Merge remote-tracking branch 'origin/main' into fix_debug_upload_url_from_mgmt

# Conflicts:
#	shared/management/proto/management.pb.go

client/cmd/posture.go

@@ -0,0 +1,30 @@
+package cmd
+
+import (
+	"github.com/spf13/cobra"
+
+	log "github.com/sirupsen/logrus"
+
+	"github.com/netbirdio/netbird/client/internal/certproof"
+)
+
+var postureCmd = &cobra.Command{
+	Use:    "posture",
+	Short:  "Posture helpers invoked by the NetBird daemon",
+	Hidden: true,
+}
+
+var postureCertProofCmd = &cobra.Command{
+	Use:   "cert-proof",
+	Short: "Answer certificate posture challenges from the calling user's keychain",
+	Long: "Reads a certificate challenge set as JSON on stdin and writes the proofs as JSON on stdout.\n" +
+		"The daemon launches this in the console user's desktop session, because a login keychain\n" +
+		"cannot be reached from a root daemon. Not intended to be run by hand.",
+	Hidden:       true,
+	SilenceUsage: true,
+	RunE: func(cmd *cobra.Command, args []string) error {
+		// Proofs travel on stdout, so every log line has to go elsewhere.
+		log.SetOutput(cmd.ErrOrStderr())
+		return certproof.RunHelper(cmd.Context(), cmd.InOrStdin(), cmd.OutOrStdout())
+	},
+}

client/cmd/root.go

@@ -180,6 +180,9 @@ func init() {
 	rootCmd.AddCommand(debugCmd)
 	rootCmd.AddCommand(profileCmd)
 	rootCmd.AddCommand(exposeCmd)
+	rootCmd.AddCommand(postureCmd)
+
+	postureCmd.AddCommand(postureCertProofCmd)
 
 	networksCMD.AddCommand(routesListCmd)
 	networksCMD.AddCommand(routesSelectCmd, routesDeselectCmd)

client/internal/auth/sessionwatch/watcher.go

@@ -4,12 +4,20 @@
 // T-WarningLead notification and a dismiss-gated T-FinalWarningLead
 // fallback dialog.
 //
+// The deadline is an absolute wall-clock instant, so the watcher compares
+// it against the wall clock on a ticker rather than arming a relative
+// timer for it. A relative timer runs on the monotonic clock, which does
+// not advance while a device is suspended: it fires once that much awake
+// time has passed, which can be long after the deadline, and nothing
+// re-evaluates when the device wakes up. Polling makes every tick after a
+// resume (or after an NTP correction) see the real remaining time.
+//
 // The watcher is idempotent: Update may be called as often as the network
 // map snapshots arrive. Repeating the same deadline is a no-op; a new
-// deadline reschedules the timers and arms a fresh warning cycle.
+// deadline starts a fresh warning cycle.
 //
-// Warning firing is edge-detected. Each unique deadline value fires each
-// warning callback at most once.
+// Warning firing is edge-detected. Each unique deadline value publishes
+// each warning at most once.
 package sessionwatch
 
 import (
@@ -28,9 +36,15 @@ const (
 
 	// maxDeadlineHorizon caps how far in the future an accepted deadline
 	// can sit. A timestamp beyond this is almost certainly a protocol
-	// glitch, and silently arming a 100-year timer would hide the bug.
+	// glitch, and silently tracking a 100-year deadline would hide the bug.
 	maxDeadlineHorizon = 10 * 365 * 24 * time.Hour
 
+	// defaultEvalInterval is how often the tracked deadline is compared
+	// against the wall clock. The leads are minutes, so a coarse tick costs
+	// nothing in accuracy and keeps the wakeup cheap on battery-powered
+	// devices.
+	defaultEvalInterval = 10 * time.Second
+
 	// WarningLead is how far before expiry the first (interactive)
 	// warning fires. Drives the T-10 OS notification with
 	// Extend/Dismiss actions.
@@ -92,18 +106,21 @@ type StatusRecorder interface {
 type Watcher struct {
 	lead         time.Duration
 	finalLead    time.Duration
-	deadlineOnly bool
+	interval     time.Duration
+	deadlineOnly bool // record the deadline, leave the warnings to the caller
 
 	mu           sync.Mutex
 	current      time.Time
-	timer        *time.Timer
-	finalTimer   *time.Timer
-	firedAt      time.Time // deadline value the T-WarningLead callback last fired against
-	finalFiredAt time.Time // deadline value the T-FinalWarningLead callback last fired against
-	dismissedAt  time.Time // deadline value the user dismissed via Dismiss(); gates fireFinal
+	firedAt      time.Time // deadline value the T-WarningLead warning last published for
+	finalFiredAt time.Time // deadline value the T-FinalWarningLead warning last published for
+	dismissedAt  time.Time // deadline value the user dismissed via Dismiss(); gates the final warning
+	announcedAt  time.Time // deadline value the recorder has been told about; gates publishing
 	closed       bool
 	recorder     StatusRecorder
 	nowFn        func() time.Time
+	stop         chan struct{} // closed to stop the evaluation loop; nil while it is not running
+	done         chan struct{} // closed by the loop on its way out
+	wake         chan struct{} // buffered nudge asking the loop to evaluate before its next tick
 }
 
 // New returns a watcher with the package defaults WarningLead and
@@ -115,20 +132,24 @@ func New(recorder StatusRecorder) *Watcher {
 }
 
 // NewWithLeads returns a watcher with custom lead times. Useful for tests.
-// final must be strictly less than lead; otherwise both timers fire in the
-// wrong order or simultaneously and the UI flow breaks. A zero final lead
-// disables the final-warning timer entirely (see armTimerLocked) so a
-// millisecond-scale deadline doesn't flush both timers in one tick.
+// final must be strictly less than lead; otherwise the final warning takes
+// over the whole warning window and the interactive notification never
+// shows. A zero final lead disables the final warning entirely (see
+// evaluate), leaving the interactive one as the only warning.
 func NewWithLeads(lead, final time.Duration, recorder StatusRecorder) *Watcher {
 	return &Watcher{
 		lead:      lead,
 		finalLead: final,
+		interval:  defaultEvalInterval,
 		recorder:  recorder,
 		nowFn:     time.Now,
 	}
 }
 
