Merge remote-tracking branch 'origin/main' into dmitri-catch-disconnected-peer

Signed-off-by: Dmitri Dolguikh <dmitri.external@netbird.io>

management/server/account.go

@@ -14,10 +14,6 @@ import (
 	"sync"
 	"time"
 
-	"github.com/netbirdio/netbird/management/internals/modules/reverseproxy/service"
-	"github.com/netbirdio/netbird/management/server/job"
-	"github.com/netbirdio/netbird/shared/auth"
-
 	cacheStore "github.com/eko/gocache/lib/v4/store"
 	"github.com/eko/gocache/store/redis/v4"
 	"github.com/rs/xid"
@@ -29,6 +25,7 @@ import (
 	"github.com/netbirdio/netbird/formatter/hook"
 	"github.com/netbirdio/netbird/idp/dex"
 	"github.com/netbirdio/netbird/management/internals/controllers/network_map"
+	"github.com/netbirdio/netbird/management/internals/modules/reverseproxy/service"
 	nbconfig "github.com/netbirdio/netbird/management/internals/server/config"
 	"github.com/netbirdio/netbird/management/server/account"
 	"github.com/netbirdio/netbird/management/server/activity"
@@ -39,6 +36,7 @@ import (
 	"github.com/netbirdio/netbird/management/server/idp"
 	"github.com/netbirdio/netbird/management/server/integrations/integrated_validator"
 	"github.com/netbirdio/netbird/management/server/integrations/port_forwarding"
+	"github.com/netbirdio/netbird/management/server/job"
 	nbpeer "github.com/netbirdio/netbird/management/server/peer"
 	"github.com/netbirdio/netbird/management/server/permissions"
 	"github.com/netbirdio/netbird/management/server/permissions/modules"
@@ -50,6 +48,7 @@ import (
 	"github.com/netbirdio/netbird/management/server/types"
 	"github.com/netbirdio/netbird/management/server/util"
 	"github.com/netbirdio/netbird/route"
+	"github.com/netbirdio/netbird/shared/auth"
 	nbdomain "github.com/netbirdio/netbird/shared/management/domain"
 	"github.com/netbirdio/netbird/shared/management/networkmap/nmdata"
 	"github.com/netbirdio/netbird/shared/management/status"
@@ -238,6 +237,10 @@ func BuildManager(
 		log.WithContext(ctx).Error(err)
 	}
 
+	if IsEmbeddedIdp(idpManager) && accountsCounter > 1 {
+		log.WithContext(ctx).Warnf("embedded IdP requires a single account, found %d", accountsCounter)
+	}
+
 	// enable single account mode only if configured by user and number of existing accounts is not grater than 1
 	am.singleAccountMode = singleAccountModeDomain != "" && accountsCounter <= 1
 	if am.singleAccountMode {
@@ -716,8 +719,10 @@ func (am *DefaultAccountManager) schedulePeerLoginExpiration(ctx context.Context
 		log.WithContext(ctx).Tracef("peer login expiration job for account %s is already scheduled", accountID)
 		return
 	}
+	// The job outlives the request that arms it, so it must not inherit the request's cancellation.
+	jobCtx := context.WithoutCancel(ctx)
 	if nextRun, ok := am.getNextPeerExpiration(ctx, accountID); ok {
-		go am.peerLoginExpiry.Schedule(ctx, nextRun, accountID, am.peerLoginExpirationJob(ctx, accountID))
+		go am.peerLoginExpiry.Schedule(jobCtx, nextRun, accountID, am.peerLoginExpirationJob(jobCtx, accountID))
 	}
 }
 
@@ -749,8 +754,9 @@ func (am *DefaultAccountManager) peerInactivityExpirationJob(ctx context.Context
 // checkAndSchedulePeerInactivityExpiration periodically checks for inactive peers to end their sessions
 func (am *DefaultAccountManager) checkAndSchedulePeerInactivityExpiration(ctx context.Context, accountID string) {
 	am.peerInactivityExpiry.Cancel(ctx, []string{accountID})
+	jobCtx := context.WithoutCancel(ctx)
 	if nextRun, ok := am.getNextInactivePeerExpiration(ctx, accountID); ok {
-		go am.peerInactivityExpiry.Schedule(ctx, nextRun, accountID, am.peerInactivityExpirationJob(ctx, accountID))
+		go am.peerInactivityExpiry.Schedule(jobCtx, nextRun, accountID, am.peerInactivityExpirationJob(jobCtx, accountID))
 	}
 }
 
@@ -1592,7 +1598,10 @@ func (am *DefaultAccountManager) updateUserAuthWithSingleMode(ctx context.Contex
 	if err != nil {
 		return err
 	}
-	userAuth.Domain = domain
+	// Keep the configured single account domain when the existing account has none
+	if domain != "" {
+		userAuth.Domain = domain
+	}
 
 	log.WithContext(ctx).Debugf("overriding JWT Domain and DomainCategory claims since single account mode is enabled")
 	return nil
@@ -1837,6 +1846,7 @@ func (am *DefaultAccountManager) getAccountIDWithAuthorizationClaims(ctx context
 
 	return am.addNewPrivateAccount(ctx, domainAccountID, userAuth)
 }
+
 func (am *DefaultAccountManager) getPrivateDomainWithGlobalLock(ctx context.Context, domain string) (string, context.CancelFunc, error) {
 	domainAccountID, err := am.Store.GetAccountIDByPrivateDomain(ctx, store.LockingStrengthNone, domain)
 	if handleNotFound(err) != nil {

management/server/account_test.go

@@ -1920,6 +1920,154 @@ func TestDefaultAccountManager_MarkPeerConnected_PeerLoginExpiration(t *testing.
 	}
 }
 
