fix(identity): 完善统一认证配对恢复与安全退役

修复 credentials_saved 状态无法恢复、配对与激活并发冲突,以及 SSF 和身份 Secret 生命周期不完整的问题。新增持久化协调器、取消与清理状态机、事务级并发门禁、受控 SSF 凭据交接、禁用后的延迟 Secret 清理,并对生产环境统一认证及 Discovery 端点强制 HTTPS。

验证:go test ./...;go test -race ./internal/auth ./internal/identity ./internal/identityruntime ./internal/securityevents ./internal/httpapi ./internal/store -count=1;go vet ./...;真实 PostgreSQL 并发及清理成功/冲突回滚测试;pnpm openapi。
This commit is contained in:
2026-07-17 18:31:12 +08:00
parent cdfca61304
commit a312ad880d
55 changed files with 9225 additions and 419 deletions
+231 -21
View File
@@ -3,8 +3,10 @@ package identityruntime
import (
"context"
"errors"
"net/http"
"net/url"
"slices"
"strings"
"sync"
"time"
@@ -24,10 +26,23 @@ type RuntimeBuilderConfig struct {
}
type preparedSecurityRuntime struct {
manager *securityevents.ConnectionManager
cancel context.CancelFunc
manager *securityevents.ConnectionManager
cancel context.CancelFunc
receiverReady bool
blockedCategory string
}
type safeRuntimeError struct {
category string
cause error
}
func (err safeRuntimeError) Error() string {
return "identity runtime operation failed: " + err.category
}
func (err safeRuntimeError) Unwrap() error { return err.cause }
func (err safeRuntimeError) SafeErrorCategory() string { return err.category }
type RuntimeBuilder struct {
ctx context.Context
store *store.Store
@@ -54,7 +69,36 @@ func NewRuntimeBuilder(ctx context.Context, data *store.Store, secrets PairingSe
return &RuntimeBuilder{ctx: ctx, store: data, secrets: secrets, config: config, metrics: metrics, prepared: map[string]preparedSecurityRuntime{}}
}
func (builder *RuntimeBuilder) PreparedSecurityEventReceiver() http.Handler {
builder.mutex.Lock()
defer builder.mutex.Unlock()
if len(builder.prepared) != 1 {
return nil
}
for _, prepared := range builder.prepared {
if prepared.receiverReady {
return prepared.manager
}
}
return nil
}
func (builder *RuntimeBuilder) PreparedSecurityEventManager() *securityevents.ConnectionManager {
builder.mutex.Lock()
defer builder.mutex.Unlock()
if len(builder.prepared) != 1 {
return nil
}
for _, prepared := range builder.prepared {
return prepared.manager
}
return nil
}
func (builder *RuntimeBuilder) Build(ctx context.Context, revision identity.Revision) (*Runtime, error) {
if err := identity.ValidateRevisionURLs(revision, builder.config.AppEnv); err != nil {
return nil, err
}
if builder.store == nil || builder.secrets == nil || revision.Issuer == "" || revision.TenantID == "" ||
revision.Audience == "" || revision.RolePrefix == "" {
return nil, errors.New("identity runtime configuration is incomplete")
@@ -69,23 +113,28 @@ func (builder *RuntimeBuilder) Build(ctx context.Context, revision identity.Revi
var securityManager *securityevents.ConnectionManager
if revision.SessionRevocation {
prepared := builder.takePreparedSecurityRuntime(revision.ID)
prepared := builder.preparedSecurityRuntime(revision.ID)
securityManager = prepared.manager
if prepared.cancel != nil {
runtime.close = func() {
if securityManager != nil {
if err := securityManager.ValidateConfiguredConnectionBinding(ctx); err != nil {
cancel()
prepared.cancel()
return nil, err
}
if !prepared.receiverReady {
cancel()
return nil, safeRuntimeError{category: "security_event_not_ready", cause: store.ErrSecurityEventConnectionConflict}
}
}
if securityManager == nil {
var err error
securityManager, err = builder.newSecurityEventManager(runtimeCtx, revision)
