story/internal/onboarding/integration_test.go

package onboarding

import (
	"bytes"
	"context"
	"crypto/sha256"
	"encoding/base64"
	"encoding/json"
	"errors"
	"net/http"
	"net/http/httptest"
	"os"
	"strconv"
	"strings"
	"sync"
	"sync/atomic"
	"testing"
	"time"

	"data-terminal-cloud/internal/hardwareprovision"
	"data-terminal-cloud/internal/identity"
	"data-terminal-cloud/internal/outbox"
	"data-terminal-cloud/internal/saas"
	"github.com/google/uuid"
	"github.com/jackc/pgx/v5/pgxpool"
	"github.com/redis/go-redis/v9"
)

type testRegistrar struct {
	calls   atomic.Int32
	err     error
	started chan struct{}
	release chan struct{}
}

func (r *testRegistrar) Register(ctx context.Context, in hardwareprovision.RegisterInput) (hardwareprovision.RegisterResult, error) {
	r.calls.Add(1)
	if r.started != nil {
		close(r.started)
		select {
		case <-r.release:
		case <-ctx.Done():
			return hardwareprovision.RegisterResult{}, ctx.Err()
		}
	}
	return hardwareprovision.RegisterResult{DeviceSecret: "fixture-device-secret-at-least-32-bytes", RTCAppID: "test-app"}, r.err
}

type fixture struct {
	t                                *testing.T
	s                                *Service
	ctx                              context.Context
	tenant, tenantPublic, operatorID string
	p, user                          identity.Principal
	d                                Device
	web                              WebSession
	setup, elder, claim              string
	confirm                          ConfirmationInput
	registrar                        *testRegistrar
}

func newFixture(t *testing.T) *fixture {
	t.Helper()
	dsn := os.Getenv("TEST_DATABASE_URL")
	if dsn == "" {
		t.Skip("TEST_DATABASE_URL required")
	}
	ctx := context.Background()
	pool, err := pgxpool.New(ctx, dsn)
	must(t, err)
	t.Cleanup(pool.Close)
	addr := os.Getenv("TEST_REDIS_ADDR")
	if addr == "" {
		addr = "127.0.0.1:56379"
	}
	cache := redis.NewClient(&redis.Options{Addr: addr})
	must(t, cache.Ping(ctx).Err())
	t.Cleanup(func() { _ = cache.Close() })
	f := &fixture{t: t, ctx: ctx, tenant: uuid.NewString(), tenantPublic: uuid.NewString(), operatorID: uuid.NewString(), p: identity.Principal{Issuer: "urn:onboarding:test", Subject: uuid.NewString()}, user: identity.Principal{Issuer: "urn:onboarding:test", Subject: uuid.NewString(), DisplayName: "测试用户1234"}, registrar: &testRegistrar{}}
	f.s, err = NewService(pool, cache, Options{Enabled: true, MasterKey: strings.Repeat("15", 32), ProviderMasterKey: strings.Repeat("25", 32), EntryURL: "https://platform.example/onboarding/entry", SelfSignupTenantIDs: []string{f.tenantPublic}, ActivationEnabled: true, Registrar: f.registrar, InstanceID: "test-instance", ProductKey: "test-product", ProductSecret: "fixture-product-secret-at-least-32-bytes", BotID: "test-bot", Transport: "high_quality", MinimumSDKVersion: "1.0.0"})
	must(t, err)
	_, err = pool.Exec(ctx, `INSERT INTO tenant(id,public_id,name) VALUES($1::uuid,$2::uuid,'onboarding integration')`, f.tenant, f.tenantPublic)
	f.s.options.LegacyDeviceRegistration = true // Existing tests exercise old-device migration contracts.