+func TestDefaultAccountManager_SchedulePeerLoginExpiration_IncludesOfflinePeers(t *testing.T) {
+	manager, updateManager, err := createManager(t)
+	require.NoError(t, err, "unable to create account manager")
+
+	accountID, err := manager.GetAccountIDByUserID(context.Background(), auth.UserAuth{UserId: userID})
+	require.NoError(t, err, "unable to create an account")
+
+	connectedKey, offlineKey := addExpiringPeers(t, manager)
+	_, err = manager.UpdateAccountSettings(context.Background(), accountID, userID, &types.Settings{
+		PeerLoginExpiration:        time.Hour,
+		PeerLoginExpirationEnabled: true,
+		Extra:                      &types.ExtraSettings{},
+	})
+	require.NoError(t, err, "expecting to update account settings successfully but got error")
+	manager.peerLoginExpiry.CancelAll(context.Background())
+
+	// The connected peer logged in just now, so a job computed from connected peers alone
+	// would be armed for an hour. The offline peer's login expires in two seconds; a
+	// reconnect of that peer must not have to wait for the connected peer's tick.
+	now := time.Now().UTC()
+	setPeerLogin(t, manager, accountID, connectedKey, true, now)
+	setPeerLogin(t, manager, accountID, offlineKey, false, now.Add(-time.Hour+2*time.Second))
+
+	offlinePeer, err := manager.Store.GetPeerByPeerPubKey(context.Background(), store.LockingStrengthNone, offlineKey)
+	require.NoError(t, err)
+	updateManager.CreateChannel(context.Background(), offlinePeer.ID)
+
+	manager.peerLoginExpiry = NewDefaultScheduler()
+	t.Cleanup(func() { manager.peerLoginExpiry.CancelAll(context.Background()) })
+	manager.schedulePeerLoginExpiration(context.Background(), accountID)
+
+	// The flag is committed per peer before the disconnect fans out, so wait for both.
+	require.Eventually(t, func() bool {
+		peer, err := manager.Store.GetPeerByPeerPubKey(context.Background(), store.LockingStrengthNone, offlineKey)
+		return err == nil && peer.Status.LoginExpired && !updateManager.HasChannel(offlinePeer.ID)
+	}, 10*time.Second, 100*time.Millisecond, "offline peer should be expired and disconnected at its own deadline")
+
+	connectedPeer, err := manager.Store.GetPeerByPeerPubKey(context.Background(), store.LockingStrengthNone, connectedKey)
+	require.NoError(t, err)
+	assert.False(t, connectedPeer.Status.LoginExpired, "connected peer with a fresh login must not expire")
+}
+
+func TestDefaultAccountManager_SchedulePeerLoginExpiration_DetachesRequestContext(t *testing.T) {
+	manager, _, err := createManager(t)
+	require.NoError(t, err, "unable to create account manager")
+
+	accountID, err := manager.GetAccountIDByUserID(context.Background(), auth.UserAuth{UserId: userID})
+	require.NoError(t, err, "unable to create an account")
+	connectedKey, _ := addExpiringPeers(t, manager)
+	setPeerLogin(t, manager, accountID, connectedKey, true, time.Now().UTC())
+
+	scheduled := make(chan context.Context, 1)
+	manager.peerLoginExpiry = &MockScheduler{
+		IsSchedulerRunningFunc: func(string) bool { return false },
+		ScheduleFunc: func(ctx context.Context, _ time.Duration, _ string, _ func() (time.Duration, bool)) {
+			scheduled <- ctx
+		},
+	}
+
+	requestCtx, cancel := context.WithCancel(context.Background())
+	manager.schedulePeerLoginExpiration(requestCtx, accountID)
+	cancel()
+
+	select {
+	case jobCtx := <-scheduled:
+		assert.NoError(t, jobCtx.Err(), "the expiration job must outlive the request that armed it")
+	case <-time.After(time.Second):
+		t.Fatal("timeout while waiting for the job to be scheduled")
+	}
+}
+
+func TestDefaultAccountManager_ExpireAndUpdatePeers_SkipsPeerThatLoggedInAgain(t *testing.T) {
+	manager, updateManager, err := createManager(t)
+	require.NoError(t, err, "unable to create account manager")
+
+	accountID, err := manager.GetAccountIDByUserID(context.Background(), auth.UserAuth{UserId: userID})
+	require.NoError(t, err, "unable to create an account")
+
+	reloggedKey, staleKey := addExpiringPeers(t, manager)
+	_, err = manager.UpdateAccountSettings(context.Background(), accountID, userID, &types.Settings{
+		PeerLoginExpiration:        time.Hour,
+		PeerLoginExpirationEnabled: true,
+		Extra:                      &types.ExtraSettings{},
+	})
+	require.NoError(t, err, "expecting to update account settings successfully but got error")
+	manager.peerLoginExpiry.CancelAll(context.Background())
+
+	expiredLogin := time.Now().UTC().Add(-2 * time.Hour)
+	setPeerLogin(t, manager, accountID, reloggedKey, true, expiredLogin)
+	setPeerLogin(t, manager, accountID, staleKey, true, expiredLogin)
+
+	expiredPeers, err := manager.getExpiredPeers(context.Background(), accountID)
+	require.NoError(t, err)
+	require.Len(t, expiredPeers, 2, "both peers should be due for expiration")
+
+	// The job holds the candidate list while one peer completes a fresh login, which
+	// moves its deadline into the future and must win over the stale candidate entry.
+	setPeerLogin(t, manager, accountID, reloggedKey, true, time.Now().UTC())
+
+	reloggedPeer, err := manager.Store.GetPeerByPeerPubKey(context.Background(), store.LockingStrengthNone, reloggedKey)
+	require.NoError(t, err)
+	stalePeer, err := manager.Store.GetPeerByPeerPubKey(context.Background(), store.LockingStrengthNone, staleKey)
+	require.NoError(t, err)
+	updateManager.CreateChannel(context.Background(), reloggedPeer.ID)
+	updateManager.CreateChannel(context.Background(), stalePeer.ID)
+
+	err = manager.expireAndUpdatePeers(context.Background(), accountID, expiredPeers, peerExpirationSessionExpired)
+	require.NoError(t, err)
+
+	reloggedPeer, err = manager.Store.GetPeerByPeerPubKey(context.Background(), store.LockingStrengthNone, reloggedKey)
+	require.NoError(t, err)
+	assert.False(t, reloggedPeer.Status.LoginExpired, "a peer that logged in again must not be flagged from the stale candidate list")
+	assert.True(t, reloggedPeer.Status.Connected, "the re-logged peer must keep its connected status")
+	assert.True(t, updateManager.HasChannel(reloggedPeer.ID), "the re-logged peer's update channel must stay open")
+
+	stalePeer, err = manager.Store.GetPeerByPeerPubKey(context.Background(), store.LockingStrengthNone, staleKey)
+	require.NoError(t, err)
+	assert.True(t, stalePeer.Status.LoginExpired, "a peer that is still due must be flagged")
+	assert.False(t, updateManager.HasChannel(stalePeer.ID), "the expired peer's update channel must be closed")
+}
+
+// addExpiringPeers registers two SSO peers with login expiration enabled and returns their public keys.
+func addExpiringPeers(t *testing.T, manager *DefaultAccountManager) (string, string) {
+	t.Helper()
+	keys := make([]string, 0, 2)
+	for _, hostname := range []string{"connected-peer", "offline-peer"} {
+		key, err := wgtypes.GenerateKey()
+		require.NoError(t, err, "unable to generate WireGuard key")
+		_, _, _, _, err = manager.AddPeer(context.Background(), "", "", userID, &nbpeer.Peer{
+			Key:                    key.PublicKey().String(),
+			Meta:                   nbpeer.PeerSystemMeta{Hostname: hostname},
+			LoginExpirationEnabled: true,
+		}, false)
+		require.NoError(t, err, "unable to add peer")
+		keys = append(keys, key.PublicKey().String())
+	}
+	return keys[0], keys[1]
+}
+
+func setPeerLogin(t *testing.T, manager *DefaultAccountManager, accountID, peerKey string, connected bool, lastLogin time.Time) {
+	t.Helper()
+	peer, err := manager.Store.GetPeerByPeerPubKey(context.Background(), store.LockingStrengthNone, peerKey)
+	require.NoError(t, err)
+	peer.Status.Connected = connected
+	peer.LastLogin = &lastLogin
+	require.NoError(t, manager.Store.SavePeer(context.Background(), accountID, peer))
+}
+
 func TestDefaultAccountManager_MarkPeerDisconnected_SchedulesInactivityExpiration(t *testing.T) {
 	manager, _, err := createManager(t)
 	require.NoError(t, err, "unable to create account manager")
@@ -2702,7 +2850,7 @@ func TestAccount_GetNextPeerExpiration(t *testing.T) {
 			expectedNextExpiration: time.Duration(0),
 		},
 		{
-			name: "No connected peers, no expiration",
+			name: "Offline peer with expiration, return expiration",
 			peers: map[string]*nbpeer.Peer{
 				"peer-1": {
 					Status: &nbpeer.PeerStatus{
@@ -2721,8 +2869,33 @@ func TestAccount_GetNextPeerExpiration(t *testing.T) {
 			},
 			expiration:             time.Second,
 			expirationEnabled:      false,
-			expectedNextRun:        false,
-			expectedNextExpiration: time.Duration(0),
+			expectedNextRun:        true,
+			expectedNextExpiration: time.Second,
+		},
+		{
+			name: "Offline peer with the earliest deadline defines the next run",
+			peers: map[string]*nbpeer.Peer{
+				"peer-1": {
+					Status: &nbpeer.PeerStatus{
+						Connected: true,
+					},
+					LoginExpirationEnabled: true,
+					LastLogin:              util.ToPtr(time.Now().UTC()),
+					UserID:                 userID,
+				},
+				"peer-2": {
+					Status: &nbpeer.PeerStatus{
+						Connected: false,
+					},
+					LoginExpirationEnabled: true,
+					LastLogin:              util.ToPtr(time.Now().UTC().Add(-50 * time.Minute)),
+					UserID:                 userID,
+				},
+			},
+			expiration:             time.Hour,
+			expirationEnabled:      true,
+			expectedNextRun:        true,
+			expectedNextExpiration: 10 * time.Minute,
 		},
 		{
 			name: "Connected peers with disabled expiration, no expiration",

management/server/auth/session.go

@@ -7,9 +7,6 @@ import (
 	"errors"
 	"fmt"
 	"time"
-
-	"github.com/eko/gocache/lib/v4/cache"
-	"github.com/eko/gocache/lib/v4/store"
 )
 
 const (
@@ -22,12 +19,17 @@ var (
 	ErrTokenExpired     = errors.New("JWT expired")
 )
 
-type SessionStore struct {
-	cache *cache.Cache[string]
+// TokenCache atomically records used JWTs until their expiration.
+type TokenCache interface {
+	SetNX(ctx context.Context, key, value string, ttl time.Duration) (bool, error)
 }
 
-func NewSessionStore(cacheStore store.StoreInterface) *SessionStore {
-	return &SessionStore{cache: cache.New[string](cacheStore)}
+type SessionStore struct {
+	cache TokenCache
+}
+
+func NewSessionStore(cacheStore TokenCache) *SessionStore {
+	return &SessionStore{cache: cacheStore}
 }
 
 // RegisterToken records a JWT until its exp time and rejects reuse.
@@ -38,20 +40,14 @@ func (s *SessionStore) RegisterToken(ctx context.Context, token string, expiresA
 	}
 
 	key := usedTokenKeyPrefix + hashToken(token)
-	_, err := s.cache.Get(ctx, key)
-	if err == nil {
+	created, err := s.cache.SetNX(ctx, key, usedTokenMarker, ttl)
+	if err != nil {
+		return fmt.Errorf("store used token entry: %w", err)
+	}
+	if !created {
 		return ErrTokenAlreadyUsed
 	}
 
-	var notFound *store.NotFound
-	if !errors.As(err, &notFound) {
-		return fmt.Errorf("failed to lookup used token entry: %w", err)
-	}
-
-	if err := s.cache.Set(ctx, key, usedTokenMarker, store.WithExpiration(ttl)); err != nil {
-		return fmt.Errorf("failed to store used token entry: %w", err)
-	}
-
 	return nil
 }
 

management/server/auth/session_test.go

@@ -2,6 +2,7 @@ package auth
 
 import (
 	"context"
+	"errors"
 	"testing"
 	"time"
 
@@ -38,6 +39,39 @@ func TestSessionStore_RegisterSameTokenTwiceIsRejected(t *testing.T) {
 	assert.ErrorIs(t, err, ErrTokenAlreadyUsed)
 }
 
+func TestSessionStore_ConcurrentRegistrationAllowsOneCaller(t *testing.T) {
+	s := newTestSessionStore(t)
+	ctx := context.Background()
+	const attempts = 100
+
+	start := make(chan struct{})
+	results := make(chan error, attempts)
+	for range attempts {
+		go func() {
+			<-start
+			results <- s.RegisterToken(ctx, "token", time.Now().Add(time.Hour))
+		}()
+	}
+	close(start)
+
+	succeeded := 0
+	alreadyUsed := 0
+	for range attempts {
+		err := <-results
+		switch {
+		case err == nil:
+			succeeded++
+		case errors.Is(err, ErrTokenAlreadyUsed):
+			alreadyUsed++
+		default:
+			require.NoError(t, err, "concurrent registration returned an unexpected error")
+		}
+	}
+
+	assert.Equal(t, 1, succeeded, "exactly one concurrent caller should register the token")
+	assert.Equal(t, attempts-1, alreadyUsed, "every other caller should be rejected as already used")
+}
+
 func TestSessionStore_RegisterDifferentTokensAreIndependent(t *testing.T) {
 	s := newTestSessionStore(t)
 	ctx := context.Background()
@@ -72,6 +106,23 @@ func TestSessionStore_EntryEvictsAtTTLAndAllowsReRegistration(t *testing.T) {
 	require.NoError(t, s.RegisterToken(ctx, token, time.Now().Add(time.Hour)))
 }
 
