unibus/cmd/membershipd/main.go

// Command membershipd is the unibus control-plane service: room metadata,
// member directory, sealed key distribution, and the media blob store. The data
// plane is NATS — if --nats-url is empty it starts an embedded nats-server with
// JetStream so the whole stack runs with `go run` and nothing to install.
package main

import (
	"context"
	"crypto/tls"
	"encoding/hex"
	"flag"
	"log"
	"net/http"
	"os"
	"os/signal"
	"syscall"
	"time"

	cs "fn-registry/functions/cybersecurity"

	"github.com/nats-io/nats.go"
	"github.com/nats-io/nats.go/jetstream"
	server "github.com/nats-io/nats-server/v2/server"

	"github.com/enmanuel/unibus/pkg/blobstore"
	"github.com/enmanuel/unibus/pkg/busauth"
	"github.com/enmanuel/unibus/pkg/embeddednats"
	"github.com/enmanuel/unibus/pkg/membership"
)

func main() {
	// Subcommand dispatch: `membershipd user ...` is the local administration CLI
	// (seed/list/revoke bus users) and must be handled before the server flag set
	// parses os.Args. Running the CLI on the bus host is trusted by design (whoever
	// has a shell there already controls the service), which is how the first admin
	// is seeded without a chicken-egg auth problem.
	if len(os.Args) > 1 && os.Args[1] == "user" {
		runUserCLI(os.Args[2:])
		return
	}
	// `membershipd migrate-to-kv` is the one-time, idempotent SQLite->JetStream KV
	// data move for decentralization (issue 0003c). Like the user CLI it runs on
	// the host and is dispatched before the server flag set parses os.Args.
	if len(os.Args) > 1 && os.Args[1] == "migrate-to-kv" {
		runMigrateCLI(os.Args[2:])
		return
	}

	var (
		bind      = flag.String("bind", "127.0.0.1", "network interface to bind the HTTP API and the embedded NATS to; use 0.0.0.0 to accept LAN/remote peers")
		natsURL   = flag.String("nats-url", "", "external NATS url; empty starts an embedded server")
		httpPort  = flag.String("http-port", "8470", "HTTP port for the control-plane API")
		dbPath    = flag.String("db", "./local_files/unibus.db", "SQLite database path")
		storeDir  = flag.String("store-dir", "./local_files/blobs", "blob store directory")
		natsPort  = flag.Int("nats-port", 4250, "embedded NATS listen port (when --nats-url empty)")
		natsStore = flag.String("nats-store", "./local_files/jetstream", "embedded JetStream store dir")
		busAuth   = flag.String("bus-auth", "off", "control-plane auth rollout: off|soft|enforce (feature flag bus-auth)")
		tlsCert   = flag.String("tls-cert", "", "PATH to the NATS server certificate (deploy/tls/server.crt); enables TLS on the embedded data plane")
		tlsKey    = flag.String("tls-key", "", "path to the NATS server private key (deploy/tls/server.key); required with --tls-cert")
		// Cluster (issue 0003a): empty --cluster-name keeps the server standalone.
		clusterName = flag.String("cluster-name", "", "NATS cluster name (identical on every node); empty = standalone, no HA")
		serverName  = flag.String("server-name", "", "unique node name within the cluster (required by JetStream RAFT when clustered)")
		clusterPort = flag.Int("cluster-port", 6250, "route listener port for server-to-server cluster traffic")
		routesCSV   = flag.String("routes", "", "comma-separated nats-route URLs of the OTHER nodes, e.g. nats://user:pass@10.0.0.2:6250")
		clusterUser = flag.String("cluster-user", "", "shared route secret username (gates the route listener)")
		clusterPass = flag.String("cluster-pass", "", "shared route secret password")
		routeTLSCert = flag.String("route-tls-cert", "", "this node's route certificate (CA-signed); enables mutual route TLS with --route-tls-key/--route-tls-ca")
		routeTLSKey  = flag.String("route-tls-key", "", "this node's route private key")
		routeTLSCA   = flag.String("route-tls-ca", "", "bus CA that signs every node's route certificate (deploy/tls/ca.crt)")
		// Replicated control plane (issue 0006a/c): the JetStream replication factor
		// for the shared nonce bucket (and, with --store kv, the control-plane KV).
		// 1 for a 1-2 node rollout, 3 for real HA quorum (raise in place with
		// `nats stream update --replicas 3` when the third node joins).
		kvReplicas = flag.Int("kv-replicas", 1, "JetStream replication factor for the shared nonce/KV buckets (1..3)")
		caFile     = flag.String("ca", "", "bus CA cert; only used to pin TLS on the internal JetStream connection to an EXTERNAL --nats-url (the embedded server uses an in-process connection that needs no CA)")
		// Control-plane store backend (issue 0006c, feature flag decentralized):
		// "sqlite" (default) keeps the local single-node SQLite control plane;
		// "kv" puts rooms/members/keys/users in replicated JetStream KV so any node
		// in the cluster serves the same state.
		storeBackend = flag.String("store", "sqlite", "control-plane store backend: sqlite (default, single-node) | kv (replicated JetStream, decentralized)")
	)
	flag.Parse()

