fix(api): setup 链接在同一事务里消费并建会话，存储故障回 500 且链接仍可重试

internal/api/api_test.go

@@ -94,6 +94,7 @@ type fakeRepo struct {
 	failRevokeOthers error
 	failMarkReauth   error
 	failGetSetting   error
+	failRedeemSetup  error
 	// player email OTPs (spec §B2). Keyed by row id; the verify path scans for the
 	// newest live (user, purpose) just as the PG query does.
 	otps map[string]*fakeEmailOTP
@@ -1779,15 +1780,23 @@ func (f *fakeRepo) ApproveOpLogin(_ context.Context, id, approverUserID string,
 	return nil
 }
 
-// ConsumeSetupToken atomically marks a one-time setup token consumed and returns
-// its user_id, or ErrNotFound when absent, already consumed, or expired.
-func (f *fakeRepo) ConsumeSetupToken(_ context.Context, tokenHash string, now time.Time) (string, error) {
+// RedeemSetupToken spends a setup token and stores s for its user, or ErrNotFound
+// when the token is absent, already spent, or expired. failRedeemSetup fails the
+// whole redemption.
+func (f *fakeRepo) RedeemSetupToken(ctx context.Context, tokenHash string, now time.Time, s NewSession) (string, error) {
+	if f.failRedeemSetup != nil {
+		return "", f.failRedeemSetup
+	}
 	tok, ok := f.setupTokens[tokenHash]
 	if !ok || !tok.ConsumedAt.IsZero() || !tok.ExpiresAt.After(now) {
 		return "", ErrNotFound
 	}
 	tok.ConsumedAt = now
 	f.setupTokens[tokenHash] = tok
+	s.UserID = tok.UserID
+	if err := f.CreateSession(ctx, s); err != nil {
+		return "", err
+	}
 	return tok.UserID, nil
 }
 

internal/api/handlers_setup.go

@@ -62,26 +62,32 @@ func (a *API) handleSetupRedeem(w http.ResponseWriter, r *http.Request) {
 	sum := sha256.Sum256([]byte(token))
 	tokenHash := hex.EncodeToString(sum[:])
 
-	now := a.now()
-	userID, err := a.Repo.ConsumeSetupToken(r.Context(), tokenHash, now)
+	// A regular felis_session; the lockdown is a product-level restriction the
+	// frontend enforces until email is verified / a passkey is bound. The token and
+	// the session are written together: a new setup link takes `felis setup` on
+	// the node, so a failure here must leave this one working.
+	sessionToken, session, err := a.mintSession(r, "", provenSignIn)
 	if err != nil {
+		writeError(w, r, err)
+		return
+	}
+	userID, err := a.Repo.RedeemSetupToken(r.Context(), tokenHash, a.now(), session)
+	switch {
+	case errors.Is(err, ErrNotFound):
 		// Unknown, already-consumed, or expired — uniform 400 so the token cannot
 		// be used as an oracle.
 		a.authFailure(r, "setup_redeem", "bad_token", nil)
 		writeError(w, r, newError(http.StatusBadRequest, "setup_token_invalid",
 			"this setup link is invalid or has already been used"))
 		return
-	}
-
-	u, err := a.Repo.UserByID(r.Context(), userID)
-	if err != nil {
+	case err != nil:
 		writeError(w, r, err)
 		return
 	}
+	setSessionCookie(w, sessionToken, session.ExpiresAt)
 
-	// Mint the session — a regular felis_session; the lockdown is a product-level
-	// restriction the frontend enforces until email is verified / a passkey is bound.
-	if err := a.startSession(w, r, u.ID, provenSignIn); err != nil {
+	u, err := a.Repo.UserByID(r.Context(), userID)
+	if err != nil {
 		writeError(w, r, err)
 		return
 	}

internal/api/handlers_setup_test.go

@@ -1,9 +1,13 @@
 package api
 
 import (
+	"crypto/sha256"
+	"encoding/hex"
 	"encoding/json"
+	"errors"
 	"net/http"
 	"testing"
+	"time"
 )
 
 // TestSetupNoSMTPFlow pins the no-SMTP onboarding contract: setup completes on email
@@ -137,6 +141,55 @@ func TestSetupCompletesForNoEmailPlayer(t *testing.T) {
 	}
 }
 