+type failingTokenCache struct {
+	err error
+}
+
+func (f failingTokenCache) SetNX(context.Context, string, string, time.Duration) (bool, error) {
+	return false, f.err
+}
+
+func TestSessionStore_CacheErrorIsReturned(t *testing.T) {
+	cacheErr := errors.New("cache unavailable")
+	s := NewSessionStore(failingTokenCache{err: cacheErr})
+
+	err := s.RegisterToken(context.Background(), "token", time.Now().Add(time.Hour))
+	require.Error(t, err, "cache failure should be surfaced to the caller")
+	assert.ErrorIs(t, err, cacheErr, "cache error should be wrapped, not replaced")
+}
+
 func TestHashToken_StableAndDoesNotLeak(t *testing.T) {
 	a := hashToken("tokenA")
 	b := hashToken("tokenB")

management/server/cache/memory.go

@@ -0,0 +1,57 @@
+package cache
+
+import (
+	"context"
+	"fmt"
+	"sync"
+	"time"
+
+	"github.com/eko/gocache/lib/v4/store"
+	gocachestore "github.com/eko/gocache/store/go_cache/v4"
+	gocache "github.com/patrickmn/go-cache"
+)
+
+type goCacheStore struct {
+	store.StoreInterface
+	client *gocache.Cache
+	mu     sync.Mutex
+}
+
+func newMemoryStore(maxTimeout, cleanupInterval time.Duration) Store {
+	client := gocache.New(maxTimeout, cleanupInterval)
+	return &goCacheStore{
+		StoreInterface: gocachestore.NewGoCache(client),
+		client:         client,
+	}
+}
+
+func (s *goCacheStore) SetNX(_ context.Context, key, value string, ttl time.Duration) (bool, error) {
+	// Add only returns an error when a non-expired entry already exists.
+	if err := s.client.Add(key, value, ttl); err != nil {
+		return false, nil //nolint:nilerr
+	}
+	return true, nil
+}
+
+// GetDel reads the value under key and removes it. go-cache has no native read-and-delete
+// and releases its own lock between the two calls, so mu holds the pair together and no
+// value is consumed twice.
+//
+// Writes do not take mu: a Set landing mid-pair is lost, since GetDel returns the prior
+// value and deletes the new one. Callers must write a consumed key only once.
+func (s *goCacheStore) GetDel(_ context.Context, key string) (string, bool, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	value, found := s.client.Get(key)
+	if !found {
+		return "", false, nil
+	}
+	s.client.Delete(key)
+
+	str, ok := value.(string)
+	if !ok {
+		return "", false, fmt.Errorf("cached value is %T, not a string", value)
+	}
+	return str, true, nil
+}

management/server/cache/memory_test.go

@@ -0,0 +1,76 @@
+package cache_test
+
+import (
+	"context"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/management/server/cache"
+)
+
+func TestMemoryStore(t *testing.T) {
+	memStore, err := cache.NewStore(context.Background(), 100*time.Millisecond, 300*time.Millisecond, 100)
+	require.NoError(t, err, "couldn't create memory store")
+
+	ctx := context.Background()
+	key, value := "testing", "tested"
+	err = memStore.Set(ctx, key, value)
+	assert.NoError(t, err, "couldn't set testing data")
+
+	result, err := memStore.Get(ctx, key)
+	assert.NoError(t, err, "couldn't get testing data")
+	assert.Equal(t, value, result, "value returned doesn't match testing data")
+
+	created, err := memStore.SetNX(ctx, "conditional", value, 100*time.Millisecond)
+	require.NoError(t, err, "couldn't conditionally set testing data")
+	require.True(t, created, "first conditional set should create the entry")
+
+	created, err = memStore.SetNX(ctx, "conditional", value, 100*time.Millisecond)
+	require.NoError(t, err, "couldn't conditionally check testing data")
+	require.False(t, created, "second conditional set should not replace the entry")
+
+	// test expiration
+	time.Sleep(300 * time.Millisecond)
+	_, err = memStore.Get(ctx, key)
+	assert.Error(t, err, "value should not be found")
+}
+
+func TestMemoryStoreGetDel(t *testing.T) {
+	ctx := context.Background()
+	newStore := func(t *testing.T) cache.Store {
+		t.Helper()
+		memStore, err := cache.NewStore(ctx, time.Minute, time.Minute, 100)
+		require.NoError(t, err, "couldn't create memory store")
+
+		return memStore
+	}
+
+	const (
+		key   = "consume"
+		value = "verifier"
+	)
+
+	t.Run("exactly one concurrent caller consumes the key", func(t *testing.T) {
+		memStore := newStore(t)
+		require.NoError(t, memStore.Set(ctx, key, value), "couldn't set testing data")
+
+		assertGetDelConsumedOnce(ctx, t, []cache.Store{memStore}, key, value)
+		assertGetDelMisses(ctx, t, memStore, key)
+	})
+
+	t.Run("missing key is not an error", func(t *testing.T) {
+		assertGetDelMisses(ctx, t, newStore(t), "never-set")
+	})
+
+	t.Run("expired key is not found", func(t *testing.T) {
+		memStore := newStore(t)
+		_, err := memStore.SetNX(ctx, key, value, 50*time.Millisecond)
+		require.NoError(t, err, "couldn't set testing data")
+
+		time.Sleep(100 * time.Millisecond)
+		assertGetDelMisses(ctx, t, memStore, key)
+	})
+}

management/server/cache/redis.go

@@ -0,0 +1,63 @@
+package cache
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"math"
+	"time"
+
+	"github.com/eko/gocache/lib/v4/store"
+	redisstore "github.com/eko/gocache/store/redis/v4"
+	"github.com/redis/go-redis/v9"
+	log "github.com/sirupsen/logrus"
+)
+
+type redisStore struct {
+	store.StoreInterface
+	client *redis.Client
+}
+
+func getRedisStore(ctx context.Context, redisEnvAddr string, maxConn int) (Store, error) {
+	options, err := redis.ParseURL(redisEnvAddr)
+	if err != nil {
+		return nil, fmt.Errorf("parsing redis cache url: %s", err)
+	}
+
+	options.MaxIdleConns = int(math.Ceil(float64(maxConn) * 0.5)) // 50% of max conns
+	options.MinIdleConns = int(math.Ceil(float64(maxConn) * 0.1)) // 10% of max conns
+	options.MaxActiveConns = maxConn
+	options.ConnMaxIdleTime = 30 * time.Minute
+	options.ConnMaxLifetime = 0
+	options.PoolTimeout = 10 * time.Second
+	redisClient := redis.NewClient(options)
+	subCtx, cancel := context.WithTimeout(ctx, 2*time.Second)
+	defer cancel()
+
+	_, err = redisClient.Ping(subCtx).Result()
+	if err != nil {
+		return nil, err
+	}
+
+	log.WithContext(subCtx).Infof("using redis cache at %s", redisEnvAddr)
+
+	return &redisStore{
+		StoreInterface: redisstore.NewRedis(redisClient),
+		client:         redisClient,
+	}, nil
+}
+
+func (s *redisStore) SetNX(ctx context.Context, key, value string, ttl time.Duration) (bool, error) {
+	return s.client.SetNX(ctx, key, value, ttl).Result()
+}
+
+func (s *redisStore) GetDel(ctx context.Context, key string) (string, bool, error) {
+	value, err := s.client.GetDel(ctx, key).Result()
+	if errors.Is(err, redis.Nil) {
+		return "", false, nil
+	}
+	if err != nil {
+		return "", false, err
+	}
+	return value, true, nil
+}

management/server/cache/redis_test.go

@@ -0,0 +1,153 @@
+package cache_test
+
+import (
+	"context"
+	"testing"
+	"time"
+
+	"github.com/eko/gocache/lib/v4/store"
+	"github.com/redis/go-redis/v9"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	testcontainersredis "github.com/testcontainers/testcontainers-go/modules/redis"
+
+	"github.com/netbirdio/netbird/management/server/cache"
+)
+
+func startRedis(t *testing.T) string {
+	t.Helper()
+
+	ctx := context.Background()
+	redisContainer, err := testcontainersredis.Run(ctx, "redis:7")
+	require.NoError(t, err, "couldn't start redis container")
+
+	t.Cleanup(func() {
+		if err := redisContainer.Terminate(ctx); err != nil {
+			t.Logf("failed to terminate container: %s", err)
+		}
+	})
+
+	redisURL, err := redisContainer.ConnectionString(ctx)
+	require.NoError(t, err, "couldn't get connection string")
+
+	t.Setenv(cache.RedisStoreEnvVar, redisURL)
+	return redisURL
+}
+
+func newRedisStore(t *testing.T) cache.Store {
+	t.Helper()
+
+	redisStore, err := cache.NewStore(context.Background(), 100*time.Millisecond, 300*time.Millisecond, 100)
+	require.NoError(t, err)
+
+	return redisStore
+}
+
+func TestRedisStoreConnectionFailure(t *testing.T) {
+	t.Setenv(cache.RedisStoreEnvVar, "redis://127.0.0.1:6379")
+	_, err := cache.NewStore(context.Background(), 10*time.Millisecond, 30*time.Millisecond, 100)
+	require.Error(t, err, "getting redis cache store should return error")
+}
+
+func TestRedisStoreConnectionSuccess(t *testing.T) {
+	ctx := context.Background()
+	redisURL := startRedis(t)
+	redisStore := newRedisStore(t)
+
+	key, value := "testing", "tested"
+	err := redisStore.Set(ctx, key, value, store.WithExpiration(100*time.Millisecond))
+	assert.NoError(t, err, "couldn't set testing data")
+
+	result, err := redisStore.Get(ctx, key)
+	assert.NoError(t, err, "couldn't get testing data")
+	assert.Equal(t, value, result, "value returned doesn't match testing data")
+
+	options, err := redis.ParseURL(redisURL)
+	require.NoError(t, err, "parsing redis cache url")
+
+	redisClient := redis.NewClient(options)
+	r, err := redisClient.Get(ctx, key).Result()
+	assert.NoError(t, err, "couldn't get testing data from redis")
+	assert.Equal(t, value, r, "value returned from redis doesn't match testing data")
+
+	// test expiration
+	time.Sleep(300 * time.Millisecond)
+	_, err = redisStore.Get(ctx, key)
+	assert.Error(t, err, "value should not be found")
+}
+
+func TestRedisStoreSetNX(t *testing.T) {
+	ctx := context.Background()
+	redisURL := startRedis(t)
+	redisStore, secondRedisStore := newRedisStore(t), newRedisStore(t)
+
+	const (
+		key   = "conditional"
+		value = "tested"
+	)
+
+	start := make(chan struct{})
+	type setResult struct {
+		created bool
+		err     error
+	}
+	results := make(chan setResult, 2)
+	for _, cacheStore := range []cache.Store{redisStore, secondRedisStore} {
+		go func() {
+			<-start
+			created, err := cacheStore.SetNX(ctx, key, value, time.Minute)
+			results <- setResult{created: created, err: err}
+		}()
+	}
+	close(start)
+
+	created := 0
+	for range 2 {
+		result := <-results
+		require.NoError(t, result.err, "conditional redis set failed")
+		if result.created {
+			created++
+		}
+	}
+	require.Equal(t, 1, created, "expected exactly one redis client to create the entry")
+
+	options, err := redis.ParseURL(redisURL)
+	require.NoError(t, err, "parsing redis cache url")
+
+	ttl, err := redis.NewClient(options).PTTL(ctx, key).Result()
+	require.NoError(t, err, "couldn't read entry TTL")
+	require.Positive(t, ttl, "created entry should have a positive TTL")
+}
+
+func TestRedisStoreGetDel(t *testing.T) {
+	ctx := context.Background()
+	startRedis(t)
+	redisStore, secondRedisStore := newRedisStore(t), newRedisStore(t)
+
+	const (
+		key   = "consume"
+		value = "verifier"
+	)
+
+	t.Run("exactly one caller across independent clients consumes the key", func(t *testing.T) {
+		// A generous TTL: the key is consumed explicitly, so expiry racing the
+		// concurrent callers would only make the test flaky on a loaded runner.
+		err := redisStore.Set(ctx, key, value, store.WithExpiration(time.Minute))
+		require.NoError(t, err, "couldn't set value to consume")
+
+		assertGetDelConsumedOnce(ctx, t, []cache.Store{redisStore, secondRedisStore}, key, value)
+		assertGetDelMisses(ctx, t, secondRedisStore, key)
+	})
+
+	t.Run("missing key is not an error", func(t *testing.T) {
+		assertGetDelMisses(ctx, t, redisStore, "never-set")
+	})
+
+	t.Run("expired key is not found", func(t *testing.T) {
+		err := redisStore.Set(ctx, key, value, store.WithExpiration(50*time.Millisecond))
+		require.NoError(t, err, "couldn't set value to consume")
+
+		time.Sleep(100 * time.Millisecond)
+		assertGetDelMisses(ctx, t, redisStore, key)
+	})
+}