-// NewDeadlineOnly returns a watcher that validates and records deadlines but arms no warning timers.
+// NewDeadlineOnly returns a watcher that validates and records deadlines
+// but publishes no warnings about them, and runs no evaluation loop to
+// decide. Used where the deadline is handed on to something that schedules
+// the warnings itself, such as the Android app.
 func NewDeadlineOnly(recorder StatusRecorder) *Watcher {
 	w := New(recorder)
 	w.deadlineOnly = true
@@ -139,11 +160,12 @@ func NewDeadlineOnly(recorder StatusRecorder) *Watcher {
 // a Sync push from the server omits the field because login expiration
 // was disabled).
 //
-// Same-value updates are no-ops. A different non-zero value cancels any
-// pending timer, resets the "already fired" guards, and — when the
-// deadline lies in the future — arms fresh warning timers. A deadline
-// already in the past (within maxPastHorizon) is recorded as-is with no
-// timers: the session has expired and consumers render it that way.
+// Same-value updates are no-ops. A different non-zero value resets the
+// "already fired" guards and starts a fresh warning cycle, evaluated
+// immediately so a deadline that already sits inside a warning window
+// warns without waiting for the next tick. A deadline already in the past
+// (within maxPastHorizon) is recorded as-is and warns nothing: the session
+// has expired and consumers render it that way.
 //
 // Returns one of the sentinel Err* values when the deadline fails the
 // sanity checks (pre-epoch, far future, or past beyond maxPastHorizon).
@@ -163,7 +185,7 @@ func (w *Watcher) Update(deadline time.Time) error {
 		return nil
 	}
 
-	now := time.Now()
+	now := w.nowFn()
 	switch {
 	case deadline.Before(time.Unix(0, 0)):
 		w.clearLocked()
@@ -181,18 +203,22 @@ func (w *Watcher) Update(deadline time.Time) error {
 		return nil
 	}
 
-	w.stopTimerLocked()
 	w.current = deadline
-	// Reset every per-deadline guard so a refreshed deadline arms a fresh
+	// Reset every per-deadline guard so a refreshed deadline starts a fresh
 	// warning cycle: both edge triggers and the user Dismiss decision
 	// (the user agreed to the old deadline expiring; a new deadline
 	// restarts the contract).
 	w.firedAt = time.Time{}
 	w.finalFiredAt = time.Time{}
 	w.dismissedAt = time.Time{}
+	w.announcedAt = time.Time{}
 
-	if deadline.After(now) && !w.deadlineOnly {
-		w.armTimerLocked(deadline)
+	// Poll every accepted deadline, including one that reads as already
+	// expired: the clock may be running ahead of real time and get
+	// corrected later, and evaluate ignores a deadline that has genuinely
+	// passed anyway.
+	if !w.deadlineOnly {
+		w.startPollLocked()
 	}
 	recorder := w.recorder
 	w.mu.Unlock()
@@ -200,6 +226,28 @@ func (w *Watcher) Update(deadline time.Time) error {
 		recorder.SetSessionExpiresAt(deadline)
 	}
 	log.Infof("auth session deadline set to: %s (in %s)", deadline.Format(time.RFC3339), time.Until(deadline).Round(time.Second))
+
+	// Open the gate only once the recorder knows the new deadline, so a
+	// warning that refers to it can never reach consumers before the state
+	// change itself. A tick landing in between finds the gate shut.
+	w.mu.Lock()
+	if w.closed || !w.current.Equal(deadline) {
+		w.mu.Unlock()
+		return nil
+	}
+	w.announcedAt = deadline
+	wake := w.wake
+	w.mu.Unlock()
+
+	// Hand the evaluation to the loop rather than running it here, so a
+	// deadline that already sits inside a warning window is published at
+	// once without this goroutine ever touching the recorder.
+	if wake != nil {
+		select {
+		case wake <- struct{}{}:
+		default:
+		}
+	}
 	return nil
 }
 
@@ -212,9 +260,9 @@ func (w *Watcher) Deadline() time.Time {
 }
 
 // Dismiss records the user's "Dismiss" action against the current deadline
-// and suppresses the upcoming final-warning callback for that deadline.
-// Idempotent: repeated calls are no-ops. A subsequent Update with a fresh
-// deadline resets the dismissal so the final-warning cycle re-arms.
+// and suppresses the final warning for that deadline. Idempotent: repeated
+// calls are no-ops. A subsequent Update with a fresh deadline resets the
+// dismissal so the final-warning cycle starts over.
 //
 // No-op when the watcher holds no deadline or has been closed.
 func (w *Watcher) Dismiss() {
@@ -227,35 +275,48 @@ func (w *Watcher) Dismiss() {
 		return
 	}
 	w.dismissedAt = w.current
-	// Cancel the armed final-warning timer eagerly. fireFinal would also
-	// gate on dismissedAt, but stopping the timer avoids a wakeup with
-	// nothing to do and makes the intent visible.
-	if w.finalTimer != nil {
-		w.finalTimer.Stop()
-		w.finalTimer = nil
-	}
 	log.Infof("auth session final-warning dismissed for deadline %s", w.current.Format(time.RFC3339))
 }
 
-// Close stops any pending timer. Update calls after Close are ignored.
-// The recorder keeps its deadline: the watcher is engine-scoped and closes
-// on every engine restart (network change, sleep/wake, stream errors)
-// while the SSO deadline stays valid across those, so clearing here would
-// blank the UI's "expires in" row on every transient reconnect. The
-// client run loop clears the server-scoped recorder when it exits for
-// real (Down, profile switch, permanent login failure).
+// Close stops the evaluation loop and waits for it to exit. Update calls
+// after Close are ignored. The recorder keeps its deadline: the watcher is
+// engine-scoped and closes on every engine restart (network change,
+// sleep/wake, stream errors) while the SSO deadline stays valid across
+// those, so clearing here would blank the UI's "expires in" row on every
+// transient reconnect. The client run loop clears the server-scoped
+// recorder when it exits for real (Down, profile switch, permanent login
+// failure).
 func (w *Watcher) Close() {
 	w.mu.Lock()
-	defer w.mu.Unlock()
 	if w.closed {
+		// A concurrent Close is already tearing down. w.done outlives it,
+		// so this caller waits for the same loop rather than returning
+		// while a warning is still on its way to the recorder.
+		done := w.done
+		w.mu.Unlock()
+		if done != nil {
+			<-done
+		}
 		return
 	}
 	w.closed = true
-	w.stopTimerLocked()
 	w.current = time.Time{}
 	w.firedAt = time.Time{}
 	w.finalFiredAt = time.Time{}
 	w.dismissedAt = time.Time{}
+	w.announcedAt = time.Time{}
+	// Copy the channels out before releasing the lock: the loop takes w.mu
+	// on every tick, so waiting for it while holding the lock would
+	// deadlock. w.done stays on the receiver for the branch above.
+	stop, done := w.stop, w.done
+	w.stop, w.wake = nil, nil
+	w.mu.Unlock()
+
+	if stop == nil {
+		return
+	}
+	close(stop)
+	<-done
 }
 