	authMode, err := membership.ParseAuthMode(*busAuth)
	if err != nil {
		log.Fatalf("%v", err)
	}
	if *storeBackend != "sqlite" && *storeBackend != "kv" {
		log.Fatalf("--store must be \"sqlite\" or \"kv\", got %q", *storeBackend)
	}

	// Fail-open guard (audit H2): a non-loopback bind, or any TLS flag, demands
	// --bus-auth enforce. This makes an insecure public startup impossible rather
	// than silently exposing the bus with the appearance of security.
	if err := validateBootConfig(*bind, authMode, *tlsCert, *tlsKey); err != nil {
		log.Fatalf("%v", err)
	}
	// Cluster route guard (issue 0003a): a public cluster needs a route secret
	// and mutual route TLS, and the route-TLS flags are all-or-nothing.
	if err := validateClusterConfig(*clusterName, *bind, *clusterUser, *clusterPass, *routeTLSCert, *routeTLSKey, *routeTLSCA); err != nil {
		log.Fatalf("%v", err)
	}

	log.SetFlags(log.LstdFlags | log.Lmsgprefix)
	log.SetPrefix("[membershipd] ")

	// A clustered node shares its control plane with peers, so it needs a JetStream
	// client to manage the replicated nonce bucket (issue 0006a). --store kv (issue
	// 0006c) also needs JetStream, for the control-plane KV itself. A standalone
	// single-node SQLite deployment needs none of this and keeps the in-process,
	// in-memory behavior unchanged.
	clustered := *clusterName != ""
	decentralized := *storeBackend == "kv"
	needJS := clustered || decentralized
	enforce := authMode == membership.AuthEnforce

	// Internal service identity (issue 0006a): when the embedded data plane enforces
	// auth, membershipd must still connect to its OWN server to manage JetStream.
	// It does so with this ephemeral identity, which the authenticator is built to
	// recognize and grant full permissions (it never enters the user allowlist). It
	// is only generated when actually needed (JetStream required AND enforce on AND
	// the server is embedded), so a standalone or non-enforce node is unchanged.
	var internalID cs.Identity
	var internalPubHex string
	if needJS && enforce && *natsURL == "" {
		internalID, err = cs.GenerateIdentity()
		if err != nil {
			log.Fatalf("generate internal identity: %v", err)
		}
		internalPubHex = hex.EncodeToString(internalID.SignPub)
	}

	// The authenticator consults the store through a holder so it can be built
	// before the store exists: with --store kv the JetStream KV store opens only
	// after NATS is up (the bootstrap cycle). In the default SQLite path the store
	// is opened and set into the holder right here, before the server starts, so
	// behavior is identical to the pre-0006c baseline. `store` is the final store
	// used by the HTTP server (set below for the KV path).
	holder := &storeHolder{}
	var store membership.Store
	if !decentralized {
		store, err = membership.Open(*dbPath)
		if err != nil {
			log.Fatalf("open membership store: %v", err)
		}
		holder.set(store)
		log.Printf("membership store: sqlite %s", *dbPath)
	}
	// Close whichever store ends up final (SQLite closes its file; the JetStream KV
	// store's Close is a no-op — its NATS connection is closed separately).
	defer func() {
		if store != nil {
			store.Close()
		}
	}()

	blobs, err := blobstore.New(*storeDir)
	if err != nil {
		log.Fatalf("open blob store: %v", err)
	}
	log.Printf("blob store: %s", *storeDir)

