[management] Expire unvalidated custom domain registrations (#7497)

Prevent unvalidated registrations from reserving domain names indefinitely.

Give pending registrations a 48-hour validation window and clean up expired entries at startup and every 60 minutes. Emit CustomDomainValidationExpired for each deletion and preserve registrations referenced by services.

Reject validation after expiry and prevent concurrent validation from recreating deleted registrations. Normalize domain names with the shared parser before registration.

Migrate existing pending registrations to receive a fresh 48-hour validation window.

management/server/store/sql_store.go

@@ -5712,6 +5712,10 @@ func (s *SqlStore) CreateCustomDomain(ctx context.Context, accountID string, dom
 		Type:          domain.TypeCustom,
 		Validated:     validated,
 	}
+	if !validated {
+		expiresAt := time.Now().UTC().Add(domain.ValidationTTL)
+		newDomain.ValidationExpiresAt = &expiresAt
+	}
 	result := s.db.Create(newDomain)
 	if result.Error != nil {
 		// The unique index is the last guard when two requests clear the
@@ -5733,12 +5737,21 @@ func (s *SqlStore) CreateCustomDomain(ctx context.Context, accountID string, dom
 	return newDomain, nil
 }
 
+// UpdateCustomDomain completes validation only while the original registration is pending.
 func (s *SqlStore) UpdateCustomDomain(ctx context.Context, accountID string, d *domain.Domain) (*domain.Domain, error) {
-	d.AccountID = accountID
-	result := s.db.Select("*").Save(d)
+	if !d.Validated {
+		return nil, status.Errorf(status.InvalidArgument, "custom domain update must complete validation")
+	}
+	result := s.db.WithContext(ctx).Model(&domain.Domain{}).
+		Where(accountAndIDQueryCondition, accountID, d.ID).
+		Where("domain = ? AND target_cluster = ?", d.Domain, d.TargetCluster).
+		Where("validated = ? AND validation_expires_at > ?", false, time.Now().UTC()).
+		Update("validated", true)
 	if result.Error != nil {
-		log.WithContext(ctx).Errorf("failed to update reverse proxy custom domain to store: %v", result.Error)
-		return nil, status.Errorf(status.Internal, "failed to update reverse proxy custom domain to store")
+		return nil, fmt.Errorf("validate custom domain in store: %w", result.Error)
+	}
+	if result.RowsAffected == 0 {
+		return nil, status.Errorf(status.PreconditionFailed, "custom domain registration is no longer pending validation")
 	}
 
 	return d, nil

management/server/store/sql_store_domain_expiration.go

@@ -0,0 +1,60 @@
+package store
+
+import (
+	"context"
+	"fmt"
+	"strings"
+	"time"
+
+	"gorm.io/gorm"
+
+	"github.com/netbirdio/netbird/management/internals/modules/reverseproxy/domain"
+	rpservice "github.com/netbirdio/netbird/management/internals/modules/reverseproxy/service"
+	"github.com/netbirdio/netbird/shared/management/status"
+)
+
+// GetExpiredCustomDomains lists pending registrations in stable batches across accounts.
+func (s *SqlStore) GetExpiredCustomDomains(ctx context.Context, now time.Time, afterID domain.ID, limit int) ([]*domain.Domain, error) {
+	var domains []*domain.Domain
+	result := s.db.WithContext(ctx).
+		Where("validated = ? AND validation_expires_at <= ? AND id > ?", false, now, string(afterID)).
+		Order("id").Limit(limit).Find(&domains)
+	if result.Error != nil {
+		return nil, fmt.Errorf("list expired custom domains: %w", result.Error)
+	}
+	return domains, nil
+}
+
+// DeleteExpiredCustomDomain deletes an expired registration only if no service uses its namespace.
+func (s *SqlStore) DeleteExpiredCustomDomain(ctx context.Context, d *domain.Domain, now time.Time) (bool, error) {
+	db := s.db.WithContext(ctx)
+	services := customDomainServices(db, d)
+	result := db.Where(accountAndIDQueryCondition, d.AccountID, d.ID).
+		Where("domain = ? AND validated = ? AND validation_expires_at <= ?", d.Domain, false, now).
+		Where("NOT EXISTS (?)", services.Select("1")).Delete(&domain.Domain{})
+	if result.Error != nil {
+		return false, fmt.Errorf("delete expired custom domain: %w", result.Error)
+	}
+	if result.RowsAffected > 0 {
+		return true, nil
+	}
+	var count int64
+	if err := customDomainServices(db, d).Count(&count).Error; err != nil {
+		return false, fmt.Errorf("check expired custom domain services: %w", err)
+	}
+	if count > 0 {
+		return false, status.Errorf(status.PreconditionFailed, "expired custom domain still has dependent services")
+	}
+	return false, nil
+}
+
+func customDomainServices(db *gorm.DB, d *domain.Domain) *gorm.DB {
+	name := strings.ToLower(strings.TrimSuffix(d.Domain, "."))
+	// Shared domain validation permits underscores, and older rows may contain
+	// other LIKE metacharacters.
+	escaped := strings.NewReplacer("!", "!!", "%", "!%", "_", "!_").Replace(name)
+	return db.Model(&rpservice.Service{}).Where(
+		"LOWER(domain) IN ? OR LOWER(domain) LIKE ? ESCAPE '!' OR LOWER(domain) LIKE ? ESCAPE '!'",
+		[]string{name, name + "."}, "%."+escaped, "%."+escaped+".",
+	)
+}