 // clearLocked drops the tracked deadline and notifies the recorder so
@@ -267,11 +328,11 @@ func (w *Watcher) clearLocked() {
 		w.mu.Unlock()
 		return
 	}
-	w.stopTimerLocked()
 	w.current = time.Time{}
 	w.firedAt = time.Time{}
 	w.finalFiredAt = time.Time{}
 	w.dismissedAt = time.Time{}
+	w.announcedAt = time.Time{}
 	recorder := w.recorder
 	w.mu.Unlock()
 	if recorder != nil {
@@ -280,133 +341,121 @@ func (w *Watcher) clearLocked() {
 	log.Infof("auth session deadline cleared")
 }
 
-func (w *Watcher) stopTimerLocked() {
-	if w.timer != nil {
-		w.timer.Stop()
-		w.timer = nil
+// startPollLocked starts the evaluation loop unless it is already
+// running. The loop starts lazily on the first accepted deadline, so a
+// client whose server never publishes a session expiry never pays for a
+// ticker, and it runs until Close: the watcher is engine-scoped, and a
+// cleared deadline is normally followed by a fresh one on the next sync.
+// Caller must hold w.mu.
+func (w *Watcher) startPollLocked() {
+	if w.stop != nil {
+		return
 	}
-	if w.finalTimer != nil {
-		w.finalTimer.Stop()
-		w.finalTimer = nil
+	stop := make(chan struct{})
+	done := make(chan struct{})
+	wake := make(chan struct{}, 1)
+	w.stop, w.done, w.wake = stop, done, wake
+	go w.poll(stop, done, wake, w.interval)
+}
+
+// poll re-evaluates the tracked deadline every interval, and as soon as a
+// new deadline is announced. It is the only caller of evaluate, so Close
+// waiting for it to exit is enough to know no warning is still on its way
+// to the recorder. Its channels and interval are passed in rather than
+// read off the receiver, so Close can clear them without racing it.
+func (w *Watcher) poll(stop <-chan struct{}, done chan<- struct{}, wake <-chan struct{}, interval time.Duration) {
+	defer close(done)
+
+	ticker := time.NewTicker(interval)
+	defer ticker.Stop()
+
+	for {
+		select {
+		case <-stop:
+			return
+		case <-wake:
+			w.evaluate()
+		case <-ticker.C:
+			w.evaluate()
+		}
 	}
 }
 
-func (w *Watcher) armTimerLocked(deadline time.Time) {
-	w.timer = armOneShotLocked(deadline.Add(-w.lead), func() { w.fire(deadline) })
-	// finalLead <= 0 disables the final-warning timer entirely. Used by
-	// tests that predate the final-warning fallback so a millisecond-scale
-	// deadline does not flush both timers at once.
-	if w.finalLead > 0 {
-		w.finalTimer = armOneShotLocked(deadline.Add(-w.finalLead), func() { w.fireFinal(deadline) })
-	}
-}
-
-func (w *Watcher) fire(armedFor time.Time) {
+// evaluate compares the tracked deadline against the wall clock and
+// publishes whichever warning the remaining time calls for. Inside the
+// final-warning window the interactive warning is stale, so the final one
+// is published in its place: a device that resumes there never saw the
+// T-WarningLead notification.
+func (w *Watcher) evaluate() {
 	w.mu.Lock()
-	if w.closed || !w.current.Equal(armedFor) {
-		// Deadline moved while we were waiting (e.g. a successful extend).
-		// The reschedule path armed a fresh timer; this one is stale.
+	if w.closed || w.deadlineOnly || w.current.IsZero() {
 		w.mu.Unlock()
 		return
 	}
-	if !w.firedAt.IsZero() && w.firedAt.Equal(armedFor) {
-		w.mu.Unlock()
-		return
-	}
-	now := w.nowFn()
-	if isLate(now, armedFor, max(w.finalLead, 0)) {
-		w.fireLateLocked(armedFor, now)
-		return
-	}
-	w.firedAt = armedFor
-	recorder := w.recorder
-	w.mu.Unlock()
-	if recorder == nil {
-		return
-	}
-	log.Infof("auth session expiry soon warning fired")
-	publishWarning(recorder, armedFor, false)
-}
 
-// fireFinal mirrors fire for the T-FinalWarningLead timer with an extra
-// dismiss-gate: if the user dismissed the T-WarningLead notification for
-// this deadline, the final warning is suppressed entirely.
-func (w *Watcher) fireFinal(armedFor time.Time) {
-	w.mu.Lock()
-	if w.closed || !w.current.Equal(armedFor) {
+	deadline := w.current
+	if !w.announcedAt.Equal(deadline) {
+		// Update is still on its way to the recorder with this deadline.
 		w.mu.Unlock()
 		return
 	}
-	if !w.finalFiredAt.IsZero() && w.finalFiredAt.Equal(armedFor) {
-		w.mu.Unlock()
-		return
-	}
-	if w.dismissedAt.Equal(armedFor) {
-		w.mu.Unlock()
-		log.Infof("auth session final-warning skipped (dismissed by user)")
-		return
-	}
-	now := w.nowFn()
-	if isLate(now, armedFor, 0) {
-		w.finalFiredAt = armedFor
-		w.mu.Unlock()
-		log.Infof("auth session final-warning skipped for deadline %s (passed %s ago)",
-			armedFor.Format(time.RFC3339), now.Round(0).Sub(armedFor).Round(time.Second))
-		return
-	}
-	w.finalFiredAt = armedFor
-	recorder := w.recorder
-	w.mu.Unlock()
-	if recorder == nil {
-		return
-	}
-	log.Infof("auth session final-warning fired")
-	publishWarning(recorder, armedFor, true)
-}
+	// Round(0) strips the monotonic reading so the comparison is wall
+	// clock on both sides, whether the deadline came off the wire or from
+	// a caller that derived it from time.Now.
+	remaining := deadline.Round(0).Sub(w.nowFn().Round(0))
 
-// fireLateLocked handles a T-WarningLead callback that fired inside the
-// final-warning window: it sends the final warning in its place while the
-// deadline has not passed and the user has not dismissed it, so a resume
-// with time left still warns. The caller must hold w.mu; this helper
-// releases it.
-func (w *Watcher) fireLateLocked(armedFor, now time.Time) {
-	w.firedAt = armedFor
 	switch {
-	case w.dismissedAt.Equal(armedFor):
+	case remaining <= 0:
+		// Already expired: the post-mortem SessionExpired flow owns it.
 		w.mu.Unlock()
-		log.Infof("auth session expiry soon warning skipped (dismissed by user)")
-		return
-	case w.finalFiredAt.Equal(armedFor):
+	case w.finalLead > 0 && remaining <= w.finalLead:
+		w.publishFinalLocked(deadline, remaining)
+	case remaining <= w.lead:
+		w.publishWarningLocked(deadline, remaining)
+	default:
 		w.mu.Unlock()
-		log.Infof("auth session expiry soon warning skipped (final warning already fired)")
-		return
-	case isLate(now, armedFor, 0):
+	}
+}
+
+// publishWarningLocked emits the interactive T-WarningLead warning, at
+// most once per deadline value. Caller must hold w.mu; this helper
+// releases it.
+func (w *Watcher) publishWarningLocked(deadline time.Time, remaining time.Duration) {
+	if w.firedAt.Equal(deadline) {
 		w.mu.Unlock()
-		log.Infof("auth session expiry soon warning skipped for deadline %s (passed %s ago)",
-			armedFor.Format(time.RFC3339), now.Round(0).Sub(armedFor).Round(time.Second))
 		return
 	}
-	w.finalFiredAt = armedFor
+	w.firedAt = deadline
 	recorder := w.recorder
 	w.mu.Unlock()
 	if recorder == nil {
 		return
 	}
-	log.Infof("auth session expiry soon warning fired inside the final-warning window, sending final warning for deadline %s",
-		armedFor.Format(time.RFC3339))
-	publishWarning(recorder, armedFor, true)
+	log.Infof("auth session expiry soon warning fired for deadline %s (in %s)",
+		deadline.Format(time.RFC3339), remaining.Round(time.Second))
+	publishWarning(recorder, deadline, false)
 }
 