securityManager, err = builder.newSecurityEventManager(runtimeCtx, revision, true)
if err != nil {
cancel()
return nil, err
}
}
runtime.SecurityEvents = securityManager
runtime.securityEventDisconnector = securityManager
}
var evaluator func(context.Context, auth.OIDCSecurityEventIdentity) (auth.OIDCSecurityEventEvaluation, error)
@@ -100,6 +149,7 @@ func (builder *RuntimeBuilder) Build(ctx context.Context, revision identity.Revi
return revision.MachineClientID, secret, err
}
verifier, err := auth.NewOIDCVerifier(auth.OIDCConfig{
AppEnv: builder.config.AppEnv,
Issuer: revision.Issuer, Audience: revision.Audience, TenantID: revision.TenantID,
RolePrefix: revision.RolePrefix, RequiredScopes: append([]string(nil), revision.Scopes...),
JWKSCacheTTL: builder.config.JWKSCacheTTL, IntrospectionEnabled: revision.TokenIntrospection,
@@ -134,6 +184,7 @@ func (builder *RuntimeBuilder) Build(ctx context.Context, revision identity.Revi
return nil, err
}
client, err := auth.NewOIDCPublicClient(auth.OIDCPublicClientConfig{
AppEnv: builder.config.AppEnv,
Issuer: revision.Issuer, ClientID: revision.BrowserClientID,
RedirectURI: revision.PublicBaseURL + "/api/v1/auth/oidc/callback",
PostLogoutRedirectURI: revision.WebBaseURL + "/", Scopes: append([]string{"openid", "profile"}, revision.Scopes...),
@@ -161,46 +212,205 @@ func (builder *RuntimeBuilder) Build(ctx context.Context, revision identity.Revi
}
func (builder *RuntimeBuilder) PrepareSecurityEvents(ctx context.Context, revision identity.Revision, managementSecret []byte) error {
if !revision.SessionRevocation || revision.SecurityEventIssuer == "" || revision.MachineClientID == "" {
if err := identity.ValidateRevisionURLs(revision, builder.config.AppEnv); err != nil {
return safeRuntimeError{category: "configuration_invalid", cause: err}
}
if !revision.SessionRevocation || revision.SecurityEventIssuer == "" || revision.SecurityEventAudience == "" || revision.MachineClientID == "" {
return errors.New("security event configuration is incomplete")
}
if prepared := builder.preparedSecurityRuntime(revision.ID); prepared.manager != nil && prepared.blockedCategory != "" {
view, err := prepared.manager.Get(ctx)
if err == nil {
category := prepared.blockedCategory
if view.LifecycleStatus == "retiring" || view.LifecycleStatus == "disconnect_pending" {
category = "retirement_pending"
}
return safeRuntimeError{category: category, cause: store.ErrSecurityEventConnectionConflict}
}
if !errors.Is(err, store.ErrSecurityEventConnectionNotFound) {
return safeSecurityEventRuntimeError(err)
}
builder.removePreparedSecurityRuntime(revision.ID, prepared)
}
runtimeCtx, cancel := context.WithCancel(builder.ctx)
manager, err := builder.newSecurityEventManager(runtimeCtx, revision)
manager, err := builder.newSecurityEventManager(runtimeCtx, revision, false)
if err != nil {
cancel()
return err
return safeRuntimeError{category: "configuration_invalid", cause: err}
}
secretCopy := append([]byte(nil), managementSecret...)
_, err = manager.Connect(ctx, revision.SecurityEventIssuer, revision.MachineClientID, secretCopy, "identity-pairing-ssf-"+revision.ID)
clear(secretCopy)
if err != nil {
// A conflicting persisted connection is itself the resource an
// administrator must safely retire. Keep this manager reachable by the
// recovery API even when there is no Active identity Runtime yet.