management/server/cache/store.go

@@ -2,17 +2,10 @@ package cache
 
 import (
 	"context"
-	"fmt"
-	"math"
 	"os"
 	"time"
 
 	"github.com/eko/gocache/lib/v4/store"
-	gocache_store "github.com/eko/gocache/store/go_cache/v4"
-	redis_store "github.com/eko/gocache/store/redis/v4"
-	gocache "github.com/patrickmn/go-cache"
-	"github.com/redis/go-redis/v9"
-	log "github.com/sirupsen/logrus"
 )
 
 // RedisStoreEnvVar is the environment variable that determines if a redis store should be used.
@@ -31,15 +24,23 @@ const (
 	DefaultStoreMaxConn = 1000
 )
 
+// Store extends the shared cache interface with conditional and consuming operations.
+type Store interface {
+	store.StoreInterface
+	// SetNX stores a value with a TTL only when the key does not exist.
+	SetNX(ctx context.Context, key, value string, ttl time.Duration) (bool, error)
+	// GetDel reads a value and removes it, so only one caller can consume a key.
+	GetDel(ctx context.Context, key string) (value string, found bool, err error)
+}
+
 // NewStore creates a new cache store with the given max timeout and cleanup interval. It checks for the environment Variable RedisStoreEnvVar
 // to determine if a redis store should be used. If the environment variable is set, it will attempt to connect to the redis store.
-func NewStore(ctx context.Context, maxTimeout, cleanupInterval time.Duration, maxConn int) (store.StoreInterface, error) {
+func NewStore(ctx context.Context, maxTimeout, cleanupInterval time.Duration, maxConn int) (Store, error) {
 	redisAddr := GetAddrFromEnv()
 	if redisAddr != "" {
 		return getRedisStore(ctx, redisAddr, maxConn)
 	}
-	goc := gocache.New(maxTimeout, cleanupInterval)
-	return gocache_store.NewGoCache(goc), nil
+	return newMemoryStore(maxTimeout, cleanupInterval), nil
 }
 
 // GetAddrFromEnv returns the redis address from the environment variable RedisStoreEnvVar or its legacy counterpart.
@@ -50,29 +51,3 @@ func GetAddrFromEnv() string {
 	}
 	return addr
 }
-
-func getRedisStore(ctx context.Context, redisEnvAddr string, maxConn int) (store.StoreInterface, error) {
-	options, err := redis.ParseURL(redisEnvAddr)
-	if err != nil {
-		return nil, fmt.Errorf("parsing redis cache url: %s", err)
-	}
-
-	options.MaxIdleConns = int(math.Ceil(float64(maxConn) * 0.5)) // 50% of max conns
-	options.MinIdleConns = int(math.Ceil(float64(maxConn) * 0.1)) // 10% of max conns
-	options.MaxActiveConns = maxConn
-	options.ConnMaxIdleTime = 30 * time.Minute
-	options.ConnMaxLifetime = 0
-	options.PoolTimeout = 10 * time.Second
-	redisClient := redis.NewClient(options)
-	subCtx, cancel := context.WithTimeout(ctx, 2*time.Second)
-	defer cancel()
-
-	_, err = redisClient.Ping(subCtx).Result()
-	if err != nil {
-		return nil, err
-	}
-
-	log.WithContext(subCtx).Infof("using redis cache at %s", redisEnvAddr)
-
-	return redis_store.NewRedis(redisClient), nil
-}

management/server/cache/store_test.go

@@ -3,101 +3,53 @@ package cache_test
 import (
 	"context"
 	"testing"
-	"time"
 
-	"github.com/eko/gocache/lib/v4/store"
-	"github.com/redis/go-redis/v9"
-	testcontainersredis "github.com/testcontainers/testcontainers-go/modules/redis"
+	"github.com/stretchr/testify/require"
 
 	"github.com/netbirdio/netbird/management/server/cache"
 )
 
-func TestMemoryStore(t *testing.T) {
-	memStore, err := cache.NewStore(context.Background(), 100*time.Millisecond, 300*time.Millisecond, 100)
-	if err != nil {
-		t.Fatalf("couldn't create memory store: %s", err)
-	}
-	ctx := context.Background()
-	key, value := "testing", "tested"
-	err = memStore.Set(ctx, key, value)
-	if err != nil {
-		t.Errorf("couldn't set testing data: %s", err)
-	}
-	result, err := memStore.Get(ctx, key)
-	if err != nil {
-		t.Errorf("couldn't get testing data: %s", err)
-	}
-	if value != result.(string) {
-		t.Errorf("value returned doesn't match testing data, got %s, expected %s", result, value)
-	}
-	// test expiration
-	time.Sleep(300 * time.Millisecond)
-	_, err = memStore.Get(ctx, key)
-	if err == nil {
-		t.Error("value should not be found")
-	}
-}
+func assertGetDelConsumedOnce(ctx context.Context, t *testing.T, stores []cache.Store, key, value string) {
+	t.Helper()
 
-func TestRedisStoreConnectionFailure(t *testing.T) {
-	t.Setenv(cache.RedisStoreEnvVar, "redis://127.0.0.1:6379")
-	_, err := cache.NewStore(context.Background(), 10*time.Millisecond, 30*time.Millisecond, 100)
-	if err == nil {
-		t.Fatal("getting redis cache store should return error")
-	}
-}
+	const getDelAttempts = 64
 
-func TestRedisStoreConnectionSuccess(t *testing.T) {
-	ctx := context.Background()
-	redisContainer, err := testcontainersredis.Run(ctx, "redis:7")
-	if err != nil {
-		t.Fatalf("couldn't start redis container: %s", err)
+	type getDelResult struct {
+		value string
+		found bool
+		err   error
 	}
-	defer func() {
-		if err := redisContainer.Terminate(ctx); err != nil {
-			t.Logf("failed to terminate container: %s", err)
+
+	start := make(chan struct{})
+	results := make(chan getDelResult, getDelAttempts)
+	for i := range getDelAttempts {
+		cacheStore := stores[i%len(stores)]
+		go func() {
+			<-start
+			value, found, err := cacheStore.GetDel(ctx, key)
+			results <- getDelResult{value: value, found: found, err: err}
+		}()
+	}
+	close(start)
+
+	consumers := 0
+	for range getDelAttempts {
+		result := <-results
+		require.NoError(t, result.err, "concurrent GetDel failed")
+		if !result.found {
+			continue
 		}
-	}()
-	redisURL, err := redisContainer.ConnectionString(ctx)
-	if err != nil {
-		t.Fatalf("couldn't get connection string: %s", err)
-	}
-
-	t.Setenv(cache.RedisStoreEnvVar, redisURL)
-	redisStore, err := cache.NewStore(context.Background(), 100*time.Millisecond, 300*time.Millisecond, 100)
-	if err != nil {
-		t.Fatalf("couldn't create redis store: %s", err)
-	}
-
-	key, value := "testing", "tested"
-	err = redisStore.Set(ctx, key, value, store.WithExpiration(100*time.Millisecond))
-	if err != nil {
-		t.Errorf("couldn't set testing data: %s", err)
-	}
-	result, err := redisStore.Get(ctx, key)
-	if err != nil {
-		t.Errorf("couldn't get testing data: %s", err)
-	}
-	if value != result.(string) {
-		t.Errorf("value returned doesn't match testing data, got %s, expected %s", result, value)
-	}
-
-	options, err := redis.ParseURL(redisURL)
-	if err != nil {
-		t.Errorf("parsing redis cache url: %s", err)
-	}
-
-	redisClient := redis.NewClient(options)
-	r, e := redisClient.Get(ctx, key).Result()
-	if e != nil {
-		t.Errorf("couldn't get testing data from redis: %s", e)
-	}
-	if value != r {
-		t.Errorf("value returned from redis doesn't match testing data, got %s, expected %s", r, value)
-	}
-	// test expiration
-	time.Sleep(300 * time.Millisecond)
-	_, err = redisStore.Get(ctx, key)
-	if err == nil {
-		t.Error("value should not be found")
+		consumers++
+		require.Equal(t, value, result.value, "consumed value doesn't match testing data")
 	}
+	require.Equal(t, 1, consumers, "expected exactly one consumer")
+}
+
+func assertGetDelMisses(ctx context.Context, t *testing.T, cacheStore cache.Store, key string) {
+	t.Helper()
+
+	value, found, err := cacheStore.GetDel(ctx, key)
+	require.NoError(t, err, "GetDel on a missing key should not error")
+	require.False(t, found, "GetDel should not find key %q, got value %q", key, value)
+	require.Empty(t, value, "GetDel should return an empty value when not found")
 }

management/server/http/handlers/proxy/auth.go

@@ -100,9 +100,10 @@ func (h *AuthCallbackHandler) handleCallback(w http.ResponseWriter, r *http.Requ
 		return
 	}
 