-// armOneShotLocked schedules cb at fireAt. When fireAt is already in the
-// past it dispatches on the next scheduler tick so a state-change recorder
-// notification (invoked after w.mu is released) lands first. Caller must
-// hold w.mu.
-func armOneShotLocked(fireAt time.Time, cb func()) *time.Timer {
-	delay := time.Until(fireAt)
-	if delay <= 0 {
-		return time.AfterFunc(0, cb)
+// publishFinalLocked emits the final warning, at most once per deadline
+// value and never once the user dismissed that deadline. It marks the
+// interactive warning as handled too: the final window is open, so a
+// "expires in WarningLead minutes" notification would be wrong. Caller
+// must hold w.mu; this helper releases it.
+func (w *Watcher) publishFinalLocked(deadline time.Time, remaining time.Duration) {
+	if w.finalFiredAt.Equal(deadline) || w.dismissedAt.Equal(deadline) {
+		w.mu.Unlock()
+		return
 	}
-	return time.AfterFunc(delay, cb)
+	w.firedAt = deadline
+	w.finalFiredAt = deadline
+	recorder := w.recorder
+	w.mu.Unlock()
+	if recorder == nil {
+		return
+	}
+	log.Infof("auth session final-warning fired for deadline %s (in %s)",
+		deadline.Format(time.RFC3339), remaining.Round(time.Second))
+	publishWarning(recorder, deadline, true)
 }
 
 // publishWarning composes the SystemEvent for a watcher-fired warning and
@@ -436,11 +485,3 @@ func publishWarning(recorder StatusRecorder, deadline time.Time, final bool) {
 		meta,
 	)
 }
-
-// isLate reports whether the wall clock now has already reached armedFor
-// minus cutoffLead. The timers run on the monotonic clock, which can stall
-// while the host sleeps, so a timer can fire long after the window it was
-// armed for.
-func isLate(now, armedFor time.Time, cutoffLead time.Duration) bool {
-	return !now.Round(0).Before(armedFor.Add(-cutoffLead).Round(0))
-}

client/internal/auth/sessionwatch/watcher_test.go

@@ -19,6 +19,9 @@ type fakeRecorder struct {
 	mu           sync.Mutex
 	events       []event
 	lastDeadline time.Time
+	// setDelay stalls SetSessionExpiresAt, widening the window in which a
+	// concurrent evaluation could publish a warning out of order.
+	setDelay time.Duration
 }
 
 type eventKind int
@@ -43,6 +46,12 @@ type event struct {
 // is the zero time, so an initial clear before any deadline is set emits
 // nothing — matching the real recorder.
 func (r *fakeRecorder) SetSessionExpiresAt(deadline time.Time) {
+	r.mu.Lock()
+	delay := r.setDelay
+	r.mu.Unlock()
+	// Stall without the lock held, so a concurrent publish can still record.
+	time.Sleep(delay)
+
 	r.mu.Lock()
 	defer r.mu.Unlock()
 	if r.lastDeadline.Equal(deadline) {
@@ -116,10 +125,54 @@ func waitForEvents(t *testing.T, r *fakeRecorder, want int) []event {
 	return nil
 }
 
-// newWatcher builds a watcher with the final timer disabled (finalLead=0),
+// testInterval keeps the ticker-driven tests fast; the leads they use are
+// in the same millisecond scale.
+const testInterval = 2 * time.Millisecond
+
+// newWatcher builds a watcher with the final warning disabled (finalLead=0),
 // matching the lead-only behaviour the pre-final-warning tests assume.
 func newWatcher(lead time.Duration, r *fakeRecorder) *Watcher {
-	return NewWithLeads(lead, 0, r)
+	return newWatcherWithLeads(lead, 0, r)
+}
+
+// newWatcherWithLeads builds a watcher that evaluates on testInterval. The
+// interval is set before Update, so the evaluation loop does not exist yet
+// and the write cannot race it.
+func newWatcherWithLeads(lead, final time.Duration, r StatusRecorder) *Watcher {
+	w := NewWithLeads(lead, final, r)
+	w.interval = testInterval
+	return w
+}
+
+// fakeClock is the watcher's wall clock under test. Reads come from the
+// evaluation goroutine while the test writes, so both go through the mutex.
+type fakeClock struct {
+	mu sync.Mutex
+	t  time.Time
+}
+
+func newFakeClock(t time.Time) *fakeClock {
+	return &fakeClock{t: t}
+}
+
+func (c *fakeClock) now() time.Time {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	return c.t
+}
+
+// set jumps the clock, standing in for a resume from suspension or for an
+// NTP correction.
+func (c *fakeClock) set(t time.Time) {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	c.t = t
+}
+
+// settle waits out a handful of evaluation ticks so a test can assert that
+// nothing was published.
+func settle() {
+	time.Sleep(20 * testInterval)
 }
 
 func TestUpdateZeroBeforeAnythingIsNoop(t *testing.T) {
@@ -410,7 +463,7 @@ func TestCloseWithoutDeadlineLeavesRecorderUntouched(t *testing.T) {
 func TestFinalWarningFiresAfterRegularWarning(t *testing.T) {
 	r := &fakeRecorder{}
 	// Warning fires at deadline-80ms, final at deadline-30ms.
-	w := NewWithLeads(80*time.Millisecond, 30*time.Millisecond, r)
+	w := newWatcherWithLeads(80*time.Millisecond, 30*time.Millisecond, r)
 	defer w.Close()
 
 	d := time.Now().Add(100 * time.Millisecond)
@@ -446,7 +499,7 @@ func TestFinalWarningFiresAfterRegularWarning(t *testing.T) {
 
