Merge remote-tracking branch 'origin/main' into refactor/permissions-manager

management/internals/modules/reverseproxy/domain/domain.go

@@ -1,5 +1,13 @@
 package domain
 
+import "time"
+
+// ValidationTTL is the time available to validate a custom domain registration.
+const ValidationTTL = 48 * time.Hour
+
+// ID identifies a custom domain registration.
+type ID string
+
 type Type string
 
 const (
@@ -8,12 +16,13 @@ const (
 )
 
 type Domain struct {
-	ID            string `gorm:"unique;primaryKey;autoIncrement"`
-	Domain        string `gorm:"unique"` // Domain records must be unique, this avoids domain reuse across accounts.
-	AccountID     string `gorm:"index"`
-	TargetCluster string // The proxy cluster this domain should be validated against
-	Type          Type   `gorm:"-"`
-	Validated     bool
+	ID                  string `gorm:"unique;primaryKey;autoIncrement"`
+	Domain              string `gorm:"unique"` // Domain records must be unique, this avoids domain reuse across accounts.
+	AccountID           string `gorm:"index"`
+	TargetCluster       string // The proxy cluster this domain should be validated against
+	Type                Type   `gorm:"-"`
+	Validated           bool
+	ValidationExpiresAt *time.Time `gorm:"index"`
 	// SupportsCustomPorts is populated at query time for free domains from the
 	// proxy cluster capabilities. Not persisted.
 	SupportsCustomPorts *bool `gorm:"-"`
@@ -36,7 +45,12 @@ func (d *Domain) EventMeta() map[string]any {
 	}
 }
 
+// Copy returns a copy with an independent validation deadline.
 func (d *Domain) Copy() *Domain {
 	dCopy := *d
+	if d.ValidationExpiresAt != nil {
+		expiresAt := *d.ValidationExpiresAt
+		dCopy.ValidationExpiresAt = &expiresAt
+	}
 	return &dCopy
 }

management/internals/modules/reverseproxy/domain/manager/expiration.go

@@ -0,0 +1,73 @@
+package manager
+
+import (
+	"context"
+	"time"
+
+	log "github.com/sirupsen/logrus"
+
+	"github.com/netbirdio/netbird/management/internals/modules/reverseproxy/domain"
+	"github.com/netbirdio/netbird/management/server/activity"
+)
+
+const (
+	validationCleanupInterval = 60 * time.Minute
+	validationCleanupBatch    = 100
+)
+
+// RunValidationCleanup removes expired registrations on startup and hourly until cancellation.
+func (m Manager) RunValidationCleanup(ctx context.Context) {
+	ticker := time.NewTicker(validationCleanupInterval)
+	defer ticker.Stop()
+	for {
+		m.cleanupExpiredDomains(ctx, time.Now().UTC())
+		select {
+		case <-ctx.Done():
+			return
+		case <-ticker.C:
+		}
+	}
+}
+
+func (m Manager) cleanupExpiredDomains(ctx context.Context, now time.Time) {
+	var afterID domain.ID
+	for ctx.Err() == nil {
+		domains, err := m.store.GetExpiredCustomDomains(ctx, now, afterID, validationCleanupBatch)
+		if err != nil {
+			if ctx.Err() == nil {
+				log.WithContext(ctx).WithError(err).Error("list expired custom domain registrations")
+			}
+			return
+		}
+		for _, d := range domains {
+			if ctx.Err() != nil {
+				return
+			}
+			m.deleteExpiredDomain(ctx, d, now)
+			afterID = domain.ID(d.ID)
+		}
+		if len(domains) < validationCleanupBatch {
+			return
+		}
+	}
+}
+
+func (m Manager) deleteExpiredDomain(ctx context.Context, d *domain.Domain, now time.Time) {
+	deleted, err := m.store.DeleteExpiredCustomDomain(ctx, d, now)
+	if err != nil {
+		if ctx.Err() == nil {
+			log.WithContext(ctx).WithFields(log.Fields{"accountID": d.AccountID, "domainID": d.ID}).
+				WithError(err).Warn("could not expire custom domain registration")
+		}
+		return
+	}
+	if !deleted {
+		return
+	}
+	meta := d.EventMeta()
+	if d.ValidationExpiresAt != nil {
+		meta["validation_expires_at"] = d.ValidationExpiresAt.UTC().Format(time.RFC3339)
+	}
+	m.accountManager.StoreEvent(ctx, activity.SystemInitiator, d.ID, d.AccountID,
+		activity.CustomDomainValidationExpired, meta)
+}