management/server/store/sql_store_domain_expiration_test.go

@@ -0,0 +1,78 @@
+package store
+
+import (
+	"context"
+	"testing"
+	"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"
+)
+
+func TestDeleteExpiredCustomDomain_ServiceDependencies(t *testing.T) {
+	runTestForAllEngines(t, "", func(t *testing.T, store Store) {
+		ctx := context.Background()
+		now := time.Now().UTC()
+		db := store.(*SqlStore).db
+		require.NoError(t, store.SaveAccount(ctx, newAccountWithId(ctx, "owner", "admin", "")))
+		for _, tt := range []struct {
+			name        string
+			domainName  string
+			serviceHost string
+			protected   bool
+		}{
+			{"exact", "example.com", "example.com", true},
+			{"subdomain", "example.com", "deep.app.example.com", true},
+			{"case", "example.com", "APP.EXAMPLE.COM.", true},
+			{"suffix-boundary", "example.com", "notexample.com", false},
+			{"literal underscore", "a_b.example.com", "app.a_b.example.com", true},
+			{"underscore wildcard", "a_b.example.com", "app.axb.example.com", false},
+			{"legacy percent wildcard", "a%b.example.com", "app.axxb.example.com", false},
+			{"legacy escape character", "a!b.example.com", "app.ab.example.com", false},
+		} {
+			t.Run(tt.name, func(t *testing.T) {
+				d, err := store.CreateCustomDomain(ctx, "owner", tt.domainName, "cluster", false)
+				require.NoError(t, err)
+				require.NoError(t, db.Model(d).Update("validation_expires_at", now.Add(-time.Hour)).Error)
+				svc := &rpservice.Service{ID: "legacy", AccountID: "owner", Domain: tt.serviceHost}
+				require.NoError(t, store.CreateService(ctx, svc))
+				deleted, err := store.DeleteExpiredCustomDomain(ctx, d, now)
+				if tt.protected {
+					require.Error(t, err)
+					assert.False(t, deleted, "service namespaces must remain reserved")
+				} else {
+					require.NoError(t, err)
+					assert.True(t, deleted, "a hostname outside the namespace must not prevent cleanup")
+				}
+				require.NoError(t, db.Delete(svc).Error)
+				require.NoError(t, db.Delete(d).Error)
+			})
+		}
+	})
+}
+
+func TestDeleteExpiredCustomDomain_RechecksValidation(t *testing.T) {
+	runTestForAllEngines(t, "", func(t *testing.T, store Store) {
+		ctx := context.Background()
+		require.NoError(t, store.SaveAccount(ctx, newAccountWithId(ctx, "owner", "admin", "")))
+		d, err := store.CreateCustomDomain(ctx, "owner", "validated.example.com", "cluster", false)
+		require.NoError(t, err)
+		d, err = store.GetCustomDomain(ctx, "owner", d.ID)
+		require.NoError(t, err)
+		stale := d.Copy()
+		d.Validated = true
+		_, err = store.UpdateCustomDomain(ctx, "owner", d)
+		require.NoError(t, err)
+		deleted, err := store.DeleteExpiredCustomDomain(ctx, stale, time.Now().Add(domain.ValidationTTL))
+		require.NoError(t, err)
+		assert.False(t, deleted, "a stale cleanup candidate must not delete a validated registration")
+		stored, err := store.GetCustomDomain(ctx, "owner", d.ID)
+		require.NoError(t, err)
+		assert.True(t, stored.Validated, "the validated registration must remain usable")
+		require.NotNil(t, stored.ValidationExpiresAt)
+		assert.Equal(t, stale.ValidationExpiresAt, stored.ValidationExpiresAt, "validation must preserve the original deadline")
+	})
+}

management/server/store/sql_store_get_account_test.go

@@ -64,7 +64,7 @@ func assertGetAccountLoadsCustomDomains(t *testing.T, store Store) {
 