 func TestDismissSuppressesFinalWarning(t *testing.T) {
 	r := &fakeRecorder{}
-	w := NewWithLeads(80*time.Millisecond, 30*time.Millisecond, r)
+	w := newWatcherWithLeads(80*time.Millisecond, 30*time.Millisecond, r)
 	defer w.Close()
 
 	d := time.Now().Add(100 * time.Millisecond)
@@ -477,7 +530,7 @@ func TestDismissSuppressesFinalWarning(t *testing.T) {
 
 func TestDismissResetByNewDeadline(t *testing.T) {
 	r := &fakeRecorder{}
-	w := NewWithLeads(80*time.Millisecond, 30*time.Millisecond, r)
+	w := newWatcherWithLeads(80*time.Millisecond, 30*time.Millisecond, r)
 	defer w.Close()
 
 	first := time.Now().Add(100 * time.Millisecond)
@@ -507,7 +560,7 @@ func TestDismissResetByNewDeadline(t *testing.T) {
 
 func TestDismissBeforeUpdateIsNoop(t *testing.T) {
 	r := &fakeRecorder{}
-	w := NewWithLeads(80*time.Millisecond, 30*time.Millisecond, r)
+	w := newWatcherWithLeads(80*time.Millisecond, 30*time.Millisecond, r)
 	defer w.Close()
 
 	// No deadline tracked yet; Dismiss must be a no-op (no panic, no state).
@@ -528,157 +581,139 @@ func TestDismissBeforeUpdateIsNoop(t *testing.T) {
 	t.Fatalf("final-warning did not publish after no-op pre-Update Dismiss, events=%+v", r.snapshot())
 }
 
-func TestIsLate(t *testing.T) {
-	armedFor := time.Date(2026, 10, 1, 12, 0, 0, 0, time.UTC)
-	lead := 2 * time.Minute
-	tests := []struct {
-		name       string
-		now        time.Time
-		cutoffLead time.Duration
-		want       bool
-	}{
-		{"before cutoff", armedFor.Add(-3 * time.Minute), lead, false},
-		{"at cutoff", armedFor.Add(-lead), lead, true},
-		{"after cutoff", armedFor.Add(-time.Minute), lead, true},
-		{"zero lead before deadline", armedFor.Add(-time.Second), 0, false},
-		{"zero lead at deadline", armedFor, 0, true},
-		{"zero lead after deadline", armedFor.Add(time.Second), 0, true},
-	}
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			if got := isLate(tt.now, armedFor, tt.cutoffLead); got != tt.want {
-				t.Fatalf("isLate(%s, %s, %s) = %v, want %v", tt.now, armedFor, tt.cutoffLead, got, tt.want)
-			}
-		})
-	}
-}
+// The tests below drive the watcher's wall clock directly. The deadline sits
+// an hour out in real time and the fake clock jumps, standing in for a device
+// that was suspended and resumed somewhere inside — or past — the warning
+// windows. The evaluation loop is what reacts to the jump, so these exercise
+// the same path production takes on a resume.
+func newResumeWatcher(t *testing.T, r *fakeRecorder) (*Watcher, *fakeClock, time.Time) {
+	t.Helper()
 
-func TestIsLateIgnoresMonotonicReading(t *testing.T) {
-	now := time.Now()
-	wallOnly := now.Round(0)
-	if isLate(now, wallOnly.Add(time.Second), 0) {
-		t.Fatalf("now with monotonic reading must compare as wall clock before a later wall-only deadline")
-	}
-	if !isLate(now, wallOnly, 0) {
-		t.Fatalf("now with monotonic reading must compare as wall clock at an equal wall-only deadline")
-	}
-}
+	w := newWatcherWithLeads(WarningLead, FinalWarningLead, r)
+	start := time.Now()
+	clock := newFakeClock(start)
+	// Set before Update: the evaluation loop does not exist yet.
+	w.nowFn = clock.now
+	t.Cleanup(w.Close)
 
-func TestLateTimerFiring(t *testing.T) {
-	tests := []struct {
-		name       string
-		final      bool
-		beforeDl   time.Duration
-		wantWarns  int
-		wantFinals int
-	}{
-		{"warning on resume inside window", false, 3 * time.Minute, 1, 0},
-		{"warning promoted to final inside final window", false, time.Minute, 0, 1},
-		{"warning skipped past deadline", false, -time.Minute, 0, 0},
-		{"final on resume before deadline", true, time.Minute, 0, 1},
-		{"final skipped past deadline", true, -time.Minute, 0, 0},
-	}
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			r := &fakeRecorder{}
-			w := New(r)
-			defer w.Close()
-
-			// The deadline is an hour out so the real timers never fire
-			// during the test; the late callback is invoked directly with an
-			// injected clock that simulates a resume near the deadline.
-			d := time.Now().Add(time.Hour).Round(0)
-			w.nowFn = func() time.Time { return d.Add(-tt.beforeDl) }
-			if err := w.Update(d); err != nil {
-				t.Fatalf("Update: %v", err)
-			}
-
-			if tt.final {
-				w.fireFinal(d)
-			} else {
-				w.fire(d)
-			}
-
-			events := r.snapshot()
-			if got := countWhere(events, event.isWarning); got != tt.wantWarns {
-				t.Fatalf("expected %d warning publishes, got %d: %+v", tt.wantWarns, got, events)
-			}
-			if got := countWhere(events, event.isFinalWarning); got != tt.wantFinals {
-				t.Fatalf("expected %d final-warning publishes, got %d: %+v", tt.wantFinals, got, events)
-			}
-		})
-	}
-}
-
-func TestPromotedFinalWarningIsNotRepeated(t *testing.T) {
-	r := &fakeRecorder{}
-	w := New(r)
-	defer w.Close()
-
-	d := time.Now().Add(time.Hour).Round(0)
-	now := d.Add(-time.Minute)
-	w.nowFn = func() time.Time { return now }
-	if err := w.Update(d); err != nil {
+	deadline := start.Add(time.Hour).Round(0)
+	if err := w.Update(deadline); err != nil {
 		t.Fatalf("Update: %v", err)
 	}
+	return w, clock, deadline
+}
 
-	w.fire(d)
-	// The final timer was suspended too, so it fires even later than the
-	// warning timer, here still just before the deadline.
-	now = d.Add(-30 * time.Second)
-	w.fireFinal(d)
+func TestResumeInsideWarningWindowWarns(t *testing.T) {
+	r := &fakeRecorder{}
+	_, clock, deadline := newResumeWatcher(t, r)
 
-	events := r.snapshot()
-	if got := countWhere(events, event.isFinalWarning); got != 1 {
-		t.Fatalf("expected exactly 1 final-warning publish, got %d: %+v", got, events)
+	clock.set(deadline.Add(-5 * time.Minute))
+
+	events := waitForEvents(t, r, 2)
+	if !events[1].isWarning() {
+		t.Fatalf("expected the interactive warning after the resume, got %+v", events[1])
 	}
-	if got := countWhere(events, event.isWarning); got != 0 {
-		t.Fatalf("expected no regular warning publish, got %d: %+v", got, events)
+	if n := countWhere(events, event.isFinalWarning); n != 0 {
+		t.Fatalf("final-warning must wait for its own window, got %d: %+v", n, events)
 	}
 }
 