if errors.Is(err, store.ErrSecurityEventConnectionConflict) {
builder.replacePreparedSecurityRuntime(revision.ID, preparedSecurityRuntime{
manager: manager, cancel: cancel, blockedCategory: safeSecurityEventCategory(err, "connection_conflict"),
})
return safeSecurityEventRuntimeError(err)
}
cancel()
return err
return safeSecurityEventRuntimeError(err)
}
if err := manager.ValidateConfiguredConnectionBinding(ctx); err != nil {
builder.replacePreparedSecurityRuntime(revision.ID, preparedSecurityRuntime{
manager: manager, cancel: cancel, blockedCategory: safeSecurityEventCategory(err, "connection_binding_mismatch"),
})
return safeSecurityEventRuntimeError(err)
}
builder.replacePreparedSecurityRuntime(revision.ID, preparedSecurityRuntime{manager: manager, cancel: cancel, receiverReady: true})
return nil
}
func (builder *RuntimeBuilder) removePreparedSecurityRuntime(revisionID string, prepared preparedSecurityRuntime) {
builder.mutex.Lock()
previous := builder.prepared[revision.ID]
builder.prepared[revision.ID] = preparedSecurityRuntime{manager: manager, cancel: cancel}
current := builder.prepared[revisionID]
if current.manager == prepared.manager {
delete(builder.prepared, revisionID)
}
builder.mutex.Unlock()
if prepared.cancel != nil {
prepared.cancel()
}
}
func (builder *RuntimeBuilder) replacePreparedSecurityRuntime(revisionID string, replacement preparedSecurityRuntime) {
builder.mutex.Lock()
previous := builder.prepared[revisionID]
builder.prepared[revisionID] = replacement
builder.mutex.Unlock()
if previous.cancel != nil {
previous.cancel()
}
}
func (builder *RuntimeBuilder) CleanupPreparedSecurityEvents(ctx context.Context, revision identity.Revision) error {
builder.mutex.Lock()
prepared, exists := builder.prepared[revision.ID]
builder.mutex.Unlock()
if !exists {
runtimeCtx, cancel := context.WithCancel(builder.ctx)
manager, err := builder.newSecurityEventManager(runtimeCtx, revision, false)
if err != nil {
cancel()
return safeRuntimeError{category: "configuration_invalid", cause: err}
}
prepared = preparedSecurityRuntime{manager: manager, cancel: cancel}
}
err := prepared.manager.DiscardPreparedConnection(ctx, "identity-pairing-ssf-"+revision.ID)
if err != nil {
builder.mutex.Lock()
if _, alreadyStored := builder.prepared[revision.ID]; !alreadyStored {
builder.prepared[revision.ID] = prepared
}
builder.mutex.Unlock()
return safeSecurityEventRuntimeError(err)
}
builder.mutex.Lock()
current := builder.prepared[revision.ID]
if current.manager == prepared.manager {
delete(builder.prepared, revision.ID)
}
builder.mutex.Unlock()
if prepared.cancel != nil {
prepared.cancel()
}
return nil
}
func (builder *RuntimeBuilder) newSecurityEventManager(ctx context.Context, revision identity.Revision) (*securityevents.ConnectionManager, error) {
return securityevents.NewConnectionManager(ctx, builder.store, builder.secrets, securityevents.ConnectionManagerConfig{
AppEnv: builder.config.AppEnv, OIDCEnabled: true, OIDCIssuer: revision.Issuer, OIDCTenantID: revision.TenantID,
ManagementClientID: revision.MachineClientID, PublicBaseURL: revision.PublicBaseURL,
HeartbeatInterval: builder.config.HeartbeatInterval, StaleAfter: builder.config.StaleAfter, ClockSkew: builder.config.ClockSkew,
}, builder.metrics)
func (builder *RuntimeBuilder) RetireConflictingSecurityEvents(ctx context.Context, revision identity.Revision) error {
prepared := builder.preparedSecurityRuntime(revision.ID)
if prepared.manager == nil {
runtimeCtx, cancel := context.WithCancel(builder.ctx)
manager, err := builder.newSecurityEventManager(runtimeCtx, revision, false)
if err != nil {
cancel()
return safeRuntimeError{category: "configuration_invalid", cause: err}
}
prepared = preparedSecurityRuntime{manager: manager, cancel: cancel}
builder.replacePreparedSecurityRuntime(revision.ID, prepared)
}
if err := prepared.manager.RetireConflictingConnection(ctx, "identity-pairing-ssf-"+revision.ID); err != nil {
return safeSecurityEventRuntimeError(err)
}
return nil
}
func (builder *RuntimeBuilder) takePreparedSecurityRuntime(revisionID string) preparedSecurityRuntime {
// RecoverSecurityEventDisconnector reconstructs only the SSF management
// surface needed to disable a fail-closed Active Revision. It deliberately
// avoids OIDC discovery, JIT, BFF Session and tenant Runtime construction.