	must(t, err)
	_, err = pool.Exec(ctx, `INSERT INTO user_account(id,tenant_id,oidc_issuer,oidc_subject,display_name,platform_role,account_kind) VALUES($1::uuid,$2::uuid,$3,$4,'生产管理员','ops_admin','operator')`, f.operatorID, f.tenant, f.p.Issuer, f.p.Subject)
	must(t, err)
	mac := strings.ToUpper(strings.ReplaceAll(uuid.NewString(), "-", ""))[:12]
	in := InventoryInput{FactoryMAC: mac, Model: "KP01", HardwareRevision: "A1", BatchNo: "test"}
	raw := requestJSON(in)
	key := uuid.NewString()
	inv, err := f.s.CreateInventory(ctx, f.p, f.tenantPublic, uuid.NewString(), key, raw, in)
	must(t, err)
	again, err := f.s.CreateInventory(ctx, f.p, f.tenantPublic, uuid.NewString(), key, raw, in)
	must(t, err)
	if inv["platform_device_id"] != again["platform_device_id"] {
		t.Fatal("inventory identity changed")
	}
	public := inv["platform_device_id"].(string)
	job := inv["enrollment_job_id"].(string)
	if inv["released"] != true {
		t.Fatal("new inventory did not become immediately claimable")
	}
	var lifecycle, keyStatus, jobStatus string
	must(t, f.s.pool.QueryRow(ctx, `SELECT d.activation_state,k.status,ej.status FROM device d JOIN device_onboarding_key k ON k.device_id=d.id AND k.key_version=1 JOIN device_enrollment_job ej ON ej.device_id=d.id AND ej.id=$2::uuid WHERE d.public_id=$1::uuid`, public, job).Scan(&lifecycle, &keyStatus, &jobStatus))
	if lifecycle != "CLAIMABLE" || keyStatus != "ACTIVE" || jobStatus != "RELEASED" {
		t.Fatalf("direct inventory state lifecycle=%s key=%s job=%s", lifecycle, keyStatus, jobStatus)
	}
	f.d, err = f.s.findFactoryDevice(ctx, "KP01-"+mac, 1)
	must(t, err)
	boot := uuid.NewString()
	challenge, err := f.s.TimeChallenge(ctx, TimeChallengeInput{DeviceID: f.d.FactoryID, KeyVersion: 1, BootID: boot, ClientNonce: randomToken(16)})
	must(t, err)
	decoded, err := base64.RawURLEncoding.DecodeString(challenge["payload_b64"].(string))
	must(t, err)
	var packet map[string]any
	must(t, json.Unmarshal(decoded, &packet))
	reg := RegisterInput{ChallengeID: packet["challenge_id"].(string), BootID: boot, Model: f.d.Model, HardwareRevision: f.d.HardwareRevision, FirmwareVersion: "1.2.0", SDKVersion: "1.0.0"}
	requestID := uuid.NewString()
	_, err = f.s.Register(ctx, f.d, requestID, requestJSON(reg), reg)
	must(t, err)
	f.refresh()
	_, err = f.s.Register(ctx, f.d, requestID, requestJSON(reg), reg)
	must(t, err)
	setupIn := SetupInput{BootID: boot, LocalSetupID: uuid.NewString(), Transport: "LOCAL_WEB"}
	setup, err := f.s.Setup(ctx, f.d, uuid.NewString(), requestJSON(setupIn), setupIn)
	must(t, err)
	f.setup = setup["setup_id"].(string)
	verifier := randomToken(32)
	sum := sha256.Sum256([]byte(verifier))
	handoffIn := HandoffInput{SetupID: f.setup, LocalID: setupIn.LocalSetupID, Challenge: encode64(sum[:])}
	handoff, err := f.s.Handoff(ctx, f.d, uuid.NewString(), requestJSON(handoffIn), handoffIn)
	must(t, err)
	credentials, err := f.s.Exchange(ctx, ExchangeInput{Token: handoff["handoff_token"].(string), Verifier: verifier})
	must(t, err)
	_, err = f.s.Exchange(ctx, ExchangeInput{Token: handoff["handoff_token"].(string), Verifier: verifier})
	if err == nil {
		t.Fatal("handoff redeemed twice")
	}
	ws, err := f.s.webSession(ctx, pool, digest([]byte(credentials.Token)))
	must(t, err)
	credentials, err = f.s.AuthLink(ctx, ws, f.user)