-func TestPromotionRespectsDismiss(t *testing.T) {
+func TestResumeInsideFinalWindowSendsFinalWarningOnly(t *testing.T) {
 	r := &fakeRecorder{}
-	w := New(r)
-	defer w.Close()
+	_, clock, deadline := newResumeWatcher(t, r)
 
-	d := time.Now().Add(time.Hour).Round(0)
-	w.nowFn = func() time.Time { return d.Add(-time.Minute) }
-	if err := w.Update(d); err != nil {
-		t.Fatalf("Update: %v", err)
+	clock.set(deadline.Add(-time.Minute))
+
+	events := waitForEvents(t, r, 2)
+	if !events[1].isFinalWarning() {
+		t.Fatalf("expected the final warning after the resume, got %+v", events[1])
+	}
+	settle()
+	if n := countWhere(r.snapshot(), event.isWarning); n != 0 {
+		t.Fatalf("the interactive warning is stale inside the final window, got %d: %+v", n, r.snapshot())
+	}
+}
+
+func TestResumePastDeadlinePublishesNothing(t *testing.T) {
+	r := &fakeRecorder{}
+	_, clock, deadline := newResumeWatcher(t, r)
+
+	clock.set(deadline.Add(time.Minute))
+
+	settle()
+	if n := countWhere(r.snapshot(), func(e event) bool { return e.kind == publish }); n != 0 {
+		t.Fatalf("an expired session must not warn, got %d publishes: %+v", n, r.snapshot())
+	}
+}
+
+func TestWarningPublishesOncePerDeadline(t *testing.T) {
+	r := &fakeRecorder{}
+	_, clock, deadline := newResumeWatcher(t, r)
+
+	clock.set(deadline.Add(-5 * time.Minute))
+	waitForEvents(t, r, 2)
+
+	// Many ticks pass inside the same window.
+	settle()
+	if n := countWhere(r.snapshot(), event.isWarning); n != 1 {
+		t.Fatalf("expected exactly 1 warning publish across ticks, got %d: %+v", n, r.snapshot())
+	}
+}
+
+// TestWarningRecoversFromAClockRunningAhead covers the device that boots
+// before NTP has corrected it: the deadline looks long gone, nothing is
+// published, and the warning still arrives once the clock is fixed.
+func TestWarningRecoversFromAClockRunningAhead(t *testing.T) {
+	r := &fakeRecorder{}
+	_, clock, deadline := newResumeWatcher(t, r)
+
+	clock.set(deadline.Add(2 * time.Hour))
+	settle()
+	if n := countWhere(r.snapshot(), func(e event) bool { return e.kind == publish }); n != 0 {
+		t.Fatalf("a deadline that looks expired must not warn, got %d publishes: %+v", n, r.snapshot())
 	}
 
+	clock.set(deadline.Add(-5 * time.Minute))
+
+	events := waitForEvents(t, r, 2)
+	if !events[1].isWarning() {
+		t.Fatalf("expected the warning once the clock was corrected, got %+v", events[1])
+	}
+}
+
+func TestResumeInsideFinalWindowRespectsDismiss(t *testing.T) {
+	r := &fakeRecorder{}
+	w, clock, deadline := newResumeWatcher(t, r)
+
+	// The user dismissed the warning for this deadline before the device
+	// was suspended; resuming inside the final window must not reopen it.
 	w.Dismiss()
-	w.fire(d)
+	clock.set(deadline.Add(-time.Minute))
 
-	events := r.snapshot()
-	if got := countWhere(events, func(e event) bool { return e.kind == publish }); got != 0 {
-		t.Fatalf("expected no publish after dismiss, got %d: %+v", got, events)
+	settle()
+	if n := countWhere(r.snapshot(), func(e event) bool { return e.kind == publish }); n != 0 {
+		t.Fatalf("a dismissed deadline must not warn on resume, got %d: %+v", n, r.snapshot())
 	}
 }
 
-func TestPromotionSkippedWhenFinalAlreadyFired(t *testing.T) {
+func TestCloseStopsTheEvaluationLoop(t *testing.T) {
 	r := &fakeRecorder{}
-	w := New(r)
-	defer w.Close()
+	w := newWatcherWithLeads(WarningLead, FinalWarningLead, r)
+	start := time.Now()
+	clock := newFakeClock(start)
+	w.nowFn = clock.now
 
-	// Both timers fall in the past after a long suspend and are dispatched
-	// with a zero delay, so the final callback can run before the warning one.
-	d := time.Now().Add(time.Hour).Round(0)
-	w.nowFn = func() time.Time { return d.Add(-time.Minute) }
-	if err := w.Update(d); err != nil {
+	deadline := start.Add(time.Hour).Round(0)
+	if err := w.Update(deadline); err != nil {
 		t.Fatalf("Update: %v", err)
 	}
+	w.Close()
 
-	w.fireFinal(d)
-	w.fire(d)
+	clock.set(deadline.Add(-5 * time.Minute))
+	settle()
 
-	events := r.snapshot()
-	if got := countWhere(events, event.isFinalWarning); got != 1 {
-		t.Fatalf("expected exactly 1 final-warning publish, got %d: %+v", got, events)
-	}
-	if got := countWhere(events, event.isWarning); got != 0 {
-		t.Fatalf("expected no regular warning publish, got %d: %+v", got, events)
+	if n := countWhere(r.snapshot(), func(e event) bool { return e.kind == publish }); n != 0 {
+		t.Fatalf("a closed watcher must not publish, got %d: %+v", n, r.snapshot())
 	}
 }
 
@@ -687,8 +722,8 @@ func TestDeadlineOnlyRecordsDeadlineWithoutWarnings(t *testing.T) {
 	w := NewDeadlineOnly(r)
 	defer w.Close()
 
-	// With the default leads this deadline would otherwise fire both
-	// timers on the next tick.
+	// With the default leads this deadline sits inside the final-warning
+	// window, so a watcher that warns at all would publish on the spot.
 	d := time.Now().Add(50 * time.Millisecond).Round(0)
 	if err := w.Update(d); err != nil {
 		t.Fatalf("Update: %v", err)
@@ -703,8 +738,11 @@ func TestDeadlineOnlyRecordsDeadlineWithoutWarnings(t *testing.T) {
 	if got := countWhere(events, func(e event) bool { return e.kind == publish }); got != 0 {
 		t.Fatalf("expected no publish in deadline-only mode, got %d: %+v", got, events)
 	}
-	if w.timer != nil || w.finalTimer != nil {
-		t.Fatal("expected no timers armed in deadline-only mode")
+	w.mu.Lock()
+	polling := w.stop != nil
+	w.mu.Unlock()
+	if polling {
+		t.Fatal("expected no evaluation loop in deadline-only mode")
 	}
 }
 
