
 		created: time.Now(),
+		ready:   make(chan struct{}),
 	}
 }
 
@@ -326,34 +331,24 @@ func (m *Manager) openConnVia(ctx context.Context, serverAddress, peerKey string
 	// check if already has a connection to the desired relay server
 	m.relayClientsMutex.RLock()
 	rt, ok := m.relayClients[serverAddress]
-	if ok {
-		rt.RLock()
-		m.relayClientsMutex.RUnlock()
-		defer rt.RUnlock()
-		if rt.err != nil {
-			return nil, rt.err
-		}
-		return rt.relayClient.OpenConn(ctx, peerKey)
-	}
 	m.relayClientsMutex.RUnlock()
+	if ok {
+		return m.openConnOnTrack(ctx, rt, peerKey)
+	}
 
 	// if not, establish a new connection but check it again (because changed the lock type) before starting the
 	// connection
 	m.relayClientsMutex.Lock()
 	rt, ok = m.relayClients[serverAddress]
 	if ok {
-		rt.RLock()
 		m.relayClientsMutex.Unlock()
-		defer rt.RUnlock()
-		if rt.err != nil {
-			return nil, rt.err
-		}
-		return rt.relayClient.OpenConn(ctx, peerKey)
+		return m.openConnOnTrack(ctx, rt, peerKey)
 	}
 
-	// create a new relay client and store it in the relayClients map
+	// Publish the track and release the map lock BEFORE dialing, so the dial does
+	// not run under rt.Lock (which would block RelayStates and the cleanup loop
+	// for the full dial). Concurrent callers find this track and wait on rt.ready.
 	rt = NewRelayTrack()
-	rt.Lock()
 	m.relayClients[serverAddress] = rt
 	m.relayClientsMutex.Unlock()
 
@@ -361,8 +356,10 @@ func (m *Manager) openConnVia(ctx context.Context, serverAddress, peerKey string
 	relayClient.SetTransportFallback(m.transportFallback)
 	err := relayClient.Connect(m.ctx)
 	if err != nil {
+		rt.Lock()
 		rt.err = err
 		rt.Unlock()
+		close(rt.ready)
 		m.relayClientsMutex.Lock()
 		delete(m.relayClients, serverAddress)
 		m.relayClientsMutex.Unlock()
@@ -370,14 +367,34 @@ func (m *Manager) openConnVia(ctx context.Context, serverAddress, peerKey string
 	}
 	// if connection closed then delete the relay client from the list
 	relayClient.SetOnDisconnectListener(m.onServerDisconnected)
+	rt.Lock()
 	rt.relayClient = relayClient
 	rt.Unlock()
+	close(rt.ready)
 
-	conn, err := relayClient.OpenConn(ctx, peerKey)
-	if err != nil {
-		return nil, err
+	return relayClient.OpenConn(ctx, peerKey)
+}
+
+// openConnOnTrack opens a peer connection through an existing relay track,
+// waiting for the dial started by another openConnVia call to finish. It waits
+// on rt.ready rather than the track lock, so it neither holds nor contends the
+// track lock across the dial.
+func (m *Manager) openConnOnTrack(ctx context.Context, rt *RelayTrack, peerKey string) (net.Conn, error) {
+	select {
+	case <-rt.ready:
+	case <-ctx.Done():
+		return nil, ctx.Err()
 	}
-	return conn, nil
+
+	rt.RLock()
+	defer rt.RUnlock()
+	if rt.err != nil {
+		return nil, rt.err
+	}
+	if rt.relayClient == nil {
+		return nil, ErrRelayClientNotConnected
+	}
+	return rt.relayClient.OpenConn(ctx, peerKey)
 }
 
 func (m *Manager) onServerConnected() {
@@ -476,6 +493,13 @@ func (m *Manager) cleanUpUnusedRelays() {
 			continue
 		}
 
+		// dial still in progress (openConnVia publishes the track before Connect
+		// completes and no longer holds rt.Lock during it), nothing to clean up.
+		if rt.relayClient == nil {
+			rt.Unlock()
+			continue
+		}
+
 		if time.Since(rt.created) <= m.keepUnusedServerTime {
 			rt.Unlock()
 			continue