+// Redeeming a setup link signs the owner in once. A link the store does not know
+// (unknown, spent, expired) is the uniform 400; a store that fails is an outage,
+// answered as one, so the page does not tell the owner a link that still works
+// was used up.
+func TestSetupRedeem(t *testing.T) {
+	const raw = "setup-token-raw"
+	sum := sha256.Sum256([]byte(raw))
+	hash := hex.EncodeToString(sum[:])
+	body := `{"token":"` + raw + `"}`
+	setup := func() (*fakeRepo, http.Handler) {
+		repo := newFakeRepo()
+		repo.settings[LocalAuthEnabledKey] = []byte("true")
+		repo.staff["owner"] = &StaffUser{ID: "o1", Username: "owner", Role: "admin"}
+		repo.setupTokens[hash] = fakeSetupToken{TokenHash: hash, UserID: "o1",
+			ExpiresAt: time.Unix(1_700_000_000, 0).Add(10 * time.Minute)}
+		return repo, newTestAPI(repo, newFakeCluster()).ExternalHandler()
+	}
+
+	t.Run("redeems once", func(t *testing.T) {
+		repo, h := setup()
+		w := do(h, "POST", "/api/v1/auth/setup/redeem", body, jsonHeader)
+		if w.Code != http.StatusOK {
+			t.Fatalf("redeem: code = %d, want 200 (%s)", w.Code, w.Body.String())
+		}
+		cookies := w.Result().Cookies()
+		if len(cookies) != 1 || cookies[0].Name != sessionCookieName {
+			t.Fatalf("cookies = %v, want one %s", cookies, sessionCookieName)
+		}
+		if s, ok := repo.sessions[hashCookie(cookies[0].Value)]; !ok || s.userID != "o1" {
+			t.Fatalf("session for the cookie = %+v (found %v), want one of o1", s, ok)
+		}
+		if w := do(h, "POST", "/api/v1/auth/setup/redeem", body, jsonHeader); w.Code != http.StatusBadRequest ||
+			errCode(w.Body.Bytes()) != "setup_token_invalid" || len(w.Result().Cookies()) != 0 {
+			t.Fatalf("replay: code = %d err = %q cookies = %v, want 400 setup_token_invalid and none",
+				w.Code, errCode(w.Body.Bytes()), w.Result().Cookies())
+		}
+	})
+
+	t.Run("store outage", func(t *testing.T) {
+		repo, h := setup()
+		repo.failRedeemSetup = errors.New("connection refused")
+		w := do(h, "POST", "/api/v1/auth/setup/redeem", body, jsonHeader)
+		if w.Code != http.StatusInternalServerError || errCode(w.Body.Bytes()) != "internal" || len(w.Result().Cookies()) != 0 {
+			t.Fatalf("outage: code = %d err = %q cookies = %v, want 500 internal and no cookie",
+				w.Code, errCode(w.Body.Bytes()), w.Result().Cookies())
+		}
+	})
+}
+
 // errCode returns the error.code of a JSON error body, or "" if body is not one (a
 // non-failing decodeErr for cases where the response may be a success).
 func errCode(body []byte) string {

internal/api/pgrepo.go

@@ -1416,8 +1416,17 @@ func (p *PGRepo) InsertOperator(ctx context.Context, id, username, email string)
 // CreateSession records a minted session by the sha-256 of its cookie value
 // (spec §B). Only the hash is stored, mirroring tokens.
 func (p *PGRepo) CreateSession(ctx context.Context, s NewSession) error {
+	return insertSession(ctx, p.db, s)
+}
+
+// sqlExecer is the write half shared by *sql.DB and *sql.Tx.
+type sqlExecer interface {
+	ExecContext(ctx context.Context, query string, args ...any) (sql.Result, error)
+}
+
+func insertSession(ctx context.Context, db sqlExecer, s NewSession) error {
 	reauth := sql.NullTime{Time: s.ReauthAt, Valid: !s.ReauthAt.IsZero()}
-	_, err := p.db.ExecContext(ctx,
+	_, err := db.ExecContext(ctx,
 		`INSERT INTO sessions (token_hash, user_id, expires_at, user_agent, client_ip, reauth_at)
 		 VALUES ($1, $2, $3, $4, $5, $6)`,
 		s.TokenHash, s.UserID, s.ExpiresAt, s.UserAgent, s.ClientIP, reauth)
@@ -2956,12 +2965,19 @@ func (p *PGRepo) ConsumeOpLoginRequest(ctx context.Context, id string, now time.
 