-	// Group validation is performed by the proxy via ValidateSession gRPC call.
-	// This allows the proxy to show 403 pages directly without redirect dance.
-
+	// GenerateSessionToken applies the service's group and account-status gates,
+	// so a user without access never receives a token. The proxy re-checks the
+	// installed cookie against the service's allowed groups, and renders the
+	// denial page from the error carried back in the redirect.
 	sessionToken, err := h.proxyService.GenerateSessionToken(r.Context(), redirectURL.Hostname(), userID, auth.MethodOIDC)
 	if err != nil {
 		log.WithError(err).Error("Failed to create session token")
@@ -136,6 +137,9 @@ func sessionTokenErrorDescription(err error) string {
 	if errors.Is(err, nbgrpc.ErrUserBlocked) {
 		return "Your account is blocked"
 	}
+	if errors.Is(err, nbgrpc.ErrUserNotInGroup) {
+		return "You are not authorized to access this service"
+	}
 	return "Service configuration error"
 }
 

management/server/identity_provider_test.go

@@ -10,9 +10,9 @@ import (
 	"testing"
 	"time"
 
-	"go.uber.org/mock/gomock"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
+	"go.uber.org/mock/gomock"
 
 	"github.com/netbirdio/netbird/management/internals/controllers/network_map/controller"
 	"github.com/netbirdio/netbird/management/internals/controllers/network_map/update_channel"
@@ -34,6 +34,20 @@ import (
 
 func createManagerWithEmbeddedIdP(t testing.TB) (*DefaultAccountManager, *update_channel.PeersUpdateManager, error) {
 	t.Helper()
+	return createManagerWithEmbeddedIdPMode(t, "netbird.selfhosted")
+}
+
+func createManagerWithEmbeddedIdPMode(t testing.TB, singleAccountModeDomain string) (*DefaultAccountManager, *update_channel.PeersUpdateManager, error) {
+	t.Helper()
+	return createManagerWithEmbeddedIdPModeAndSetup(t, singleAccountModeDomain, nil)
+}
+
+func createManagerWithEmbeddedIdPModeAndSetup(
+	t testing.TB,
+	singleAccountModeDomain string,
+	setupStore func(context.Context, store.Store) error,
+) (*DefaultAccountManager, *update_channel.PeersUpdateManager, error) {
+	t.Helper()
 
 	ctx := context.Background()
 
@@ -43,6 +57,11 @@ func createManagerWithEmbeddedIdP(t testing.TB) (*DefaultAccountManager, *update
 		return nil, nil, err
 	}
 	t.Cleanup(cleanUp)
+	if setupStore != nil {
+		if err := setupStore(ctx, testStore); err != nil {
+			return nil, nil, err
+		}
+	}
 
 	// Create embedded IdP manager
 	embeddedConfig := &idp.EmbeddedIdPConfig{
@@ -93,7 +112,7 @@ func createManagerWithEmbeddedIdP(t testing.TB) (*DefaultAccountManager, *update
 	updateManager := update_channel.NewPeersUpdateManager(metrics)
 	requestBuffer := NewAccountRequestBuffer(ctx, testStore)
 	networkMapController := controller.NewController(ctx, testStore, metrics, updateManager, requestBuffer, MockIntegratedValidator{}, settingsMockManager, "netbird.cloud", port_forwarding.NewControllerMock(), ephemeral_manager.NewEphemeralManager(testStore, peersManager), &config.Config{}, nil)
-	manager, err := BuildManager(ctx, &config.Config{}, testStore, networkMapController, job.NewJobManager(nil, testStore, peersManager), idpManager, "", eventStore, nil, false, MockIntegratedValidator{}, metrics, port_forwarding.NewControllerMock(), settingsMockManager, permissionsManager, false, cacheStore)
+	manager, err := BuildManager(ctx, &config.Config{}, testStore, networkMapController, job.NewJobManager(nil, testStore, peersManager), idpManager, singleAccountModeDomain, eventStore, nil, false, MockIntegratedValidator{}, metrics, port_forwarding.NewControllerMock(), settingsMockManager, permissionsManager, false, cacheStore)
 	if err != nil {
 		return nil, nil, err
 	}
@@ -196,6 +215,23 @@ func TestDefaultAccountManager_GetIdentityProvider_NotFound(t *testing.T) {
 	assert.Contains(t, err.Error(), "not found")
 }
 
+func TestUpdateUserAuthWithSingleModeKeepsConfiguredDomain(t *testing.T) {
+	ctx := context.Background()
+	manager, _, err := createManagerWithEmbeddedIdPModeAndSetup(t, "netbird.selfhosted", func(ctx context.Context, testStore store.Store) error {
+		// An account with no domain, as left behind by an IdP that emitted no domain claims.
+		return testStore.SaveAccount(ctx, newAccountWithId(ctx, "account-1", "user-1", "", "", "", false))
+	})
+	require.NoError(t, err)
+	require.True(t, manager.singleAccountMode)
+
+	userAuth := auth.UserAuth{UserId: "user-2"}
+	require.NoError(t, manager.updateUserAuthWithSingleMode(ctx, &userAuth))
+
+	assert.Equal(t, "netbird.selfhosted", userAuth.Domain,
+		"An empty account domain must not clear the configured single account domain")
+	assert.Equal(t, types.PrivateCategory, userAuth.DomainCategory)
+}
+
 func TestDefaultAccountManager_UpdateIdentityProvider_Validation(t *testing.T) {
 	manager, _, err := createManager(t)
 	require.NoError(t, err)

management/server/idp/migration/migration.go

@@ -10,6 +10,8 @@ import (
 	"errors"
 	"fmt"
 	"os"
+	"regexp"
+	"strings"
 
 	log "github.com/sirupsen/logrus"
 
@@ -25,8 +27,10 @@ type Server interface {
 	EventStore() EventStore // may return nil
 }
 
-const idpSeedInfoKey = "IDP_SEED_INFO"
-const dryRunEnvKey = "NB_IDP_MIGRATION_DRY_RUN"
+const (
+	idpSeedInfoKey = "IDP_SEED_INFO"
+	dryRunEnvKey   = "NB_IDP_MIGRATION_DRY_RUN"
+)
 
 func isDryRun() bool {
 	return os.Getenv(dryRunEnvKey) == "true"
@@ -233,3 +237,163 @@ func PopulateUserInfo(s Server, idpManager idp.Manager, dryRun bool) error {
 
 	return nil
 }
+
+const DefaultSingleAccountDomain = "netbird.selfhosted"
+
+var (
+	ErrMultipleAccounts = errors.New("the embedded IdP supports a single account only")
+	ErrUnusableDomain   = errors.New("domain cannot be resolved in single account mode")
+	ErrDomainConflict   = errors.New("requested domain conflicts with the account domain")
+)
+
+var resolvableDomainRegexp = regexp.MustCompile(`^([a-z0-9]+(-[a-z0-9]+)*\.)+[a-z]{2,}$`)
+
+// RequireSingleAccount refuses to migrate an instance that holds more than one account.
+func RequireSingleAccount(s Server) error {
+	accountsCounter, err := s.Store().GetAccountsCounter(context.Background())
+	if err != nil {
+		return fmt.Errorf("failed to count accounts: %w", err)
+	}
+
+	if accountsCounter > 1 {
+		return errMultipleAccounts(accountsCounter)
+	}
+
+	return nil
+}
+
+func errMultipleAccounts(accountsCounter int64) error {
+	return fmt.Errorf("%w: this instance has %d accounts. Identity provider connectors are stored without "+
+		"an account scope, so every account would share and be able to manage the same connectors. "+
+		"Consolidate this instance to a single account, or keep using an external IdP, before migrating",
+		ErrMultipleAccounts, accountsCounter)
+}
+
+func NormalizeSingleAccountDomain(singleAccountDomain string) (string, error) {
+	if singleAccountDomain == "" {
+		singleAccountDomain = DefaultSingleAccountDomain
+	}
+
+	singleAccountDomain = strings.ToLower(singleAccountDomain)
+	if !resolvableDomainRegexp.MatchString(singleAccountDomain) {
+		return "", fmt.Errorf("%w: %q must contain at least one dot and only lowercase letters, digits and "+
+			"hyphens, otherwise users cannot join the existing account", ErrUnusableDomain, singleAccountDomain)
+	}
+
+	return singleAccountDomain, nil
+}
+
+// resolveAccountDomain picks the domain the account should end up with. The account keeps a usable
+// domain of its own, the configured one only fills a blank. Anything else is a conflict to report.
+func resolveAccountDomain(accountID, accountDomain, singleAccountDomain string, requested bool) (string, error) {
+	accountDomain = strings.ToLower(accountDomain)
+
+	if accountDomain == "" {
+		return singleAccountDomain, nil
+	}
+
+	if !resolvableDomainRegexp.MatchString(accountDomain) {
+		return "", fmt.Errorf("%w: account %s has domain %q, which must contain at least one dot and only "+
+			"lowercase letters, digits and hyphens. Correct the account domain before migrating",
+			ErrUnusableDomain, accountID, accountDomain)
+	}
+
+	if requested && accountDomain != singleAccountDomain {
+		return "", fmt.Errorf("%w: account %s already uses domain %q but %q was requested. Re-run without "+
+			"--single-account-mode-domain to keep %q, or correct the account domain first",
+			ErrDomainConflict, accountID, accountDomain, singleAccountDomain, accountDomain)
+	}
+
+	return accountDomain, nil
+}
+
+// EnsureSingleAccountDomain gives the remaining account the domain attributes single account mode
+// resolves against, so users can still join it after the migration.
+func EnsureSingleAccountDomain(s Server, singleAccountDomain string) error {
+	plan, err := planSingleAccountDomain(s, singleAccountDomain)
+	if err != nil {
+		return err
+	}
+	if plan.skip {
+		return nil
+	}
+
+	if isDryRun() {
+		log.Infof("[DRY RUN] would set account %s domain to %q, category to %q and mark it as the primary domain account "+
+			"(currently domain=%q primary=%v)", plan.accountID, plan.domain, types.PrivateCategory,
+			plan.currentDomain, plan.isPrimary)
+		return nil
+	}
+
+	if err := s.Store().UpdateAccountDomainAttributes(context.Background(), plan.accountID, plan.domain,
+		types.PrivateCategory, true); err != nil {
+		return fmt.Errorf("failed to update domain attributes of account %s: %w", plan.accountID, err)
+	}
+
+	log.Infof("account %s now resolves in single account mode with domain %q", plan.accountID, plan.domain)
+	return nil
+}
+
+// CheckSingleAccountDomain reports whether EnsureSingleAccountDomain would succeed, without writing.
+func CheckSingleAccountDomain(s Server, singleAccountDomain string) error {
+	_, err := planSingleAccountDomain(s, singleAccountDomain)
+	return err
+}
+
+type singleAccountDomainPlan struct {
+	accountID     string
+	domain        string
+	currentDomain string
+	isPrimary     bool
+	skip          bool
+}
+
+// planSingleAccountDomain decides what the account's domain attributes should become. It reads
+// only, so it can run both as a preflight and as the first half of the update.
+func planSingleAccountDomain(s Server, singleAccountDomain string) (singleAccountDomainPlan, error) {
+	ctx := context.Background()
+
+	// An empty value means the operator did not pick a domain, so the default is only a fallback.
+	requested := singleAccountDomain != ""
+
+	singleAccountDomain, err := NormalizeSingleAccountDomain(singleAccountDomain)
+	if err != nil {
+		return singleAccountDomainPlan{}, err
+	}
+
+	accountsCounter, err := s.Store().GetAccountsCounter(ctx)
+	if err != nil {
+		return singleAccountDomainPlan{}, fmt.Errorf("failed to count accounts: %w", err)
+	}
+	// The count is checked again here: it is read long after RequireSingleAccount, and marking an
+	// arbitrary account as the primary one for the domain would be wrong.
+	switch {
+	case accountsCounter == 0:
+		log.Info("no accounts yet, nothing to prepare for single account mode")
+		return singleAccountDomainPlan{skip: true}, nil
+	case accountsCounter > 1:
+		return singleAccountDomainPlan{}, errMultipleAccounts(accountsCounter)
+	}
+
+	accountID, err := s.Store().GetAnyAccountID(ctx)
+	if err != nil {
+		return singleAccountDomainPlan{}, fmt.Errorf("failed to get the existing account: %w", err)
+	}
+
+	isPrimary, accountDomain, err := s.Store().IsPrimaryAccount(ctx, accountID)
+	if err != nil {
+		return singleAccountDomainPlan{}, fmt.Errorf("failed to read domain attributes of account %s: %w", accountID, err)
+	}
+
+	domain, err := resolveAccountDomain(accountID, accountDomain, singleAccountDomain, requested)
+	if err != nil {
+		return singleAccountDomainPlan{}, err
+	}
+
+	return singleAccountDomainPlan{
+		accountID:     accountID,
+		domain:        domain,
+		currentDomain: accountDomain,
+		isPrimary:     isPrimary,
+	}, nil
+}