management/internals/modules/reverseproxy/domain/manager/expiration_test.go

@@ -0,0 +1,274 @@
+package manager
+
+import (
+	"context"
+	"fmt"
+	"sync"
+	"testing"
+	"testing/synctest"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/management/internals/modules/reverseproxy/domain"
+	rpservice "github.com/netbirdio/netbird/management/internals/modules/reverseproxy/service"
+	"github.com/netbirdio/netbird/management/server/activity"
+	"github.com/netbirdio/netbird/management/server/mock_server"
+	nbstore "github.com/netbirdio/netbird/management/server/store"
+)
+
+func TestValidateDomain_ExpiredRegistration(t *testing.T) {
+	env := setupDomainTest(t)
+	ctx := context.Background()
+	d, err := env.manager.CreateDomain(ctx, accountA, accountAUser, "expired.example.com", testCluster)
+	require.NoError(t, err)
+	expiresAt := time.Now().Add(-time.Second)
+	db := env.store.(*nbstore.SqlStore).GetDB()
+	require.NoError(t, db.Model(&domain.Domain{}).Where("id = ?", d.ID).
+		Update("validation_expires_at", expiresAt).Error)
+	env.resolver.set("validation.expired.example.com", testCluster)
+
+	env.manager.ValidateDomain(ctx, accountA, accountAUser, d.ID)
+
+	stored := storedDomain(t, env.store, accountA, d.Domain)
+	require.NotNil(t, stored)
+	assert.False(t, stored.Validated, "an expired registration must not become usable before cleanup runs")
+}
+
+func TestCreateDomain_ValidationDeadline(t *testing.T) {
+	env := setupClockDomainTest(t)
+	synctest.Test(t, func(t *testing.T) {
+		ctx := context.Background()
+		createdAt := time.Now().UTC()
+		d, err := env.manager.CreateDomain(ctx, accountA, accountAUser, "pending.example.com", testCluster)
+		require.NoError(t, err)
+		require.NotNil(t, d.ValidationExpiresAt)
+		assert.Equal(t, createdAt.Add(48*time.Hour), *d.ValidationExpiresAt, "new registrations get 48 hours")
+
+		time.Sleep(time.Hour)
+		env.manager.ValidateDomain(ctx, accountA, accountAUser, d.ID)
+		stored := storedDomain(t, env.store, accountA, d.Domain)
+		require.NotNil(t, stored)
+		require.NotNil(t, stored.ValidationExpiresAt)
+		assert.WithinDuration(t, *d.ValidationExpiresAt, *stored.ValidationExpiresAt, 0, "failed validation must not extend the deadline")
+	})
+}
+
+func TestCleanupExpiredDomains_Boundaries(t *testing.T) {
+	env := setupDomainTest(t)
+	events := captureDomainEvents(env)
+	ctx := context.Background()
+	now := time.Now().UTC().Truncate(time.Second)
+	tests := []struct {
+		name      string
+		expiresAt time.Time
+		validated bool
+		deleted   bool
+	}{
+		{"expired", now.Add(-time.Second), false, true},
+		{"deadline", now, false, true},
+		{"pending", now.Add(time.Second), false, false},
+		{"validated", now.Add(-time.Hour), true, false},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			d := createExpiringDomain(t, env, tt.name+".example.com", tt.expiresAt)
+			if tt.validated {
+				require.NoError(t, env.store.(*nbstore.SqlStore).GetDB().Model(d).Update("validated", true).Error)
+			}
+			env.manager.cleanupExpiredDomains(ctx, now)
+			stored := storedDomain(t, env.store, accountA, d.Domain)
+			if !tt.deleted {
+				assert.NotNil(t, stored, "pending and validated registrations must survive cleanup")
+				return
+			}
+			assert.Nil(t, stored, "expired unused registrations must be removed")
+			replacement, err := env.manager.CreateDomain(ctx, accountB, accountBUser, d.Domain, testCluster)
+			require.NoError(t, err)
+			assert.NotEqual(t, d.ID, replacement.ID, "the released name must receive a fresh registration")
+			assert.False(t, replacement.Validated, "the new account must validate its own registration")
+		})
+	}
+	got := events.get()