 // ---- setup token redemption (spec §B) ----
 
-// ConsumeSetupToken atomically marks a one-time setup token consumed and returns
-// its user_id, or ErrNotFound when the token is absent, already consumed, or
-// expired. The /setup?token=... web flow redeems it for a lockdown session.
-func (p *PGRepo) ConsumeSetupToken(ctx context.Context, tokenHash string, now time.Time) (string, error) {
+// RedeemSetupToken spends a one-time setup token and stores s as a session of the
+// token's user in one transaction, so a failure leaves the token unspent. It
+// returns the user id, or ErrNotFound when the token is absent, already spent, or
+// expired.
+func (p *PGRepo) RedeemSetupToken(ctx context.Context, tokenHash string, now time.Time, s NewSession) (string, error) {
+	tx, err := p.db.BeginTx(ctx, nil)
+	if err != nil {
+		return "", err
+	}
+	defer tx.Rollback() //nolint:errcheck // no-op after commit
+
 	var userID string
-	switch err := p.db.QueryRowContext(ctx,
+	switch err := tx.QueryRowContext(ctx,
 		`UPDATE setup_tokens SET consumed_at = $2
 		 WHERE token_hash = $1 AND consumed_at IS NULL AND expires_at > $2
 		 RETURNING user_id`,
@@ -2971,6 +2987,13 @@ func (p *PGRepo) ConsumeSetupToken(ctx context.Context, tokenHash string, now ti
 	case err != nil:
 		return "", err
 	}
+	s.UserID = userID
+	if err := insertSession(ctx, tx, s); err != nil {
+		return "", err
+	}
+	if err := tx.Commit(); err != nil {
+		return "", err
+	}
 	return userID, nil
 }
 
@@ -2978,7 +3001,7 @@ func (p *PGRepo) ConsumeSetupToken(ctx context.Context, tokenHash string, now ti
 // hash (the raw value rides in the /setup?token=... URL). The setup Owner-bind
 // path uses CompleteOwnerSetup so identity binding, local auth, and this token
 // commit atomically; this lower-level helper remains for callers that already
-// established the user. The token is redeemed exactly once by ConsumeSetupToken.
+// established the user. The token is redeemed exactly once by RedeemSetupToken.
 func (p *PGRepo) CreateSetupToken(ctx context.Context, tokenHash, userID string, expiresAt time.Time) error {
 	_, err := p.db.ExecContext(ctx,
 		`INSERT INTO setup_tokens (token_hash, user_id, expires_at) VALUES ($1, $2, $3)`,

internal/api/repo.go

@@ -722,11 +722,13 @@ type Repo interface {
 	// it ended.
 	RevokeOtherUserSessions(ctx context.Context, userID, keepTokenHash string) (int, error)
 
-	// ConsumeSetupToken atomically marks a one-time setup token consumed and returns
-	// its user_id, or ErrNotFound when the token is absent, already consumed, or
-	// expired. The /setup?token=... web flow redeems it for a lockdown session that
-	// can only complete passwordless login setup (verify email / enroll passkey).
-	ConsumeSetupToken(ctx context.Context, tokenHash string, now time.Time) (userID string, err error)
+	// RedeemSetupToken spends a one-time setup token and stores s as a session of
+	// the token's user (s.UserID is ignored) in one transaction, so a failure
+	// leaves the token unspent for another try. It returns the user id, or
+	// ErrNotFound when the token is absent, already spent, or expired. The /setup?token=... web flow redeems it for a lockdown
+	// session that can only complete passwordless login setup (verify email /
+	// enroll passkey).
+	RedeemSetupToken(ctx context.Context, tokenHash string, now time.Time, s NewSession) (userID string, err error)
 
 	// ---- runtime platform settings (spec §B platform_settings) ----
 

internal/api/session.go

@@ -80,12 +80,27 @@ const (
 )
 
 // startSession mints a session for userID and sets its cookie. Every sign-in door
-// ends here, so every session records the device it was minted for.
+// ends here (or at mintSession), so every session records the device it was
+// minted for.
 func (a *API) startSession(w http.ResponseWriter, r *http.Request, userID string, proof signInProof) error {
-	token, err := newSessionToken()
+	token, s, err := a.mintSession(r, userID, proof)
 	if err != nil {
 		return err
 	}
+	if err := a.Repo.CreateSession(r.Context(), s); err != nil {
+		return err
+	}
+	setSessionCookie(w, token, s.ExpiresAt)
+	return nil
+}
+
+// mintSession makes a session cookie value and the row that stores it, for a door
+// that writes the row itself.
+func (a *API) mintSession(r *http.Request, userID string, proof signInProof) (string, NewSession, error) {
+	token, err := newSessionToken()
+	if err != nil {
+		return "", NewSession{}, err
+	}
 	now := a.now()
 	expires := now.Add(sessionTTL)
 	ip := ""
@@ -96,18 +111,14 @@ func (a *API) startSession(w http.ResponseWriter, r *http.Request, userID string
 	if proof == provenSignIn {
 		reauth = now
 	}
-	if err := a.Repo.CreateSession(r.Context(), NewSession{
+	return token, NewSession{
 		TokenHash: hashCookie(token),
 		UserID:    userID,
 		ExpiresAt: expires,
 		UserAgent: truncateUTF8(r.UserAgent(), maxSessionUserAgent),
 		ClientIP:  ip,
 		ReauthAt:  reauth,
-	}); err != nil {
-		return err
-	}
-	setSessionCookie(w, token, expires)
-	return nil
+	}, nil
 }
 
 // currentSessionHash is the storage key of the session cookie r carries, or ""