management/server/idp/migration/migration_test.go

@@ -24,6 +24,17 @@ type testStore struct {
 	checkSchemaFunc    func(checks []SchemaCheck) []SchemaError
 	updateCalls        []updateUserIDCall
 	updateInfoCalls    []updateUserInfoCall
+
+	accountsCounter int64
+	accounts        map[string]*types.Account
+	domainAttrCalls []domainAttrCall
+}
+
+type domainAttrCall struct {
+	AccountID string
+	Domain    string
+	Category  string
+	IsPrimary bool
 }
 
 type updateUserIDCall struct {
@@ -38,6 +49,35 @@ type updateUserInfoCall struct {
 	Name   string
 }
 
+func (s *testStore) GetAccountsCounter(context.Context) (int64, error) {
+	return s.accountsCounter, nil
+}
+
+func (s *testStore) GetAnyAccountID(context.Context) (string, error) {
+	for id := range s.accounts {
+		return id, nil
+	}
+	return "", fmt.Errorf("no accounts")
+}
+
+func (s *testStore) IsPrimaryAccount(_ context.Context, accountID string) (bool, string, error) {
+	account, ok := s.accounts[accountID]
+	if !ok {
+		return false, "", fmt.Errorf("account %s not found", accountID)
+	}
+	return account.IsDomainPrimaryAccount, account.Domain, nil
+}
+
+func (s *testStore) UpdateAccountDomainAttributes(_ context.Context, accountID, domain, category string, isPrimaryDomain bool) error {
+	s.domainAttrCalls = append(s.domainAttrCalls, domainAttrCall{accountID, domain, category, isPrimaryDomain})
+	if account, ok := s.accounts[accountID]; ok {
+		account.Domain = domain
+		account.DomainCategory = category
+		account.IsDomainPrimaryAccount = isPrimaryDomain
+	}
+	return nil
+}
+
 func (s *testStore) ListUsers(ctx context.Context) ([]*types.User, error) {
 	return s.listUsersFunc(ctx)
 }
@@ -826,3 +866,212 @@ func TestCheckSchema_MockStore(t *testing.T) {
 		assert.Equal(t, "email", errs[0].Column)
 	})
 }
+
+func TestRequireSingleAccount(t *testing.T) {
+	tests := []struct {
+		name      string
+		accounts  int64
+		expectErr bool
+	}{
+		{name: "fresh install", accounts: 0},
+		{name: "single account", accounts: 1},
+		{name: "multiple accounts", accounts: 3, expectErr: true},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			srv := &testServer{store: &testStore{accountsCounter: tt.accounts}}
+
+			err := RequireSingleAccount(srv)
+			if !tt.expectErr {
+				require.NoError(t, err)
+				return
+			}
+
+			require.Error(t, err)
+			assert.ErrorIs(t, err, ErrMultipleAccounts)
+		})
+	}
+}
+
+func TestEnsureSingleAccountDomain(t *testing.T) {
+	tests := []struct {
+		name            string
+		account         *types.Account
+		requestedDomain string
+		expectedDomain  string
+	}{
+		{
+			name:           "account migrated from an IdP without domain claims",
+			account:        &types.Account{Id: "account-1"},
+			expectedDomain: DefaultSingleAccountDomain,
+		},
+		{
+			name:            "requested domain is applied to an account without one",
+			account:         &types.Account{Id: "account-1"},
+			requestedDomain: "corp.example.com",
+			expectedDomain:  "corp.example.com",
+		},
+		{
+			name:           "account keeps its own domain",
+			account:        &types.Account{Id: "account-1", Domain: "acme.com"},
+			expectedDomain: "acme.com",
+		},
+		{
+			name:            "requesting the domain the account already has is not a conflict",
+			account:         &types.Account{Id: "account-1", Domain: "acme.com"},
+			requestedDomain: "acme.com",
+			expectedDomain:  "acme.com",
+		},
+		{
+			name: "already resolvable account is rewritten with the same values",
+			account: &types.Account{
+				Id:                     "account-1",
+				Domain:                 "acme.com",
+				DomainCategory:         types.PrivateCategory,
+				IsDomainPrimaryAccount: true,
+			},
+			expectedDomain: "acme.com",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			store := &testStore{
+				accountsCounter: 1,
+				accounts:        map[string]*types.Account{tt.account.Id: tt.account},
+			}
+
+			require.NoError(t, EnsureSingleAccountDomain(&testServer{store: store}, tt.requestedDomain))
+
+			require.Len(t, store.domainAttrCalls, 1)
+			assert.Equal(t, domainAttrCall{
+				AccountID: tt.account.Id,
+				Domain:    tt.expectedDomain,
+				Category:  types.PrivateCategory,
+				IsPrimary: true,
+			}, store.domainAttrCalls[0])
+		})
+	}
+}
+
+func TestEnsureSingleAccountDomainDryRun(t *testing.T) {
+	t.Setenv(dryRunEnvKey, "true")
+
+	store := &testStore{
+		accountsCounter: 1,
+		accounts:        map[string]*types.Account{"account-1": {Id: "account-1"}},
+	}
+
+	require.NoError(t, EnsureSingleAccountDomain(&testServer{store: store}, ""))
+	assert.Empty(t, store.domainAttrCalls, "Dry run must not write anything")
+}
+
+func TestEnsureSingleAccountDomainRejectsUnresolvableDomains(t *testing.T) {
+	t.Run("account domain that cannot resolve is reported", func(t *testing.T) {
+		account := &types.Account{Id: "account-1", Domain: "corp"}
+		store := &testStore{
+			accountsCounter: 1,
+			accounts:        map[string]*types.Account{account.Id: account},
+		}
+
+		err := EnsureSingleAccountDomain(&testServer{store: store}, "")
+		require.Error(t, err)
+		assert.ErrorIs(t, err, ErrUnusableDomain)
+		assert.Empty(t, store.domainAttrCalls, "A broken account domain must not be replaced silently")
+	})
+
+	t.Run("requested domain conflicting with the account domain is reported", func(t *testing.T) {
+		account := &types.Account{Id: "account-1", Domain: "acme.com"}
+		store := &testStore{
+			accountsCounter: 1,
+			accounts:        map[string]*types.Account{account.Id: account},
+		}
+
+		err := EnsureSingleAccountDomain(&testServer{store: store}, "corp.example.com")
+		require.Error(t, err)
+		assert.ErrorIs(t, err, ErrDomainConflict)
+		assert.Empty(t, store.domainAttrCalls, "A conflict must not overwrite the account domain")
+	})
+
+	t.Run("configured domain that cannot resolve is rejected", func(t *testing.T) {
+		store := &testStore{
+			accountsCounter: 1,
+			accounts:        map[string]*types.Account{"account-1": {Id: "account-1"}},
+		}
+
+		err := EnsureSingleAccountDomain(&testServer{store: store}, "corp")
+		require.Error(t, err)
+		assert.ErrorIs(t, err, ErrUnusableDomain)
+		assert.Empty(t, store.domainAttrCalls)
+	})
+
+	t.Run("account appearing after the preflight is rejected", func(t *testing.T) {
+		store := &testStore{
+			accountsCounter: 2,
+			accounts: map[string]*types.Account{
+				"account-1": {Id: "account-1"},
+				"account-2": {Id: "account-2"},
+			},
+		}
+
+		err := EnsureSingleAccountDomain(&testServer{store: store}, "")
+		require.Error(t, err)
+		assert.ErrorIs(t, err, ErrMultipleAccounts)
+		assert.Empty(t, store.domainAttrCalls, "No account may be marked primary when several exist")
+	})
+
+	t.Run("fresh install with no accounts is a no-op", func(t *testing.T) {
+		store := &testStore{accountsCounter: 0, accounts: map[string]*types.Account{}}
+
+		require.NoError(t, EnsureSingleAccountDomain(&testServer{store: store}, ""))
+		assert.Empty(t, store.domainAttrCalls)
+	})
+}
+
+func TestCheckSingleAccountDomain(t *testing.T) {
+	tests := []struct {
+		name      string
+		account   *types.Account
+		requested string
+		expectErr error
+	}{
+		{
+			name:    "usable account domain passes",
+			account: &types.Account{Id: "account-1", Domain: "acme.com"},
+		},
+		{
+			name:    "empty account domain passes",
+			account: &types.Account{Id: "account-1"},
+		},
+		{
+			name:      "unresolvable account domain fails",
+			account:   &types.Account{Id: "account-1", Domain: "corp"},
+			expectErr: ErrUnusableDomain,
+		},
+		{
+			name:      "conflicting request fails",
+			account:   &types.Account{Id: "account-1", Domain: "acme.com"},
+			requested: "corp.example.com",
+			expectErr: ErrDomainConflict,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			store := &testStore{
+				accountsCounter: 1,
+				accounts:        map[string]*types.Account{tt.account.Id: tt.account},
+			}
+
+			err := CheckSingleAccountDomain(&testServer{store: store}, tt.requested)
+			if tt.expectErr == nil {
+				require.NoError(t, err)
+			} else {
+				require.ErrorIs(t, err, tt.expectErr)
+			}
+
+			assert.Empty(t, store.domainAttrCalls, "The preflight must not write anything")
+		})
+	}
+}