+	require.Len(t, got, 2, "only successful expiration deletions emit events")
+	for _, event := range got {
+		assert.Equal(t, activity.CustomDomainValidationExpired, event.Activity, "use the requested expiration event")
+		assert.Equal(t, activity.SystemInitiator, event.InitiatorID, "cleanup is attributed to the system")
+		assert.Equal(t, accountA, event.AccountID, "expiration belongs to the original account")
+		assert.NotEmpty(t, event.TargetID, "retain the deleted domain ID")
+		assert.NotEmpty(t, event.Meta["domain"], "retain the deleted domain name")
+		assert.NotEmpty(t, event.Meta["validation_expires_at"], "include the validation deadline")
+	}
+}
+
+func TestCleanupExpiredDomains_ContinuesPastProtectedBatch(t *testing.T) {
+	env := setupDomainTest(t)
+	ctx := context.Background()
+	now := time.Now().UTC()
+	for i := range validationCleanupBatch {
+		d := createExpiringDomain(t, env, fmt.Sprintf("protected-%d.example.com", i), now.Add(-time.Hour))
+		require.NoError(t, env.store.CreateService(ctx, &rpservice.Service{
+			ID: fmt.Sprintf("service-%d", i), AccountID: accountA, Domain: "app." + d.Domain,
+		}))
+	}
+	unprotected := createExpiringDomain(t, env, "unused.example.com", now.Add(-time.Hour))
+	env.manager.cleanupExpiredDomains(ctx, now)
+	assert.Nil(t, storedDomain(t, env.store, accountA, unprotected.Domain), "protected registrations must not starve later batches")
+	remaining, err := env.store.ListCustomDomains(ctx, accountA)
+	require.NoError(t, err)
+	assert.Len(t, remaining, validationCleanupBatch, "all registrations with dependent services must survive")
+}
+
+func TestCleanupExpiredDomains_ConcurrentWorkers(t *testing.T) {
+	env := setupDomainTest(t)
+	events := captureDomainEvents(env)
+	now := time.Now().UTC()
+	d := createExpiringDomain(t, env, "concurrent.example.com", now.Add(-time.Hour))
+	var workers sync.WaitGroup
+	for range 2 {
+		workers.Go(func() { env.manager.cleanupExpiredDomains(context.Background(), now) })
+	}
+	workers.Wait()
+	assert.Nil(t, storedDomain(t, env.store, accountA, d.Domain), "one worker must remove the expired registration")
+	assert.Len(t, events.get(), 1, "only the worker that deletes the row may emit the event")
+}
+
+func TestRunValidationCleanup_HourlyAndRestart(t *testing.T) {
+	env := setupClockDomainTest(t)
+	synctest.Test(t, func(t *testing.T) {
+		events := captureDomainEvents(env)
+		now := time.Now().UTC()
+		startup := createExpiringDomain(t, env, "startup.example.com", now.Add(-time.Hour))
+		hourly := createExpiringDomain(t, env, "hourly.example.com", now.Add(time.Minute))
+		ctx, cancel := context.WithCancel(context.Background())
+		done := make(chan struct{})
+		go func() {
+			defer close(done)
+			env.manager.RunValidationCleanup(ctx)
+		}()
+		synctest.Wait()
+		assert.Nil(t, storedDomain(t, env.store, accountA, startup.Domain), "startup must collect overdue registrations")
+		time.Sleep(59 * time.Minute)
+		synctest.Wait()
+		assert.NotNil(t, storedDomain(t, env.store, accountA, hourly.Domain), "cleanup must wait for the 60-minute interval")
+		time.Sleep(time.Minute)
+		synctest.Wait()
+		assert.Nil(t, storedDomain(t, env.store, accountA, hourly.Domain), "the hourly scan must collect expired registrations")
+		cancel()
+		<-done
+
+		offline := createExpiringDomain(t, env, "offline.example.com", time.Now().UTC().Add(time.Minute))
+		time.Sleep(2 * time.Hour)
+		assert.NotNil(t, storedDomain(t, env.store, accountA, offline.Domain), "a stopped worker must not continue deleting")
+		ctx, cancel = context.WithCancel(context.Background())
+		done = make(chan struct{})
+		go func() {
+			defer close(done)
+			env.manager.RunValidationCleanup(ctx)
+		}()
+		synctest.Wait()
