fix: revoke sessions after updates

Dockerfile

@@ -1,20 +1,25 @@
 # syntax=docker/dockerfile:1.7
 
-# ---- Stage 1: build the web frontend ----
-FROM node:20-alpine AS web-builder
+# Build toolchains run natively on the BuildKit host. Without BUILDPLATFORM,
+# the arm64 branch executes npm and the Go compiler through QEMU, which is much
+# slower and makes npm ci appear to hang despite producing no progress output.
+# ---- Stage 1: build the web frontend once on the native builder ----
+FROM --platform=$BUILDPLATFORM node:20-alpine AS web-builder
 WORKDIR /web
 COPY web/package.json web/package-lock.json* ./
 RUN npm ci
 COPY web/ ./
 RUN npm run build
 
-# ---- Stage 2: build the Go binary ----
-FROM golang:1.25-alpine AS go-builder
+# ---- Stage 2: cross-compile the Go binary on the native builder ----
+FROM --platform=$BUILDPLATFORM golang:1.25-alpine AS go-builder
 RUN apk add --no-cache git
 WORKDIR /src
 
 ARG VERSION=0.1.0-dev
 ARG BUILD_TIME=""
+ARG TARGETOS
+ARG TARGETARCH
 
 COPY go.mod go.sum ./
 RUN go mod download
@@ -23,7 +28,7 @@ COPY . .
 # Overlay the freshly built frontend so go:embed web/dist picks it up.
 COPY --from=web-builder /web/dist ./web/dist
 
-RUN CGO_ENABLED=0 GOOS=linux go build \
+RUN CGO_ENABLED=0 GOOS=${TARGETOS:-linux} GOARCH=${TARGETARCH} go build \
     -trimpath \
     -ldflags "-s -w -X vocat/internal/buildinfo.Version=${VERSION} -X vocat/internal/buildinfo.BuildTime=${BUILD_TIME}" \
     -o /out/vocat \

cmd/vocat/main.go

@@ -248,6 +248,10 @@ func run(logger *slog.Logger, logs *loghub.Hub) error {
 	case <-signalContext.Done():
 		logger.Info("shutdown signal received")
 	}
+	// Long-lived SSE and polling handlers use this context. Stop them before
+	// http.Server.Shutdown so they do not consume the entire graceful-shutdown
+	// deadline while waiting for a stream that is intentionally still active.
+	cancelPolling()
 
 	shutdownContext, cancelShutdown := context.WithTimeout(
 		context.Background(),

internal/server/device_features_api_test.go

@@ -3,11 +3,13 @@ package server
 import (
 	"context"
 	"encoding/json"
+	"errors"
 	"log/slog"
 	"net/http"
 	"net/http/httptest"
 	"strings"
 	"testing"
+	"time"
 
 	"vocat/internal/device"
 	"vocat/internal/modem"
@@ -354,7 +356,22 @@ func TestHandleUpdateCheckUsesTrustedRepository(t *testing.T) {
 }
 
 func TestHandleUpdateApplyInstallsFromTrustedRepository(t *testing.T) {
+	database, err := store.Open(context.Background(), ":memory:")
+	if err != nil {
+		t.Fatal(err)
+	}
+	t.Cleanup(func() { _ = database.Close() })
+	if err := database.SetAdmin(context.Background(), "admin", []byte("hash")); err != nil {
+		t.Fatal(err)
+	}
+	tokenHash := []byte("active-session")
+	if err := database.CreateSession(
+		context.Background(), 1, tokenHash, []byte("csrf"), time.Now().Add(time.Hour),
+	); err != nil {
+		t.Fatal(err)
+	}
 	server := &Server{
+		store:            database,
 		logger:           regionTestLogger(),
 		updateRepository: update.DefaultRepository,
 		updateApply: func(_ context.Context, _ *slog.Logger, options update.Options, restart bool) (update.CheckResult, error) {
@@ -370,9 +387,21 @@ func TestHandleUpdateApplyInstallsFromTrustedRepository(t *testing.T) {
 		t.Fatalf("status = %d, body = %s", recorder.Code, recorder.Body)
 	}
 	data := decodeData(t, recorder)
-	if data["applied"] != true || data["version"] != "9.9.9" {
+	if data["applied"] != true || data["version"] != "9.9.9" || data["reauthentication_required"] != true {
 		t.Fatalf("apply data = %#v", data)
 	}
+	if _, err := database.SessionByTokenHash(context.Background(), tokenHash); !errors.Is(err, store.ErrNotFound) {
+		t.Fatalf("session must be revoked after update, got %v", err)
+	}
+	expired := map[string]bool{}
+	for _, cookie := range recorder.Result().Cookies() {
+		if cookie.MaxAge < 0 {
+			expired[cookie.Name] = true
+		}
+	}
+	if !expired[sessionCookieName] || !expired[csrfCookieName] {
+		t.Fatalf("auth cookies were not expired: %#v", recorder.Result().Cookies())
+	}
 }
 