	must(t, err)
	f.web, err = f.s.webSession(ctx, pool, digest([]byte(credentials.Token)))
	must(t, err)
	profile := ProfileInput{HouseholdName: "测试家庭", Name: "奶奶", InitialConsents: saas.InitialConsentSelection{Types: []string{saas.ConsentTranscriptStorage, saas.ConsentPersonalMemory}, Confirmed: true}}
	result, err := f.s.CreateProfile(ctx, f.web, uuid.NewString(), requestJSON(profile), profile)
	must(t, err)
	f.elder = result["elder_id"].(string)
	return f
}
func must(t *testing.T, err error) {
	t.Helper()
	if err != nil {
		t.Fatal(err)
	}
}
func (f *fixture) refresh() {
	f.t.Helper()
	var err error
	f.d, err = f.s.findFactoryDevice(f.ctx, f.d.FactoryID, 1)
	must(f.t, err)
	if f.web.Hash != "" {
		f.web, err = f.s.webSession(f.ctx, f.s.pool, f.web.Hash)
		must(f.t, err)
	}
}
func (f *fixture) pending() {
	f.t.Helper()
	in := ClaimInput{ElderID: f.elder, Generation: f.d.Generation}
	result, err := f.s.Claim(f.ctx, f.web, uuid.NewString(), requestJSON(in), in)
	must(f.t, err)
	f.claim = result["claim_id"].(string)
	poll, err := f.s.Poll(f.ctx, f.d, PollInput{SetupID: f.setup, BootID: f.d.BootID})
	must(f.t, err)
	payload := poll["command"].(map[string]any)["payload"].(map[string]any)
	f.confirm = ConfirmationInput{ClaimID: f.claim, Challenge: payload["challenge"].(string), Digest: payload["binding_digest"].(string), ExpectedGeneration: f.d.Generation, PressEventID: uuid.NewString()}
}
func (f *fixture) bound() {
	f.t.Helper()
	f.pending()
	_, err := f.s.Confirm(f.ctx, f.d, uuid.NewString(), requestJSON(f.confirm), f.confirm)
	must(f.t, err)
	f.refresh()
}
func (f *fixture) event() outbox.Event {
	f.t.Helper()
	var e outbox.Event
	err := f.s.pool.QueryRow(f.ctx, `SELECT id::text,payload FROM outbox_event WHERE aggregate_id=$1::uuid AND event_type=$2 ORDER BY created_at DESC LIMIT 1`, f.d.ID, ActivationEvent).Scan(&e.ID, &e.Payload)
	must(f.t, err)
	return e
}

func TestIntegrationPhysicalConfirmationAndReplay(t *testing.T) {
	f := newFixture(t)
	f.pending()
	var count int
	must(t, f.s.pool.QueryRow(f.ctx, `SELECT count(*) FROM device_binding WHERE device_id=$1::uuid`, f.d.ID).Scan(&count))
	if count != 0 {
		t.Fatal("claim established ownership before physical confirmation")
	}
	id := uuid.NewString()
	_, err := f.s.Confirm(f.ctx, f.d, id, requestJSON(f.confirm), f.confirm)
	must(t, err)
	f.refresh()
	if f.d.Generation != 1 {
		t.Fatal("wrong binding generation")
	}
	_, err = f.s.Confirm(f.ctx, f.d, id, requestJSON(f.confirm), f.confirm)
	must(t, err)
	status, err := f.s.WebStatus(f.ctx, f.web)
	must(t, err)
	if status["binding_id"] == nil {
		t.Fatal("binding result not recoverable")
	}
	_, err = f.s.pool.Exec(f.ctx, `UPDATE household_member SET status='removed' WHERE user_id=$1::uuid`, f.web.UserID)
	must(t, err)
	_, err = f.s.Confirm(f.ctx, f.d, id, requestJSON(f.confirm), f.confirm)
	if !errors.Is(err, errAccess) {
		t.Fatalf("replay bypassed revoked permission: %v", err)
	}
	_, err = f.s.pool.Exec(f.ctx, `UPDATE device_binding SET unbound_at=now() WHERE device_id=$1::uuid AND unbound_at IS NULL`, f.d.ID)
	must(t, err)
	current, err := readDevice(f.ctx, f.s.pool, f.d.ID, false)
	must(t, err)
	if current.Generation != 2 {
		t.Fatal("unbind did not increment generation")
	}
	_, err = f.s.Confirm(f.ctx, current, id, requestJSON(f.confirm), f.confirm)