management/server/idp/migration/store.go

@@ -60,6 +60,20 @@ type Store interface {
 	// CheckSchema verifies that all tables and columns required by the migration
 	// exist in the database. Returns a list of problems; an empty slice means OK.
 	CheckSchema(checks []SchemaCheck) []SchemaError
+
+	// GetAccountsCounter returns the total number of accounts in the store.
+	GetAccountsCounter(ctx context.Context) (int64, error)
+
+	// GetAnyAccountID returns the ID of one of the existing accounts.
+	GetAnyAccountID(ctx context.Context) (string, error)
+
+	// IsPrimaryAccount returns whether the account is the primary account for its domain,
+	// along with that domain.
+	IsPrimaryAccount(ctx context.Context, accountID string) (bool, string, error)
+
+	// UpdateAccountDomainAttributes sets the domain, domain category and primary
+	// domain flag of an account.
+	UpdateAccountDomainAttributes(ctx context.Context, accountID string, domain string, category string, isPrimaryDomain bool) error
 }
 
 // RequiredEventSchema lists all tables and columns that the migration tool needs

management/server/mock_server/management_server_mock.go

@@ -13,7 +13,7 @@ type ManagementServiceServerMock struct {
 	proto.UnimplementedManagementServiceServer
 
 	LoginFunc                      func(context.Context, *proto.EncryptedMessage) (*proto.EncryptedMessage, error)
-	SyncFunc                       func(*proto.EncryptedMessage, proto.ManagementService_SyncServer)
+	SyncFunc                       func(*proto.EncryptedMessage, proto.ManagementService_SyncServer) error
 	GetServerKeyFunc               func(context.Context, *proto.Empty) (*proto.ServerKeyResponse, error)
 	IsHealthyFunc                  func(context.Context, *proto.Empty) (*proto.Empty, error)
 	GetDeviceAuthorizationFlowFunc func(ctx context.Context, req *proto.EncryptedMessage) (*proto.EncryptedMessage, error)
@@ -30,7 +30,7 @@ func (m ManagementServiceServerMock) Login(ctx context.Context, req *proto.Encry
 
 func (m ManagementServiceServerMock) Sync(msg *proto.EncryptedMessage, sync proto.ManagementService_SyncServer) error {
 	if m.SyncFunc != nil {
-		return m.Sync(msg, sync)
+		return m.SyncFunc(msg, sync)
 	}
 	return status.Errorf(codes.Unimplemented, "method Sync not implemented")
 }

management/server/peer.go

@@ -1499,9 +1499,12 @@ func checkAuth(ctx context.Context, loginUserID string, peer *nbpeer.Peer) error
 
 func peerLoginExpired(ctx context.Context, peer *nbpeer.Peer, settings *types.Settings) bool {
 	expired, expiresIn := peer.LoginExpired(settings.PeerLoginExpiration)
-	expired = settings.PeerLoginExpirationEnabled && expired
-	if expired || peer.Status.LoginExpired {
-		log.WithContext(ctx).Debugf("peer's %s login expired %v ago", peer.ID, expiresIn)
+	if settings.PeerLoginExpirationEnabled && expired {
+		log.WithContext(ctx).Debugf("peer's %s login expired %v ago", peer.ID, -expiresIn)
+		return true
+	}
+	if peer.Status.LoginExpired {
+		log.WithContext(ctx).Debugf("peer's %s login is marked as expired", peer.ID)
 		return true
 	}
 	return false
@@ -1648,7 +1651,9 @@ func (am *DefaultAccountManager) UpdateAccountPeer(ctx context.Context, accountI
 
 // getNextPeerExpiration returns the minimum duration in which the next peer of the account will expire if it was found.
 // If there is no peer that expires this function returns false and a duration of 0.
-// This function only considers peers that haven't been expired yet and that are connected.
+// This function only considers peers that haven't been expired yet. Offline peers count too:
+// a running job is never re-armed on connect, so a peer that reconnects with an old login
+// must already be part of the scheduled run.
 func (am *DefaultAccountManager) getNextPeerExpiration(ctx context.Context, accountID string) (time.Duration, bool) {
 	peersWithExpiry, err := am.Store.GetAccountPeersWithExpiration(ctx, store.LockingStrengthNone, accountID)
 	if err != nil {
@@ -1668,8 +1673,7 @@ func (am *DefaultAccountManager) getNextPeerExpiration(ctx context.Context, acco
 
 	var nextExpiry *time.Duration
 	for _, peer := range peersWithExpiry {
-		// consider only connected peers because others will require login on connecting to the management server
-		if peer.Status.LoginExpired || !peer.Status.Connected {
+		if peer.Status.LoginExpired {
 			continue
 		}
 		_, duration := peer.LoginExpired(settings.PeerLoginExpiration)

management/server/scheduler.go

@@ -117,6 +117,7 @@ func (wm *DefaultScheduler) Schedule(ctx context.Context, in time.Duration, ID s
 	}
 
 	ticker := time.NewTicker(in)
+	period := in
 
 	wm.jobs[ID] = cancel
 	log.WithContext(ctx).Debugf("scheduled a job %s to run in %s. There are %d total jobs scheduled.", ID, in.String(), len(wm.jobs))
@@ -136,14 +137,18 @@ func (wm *DefaultScheduler) Schedule(ctx context.Context, in time.Duration, ID s
 				if !reschedule {
 					wm.mu.Lock()
 					defer wm.mu.Unlock()
-					delete(wm.jobs, ID)
+					// A Cancel during job() may have registered a replacement under this ID.
+					if current, ok := wm.jobs[ID]; ok && current == cancel {
+						delete(wm.jobs, ID)
+					}
 					log.WithContext(ctx).Debugf("job %s is not scheduled to run again", ID)
 					ticker.Stop()
 					return
 				}
 				// we need this comparison to avoid resetting the ticker with the same duration and missing the current elapsesed time
-				if runIn != in {
+				if runIn != period {
 					ticker.Reset(runIn)
+					period = runIn
 				}
 			case <-cancel:
 				log.WithContext(ctx).Debugf("job %s was canceled, stopping timer", ID)

management/server/scheduler_test.go

@@ -6,10 +6,12 @@ import (
 	"math/rand"
 	"runtime"
 	"sync"
+	"sync/atomic"
 	"testing"
 	"time"
 
 	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
 )
 
 func TestScheduler_Performance(t *testing.T) {
@@ -150,3 +152,90 @@ func TestScheduler_Schedule(t *testing.T) {
 	scheduler.cancel(context.Background(), jobID)
 
 }
+
+func TestScheduler_Schedule_ResetsTickerAfterReturningInitialInterval(t *testing.T) {
+	jobID := "test-scheduler-job-2"
+	scheduler := NewDefaultScheduler()
+	defer scheduler.Cancel(context.Background(), []string{jobID})
+
+	initial := 30 * time.Millisecond
+	stretched := 400 * time.Millisecond
+	runs := make(chan time.Time, 3)
+	count := 0
+	// The first run stretches the period; the second returns the initial interval again,
+	// which must shrink the period back instead of keeping the stretched one.
+	job := func() (nextRunIn time.Duration, reschedule bool) {
+		count++
+		runs <- time.Now()
+		switch count {
+		case 1:
+			return stretched, true
+		case 2:
+			return initial, true
+		default:
+			return 0, false
+		}
+	}
+	scheduler.Schedule(context.Background(), initial, jobID, job)
+
+	var stamps []time.Time
+	for len(stamps) < 3 {
+		select {
+		case ts := <-runs:
+			stamps = append(stamps, ts)
+		case <-time.After(2 * time.Second):
+			t.Fatalf("timed out after %d runs", len(stamps))
+		}
+	}
+	assert.Less(t, stamps[2].Sub(stamps[1]), stretched/2, "returning the initial interval must reset the stretched ticker")
+}
+
+func TestScheduler_Schedule_StaleCompletionKeepsReplacement(t *testing.T) {
+	jobID := "test-scheduler-job-3"
+	scheduler := NewDefaultScheduler()
+	defer scheduler.Cancel(context.Background(), []string{jobID})
+
+	started := make(chan struct{})
+	release := make(chan struct{})
+	staleJob := func() (nextRunIn time.Duration, reschedule bool) {
+		close(started)
+		<-release
+		return 0, false
+	}
+	scheduler.Schedule(context.Background(), 10*time.Millisecond, jobID, staleJob)
+	select {
+	case <-started:
+	case <-time.After(time.Second):
+		t.Fatal("timed out waiting for the first job to start")
+	}
+
+	// Cancel the job while it is still executing and register a replacement under the
+	// same ID, as the expiration paths do on a settings change.
+	scheduler.Cancel(context.Background(), []string{jobID})
+	var replacementRuns atomic.Int32
+	scheduler.Schedule(context.Background(), 20*time.Millisecond, jobID, func() (nextRunIn time.Duration, reschedule bool) {
+		replacementRuns.Add(1)
+		return 20 * time.Millisecond, true
+	})
+	require.True(t, scheduler.IsSchedulerRunning(jobID), "replacement must be registered")
+
+	// The stale job now completes without rescheduling; its cleanup must leave the
+	// replacement's entry in place.
+	close(release)
+	assert.Never(t, func() bool { return !scheduler.IsSchedulerRunning(jobID) }, 200*time.Millisecond, 10*time.Millisecond,
+		"stale completion must not drop the replacement job")
+
+	var duplicateRuns atomic.Int32
+	scheduler.Schedule(context.Background(), 10*time.Millisecond, jobID, func() (nextRunIn time.Duration, reschedule bool) {
+		duplicateRuns.Add(1)
+		return 10 * time.Millisecond, true
+	})
+	assert.Never(t, func() bool { return duplicateRuns.Load() > 0 }, 100*time.Millisecond, 10*time.Millisecond,
+		"a duplicate schedule must be refused while the replacement is registered")
+
+	scheduler.Cancel(context.Background(), []string{jobID})
+	assert.False(t, scheduler.IsSchedulerRunning(jobID), "cancel must find and remove the replacement")
+	runsAfterCancel := replacementRuns.Load()
+	assert.Never(t, func() bool { return replacementRuns.Load() > runsAfterCancel+1 }, 150*time.Millisecond, 10*time.Millisecond,
+		"the replacement must stop after cancel")
+}