 func TestE911WebsheetFlow(t *testing.T) {

internal/server/general_api.go

@@ -421,11 +421,22 @@ func (s *Server) handleUpdateApply(w http.ResponseWriter, r *http.Request) {
 		})
 		return
 	}
+	// A binary update changes the trusted server code underneath every active
+	// browser/API session. Revoke every durable token before scheduling the
+	// restart and expire this client's cookies so all users must authenticate
+	// against the newly installed version.
+	if err := s.store.DeleteAllSessions(r.Context()); err != nil {
+		s.logger.Error("revoke sessions after update failed", "error", err)
+		writeError(w, http.StatusInternalServerError, "update_session_revocation_failed", "The update was installed, but active sessions could not be revoked; restart the service and sign in again.")
+		return
+	}
+	s.clearAuthCookies(w)
 	writeJSON(w, http.StatusOK, map[string]any{
 		"data": map[string]any{
-			"applied": true,
-			"version": result.Latest,
-			"message": "Update verified and installed; the service is restarting.",
+			"applied":                   true,
+			"version":                   result.Latest,
+			"reauthentication_required": true,
+			"message":                   "Update verified and installed; all sessions were revoked and the service is restarting.",
 		},
 	})
 	if flusher, ok := w.(http.Flusher); ok {

internal/update/update.go

@@ -7,8 +7,8 @@
 // Trust model: GitHub TLS guarantees the channel; the repository owner controls
 // which assets are published; SHA256SUMS guards integrity. There is no GPG
 // signature verification — an accepted trade-off for a closed-network testing
-// tool. The web UI's check-update button remains an intentional no-op; only the
-// CLI performs code replacement.
+// tool. Both the CLI and authenticated web UI use this same verified replacement
+// path.
 package update
 
 import (
@@ -232,7 +232,10 @@ func RestartService(logger *slog.Logger) error {
 	if _, err := exec.LookPath("systemctl"); err != nil {
 		return fmt.Errorf("systemctl not found in PATH")
 	}
-	cmd := exec.Command("systemctl", "restart", "vocat")
+	// Queue the restart and let systemctl exit before systemd stops this unit.
+	// A blocking restart command becomes part of vocat.service's own cgroup and
+	// waits for that same cgroup to terminate, creating a stop-timeout cycle.
+	cmd := exec.Command("systemctl", "restart", "--no-block", "vocat")
 	if out, err := cmd.CombinedOutput(); err != nil {
 		logger.Warn("systemctl restart failed", "error", err, "output", string(out))
 		return fmt.Errorf("systemctl restart vocat: %w", err)

internal/vowifi/ims/sms_runtime_test.go

@@ -29,8 +29,9 @@ func TestSessionReceivesAndAcknowledgesSMSOverIMS(t *testing.T) {
 