	// Data plane: embedded or external NATS. For the embedded server, enforce
	// turns on the nkey authenticator (only allowlisted identities may connect)
	// and --tls-cert/--tls-key turn on TLS. An external NATS manages its own
	// auth/TLS, so those flags do not apply to it.
	var ns *server.Server
	natsClientURL := *natsURL
	if natsClientURL == "" {
		cfg := embeddednats.ServerConfig{
			// Bind the embedded NATS to the same interface as the HTTP API so a
			// single --bind flag governs reachability: 127.0.0.1 keeps the whole
			// stack loopback-only; 0.0.0.0 exposes both planes to the LAN.
			StoreDir:   *natsStore,
			Host:       *bind,
			Port:       *natsPort,
			ServerName: *serverName,
		}
		// Cluster (issue 0003a): with a cluster name, join the route layer for HA.
		if *clusterName != "" {
			cc := &embeddednats.ClusterConfig{
				Name:     *clusterName,
				Host:     *bind,
				Port:     *clusterPort,
				Routes:   splitRoutes(*routesCSV),
				Username: *clusterUser,
				Password: *clusterPass,
			}
			if *routeTLSCert != "" {
				rtls, err := busauth.RouteTLSConfig(*routeTLSCert, *routeTLSKey, *routeTLSCA)
				if err != nil {
					log.Fatalf("load route TLS: %v", err)
				}
				cc.TLS = rtls
				log.Printf("cluster route TLS: ON (mutual, CA %s)", *routeTLSCA)
			}
			cfg.Cluster = cc
			log.Printf("cluster: %q node %q, route port %d, %d peer route(s)", *clusterName, *serverName, *clusterPort, len(cc.Routes))
		}
		if authMode == membership.AuthEnforce {
			// Per-subject data-plane ACL (audit H4 / N4 residual): the authenticator
			// authorizes by the bus allowlist AND confines each connection to the
			// subjects of the rooms it belongs to (plus client-infra subjects). This
			// closes the wildcard metadata leak where a registered non-member could
			// Subscribe(">") and harvest every room's subject and JetStream activity.
			// NATS freezes permissions at connect time, so a peer that joins a room
			// after connecting must client.RefreshSession to gain that room's subject.
			cfg.Auth = busauth.NewNkeyAuthenticatorACLInternal(
				holder.IsAuthorized,
				busauth.PermissionsFromSubjects(holder.subjectACL),
				internalPubHex,
			)
			log.Printf("NATS nkey authentication: ON (enforce, per-subject ACL)")
		}
		if *tlsCert != "" || *tlsKey != "" {
			if *tlsCert == "" || *tlsKey == "" {
				log.Fatalf("--tls-cert and --tls-key must be set together")
			}
			tlsCfg, err := busauth.ServerTLSConfig(*tlsCert, *tlsKey)
			if err != nil {
				log.Fatalf("load NATS TLS: %v", err)
			}
			cfg.TLS = tlsCfg
			log.Printf("NATS TLS: ON (%s)", *tlsCert)
		}
		ns, err = embeddednats.StartServer(cfg)
		if err != nil {
			log.Fatalf("start embedded nats: %v", err)
		}
		natsClientURL = embeddednats.ClientURL(ns)
		log.Printf("embedded NATS (JetStream) ready: %s", natsClientURL)
	} else {
		log.Printf("using external NATS: %s", natsClientURL)
	}