	if err == nil {
		t.Fatal("old confirmation revived unbound owner")
	}
}
func TestIntegrationConcurrentClaimsAndCancel(t *testing.T) {
	f := newFixture(t)
	in := ClaimInput{ElderID: f.elder, Generation: 0}
	var winners atomic.Int32
	var wg sync.WaitGroup
	for i := 0; i < 2; i++ {
		wg.Add(1)
		go func() {
			defer wg.Done()
			_, err := f.s.Claim(f.ctx, f.web, uuid.NewString(), requestJSON(in), in)
			if err == nil {
				winners.Add(1)
			}
		}()
	}
	wg.Wait()
	if winners.Load() != 1 {
		t.Fatalf("claim winners=%d", winners.Load())
	}
	var id string
	must(t, f.s.pool.QueryRow(f.ctx, `SELECT id::text FROM device_binding_claim WHERE device_id=$1::uuid AND status='PENDING'`, f.d.ID).Scan(&id))
	cancel := CancelInput{ClaimID: id, Generation: 0}
	_, err := f.s.Cancel(f.ctx, f.web, uuid.NewString(), requestJSON(cancel), cancel)
	must(t, err)
	var challenge, hash string
	var encrypted []byte
	must(t, f.s.pool.QueryRow(f.ctx, `SELECT challenge_ciphertext,binding_digest FROM device_binding_claim WHERE id=$1::uuid`, id).Scan(&encrypted, &hash))
	plain, err := f.s.crypt.open(encrypted, "claim:"+id)
	must(t, err)
	challenge = string(plain)
	confirm := ConfirmationInput{ClaimID: id, Challenge: challenge, Digest: hash, ExpectedGeneration: 0, PressEventID: uuid.NewString()}
	_, err = f.s.Confirm(f.ctx, f.d, uuid.NewString(), requestJSON(confirm), confirm)
	if err == nil {
		t.Fatal("cancelled claim confirmed")
	}
}
func (f *fixture) job() activationJob {
	f.t.Helper()
	tx, err := f.s.pool.Begin(f.ctx)
	must(f.t, err)
	defer tx.Rollback(f.ctx)
	j, err := readJob(f.ctx, tx, f.d.ID)
	must(f.t, err)
	return j
}
func (f *fixture) registerSDK() activationJob {
	f.t.Helper()
	must(f.t, f.s.ProcessActivation(f.ctx, f.event()))
	f.refresh()
	j := f.job()
	attempt := AttemptInput{ActivationID: j.ID, ConfigVersion: j.ConfigVersion, AttemptID: uuid.NewString()}
	_, err := f.s.Attempt(f.ctx, f.d, uuid.NewString(), requestJSON(attempt), attempt)
	must(f.t, err)
	result := ResultInput{ActivationID: j.ID, AttemptID: attempt.AttemptID, Stage: "SDK_REGISTER", Outcome: "SUCCESS_OBSERVED"}
	_, err = f.s.ActivationResult(f.ctx, f.d, uuid.NewString(), requestJSON(result), result)
	must(f.t, err)
	return f.job()
}
func TestIntegrationDeviceUnknownDoesNotRegisterAgain(t *testing.T) {
	f := newFixture(t)
	f.bound()
	must(t, f.s.ProcessActivation(f.ctx, f.event()))
	f.refresh()
	j := f.job()
	attempt := AttemptInput{ActivationID: j.ID, ConfigVersion: j.ConfigVersion, AttemptID: uuid.NewString()}
	permit, err := f.s.Attempt(f.ctx, f.d, uuid.NewString(), requestJSON(attempt), attempt)
	must(t, err)
	if !permit["allowed"].(bool) || time.UnixMilli(permit["permit_expires_at_ms"].(int64)).Sub(time.Now()) > 30*time.Second {
		t.Fatal("permit contract")
	}
	result := ResultInput{ActivationID: j.ID, AttemptID: attempt.AttemptID, Stage: "SDK_REGISTER", SDKResult: -1, Outcome: "UNKNOWN"}
	response, err := f.s.ActivationResult(f.ctx, f.d, uuid.NewString(), requestJSON(result), result)
	must(t, err)
	if response["reconcile_required"] != true {
		t.Fatal("unknown must require reconciliation")
	}
	attempt.AttemptID = uuid.NewString()
	_, err = f.s.Attempt(f.ctx, f.d, uuid.NewString(), requestJSON(attempt), attempt)
	if err == nil {
		t.Fatal("unknown issued another execution")
	}