 	received := make(chan ReceivedSMS, 1)
 	serverDone := make(chan error, 1)
+	readyForClose := make(chan struct{})
 	nonce := base64.StdEncoding.EncodeToString(make([]byte, 32))
-	go func() { serverDone <- serveInboundSMS(listener, nonce) }()
+	go func() { serverDone <- serveInboundSMS(listener, nonce, readyForClose) }()
 	provider, err := NewProvider(
 		smsTestAKA{&recordingAKA{result: vowifi.AKAResult{RES: []byte{1, 2, 3, 4}}}},
 		Config{
@@ -65,6 +66,11 @@ func TestSessionReceivesAndAcknowledgesSMSOverIMS(t *testing.T) {
 	case <-time.After(5 * time.Second):
 		t.Fatal("timed out waiting for inbound SMS")
 	}
+	select {
+	case <-readyForClose:
+	case <-time.After(5 * time.Second):
+		t.Fatal("timed out waiting for the inbound RP-ACK exchange")
+	}
 	if err := session.Close(context.Background()); err != nil {
 		t.Fatal(err)
 	}
@@ -102,9 +108,10 @@ func TestSessionSendsSMSOverIMS(t *testing.T) {
 	defer listener.Close()
 	_ = listener.SetDeadline(time.Now().Add(10 * time.Second))
 	serverDone := make(chan error, 1)
+	readyForClose := make(chan struct{})
 	statusReceived := make(chan ReceivedSMSStatus, 1)
 	nonce := base64.StdEncoding.EncodeToString(make([]byte, 32))
-	go func() { serverDone <- serveOutboundSMS(listener, nonce) }()
+	go func() { serverDone <- serveOutboundSMS(listener, nonce, readyForClose) }()
 	provider, err := NewProvider(
 		smsTestAKA{&recordingAKA{result: vowifi.AKAResult{RES: []byte{1, 2, 3, 4}}}},
 		Config{
@@ -145,6 +152,11 @@ func TestSessionSendsSMSOverIMS(t *testing.T) {
 	case <-time.After(5 * time.Second):
 		t.Fatal("timed out waiting for SMS delivery status")
 	}
+	select {
+	case <-readyForClose:
+	case <-time.After(5 * time.Second):
+		t.Fatal("timed out waiting for the status-report RP-ACK exchange")
+	}
 	if err := session.Close(context.Background()); err != nil {
 		t.Fatal(err)
 	}
@@ -153,7 +165,7 @@ func TestSessionSendsSMSOverIMS(t *testing.T) {
 	}
 }
 
-func serveInboundSMS(listener *net.UDPConn, nonce string) error {
+func serveInboundSMS(listener *net.UDPConn, nonce string, readyForClose chan<- struct{}) error {
 	packet := make([]byte, 65535)
 	count, remote, err := listener.ReadFromUDP(packet)
 	if err != nil {
@@ -229,6 +241,7 @@ func serveInboundSMS(listener *net.UDPConn, nonce string) error {
 	if _, err = listener.WriteToUDP(testResponse(200, "OK", report.Request.value("Call-ID"), report.Request.value("CSeq"), nil), remote); err != nil {
 		return err
 	}
+	close(readyForClose)
 
 	count, remote, err = listener.ReadFromUDP(packet)
 	if err != nil {
@@ -245,7 +258,7 @@ func serveInboundSMS(listener *net.UDPConn, nonce string) error {
 	return err
 }
 
-func serveOutboundSMS(listener *net.UDPConn, nonce string) error {
+func serveOutboundSMS(listener *net.UDPConn, nonce string, readyForClose chan<- struct{}) error {
 	packet := make([]byte, 65535)
 	count, remote, err := listener.ReadFromUDP(packet)
 	if err != nil {
@@ -355,6 +368,7 @@ func serveOutboundSMS(listener *net.UDPConn, nonce string) error {
 	if _, err = listener.WriteToUDP(testResponse(200, "OK", statusACK.Request.value("Call-ID"), statusACK.Request.value("CSeq"), nil), remote); err != nil {
 		return err
 	}
+	close(readyForClose)
 
 	count, remote, err = listener.ReadFromUDP(packet)
 	if err != nil {

scripts/install.sh

@@ -216,7 +216,9 @@ ExecStart=${BINARY_PATH}
 Restart=on-failure
 RestartSec=3s
 TimeoutStartSec=30s
-TimeoutStopSec=20s
+# HTTP, VoWiFi, and modem cleanup have bounded shutdown contexts totalling up
+# to 30 seconds. Leave a small margin before systemd resorts to SIGKILL.
+TimeoutStopSec=40s
 
 AmbientCapabilities=CAP_NET_ADMIN CAP_NET_RAW
 CapabilityBoundingSet=CAP_NET_ADMIN CAP_NET_RAW
@@ -229,7 +231,9 @@ ProtectKernelLogs=true
 ProtectKernelModules=true
 ProtectKernelTunables=true
 ProtectControlGroups=true
-ReadWritePaths=/opt/vocat/data
+# The web/CLI self-updater verifies a release in this directory and atomically
+# renames it over the running binary. Keep the rest of the host read-only.
+ReadWritePaths=/opt/vocat/data /opt/vocat/bin
 RestrictAddressFamilies=AF_UNIX AF_INET AF_INET6 AF_NETLINK
 RestrictRealtime=true
 LockPersonality=true