shared/relay/client/manager_cleanup_test.go

@@ -0,0 +1,60 @@
+package client
+
+import (
+	"context"
+	"net/netip"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/require"
+)
+
+// TestCleanUpUnusedRelays_DoesNotBlockOnRealHangingDial drives a real, hanging foreign
+// relay dial and asserts cleanUpUnusedRelays does not stall behind it.
+func TestCleanUpUnusedRelays_DoesNotBlockOnRealHangingDial(t *testing.T) {
+	serverAddr := stallingRelayListener(t)
+
+	mCtx, mCancel := context.WithCancel(context.Background())
+	t.Cleanup(mCancel)
+
+	m := NewManager(mCtx, nil, "alice", 1280)
+
+	dialDone := make(chan struct{})
+	go func() {
+		defer close(dialDone)
+		_, _ = m.openConnVia(mCtx, serverAddr, "peerKey", netip.Addr{})
+	}()
+
+	// The track appears in the map once the dial is in flight.
+	require.Eventually(t, func() bool {
+		m.relayClientsMutex.RLock()
+		defer m.relayClientsMutex.RUnlock()
+		_, ok := m.relayClients[serverAddr]
+		return ok
+	}, 5*time.Second, 5*time.Millisecond, "relay dial did not start")
+
+	cleanupDone := make(chan struct{})
+	go func() {
+		defer close(cleanupDone)
+		m.cleanUpUnusedRelays()
+	}()
+
+	select {
+	case <-cleanupDone:
+	case <-time.After(2 * time.Second):
+		t.Fatal("cleanUpUnusedRelays blocked on an in-progress relay dial while holding the relay map lock")
+	}
+
+	m.relayClientsMutex.RLock()
+	_, stillTracked := m.relayClients[serverAddr]
+	m.relayClientsMutex.RUnlock()
+	require.True(t, stillTracked, "an in-progress relay dial must not be evicted by cleanup")
+
+	// Release the hanging dial so the goroutine can exit cleanly.
+	mCancel()
+	select {
+	case <-dialDone:
+	case <-time.After(5 * time.Second):
+		t.Fatal("openConnVia did not return after context cancellation")
+	}
+}

shared/relay/client/manager_relaystates_test.go

@@ -0,0 +1,91 @@
+package client
+
+import (
+	"context"
+	"net"
+	"net/netip"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/require"
+)
+
+// stallingRelayListener accepts TCP connections and holds them open without ever
+// responding, so a relay handshake dialed against it blocks until its context is
+// cancelled. It returns the "rel://host:port" URL to dial.
+func stallingRelayListener(t *testing.T) string {
+	t.Helper()
+
+	ln, err := net.Listen("tcp", "127.0.0.1:0")
+	require.NoError(t, err)
+
+	var mu sync.Mutex
+	var conns []net.Conn
+	go func() {
+		for {
+			c, err := ln.Accept()
+			if err != nil {
+				return
+			}
+			mu.Lock()
+			conns = append(conns, c)
+			mu.Unlock()
+		}
+	}()
+	t.Cleanup(func() {
+		_ = ln.Close()
+		mu.Lock()
+		for _, c := range conns {
+			_ = c.Close()
+		}
+		mu.Unlock()
+	})
+
+	return "rel://" + ln.Addr().String()
+}
+
+// TestRelayStates_DoesNotBlockOnRealHangingDial is a regression test for
+// RelayStates() called by a "status -d command" hanging behind an in-progress
+// relay dial.
+func TestRelayStates_DoesNotBlockOnRealHangingDial(t *testing.T) {
+	serverAddr := stallingRelayListener(t)
+
+	mCtx, mCancel := context.WithCancel(context.Background())
+	t.Cleanup(mCancel)
+
+	m := NewManager(mCtx, nil, "alice", 1280)
+
+	dialDone := make(chan struct{})
+	go func() {
+		defer close(dialDone)
+		_, _ = m.openConnVia(mCtx, serverAddr, "peerKey", netip.Addr{})
+	}()
+
+	require.Eventually(t, func() bool {
+		m.relayClientsMutex.RLock()
+		defer m.relayClientsMutex.RUnlock()
+		_, ok := m.relayClients[serverAddr]
+		return ok
+	}, 5*time.Second, 5*time.Millisecond, "relay dial did not start")
+
+	done := make(chan []RelayConnState, 1)
+	go func() {
+		done <- m.RelayStates()
+	}()
+
+	select {
+	case states := <-done:
+		require.Empty(t, states, "a relay still being dialed carries no state and must be omitted")
+	case <-time.After(2 * time.Second):
+		t.Fatal("RelayStates blocked on a foreign relay whose Connect() is in progress")
+	}
+
+	// Release the hanging dial so the goroutine can exit cleanly.
+	mCancel()
+	select {
+	case <-dialDone:
+	case <-time.After(5 * time.Second):
+		t.Fatal("openConnVia did not return after context cancellation")
+	}
+}