 	_, err := store.CreateCustomDomain(ctx, accountID, "example.com", "eu.proxy.netbird.io", true)
 	require.NoError(t, err, "creating the first custom domain must succeed")
-	_, err = store.CreateCustomDomain(ctx, accountID, "apps.acme.io", "us.proxy.netbird.io", false)
+	pending, err := store.CreateCustomDomain(ctx, accountID, "apps.acme.io", "us.proxy.netbird.io", false)
 	require.NoError(t, err, "creating the second custom domain must succeed")
 
 	account, err := store.GetAccount(ctx, accountID)
@@ -75,6 +75,10 @@ func assertGetAccountLoadsCustomDomains(t *testing.T, store Store) {
 	for _, d := range account.Domains {
 		require.NotNil(t, d)
 		byDomain[d.Domain] = d.TargetCluster
+		if d.ID == pending.ID {
+			require.NotNil(t, d.ValidationExpiresAt)
+			assert.WithinDuration(t, *pending.ValidationExpiresAt, *d.ValidationExpiresAt, time.Millisecond, "both account loaders must preserve the validation deadline")
+		}
 	}
 	assert.Equal(t, "eu.proxy.netbird.io", byDomain["example.com"], "custom domain must carry its target cluster")
 	assert.Equal(t, "us.proxy.netbird.io", byDomain["apps.acme.io"], "custom domain must carry its target cluster")

management/server/store/store.go

@@ -305,6 +305,8 @@ type Store interface {
 	GetCustomDomainByName(ctx context.Context, domainName string) (*domain.Domain, error)
 	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)
+	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)
 	DeleteCustomDomain(ctx context.Context, accountID string, domainID string) error
 
 	CreateAccessLog(ctx context.Context, log *accesslogs.AccessLogEntry) error
@@ -642,6 +644,9 @@ func migratePostAuto(ctx context.Context, db *gorm.DB) error {
 
 func getMigrationsPostAuto(ctx context.Context) []migrationFunc {
 	return []migrationFunc{
+		func(db *gorm.DB) error {
+			return migration.MigrateCustomDomainValidationExpiry(ctx, db)
+		},
 		func(db *gorm.DB) error {
 			return migration.CreateIndexIfNotExists[nbpeer.Peer](ctx, db, "idx_account_ip", "account_id", "ip")
 		},

management/server/store/store_mock.go

@@ -555,6 +555,21 @@ func (mr *MockStoreMockRecorder) DeleteDNSRecord(ctx, accountID, zoneID, recordI
 	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "DeleteDNSRecord", reflect.TypeOf((*MockStore)(nil).DeleteDNSRecord), ctx, accountID, zoneID, recordID)
 }
 
+// DeleteExpiredCustomDomain mocks base method.
+func (m *MockStore) DeleteExpiredCustomDomain(ctx context.Context, d *domain.Domain, now time.Time) (bool, error) {
+	m.ctrl.T.Helper()
+	ret := m.ctrl.Call(m, "DeleteExpiredCustomDomain", ctx, d, now)
+	ret0, _ := ret[0].(bool)
+	ret1, _ := ret[1].(error)
+	return ret0, ret1
+}
+
+// DeleteExpiredCustomDomain indicates an expected call of DeleteExpiredCustomDomain.
+func (mr *MockStoreMockRecorder) DeleteExpiredCustomDomain(ctx, d, now any) *gomock.Call {
+	mr.mock.ctrl.T.Helper()
+	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "DeleteExpiredCustomDomain", reflect.TypeOf((*MockStore)(nil).DeleteExpiredCustomDomain), ctx, d, now)
+}
+
 // DeleteGroup mocks base method.
 func (m *MockStore) DeleteGroup(ctx context.Context, accountID, groupID string) error {
 	m.ctrl.T.Helper()
@@ -2002,6 +2017,21 @@ func (mr *MockStoreMockRecorder) GetEmbeddedProxyPeerIDsByCluster(ctx, accountID
 	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "GetEmbeddedProxyPeerIDsByCluster", reflect.TypeOf((*MockStore)(nil).GetEmbeddedProxyPeerIDsByCluster), ctx, accountID)
 }
 
+// GetExpiredCustomDomains mocks base method.
+func (m *MockStore) GetExpiredCustomDomains(ctx context.Context, now time.Time, afterID domain.ID, limit int) ([]*domain.Domain, error) {
+	m.ctrl.T.Helper()
+	ret := m.ctrl.Call(m, "GetExpiredCustomDomains", ctx, now, afterID, limit)
+	ret0, _ := ret[0].([]*domain.Domain)
+	ret1, _ := ret[1].(error)
+	return ret0, ret1
+}
+
+// GetExpiredCustomDomains indicates an expected call of GetExpiredCustomDomains.
+func (mr *MockStoreMockRecorder) GetExpiredCustomDomains(ctx, now, afterID, limit any) *gomock.Call {
+	mr.mock.ctrl.T.Helper()
+	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "GetExpiredCustomDomains", reflect.TypeOf((*MockStore)(nil).GetExpiredCustomDomains), ctx, now, afterID, limit)
+}
+
 // GetExpiredEphemeralServices mocks base method.
 func (m *MockStore) GetExpiredEphemeralServices(ctx context.Context, ttl time.Duration, limit int) ([]*service.Service, error) {
 	m.ctrl.T.Helper()