	// JetStream client + decentralized store (issue 0006a/c). needJS is set for a
	// clustered node (shared nonce bucket) and for --store kv (the KV control
	// plane). Open the privileged JetStream client first (in-process for the
	// embedded server, a plain client for external NATS), then — for --store kv —
	// open the replicated KV store and publish it into the holder so the
	// authenticator and HTTP server serve from it. The privileged connection is the
	// only client that can connect in this window (the holder still denies everyone
	// else; the internal identity bypasses the store).
	var js jetstream.JetStream
	if needJS {
		var internalNC *nats.Conn
		if *natsURL == "" {
			internalNC, js, err = connectInternalJS(ns, internalID, enforce)
		} else {
			internalNC, js, err = connectExternalJS(natsClientURL, *caFile)
		}
		if err != nil {
			log.Fatalf("internal JetStream connection (required by --cluster-name/--store kv): %v", err)
		}
		defer internalNC.Close()

		if decentralized {
			kvStore, err := membership.OpenJetStream(js, membership.JetStreamConfig{Replicas: *kvReplicas})
			if err != nil {
				log.Fatalf("open decentralized control-plane KV store: %v", err)
			}
			store = kvStore
			holder.set(store)
			log.Printf("membership store: jetstream KV (replicas=%d)", *kvReplicas)
		}
	}

	srv := membership.NewServer(store, blobs, authMode)
	// On a public (non-loopback) bind, disable cleartext rooms: the embedded NATS
	// has no per-subject ACL, so cleartext content would be readable by any
	// registered peer. Forcing E2E keeps message content confidential regardless
	// (audit H4 minimum defense; see dev/0004d-dataplane-acl.md).
	if !isLoopbackBind(*bind) {
		srv.RequireEncryptedRooms = true
		log.Printf("cleartext rooms: DISABLED (public bind requires end-to-end encryption)")
	}

	// Replicated anti-replay (issue 0006a, audit 0008 N3): a clustered node MUST
	// share its nonce store across the cluster, or a request accepted on one node
	// can be replayed to another. HARD requirement: if the bucket cannot be created
	// the node refuses to start rather than run with a per-process cache that leaves
	// the replay hole open.
	if needJS {
		if err := wireReplicatedNonces(srv, js, clustered, *kvReplicas); err != nil {
			log.Fatalf("%v", err)
		}
		if clustered {
			log.Printf("anti-replay: replicated nonce bucket \"KV_UNIBUS_nonces\" (replicas=%d) — cluster-safe", *kvReplicas)
		}
	}

	log.Printf("control-plane auth: %s", authMode)
	addr := *bind + ":" + *httpPort
	httpSrv := &http.Server{
		Addr:    addr,
		Handler: srv,
		// Bound request header size so a peer cannot exhaust memory with huge
		// headers before any body limit applies (the body ceilings live in the
		// membership middleware).
		MaxHeaderBytes:    membership.MaxHeaderBytes,
		ReadHeaderTimeout: 10 * time.Second,
	}

	go func() {
		var serveErr error
		if *tlsCert != "" {
			// Serve the control plane over TLS with the same CA-signed cert as the
			// data plane (audit H5): metadata (subjects, pubkeys, sealed keys, the
			// social graph) is no longer readable by a network MITM. The fail-open
			// guard already requires --bus-auth enforce alongside these flags.
			httpSrv.TLSConfig = &tls.Config{MinVersion: tls.VersionTLS12}
			log.Printf("HTTPS control-plane API: https://%s", addr)
			log.Printf("  health: https://%s/healthz", addr)
			log.Printf("control-plane TLS: ON (%s)", *tlsCert)
			serveErr = httpSrv.ListenAndServeTLS(*tlsCert, *tlsKey)
		} else {
			log.Printf("HTTP control-plane API: http://%s", addr)
			log.Printf("  health: http://%s/healthz", addr)
			serveErr = httpSrv.ListenAndServe()
		}
		if serveErr != nil && serveErr != http.ErrServerClosed {
			log.Fatalf("http server: %v", serveErr)
		}
	}()

	// Graceful shutdown on SIGINT/SIGTERM.
	stop := make(chan os.Signal, 1)
	signal.Notify(stop, syscall.SIGINT, syscall.SIGTERM)
	<-stop
	log.Printf("shutting down...")

	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
	defer cancel()
	_ = httpSrv.Shutdown(ctx)
	if ns != nil {
		ns.Shutdown()
		ns.WaitForShutdown()
	}
	log.Printf("bye")
}