func (builder *RuntimeBuilder) RecoverSecurityEventDisconnector(_ context.Context, revision identity.Revision) (SecurityEventDisconnector, error) {
prepared := builder.preparedSecurityRuntime(revision.ID)
if prepared.manager != nil {
if err := prepared.manager.ValidateConfiguredConnectionBinding(builder.ctx); err != nil {
return nil, err
}
return prepared.manager, nil
}
manager, err := builder.newSecurityEventManager(builder.ctx, revision, true)
if errors.Is(err, store.ErrSecurityEventConnectionNotFound) {
return nil, nil
}
if err != nil {
return nil, err
}
return manager, nil
}
// AdoptPreparedSecurityEvents transfers the prepared manager lifetime to an
// activated Runtime. Validation only peeks at prepared state, so Verification
// callbacks remain available between validation and activation.
func (builder *RuntimeBuilder) AdoptPreparedSecurityEvents(revisionID string) context.CancelFunc {
builder.mutex.Lock()
defer builder.mutex.Unlock()
prepared := builder.prepared[revisionID]
delete(builder.prepared, revisionID)
return prepared
return prepared.cancel
}
func safeSecurityEventRuntimeError(err error) error {
var categorized interface{ SafeErrorCategory() string }
if errors.As(err, &categorized) {
return err
}
if errors.Is(err, store.ErrSecurityEventConnectionConflict) {
return safeRuntimeError{category: "connection_conflict", cause: err}
}
return safeRuntimeError{category: "preparation_failed", cause: err}
}
func safeSecurityEventCategory(err error, fallback string) string {
var categorized interface{ SafeErrorCategory() string }
if errors.As(err, &categorized) && categorized.SafeErrorCategory() != "" {
return categorized.SafeErrorCategory()
}
return fallback
}
func (builder *RuntimeBuilder) newSecurityEventManager(ctx context.Context, revision identity.Revision, strictRevisionBinding bool) (*securityevents.ConnectionManager, error) {
return securityevents.NewConnectionManager(ctx, builder.store, builder.secrets, builder.securityEventManagerConfig(revision, strictRevisionBinding), builder.metrics)
}
func (builder *RuntimeBuilder) securityEventManagerConfig(revision identity.Revision, strictRevisionBinding bool) securityevents.ConnectionManagerConfig {
return securityevents.ConnectionManagerConfig{
AppEnv: builder.config.AppEnv, OIDCEnabled: true, OIDCIssuer: revision.Issuer, OIDCTenantID: revision.TenantID,
ManagementClientID: revision.MachineClientID, PublicBaseURL: revision.PublicBaseURL,
ExpectedTransmitterIssuer: strings.TrimRight(strings.TrimSpace(revision.SecurityEventIssuer), "/"),
ExpectedAudience: revision.SecurityEventAudience,
ExpectedOwnerKey: "identity-pairing-ssf-" + revision.ID,
StrictRevisionBinding: strictRevisionBinding,
HeartbeatInterval: builder.config.HeartbeatInterval,
StaleAfter: builder.config.StaleAfter,
ClockSkew: builder.config.ClockSkew,
}
}
func (builder *RuntimeBuilder) preparedSecurityRuntime(revisionID string) preparedSecurityRuntime {
builder.mutex.Lock()
defer builder.mutex.Unlock()
return builder.prepared[revisionID]
}
func secureCookieFor(baseURL string) bool {