@@ -725,3 +763,157 @@ func TestDeadlineOnlyStillRejectsOutOfRangeDeadlines(t *testing.T) {
 		t.Fatalf("expected recorder cleared after rejection, got %v", got)
 	}
 }
+
+// TestClockAheadAtUpdateStillWarnsOnceCorrected covers a client that connects
+// while its clock runs ahead of real time: the deadline reads as already
+// expired when it arrives, so nothing publishes, and the warning must still
+// come once NTP pulls the clock back.
+func TestClockAheadAtUpdateStillWarnsOnceCorrected(t *testing.T) {
+	r := &fakeRecorder{}
+	w := newWatcherWithLeads(WarningLead, FinalWarningLead, r)
+	deadline := time.Now().Add(time.Hour).Round(0)
+	// Two hours past the deadline from the device's point of view, well
+	// inside maxPastHorizon, so Update accepts and records it.
+	clock := newFakeClock(deadline.Add(2 * time.Hour))
+	w.nowFn = clock.now
+	t.Cleanup(w.Close)
+
+	if err := w.Update(deadline); err != nil {
+		t.Fatalf("Update: %v", err)
+	}
+
+	settle()
+	if n := countWhere(r.snapshot(), func(e event) bool { return e.kind == publish }); n != 0 {
+		t.Fatalf("a deadline that reads as expired must not warn, got %d: %+v", n, r.snapshot())
+	}
+
+	clock.set(deadline.Add(-5 * time.Minute))
+
+	events := waitForEvents(t, r, 2)
+	if !events[1].isWarning() {
+		t.Fatalf("expected the warning once the clock was corrected, got %+v", events[1])
+	}
+}
+
+// TestWarningNeverPrecedesTheDeadlineStateChange pins the ordering Update
+// documents: consumers learn the new deadline before they see a warning that
+// refers to it. The deadline here already sits inside the warning window, and
+// the recorder stalls, so the evaluation loop gets many chances to publish
+// while Update is still announcing.
+func TestWarningNeverPrecedesTheDeadlineStateChange(t *testing.T) {
+	r := &fakeRecorder{setDelay: 50 * time.Millisecond}
+	w := newWatcherWithLeads(WarningLead, FinalWarningLead, r)
+	deadline := time.Now().Add(time.Hour).Round(0)
+	clock := newFakeClock(deadline.Add(-5 * time.Minute))
+	w.nowFn = clock.now
+	t.Cleanup(w.Close)
+
+	if err := w.Update(deadline); err != nil {
+		t.Fatalf("Update: %v", err)
+	}
+
+	events := waitForEvents(t, r, 2)
+	if events[0].kind != stateChange {
+		t.Fatalf("event[0] should be the deadline state change, got %+v", events)
+	}
+	if !events[1].isWarning() {
+		t.Fatalf("event[1] should be the warning, got %+v", events)
+	}
+}
+
+// TestUpdateWakesTheLoopWithoutWaitingForATick pins the hand-off: Update
+// publishes nothing itself, it nudges the evaluation loop, so a deadline that
+// already sits inside a warning window is warned about at once even though the
+// ticker here would not fire for an hour.
+func TestUpdateWakesTheLoopWithoutWaitingForATick(t *testing.T) {
+	r := &fakeRecorder{}
+	w := newWatcherWithLeads(WarningLead, FinalWarningLead, r)
+	w.interval = time.Hour
+	t.Cleanup(w.Close)
+
+	d := time.Now().Add(5 * time.Minute).Round(0)
+	if err := w.Update(d); err != nil {
+		t.Fatalf("Update: %v", err)
+	}
+
+	events := waitForEvents(t, r, 2)
+	if !events[1].isWarning() {
+		t.Fatalf("expected the warning on the wake-up, got %+v", events[1])
+	}
+}
+
+// blockingRecorder holds a publish open until the test releases it, so the
+// evaluation loop can be parked mid-publish while Close is called.
+type blockingRecorder struct {
+	fakeRecorder
+	entered chan struct{}
+	release chan struct{}
+}
+
+func (r *blockingRecorder) PublishEvent(
+	severity cProto.SystemEvent_Severity,
+	category cProto.SystemEvent_Category,
+	message string,
+	userMessage string,
+	metadata map[string]string,
+) {
+	select {
+	case r.entered <- struct{}{}:
+	default:
+	}
+	<-r.release
+	r.fakeRecorder.PublishEvent(severity, category, message, userMessage, metadata)
+}
+
+// TestConcurrentCloseWaitsForTheLoop pins the contract for the caller that
+// loses the race: Close returns only once the loop is done, whichever of the
+// two calls got there first.
+func TestConcurrentCloseWaitsForTheLoop(t *testing.T) {
+	r := &blockingRecorder{entered: make(chan struct{}, 1), release: make(chan struct{})}
+	var releaseOnce sync.Once
+	release := func() { releaseOnce.Do(func() { close(r.release) }) }
+
+	w := newWatcherWithLeads(WarningLead, FinalWarningLead, r)
+	w.interval = time.Hour
+	// Order matters: Close waits for the loop, which stays parked in the
+	// publish until the release, so a t.Fatal below would hang the cleanup
+	// instead of reporting the failure.
+	t.Cleanup(func() {
+		release()
+		w.Close()
+	})
+
+	d := time.Now().Add(5 * time.Minute).Round(0)
+	if err := w.Update(d); err != nil {
+		t.Fatalf("Update: %v", err)
+	}
+
+	// The loop is now parked inside the warning publish.
+	select {
+	case <-r.entered:
+	case <-time.After(2 * time.Second):
+		t.Fatal("the loop never reached the publish")
+	}
+
+	first := make(chan struct{})
+	second := make(chan struct{})
+	go func() { w.Close(); close(first) }()
+	go func() { w.Close(); close(second) }()
+
+	select {
+	case <-first:
+		t.Fatal("Close returned while the loop was still publishing")
+	case <-second:
+		t.Fatal("Close returned while the loop was still publishing")
+	case <-time.After(100 * time.Millisecond):
+	}
+
+	release()
+	for _, done := range []chan struct{}{first, second} {
+		select {
+		case <-done:
+		case <-time.After(2 * time.Second):
+			t.Fatal("Close did not return after the publish completed")
+		}
+	}
+}

