Add vault seal/unseal lifecycle

- New internal/vault package: thread-safe Vault struct with seal/unseal state, key material zeroing, and key derivation - REST: POST /v1/vault/unseal, POST /v1/vault/seal, GET /v1/vault/status; health returns sealed status - UI: /unseal page with passphrase form, redirect when sealed - gRPC: sealedInterceptor rejects RPCs when sealed - Middleware: RequireUnsealed blocks all routes except exempt paths; RequireAuth reads pubkey from vault at request time - Startup: server starts sealed when passphrase unavailable - All servers share single *vault.Vault by pointer - CSRF manager derives key lazily from vault Security: Key material is zeroed on seal. Sealed middleware runs before auth. Handlers fail closed if vault becomes sealed mid-request. Unseal endpoint is rate-limited (3/s burst 5). No CSRF on unseal page (no session to protect; chicken-and-egg with master key). Passphrase never logged. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-03-14 23:55:37 -07:00
parent 5c242f8abb
commit d87b4b4042
28 changed files with 1292 additions and 119 deletions
--- a/internal/grpcserver/admin.go
+++ b/internal/grpcserver/admin.go
@@ -17,8 +17,12 @@ type adminServiceServer struct {
 	s *Server
 }

-// Health returns {"status":"ok"} to signal the server is operational.
+// Health returns {"status":"ok"} to signal the server is operational, or
+// {"status":"sealed"} when the vault is sealed.
 func (a *adminServiceServer) Health(_ context.Context, _ *mciasv1.HealthRequest) (*mciasv1.HealthResponse, error) {
+	if a.s.vault.IsSealed() {
+		return &mciasv1.HealthResponse{Status: "sealed"}, nil
+	}
 	return &mciasv1.HealthResponse{Status: "ok"}, nil
 }

@@ -26,11 +30,12 @@ func (a *adminServiceServer) Health(_ context.Context, _ *mciasv1.HealthRequest)
 // The "x" field is the raw 32-byte public key base64url-encoded without padding,
 // matching the REST /v1/keys/public response format.
 func (a *adminServiceServer) GetPublicKey(_ context.Context, _ *mciasv1.GetPublicKeyRequest) (*mciasv1.GetPublicKeyResponse, error) {
-	if len(a.s.pubKey) == 0 {
-		return nil, status.Error(codes.Internal, "public key not available")
+	pubKey, err := a.s.vault.PubKey()
+	if err != nil {
+		return nil, status.Error(codes.Unavailable, "vault sealed")
 	}
 	// Encode as base64url without padding — identical to the REST handler.
-	x := base64.RawURLEncoding.EncodeToString(a.s.pubKey)
+	x := base64.RawURLEncoding.EncodeToString(pubKey)
 	return &mciasv1.GetPublicKeyResponse{
 		Kty: "OKP",
 		Crv: "Ed25519",
--- a/internal/grpcserver/auth.go
+++ b/internal/grpcserver/auth.go
@@ -86,7 +86,11 @@ func (a *authServiceServer) Login(ctx context.Context, req *mciasv1.LoginRequest
 			a.s.db.WriteAuditEvent(model.EventLoginFail, &acct.ID, nil, ip, `{"reason":"totp_missing"}`) //nolint:errcheck
 			return nil, status.Error(codes.Unauthenticated, "TOTP code required")
 		}
-		secret, err := crypto.OpenAESGCM(a.s.masterKey, acct.TOTPSecretNonce, acct.TOTPSecretEnc)
+		masterKey, mkErr := a.s.vault.MasterKey()
+		if mkErr != nil {
+			return nil, status.Error(codes.Unavailable, "vault sealed")
+		}
+		secret, err := crypto.OpenAESGCM(masterKey, acct.TOTPSecretNonce, acct.TOTPSecretEnc)
 		if err != nil {
 			a.s.logger.Error("decrypt TOTP secret", "error", err, "account_id", acct.ID)
 			return nil, status.Error(codes.Internal, "internal error")
@@ -121,7 +125,11 @@ func (a *authServiceServer) Login(ctx context.Context, req *mciasv1.LoginRequest
 		}
 	}