	must(t, f.s.ProcessActivation(f.ctx, f.event()))
	must(t, f.s.ProcessActivation(f.ctx, f.event()))
	if f.registrar.calls.Load() != 0 {
		t.Fatal("SaaS invoked supplier registration")
	}
	var bound bool
	must(t, f.s.pool.QueryRow(f.ctx, `SELECT EXISTS(SELECT 1 FROM device_binding WHERE device_id=$1::uuid AND unbound_at IS NULL)`, f.d.ID).Scan(&bound))
	if !bound {
		t.Fatal("unknown removed binding")
	}
}
func TestIntegrationDeviceConfigAndRegistration(t *testing.T) {
	f := newFixture(t)
	f.bound()
	must(t, f.s.ProcessActivation(f.ctx, f.event()))
	f.refresh()
	response, err := f.s.Poll(f.ctx, f.d, PollInput{SetupID: f.setup, BootID: f.d.BootID})
	must(t, err)
	payload := response["command"].(map[string]any)["payload"].(map[string]any)
	if len(payload) != 2 {
		t.Fatal("extra activation command fields")
	}
	raw, err := base64.RawURLEncoding.DecodeString(payload["config_b64"].(string))
	must(t, err)
	var config map[string]any
	must(t, json.Unmarshal(raw, &config))
	if len(config) != 8 || config["contract_version"] != "ONBOARDING-1.1" {
		t.Fatal("wrong activation config")
	}
	iot := config["iot"].(map[string]any)
	if len(iot) != 4 || iot["product_secret"] != f.s.options.ProductSecret || iot["device_name"] != f.d.FactoryID {
		t.Fatal("wrong SDK registration config")
	}
	if !sameSecret(domainMAC(f.d.Key, "CFG1\n"+payload["config_b64"].(string)), payload["config_mac"].(string)) {
		t.Fatal("config MAC")
	}
	j := f.job()
	ack := ActivationAckInput{ActivationID: j.ID, ConfigVersion: j.ConfigVersion, LocalGeneration: 1}
	_, err = f.s.ActivationAck(f.ctx, f.d, uuid.NewString(), requestJSON(ack), ack)
	if err == nil {
		t.Fatal("ACK before registration was accepted")
	}
	f.registerSDK()
	_, err = f.s.ActivationAck(f.ctx, f.d, uuid.NewString(), requestJSON(ack), ack)
	must(t, err)
	if f.job().Status != "REGISTERED_OBSERVED" {
		t.Fatal("registration falsely implies first call")
	}
	if f.registrar.calls.Load() != 0 {
		t.Fatal("server performed SDK registration")
	}
}
func TestIntegrationDEVREQReplayAndTLS(t *testing.T) {
	f := newFixture(t)
	in := PollInput{SetupID: f.setup, BootID: f.d.BootID}
	raw := requestJSON(in)
	path := "/api/v1/device/poll"
	id := uuid.NewString()
	nonce := randomToken(16)
	timestamp := time.Now().UnixMilli()
	request := func() *http.Request {
		r := httptest.NewRequest("POST", "https://platform.example"+path, bytes.NewReader(raw))
		r.Header.Set("Content-Type", "application/json")
		r.Header.Set("X-Device-ID", f.d.FactoryID)
		r.Header.Set("X-Key-Version", "1")
		r.Header.Set("X-Timestamp-Ms", strconv.FormatInt(timestamp, 10))
		r.Header.Set("X-Nonce", nonce)
		r.Header.Set("X-Request-ID", id)
		r.Header.Set("Authorization", "Device-HMAC-SHA256 "+RequestMAC(f.d.Key, "POST", path, f.d.FactoryID, 1, timestamp, nonce, id, raw))
		return r
	}
	mux := http.NewServeMux()
	f.s.RegisterRoutes(mux)
	w := httptest.NewRecorder()
	mux.ServeHTTP(w, request())
	if w.Code != 200 {
		t.Fatal(w.Code, w.Body.String())
	}
	w = httptest.NewRecorder()
	mux.ServeHTTP(w, request())
	if w.Code != 401 {
		t.Fatal("nonce replay was accepted", w.Code)
	}
	r := request()
	r.TLS = nil
	r.Header.Set("X-Forwarded-Proto", "https")
	w = httptest.NewRecorder()
	mux.ServeHTTP(w, r)
	if w.Code != 400 {
		t.Fatal("untrusted forwarded HTTPS accepted")
	}
}