+		assert.Nil(t, storedDomain(t, env.store, accountA, offline.Domain), "restart must use the persisted deadline")
+		cancel()
+		<-done
+		assert.Len(t, events.get(), 3, "each deletion should emit an expiration event")
+	})
+}
+
+type blockingDomainResolver struct {
+	started chan struct{}
+	release chan struct{}
+}
+
+func (r blockingDomainResolver) LookupCNAME(context.Context, string) (string, error) {
+	close(r.started)
+	<-r.release
+	return testCluster + ".", nil
+}
+
+func TestValidateDomain_DeadlinePassesDuringLookup(t *testing.T) {
+	for _, cleanup := range []bool{false, true} {
+		t.Run(fmt.Sprintf("cleanup=%t", cleanup), func(t *testing.T) {
+			env := setupClockDomainTest(t)
+			synctest.Test(t, func(t *testing.T) {
+				events := captureDomainEvents(env)
+				ctx := context.Background()
+				d, err := env.manager.CreateDomain(ctx, accountA, accountAUser, "late.example.com", testCluster)
+				require.NoError(t, err)
+				resolver := blockingDomainResolver{started: make(chan struct{}), release: make(chan struct{})}
+				env.manager.validator.Resolver = resolver
+				done := make(chan struct{})
+				go func() {
+					defer close(done)
+					env.manager.ValidateDomain(ctx, accountA, accountAUser, d.ID)
+				}()
+				<-resolver.started
+				time.Sleep(48 * time.Hour)
+				if cleanup {
+					env.manager.cleanupExpiredDomains(ctx, time.Now().UTC())
+					_, err = env.store.CreateCustomDomain(ctx, accountB, d.Domain, testCluster, false)
+					require.NoError(t, err)
+				}
+				close(resolver.release)
+				<-done
+				owner := accountA
+				if cleanup {
+					assert.Nil(t, storedDomain(t, env.store, accountA, d.Domain), "late validation must not restore the old claim")
+					owner = accountB
+				}
+				stored := storedDomain(t, env.store, owner, d.Domain)
+				require.NotNil(t, stored)
+				assert.False(t, stored.Validated, "late validation must not validate either claim")
+				for _, event := range events.get() {
+					assert.NotEqual(t, activity.DomainValidated, event.Activity, "a rejected write must not emit a validation event")
+				}
+			})
+		})
+	}
+}
+
+func setupClockDomainTest(t *testing.T) *domainTestEnv {
+	t.Helper()
+	// Network driver watchers cannot share cancellation channels across synctest bubbles.
+	// Store boundary and concurrency tests still exercise the selected database engine.
+	t.Setenv("NETBIRD_STORE_ENGINE", "sqlite")
+	return setupDomainTest(t)
+}
+
+func createExpiringDomain(t *testing.T, env *domainTestEnv, name string, expiresAt time.Time) *domain.Domain {
+	t.Helper()
+	d, err := env.store.CreateCustomDomain(context.Background(), accountA, name, testCluster, false)
+	require.NoError(t, err)
+	require.NoError(t, env.store.(*nbstore.SqlStore).GetDB().Model(d).Update("validation_expires_at", expiresAt).Error)
+	d.ValidationExpiresAt = &expiresAt
+	return d
+}
+
+type domainEvents struct {
+	mu     sync.Mutex
+	events []*activity.Event
+}
+
+func captureDomainEvents(env *domainTestEnv) *domainEvents {
+	events := &domainEvents{}
+	env.manager.accountManager = &mock_server.MockAccountManager{
+		StoreEventFunc: func(_ context.Context, initiator, target, account string, code activity.ActivityDescriber, meta map[string]any) {
+			if code == activity.DomainAdded {
+				return
+			}
+			events.mu.Lock()
+			defer events.mu.Unlock()
+			events.events = append(events.events, &activity.Event{
+				InitiatorID: initiator, TargetID: target, AccountID: account,
+				Activity: code.(activity.Activity), Meta: meta,
+			})
+		},
+	}
+	return events
+}
+
+func (e *domainEvents) get() []*activity.Event {
+	e.mu.Lock()
+	defer e.mu.Unlock()
+	return append([]*activity.Event(nil), e.events...)
+}