-	tokenStr, claims, err := token.IssueToken(a.s.privKey, a.s.cfg.Tokens.Issuer, acct.UUID, roles, expiry)
+	privKey, pkErr := a.s.vault.PrivKey()
+	if pkErr != nil {
+		return nil, status.Error(codes.Unavailable, "vault sealed")
+	}
+	tokenStr, claims, err := token.IssueToken(privKey, a.s.cfg.Tokens.Issuer, acct.UUID, roles, expiry)
 	if err != nil {
 		a.s.logger.Error("issue token", "error", err)
 		return nil, status.Error(codes.Internal, "internal error")
@@ -186,7 +194,11 @@ func (a *authServiceServer) RenewToken(ctx context.Context, _ *mciasv1.RenewToke
 		}
 	}

-	newTokenStr, newClaims, err := token.IssueToken(a.s.privKey, a.s.cfg.Tokens.Issuer, acct.UUID, roles, expiry)
+	privKey, pkErr := a.s.vault.PrivKey()
+	if pkErr != nil {
+		return nil, status.Error(codes.Unavailable, "vault sealed")
+	}
+	newTokenStr, newClaims, err := token.IssueToken(privKey, a.s.cfg.Tokens.Issuer, acct.UUID, roles, expiry)
 	if err != nil {
 		return nil, status.Error(codes.Internal, "internal error")
 	}
@@ -245,7 +257,11 @@ func (a *authServiceServer) EnrollTOTP(ctx context.Context, req *mciasv1.EnrollT
 		return nil, status.Error(codes.Internal, "internal error")
 	}

-	secretEnc, secretNonce, err := crypto.SealAESGCM(a.s.masterKey, rawSecret)
+	masterKey, mkErr := a.s.vault.MasterKey()
+	if mkErr != nil {
+		return nil, status.Error(codes.Unavailable, "vault sealed")
+	}
+	secretEnc, secretNonce, err := crypto.SealAESGCM(masterKey, rawSecret)
 	if err != nil {
 		return nil, status.Error(codes.Internal, "internal error")
 	}
@@ -283,7 +299,11 @@ func (a *authServiceServer) ConfirmTOTP(ctx context.Context, req *mciasv1.Confir
 		return nil, status.Error(codes.FailedPrecondition, "TOTP enrollment not started")
 	}

-	secret, err := crypto.OpenAESGCM(a.s.masterKey, acct.TOTPSecretNonce, acct.TOTPSecretEnc)
+	masterKey, mkErr := a.s.vault.MasterKey()
+	if mkErr != nil {
+		return nil, status.Error(codes.Unavailable, "vault sealed")
+	}
+	secret, err := crypto.OpenAESGCM(masterKey, acct.TOTPSecretNonce, acct.TOTPSecretEnc)
 	if err != nil {
 		return nil, status.Error(codes.Internal, "internal error")
 	}
--- a/internal/grpcserver/credentialservice.go
+++ b/internal/grpcserver/credentialservice.go
@@ -47,7 +47,11 @@ func (c *credentialServiceServer) GetPGCreds(ctx context.Context, req *mciasv1.G
 	}

 	// Decrypt the password for admin retrieval.
-	password, err := crypto.OpenAESGCM(c.s.masterKey, cred.PGPasswordNonce, cred.PGPasswordEnc)
+	masterKey, mkErr := c.s.vault.MasterKey()
+	if mkErr != nil {
+		return nil, status.Error(codes.Unavailable, "vault sealed")
+	}
+	password, err := crypto.OpenAESGCM(masterKey, cred.PGPasswordNonce, cred.PGPasswordEnc)
 	if err != nil {
 		return nil, status.Error(codes.Internal, "internal error")
 	}
@@ -94,7 +98,11 @@ func (c *credentialServiceServer) SetPGCreds(ctx context.Context, req *mciasv1.S
 		return nil, status.Error(codes.Internal, "internal error")
 	}