management/server/types/account.go

@@ -404,7 +404,7 @@ func (a *Account) GetExpiredPeers() []*nbpeer.Peer {
 
 // GetNextPeerExpiration returns the minimum duration in which the next peer of the account will expire if it was found.
 // If there is no peer that expires this function returns false and a duration of 0.
-// This function only considers peers that haven't been expired yet and that are connected.
+// This function only considers peers that haven't been expired yet, whether connected or not.
 func (a *Account) GetNextPeerExpiration() (time.Duration, bool) {
 	peersWithExpiry := a.GetPeersWithExpiration()
 	if len(peersWithExpiry) == 0 {
@@ -412,8 +412,7 @@ func (a *Account) GetNextPeerExpiration() (time.Duration, bool) {
 	}
 	var nextExpiry *time.Duration
 	for _, peer := range peersWithExpiry {
-		// consider only connected peers because others will require login on connecting to the management server
-		if peer.Status.LoginExpired || !peer.Status.Connected {
+		if peer.Status.LoginExpired {
 			continue
 		}
 		_, duration := peer.LoginExpired(a.Settings.PeerLoginExpiration)

management/server/user.go

@@ -1177,28 +1177,35 @@ func (am *DefaultAccountManager) expireAndUpdatePeers(ctx context.Context, accou
 	dnsDomain := am.networkMapController.GetDNSDomain(settings)
 
 	var peerIDs []string
-	for _, peer := range peers {
+	defer func() {
+		if len(peerIDs) == 0 {
+			return
+		}
+		// this will trigger peer disconnect from the management service
+		log.Debugf("Expiring %d peers for account %s", len(peerIDs), accountID)
+		am.networkMapController.DisconnectPeers(ctx, accountID, peerIDs)
+	}()
+	for _, candidate := range peers {
 		// nolint:staticcheck
-		ctx = context.WithValue(ctx, nbcontext.PeerIDKey, peer.Key)
+		peerCtx := context.WithValue(ctx, nbcontext.PeerIDKey, candidate.Key)
 
-		if peer.UserID == "" {
+		if candidate.UserID == "" {
 			// we do not want to expire peers that are added via setup key
 			continue
 		}
 
-		if peer.Status.LoginExpired {
+		peer, err := am.expirePeerIfStillDue(peerCtx, accountID, candidate.ID, settings, reason)
+		if err != nil {
+			return err
+		}
+		if peer == nil {
 			continue
 		}
 		peerIDs = append(peerIDs, peer.ID)
-		peer.MarkLoginExpired(true)
-
-		if err := am.Store.SavePeerStatus(ctx, accountID, peer.ID, *peer.Status); err != nil {
-			return err
-		}
 		meta := peer.EventMeta(dnsDomain)
 		meta["reason"] = string(reason)
 		am.StoreEvent(
-			ctx,
+			peerCtx,
 			peer.UserID, peer.ID, accountID,
 			activity.PeerLoginExpired, meta,
 		)
@@ -1215,15 +1222,53 @@ func (am *DefaultAccountManager) expireAndUpdatePeers(ctx context.Context, accou
 	if err != nil {
 		return fmt.Errorf("notify network map controller of peer update: %w", err)
 	}
-
-	if len(peerIDs) != 0 {
-		// this will trigger peer disconnect from the management service
-		log.Debugf("Expiring %d peers for account %s", len(peerIDs), accountID)
-		am.networkMapController.DisconnectPeers(ctx, accountID, peerIDs)
-	}
 	return nil
 }
 
+// expirePeerIfStillDue flags the peer as login-expired and returns its fresh copy, or nil
+// when it no longer qualifies. The candidate list is read without a lock, so a login that
+// landed in between would otherwise be overwritten with a stale expired status.
+func (am *DefaultAccountManager) expirePeerIfStillDue(ctx context.Context, accountID, peerID string, settings *types.Settings, reason peerExpirationReason) (*nbpeer.Peer, error) {
+	var expired *nbpeer.Peer
+	err := am.Store.ExecuteInTransaction(ctx, func(transaction store.Store) error {
+		peer, err := transaction.GetPeerByID(ctx, store.LockingStrengthUpdate, accountID, peerID)
+		if err != nil {
+			if s, ok := status.FromError(err); ok && s.Type() == status.NotFound {
+				return nil
+			}
+			return err
+		}
+		if peer.Status.LoginExpired || !peerExpirationDue(peer, settings, reason) {
+			return nil
+		}
+		peer.MarkLoginExpired(true)
+		if err := transaction.SavePeerStatus(ctx, accountID, peer.ID, *peer.Status); err != nil {
+			return err
+		}
+		expired = peer
+		return nil
+	})
+	if err != nil {
+		return nil, err
+	}
+	return expired, nil
+}
+
+// peerExpirationDue re-evaluates a time-based expiry against the peer's current state.
+// Administrative reasons expire the peer unconditionally.
+func peerExpirationDue(peer *nbpeer.Peer, settings *types.Settings, reason peerExpirationReason) bool {
+	switch reason {
+	case peerExpirationSessionExpired:
+		expired, _ := peer.LoginExpired(settings.PeerLoginExpiration)
+		return settings.PeerLoginExpirationEnabled && expired
+	case peerExpirationInactivity:
+		expired, _ := peer.SessionExpired(settings.PeerInactivityExpiration)
+		return settings.PeerInactivityExpirationEnabled && expired
+	default:
+		return true
+	}
+}
+
 func (am *DefaultAccountManager) deleteUserFromIDP(ctx context.Context, targetUserID, accountID string) error {
 	if am.userDeleteFromIDPEnabled {
 		log.WithContext(ctx).Debugf("user %s deleted from IdP", targetUserID)
@@ -1337,6 +1382,10 @@ func (am *DefaultAccountManager) deleteRegularUser(ctx context.Context, accountI
 			return fmt.Errorf("failed to get user to delete: %w", err)
 		}
 
+		if targetUser.Role == types.UserRoleOwner && targetUser.Id != initiatorUserID {
+			return status.NewOwnerDeletePermissionError()
+		}
+
 		settings, err = transaction.GetAccountSettings(ctx, store.LockingStrengthNone, accountID)
 		if err != nil {
 			return fmt.Errorf("failed to get account settings: %w", err)

management/server/user_test.go

@@ -942,6 +942,49 @@ func TestUser_DeleteUser_regularUser(t *testing.T) {
 
 }
 
+func TestUser_deleteRegularUser_RejectsOwner(t *testing.T) {
+	s, cleanup, err := store.NewTestStoreFromSQL(context.Background(), "", t.TempDir())
+	require.NoError(t, err)
+	t.Cleanup(cleanup)
+
+	account := newAccountWithId(context.Background(), mockAccountID, mockUserID, "", "", "", false)
+	account.Users[mockTargetUserId] = &types.User{
+		Id:     mockTargetUserId,
+		Issued: types.UserIssuedAPI,
+		Role:   types.UserRoleOwner,
+	}
+	require.NoError(t, s.SaveAccount(context.Background(), account))
+
+	am := DefaultAccountManager{Store: s}
+
+	_, err = am.deleteRegularUser(context.Background(), mockAccountID, mockUserID, &types.UserInfo{ID: mockTargetUserId})
+	assert.EqualError(t, err, status.NewOwnerDeletePermissionError().Error())
+}
+
+func TestUser_deleteRegularUser_InitiatorOwnerDeletesThemself(t *testing.T) {
+	s, cleanup, err := store.NewTestStoreFromSQL(context.Background(), "", t.TempDir())
+	require.NoError(t, err)
+	t.Cleanup(cleanup)
+
+	account := newAccountWithId(context.Background(), mockAccountID, mockUserID, "", "", "", false)
+	require.NoError(t, s.SaveAccount(context.Background(), account))
+
+	networkMapControllerMock := network_map.NewMockController(gomock.NewController(t))
+	networkMapControllerMock.EXPECT().OnPeersDeleted(gomock.Any(), gomock.Any(), gomock.Any(), gomock.Any()).Return(nil)
+
+	am := DefaultAccountManager{
+		Store:                s,
+		eventStore:           &activity.InMemoryEventStore{},
+		networkMapController: networkMapControllerMock,
+	}
+
+	_, err = am.deleteRegularUser(context.Background(), mockAccountID, mockUserID, &types.UserInfo{ID: mockUserID})
+	require.NoError(t, err)
+
+	_, err = s.GetUserByUserID(context.Background(), store.LockingStrengthNone, mockUserID)
+	assert.Equal(t, status.NewUserNotFoundError(mockUserID), err)
+}
+
 func TestUser_DeleteUser_RegularUsers(t *testing.T) {
 	store, cleanup, err := store.NewTestStoreFromSQL(context.Background(), "", t.TempDir())
 	if err != nil {