management/internals/modules/reverseproxy/domain/manager/manager.go

@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"net"
 	"strings"
+	"time"
 
 	log "github.com/sirupsen/logrus"
 
@@ -15,6 +16,7 @@ import (
 	"github.com/netbirdio/netbird/management/server/activity"
 	nbstore "github.com/netbirdio/netbird/management/server/store"
 	"github.com/netbirdio/netbird/management/server/types"
+	nbdomain "github.com/netbirdio/netbird/shared/management/domain"
 	"github.com/netbirdio/netbird/shared/management/status"
 )
 
@@ -29,6 +31,8 @@ type store interface {
 	CreateCustomDomain(ctx context.Context, accountID string, domainName string, targetCluster string, validated bool) (*domain.Domain, error)
 	UpdateCustomDomain(ctx context.Context, accountID string, d *domain.Domain) (*domain.Domain, error)
 	DeleteCustomDomain(ctx context.Context, accountID string, domainID string) error
+	GetExpiredCustomDomains(ctx context.Context, now time.Time, afterID domain.ID, limit int) ([]*domain.Domain, error)
+	DeleteExpiredCustomDomain(ctx context.Context, d *domain.Domain, now time.Time) (bool, error)
 }
 
 type proxyManager interface {
@@ -93,12 +97,13 @@ func (m Manager) GetDomains(ctx context.Context, accountID, userID string) ([]*d
 	// Add custom domains.
 	for _, d := range domains {
 		cd := &domain.Domain{
-			ID:            d.ID,
-			Domain:        d.Domain,
-			AccountID:     accountID,
-			TargetCluster: d.TargetCluster,
-			Type:          domain.TypeCustom,
-			Validated:     d.Validated,
+			ID:                  d.ID,
+			Domain:              d.Domain,
+			AccountID:           accountID,
+			TargetCluster:       d.TargetCluster,
+			Type:                domain.TypeCustom,
+			Validated:           d.Validated,
+			ValidationExpiresAt: d.ValidationExpiresAt,
 		}
 		if d.TargetCluster != "" {
 			cd.SupportsCustomPorts = m.proxyManager.ClusterSupportsCustomPorts(ctx, d.TargetCluster)
@@ -113,7 +118,17 @@ func (m Manager) GetDomains(ctx context.Context, accountID, userID string) ([]*d
 	return ret, nil
 }
 
+// CreateDomain registers a normalized custom domain and attempts DNS validation.
 func (m Manager) CreateDomain(ctx context.Context, accountID, userID, domainName, targetCluster string) (*domain.Domain, error) {
+	parsed, err := nbdomain.FromString(strings.TrimSuffix(domainName, "."))
+	if err != nil {
+		return nil, status.Errorf(status.InvalidArgument, "invalid domain: %v", err)
+	}
+	domainName = parsed.PunycodeString()
+	if !nbdomain.IsValidDomainNoWildcard(domainName) {
+		return nil, status.Errorf(status.InvalidArgument, "invalid domain format")
+	}
+
 	// Verify the target cluster is in the available clusters for this account
 	allowList, err := m.getClusterAllowList(ctx, accountID)
 	if err != nil {
@@ -197,6 +212,14 @@ func (m Manager) ValidateDomain(ctx context.Context, accountID, userID, domainID
 		}).WithError(err).Error("get custom domain from store")
 		return
 	}
+	if d.Validated {
+		return
+	}
+	if d.ValidationExpiresAt == nil || !time.Now().Before(*d.ValidationExpiresAt) {
+		log.WithFields(log.Fields{"accountID": accountID, "domainID": domainID}).
+			Debug("custom domain validation window has expired")
+		return
+	}
 