-	enc, nonce, err := crypto.SealAESGCM(c.s.masterKey, []byte(cr.Password))
+	masterKey, mkErr := c.s.vault.MasterKey()
+	if mkErr != nil {
+		return nil, status.Error(codes.Unavailable, "vault sealed")
+	}
+	enc, nonce, err := crypto.SealAESGCM(masterKey, []byte(cr.Password))
 	if err != nil {
 		return nil, status.Error(codes.Internal, "internal error")
 	}
--- a/internal/grpcserver/grpcserver.go
+++ b/internal/grpcserver/grpcserver.go
@@ -17,7 +17,6 @@ package grpcserver

 import (
 	"context"
-	"crypto/ed25519"
 	"log/slog"
 	"net"
 	"strings"
@@ -35,6 +34,7 @@ import (
 	"git.wntrmute.dev/kyle/mcias/internal/config"
 	"git.wntrmute.dev/kyle/mcias/internal/db"
 	"git.wntrmute.dev/kyle/mcias/internal/token"
+	"git.wntrmute.dev/kyle/mcias/internal/vault"
 )

 // contextKey is the unexported context key type for this package.
@@ -57,21 +57,17 @@ type Server struct {
 	cfg         *config.Config
 	logger      *slog.Logger
 	rateLimiter *grpcRateLimiter
-	privKey     ed25519.PrivateKey
-	pubKey      ed25519.PublicKey
-	masterKey   []byte
+	vault       *vault.Vault
 }

 // New creates a Server with the given dependencies (same as the REST Server).
 // A fresh per-IP rate limiter (10 req/s, burst 10) is allocated per Server
 // instance so that tests do not share state across test cases.
-func New(database *db.DB, cfg *config.Config, priv ed25519.PrivateKey, pub ed25519.PublicKey, masterKey []byte, logger *slog.Logger) *Server {
+func New(database *db.DB, cfg *config.Config, v *vault.Vault, logger *slog.Logger) *Server {
 	return &Server{
 		db:          database,
 		cfg:         cfg,
-		privKey:     priv,
-		pubKey:      pub,
-		masterKey:   masterKey,
+		vault:       v,
 		logger:      logger,
 		rateLimiter: newGRPCRateLimiter(10, 10),
 	}
@@ -106,6 +102,7 @@ func (s *Server) buildServer(extra ...grpc.ServerOption) *grpc.Server {
 		[]grpc.ServerOption{
 			grpc.ChainUnaryInterceptor(
 				s.loggingInterceptor,
+				s.sealedInterceptor,
 				s.authInterceptor,
 				s.rateLimitInterceptor,
 			),
@@ -162,14 +159,36 @@ func (s *Server) loggingInterceptor(
 	return resp, err
 }

+// sealedInterceptor rejects all RPCs (except Health) when the vault is sealed.
+//
+// Security: This is the first interceptor in the chain (after logging). It
+// prevents any authenticated or data-serving handler from running while the
+// vault is sealed and key material is unavailable.
+func (s *Server) sealedInterceptor(
+	ctx context.Context,
+	req interface{},
+	info *grpc.UnaryServerInfo,
+	handler grpc.UnaryHandler,
+) (interface{}, error) {
+	if !s.vault.IsSealed() {
+		return handler(ctx, req)
+	}
+	// Health is always allowed — returns sealed status.
+	if info.FullMethod == "/mcias.v1.AdminService/Health" {
+		return handler(ctx, req)
+	}
+	return nil, status.Error(codes.Unavailable, "vault sealed")
+}
+
 // authInterceptor validates the Bearer JWT from gRPC metadata and injects
 // claims into the context. Public methods bypass this check.
 //
 // Security: Same validation path as the REST RequireAuth middleware:
 //  1. Extract "authorization" metadata value (case-insensitive key lookup).
-//  2. Validate JWT (alg-first, then signature, then expiry/issuer).
-//  3. Check JTI against revocation table.
-//  4. Inject claims into context.
+//  2. Read public key from vault (fail closed if sealed).
+//  3. Validate JWT (alg-first, then signature, then expiry/issuer).
+//  4. Check JTI against revocation table.
+//  5. Inject claims into context.
 func (s *Server) authInterceptor(
 	ctx context.Context,
 	req interface{},
@@ -186,7 +205,13 @@ func (s *Server) authInterceptor(
 		return nil, status.Error(codes.Unauthenticated, "missing or invalid authorization")
 	}