 	// Validate only against the domain's target cluster
 	targetCluster := d.TargetCluster
@@ -217,20 +240,21 @@ func (m Manager) ValidateDomain(ctx context.Context, accountID, userID, domainID
 	}).Info("validating domain against target cluster")
 
 	if m.validator.IsValid(context.Background(), d.Domain, []string{targetCluster}) {
-		log.WithFields(log.Fields{
-			"accountID": accountID,
-			"domainID":  domainID,
-			"domain":    d.Domain,
-		}).Info("domain validated successfully")
 		d.Validated = true
 		if _, err := m.store.UpdateCustomDomain(context.Background(), accountID, d); err != nil {
-			log.WithFields(log.Fields{
+			entry := log.WithFields(log.Fields{
 				"accountID": accountID,
 				"domainID":  domainID,
-				"domain":    d.Domain,
-			}).WithError(err).Error("update custom domain in store")
+			}).WithError(err)
+			if sErr, ok := status.FromError(err); ok && sErr.Type() == status.PreconditionFailed {
+				entry.Debug("custom domain registration is no longer pending validation")
+				return
+			}
+			entry.Error("update custom domain in store")
 			return
 		}
+		log.WithFields(log.Fields{"accountID": accountID, "domainID": domainID}).
+			Info("custom domain validated successfully")
 
 		m.accountManager.StoreEvent(context.Background(), userID, domainID, accountID, activity.DomainValidated, d.EventMeta())
 	} else {

management/internals/modules/reverseproxy/domain/manager/manager_realstore_test.go

@@ -268,11 +268,8 @@ func TestStore_DuplicateDomainRejectedByIndexAsConflict(t *testing.T) {
 	assert.Equal(t, status.AlreadyExists, sErr.Type(), "a lost race is a 409, not a 500")
 }
 
-// Validation runs asynchronously, so it can finish after the domain was
-// deleted and then write a stale row back. gorm's Save falls back to an insert
-// when an update affects no rows, which would resurrect the domain as
-// validated; UpdateCustomDomain avoids that by selecting explicit columns.
-// This pins that behaviour, since dropping the Select would reintroduce it.
+// A validation finishing after deletion must reject the stale write, without
+// restoring the registration or reporting successful validation.
 func TestUpdateCustomDomain_DoesNotResurrectDeletedDomain(t *testing.T) {
 	ctx := context.Background()
 	env := setupDomainTest(t)
@@ -287,11 +284,9 @@ func TestUpdateCustomDomain_DoesNotResurrectDeletedDomain(t *testing.T) {
 	require.Nil(t, storedDomain(t, env.store, accountA, "racy.example.com"), "the domain should be gone")
 
 	// What an in-flight validation would write once its CNAME check succeeded.
-	// The write has to succeed for the assertion below to mean anything: a
-	// rejected write would leave the domain absent for the wrong reason.
 	stale.Validated = true
 	_, err = env.store.UpdateCustomDomain(ctx, accountA, stale)
-	require.NoError(t, err, "the update itself must succeed, so absence is not just a failed write")
+	require.Error(t, err, "a deleted registration must reject a late validation")
 
 	assert.Nil(t, storedDomain(t, env.store, accountA, "racy.example.com"),
 		"a late validation write must not recreate a deleted domain")

management/internals/modules/reverseproxy/domain/manager/manager_test.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"errors"
 	"testing"
+	"time"
 
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
@@ -208,6 +209,14 @@ func (s *stubStore) DeleteCustomDomain(context.Context, string, string) error {
 	panic("not used in allow-list tests")
 }
 
+func (s *stubStore) GetExpiredCustomDomains(context.Context, time.Time, domain.ID, int) ([]*domain.Domain, error) {
+	panic("not used in allow-list tests")
+}
+
+func (s *stubStore) DeleteExpiredCustomDomain(context.Context, *domain.Domain, time.Time) (bool, error) {
+	panic("not used in allow-list tests")
+}
+
 // TestGetClusterAllowList_DedicatedGatewayAddressExcluded pins invariant (B)'s
 // chokepoint: a self-addressed settings pin reserves the account's gateway
 // address, so it is dropped from the allow list — which, because the

management/internals/modules/reverseproxy/domain/manager/normalization_test.go

@@ -0,0 +1,83 @@
+package manager
+
+import (
+	"context"
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/shared/management/status"
+)
+
+func TestCreateDomain_NormalizesName(t *testing.T) {
+	for _, tt := range []struct {
+		name      string
+		input     string
+		canonical string
+	}{
+		{"mixed case", "Apps.Example.COM", "apps.example.com"},
+		{"unicode", "münchen.example.com", "xn--mnchen-3ya.example.com"},
+		{"trailing dot", "apps.example.com.", "apps.example.com"},
+		{"underscore", "My_App.example.com", "my_app.example.com"},
+	} {
+		t.Run(tt.name, func(t *testing.T) {
+			ctx := context.Background()
+			env := setupDomainTest(t)
+			env.resolver.set("validation."+tt.canonical, testCluster)
+
+			created, err := env.manager.CreateDomain(ctx, accountA, accountAUser, tt.input, testCluster)
+			require.NoError(t, err)
+			assert.Equal(t, tt.canonical, created.Domain, "the response must use the normalized name")
+			assert.True(t, created.Validated, "the CNAME lookup must use the normalized name")
+			stored, err := env.store.GetCustomDomain(ctx, accountA, created.ID)
+			require.NoError(t, err)
+			assert.Equal(t, tt.canonical, stored.Domain, "the database must retain the normalized name")
+
+			_, err = env.manager.CreateDomain(ctx, accountB, accountBUser, tt.canonical, testCluster)
+			require.Error(t, err)
+			sErr, ok := status.FromError(err)
+			require.True(t, ok, "an equivalent name must return a typed conflict")
+			assert.Equal(t, status.AlreadyExists, sErr.Type(), "normalization must precede the availability check")
+		})
+	}
+}
+
+func TestCreateDomain_NormalizedNameCanValidateLater(t *testing.T) {
+	ctx := context.Background()
+	env := setupDomainTest(t)
+	created, err := env.manager.CreateDomain(ctx, accountA, accountAUser, "Apps.Example.COM.", testCluster)
+	require.NoError(t, err)
+	require.False(t, created.Validated, "a missing CNAME must leave the normalized registration pending")
+
+	env.resolver.set("validation.apps.example.com", testCluster)
+	env.manager.ValidateDomain(ctx, accountA, accountAUser, created.ID)
+	stored, err := env.store.GetCustomDomain(ctx, accountA, created.ID)
+	require.NoError(t, err)
+	assert.Equal(t, "apps.example.com", stored.Domain, "retrying validation must retain the normalized name")
+	assert.True(t, stored.Validated, "later validation must look up the normalized name")
+}
+
+func TestCreateDomain_RejectsInvalidName(t *testing.T) {
+	ctx := context.Background()
+	env := setupDomainTest(t)
+	for _, name := range []string{
+		"", ".", "app..example.com", "app.example.com..", "-app.example.com",
+		"app%.example.com", "app!.example.com", "*.example.com", "app example.com",
+		"https://example.com", strings.Repeat("a", 64) + ".example.com",
+	} {
+		t.Run(name, func(t *testing.T) {
+			// A matching DNS response must not make a malformed name acceptable.
+			env.resolver.set("validation."+name, testCluster)
+			_, err := env.manager.CreateDomain(ctx, accountA, accountAUser, name, testCluster)
+			require.Error(t, err)
+			sErr, ok := status.FromError(err)
+			require.True(t, ok, "invalid names must return a typed client error")
+			assert.Equal(t, status.InvalidArgument, sErr.Type(), "malformed names must be rejected before storage")
+		})
+	}
+	stored, err := env.store.ListCustomDomains(ctx, accountA)
+	require.NoError(t, err)
+	assert.Empty(t, stored, "invalid registration attempts must not reserve any names")
+}