-	claims, err := token.ValidateToken(s.pubKey, tokenStr, s.cfg.Tokens.Issuer)
+	// Security: read the public key from vault at request time.
+	pubKey, err := s.vault.PubKey()
+	if err != nil {
+		return nil, status.Error(codes.Unavailable, "vault sealed")
+	}
+
+	claims, err := token.ValidateToken(pubKey, tokenStr, s.cfg.Tokens.Issuer)
 	if err != nil {
 		return nil, status.Error(codes.Unauthenticated, "invalid or expired token")
 	}
--- a/internal/grpcserver/grpcserver_test.go
+++ b/internal/grpcserver/grpcserver_test.go
@@ -30,6 +30,7 @@ import (
 	"git.wntrmute.dev/kyle/mcias/internal/db"
 	"git.wntrmute.dev/kyle/mcias/internal/model"
 	"git.wntrmute.dev/kyle/mcias/internal/token"
+	"git.wntrmute.dev/kyle/mcias/internal/vault"
 )

 const (
@@ -73,7 +74,8 @@ func newTestEnv(t *testing.T) *testEnv {
 	cfg := config.NewTestConfig(testIssuer)
 	logger := slog.New(slog.NewTextHandler(io.Discard, nil))

-	srv := New(database, cfg, priv, pub, masterKey, logger)
+	v := vault.NewUnsealed(masterKey, priv, pub)
+	srv := New(database, cfg, v, logger)
 	grpcSrv := srv.GRPCServer()

 	lis := bufconn.Listen(bufConnSize)
--- a/internal/grpcserver/tokenservice.go
+++ b/internal/grpcserver/tokenservice.go
@@ -32,7 +32,11 @@ func (t *tokenServiceServer) ValidateToken(_ context.Context, req *mciasv1.Valid
 		return &mciasv1.ValidateTokenResponse{Valid: false}, nil
 	}

-	claims, err := token.ValidateToken(t.s.pubKey, tokenStr, t.s.cfg.Tokens.Issuer)
+	pubKey, pkErr := t.s.vault.PubKey()
+	if pkErr != nil {
+		return &mciasv1.ValidateTokenResponse{Valid: false}, nil
+	}
+	claims, err := token.ValidateToken(pubKey, tokenStr, t.s.cfg.Tokens.Issuer)
 	if err != nil {
 		return &mciasv1.ValidateTokenResponse{Valid: false}, nil
 	}
@@ -67,7 +71,11 @@ func (ts *tokenServiceServer) IssueServiceToken(ctx context.Context, req *mciasv
 		return nil, status.Error(codes.InvalidArgument, "token issue is only for system accounts")
 	}

-	tokenStr, claims, err := token.IssueToken(ts.s.privKey, ts.s.cfg.Tokens.Issuer, acct.UUID, nil, ts.s.cfg.ServiceExpiry())
+	privKey, pkErr := ts.s.vault.PrivKey()
+	if pkErr != nil {
+		return nil, status.Error(codes.Unavailable, "vault sealed")
+	}
+	tokenStr, claims, err := token.IssueToken(privKey, ts.s.cfg.Tokens.Issuer, acct.UUID, nil, ts.s.cfg.ServiceExpiry())
 	if err != nil {
 		return nil, status.Error(codes.Internal, "internal error")
 	}