client/internal/certproof/collect.go

@@ -0,0 +1,158 @@
+package certproof
+
+import (
+	"context"
+	"crypto"
+	"crypto/sha256"
+	"crypto/x509"
+	"time"
+
+	log "github.com/sirupsen/logrus"
+	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
+
+	"github.com/netbirdio/netbird/shared/management/certposture"
+	"github.com/netbirdio/netbird/shared/management/proto"
+)
+
+// Collect answers the certificate challenges in checks: for each challenge it picks a
+// stored certificate that chains to the challenge's CAs and signs the nonce with its
+// key. The same certificate is proven once even if several checks accept it.
+func Collect(ctx context.Context, store Store, checks []*proto.Checks, peerKey []byte) []certposture.Proof {
+	challenges := certificateChallenges(checks)
+	if len(challenges) == 0 {
+		logNoChallenges(checks)
+		return nil
+	}
+	return CollectChallenges(ctx, store, challenges, peerKey)
+}
+
+func logNoChallenges(checks []*proto.Checks) {
+	if len(checks) > 0 {
+		log.Debugf("certificate posture: %d posture checks received, none carries a certificate challenge", len(checks))
+	}
+}
+
+// CollectChallenges answers challenges already extracted from the posture checks, so a
+// caller that ships them across a process boundary reuses the same matching and signing.
+// Only nonces of the size management issues are signed, for a peer key of the size of
+// ours, so the keys behind the store never sign arbitrary caller-chosen data.
+func CollectChallenges(ctx context.Context, store Store, challenges []*proto.CertificateChallenge, peerKey []byte) []certposture.Proof {
+	if len(peerKey) != wgtypes.KeyLen {
+		log.Warnf("certificate posture: refusing to sign for a %d byte peer key", len(peerKey))
+		return nil
+	}
+	challenges = wellFormed(challenges)
+	if len(challenges) == 0 {
+		return nil
+	}
+	log.Debugf("certificate posture: answering %d certificate challenges from store %T", len(challenges), store)
+
+	candidates, err := store.Candidates(ctx)
+	if err != nil {
+		log.Warnf("failed loading certificates for posture checks: %v", err)
+		return nil
+	}
+	if len(candidates) == 0 {
+		log.Debug("certificate posture: certificate store holds no candidates, no proof will be sent")
+		return nil
+	}
+	log.Debugf("certificate posture: store holds %d candidate certificates", len(candidates))
+
+	now := time.Now()
+	proven := make(map[[sha256.Size]byte]struct{})
+	var proofs []certposture.Proof
+	for i, challenge := range challenges {
+		roots, err := certposture.ParseCAs(challenge.GetCaCertificates())
+		if err != nil {
+			log.Warnf("skipping certificate challenge with invalid CA certificates: %v", err)
+			continue
+		}
+		log.Debugf("certificate posture: challenge %d accepts %d CA certificates, nonce is %d bytes", i, len(challenge.GetCaCertificates()), len(challenge.GetNonce()))
+
+		matched := false
+		for _, candidate := range candidates {
+			if len(candidate.Chain) == 0 {
+				continue
+			}
+			leaf := candidate.Chain[0]
+			chain, err := certposture.VerifiedChain(leaf, candidate.issuers(), roots, now)
+			if err != nil {
+				log.Debugf("certificate posture: challenge %d rejected %q issued by %q: %v", i, leaf.Subject, leaf.Issuer, err)
+				continue
+			}
+			matched = true
+
+			// The same leaf can chain to different CAs for different challenges, and
+			// management checks each chain against each check's CAs, so a proof is
+			// deduplicated by its whole chain rather than by its leaf.
+			fingerprint := chainFingerprint(chain)
+			if _, done := proven[fingerprint]; done {
+				log.Debugf("certificate posture: challenge %d matched %q, already proven for an earlier challenge", i, leaf.Subject)
+				break
+			}
+			proof, err := prove(candidate.Signer, chain, challenge.GetNonce(), peerKey)
+			if err != nil {
+				log.Warnf("failed signing certificate proof for %s: %v", leaf.Subject, err)
+				continue
+			}
+			log.Debugf("certificate posture: challenge %d proven by %q with %s, signature %d bytes, chain of %d", i, leaf.Subject, proof.SigAlg, len(proof.Signature), len(proof.Chain))
+			proven[fingerprint] = struct{}{}
+			proofs = append(proofs, proof)
+			break
+		}
+		if !matched {
+			log.Debugf("certificate posture: challenge %d matched none of the %d candidates", i, len(candidates))
+		}
+	}
+	log.Debugf("certificate posture: %d challenges produced %d proofs", len(challenges), len(proofs))
+	return proofs
+}
+
+// HasChallenges reports whether any of checks asks for a certificate proof.
+func HasChallenges(checks []*proto.Checks) bool {
+	return len(certificateChallenges(checks)) > 0
+}
+
+func certificateChallenges(checks []*proto.Checks) []*proto.CertificateChallenge {
+	var challenges []*proto.CertificateChallenge
+	for _, check := range checks {
+		if challenge := check.GetCertificateChallenge(); challenge != nil {
+			challenges = append(challenges, challenge)
+		}
+	}
+	return wellFormed(challenges)
+}
+
+// wellFormed drops challenges whose nonce is not one management could have issued.
+func wellFormed(challenges []*proto.CertificateChallenge) []*proto.CertificateChallenge {
+	var kept []*proto.CertificateChallenge
+	for _, challenge := range challenges {
+		if len(challenge.GetNonce()) != certposture.NonceSize {
+			log.Debugf("certificate posture: skipping challenge with a %d byte nonce", len(challenge.GetNonce()))
+			continue
+		}
+		kept = append(kept, challenge)
+	}
+	return kept
+}
+
+func prove(signer crypto.Signer, chain []*x509.Certificate, nonce, peerKey []byte) (certposture.Proof, error) {
+	sigAlg, sig, err := certposture.Sign(signer, nonce, peerKey)
+	if err != nil {
+		return certposture.Proof{}, err
+	}
+	der := make([][]byte, 0, len(chain))
+	for _, cert := range chain {
+		der = append(der, cert.Raw)
+	}
+	return certposture.Proof{Nonce: nonce, Chain: der, SigAlg: sigAlg, Signature: sig}, nil
+}
+
+func chainFingerprint(chain []*x509.Certificate) [sha256.Size]byte {
+	buf := make([]byte, 0, len(chain)*sha256.Size)
+	for _, cert := range chain {
+		certHash := sha256.Sum256(cert.Raw)
+		buf = append(buf, certHash[:]...)
+	}
+	return sha256.Sum256(buf)
+}

client/internal/certproof/collect_darwin.go

@@ -0,0 +1,137 @@
+//go:build !ios
+
+package certproof
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strconv"
+	"time"
+
+	log "github.com/sirupsen/logrus"
+
+	"github.com/netbirdio/netbird/shared/management/certposture"
+	"github.com/netbirdio/netbird/shared/management/proto"
+)
+
+const helperTimeout = 30 * time.Second
+
+// userHelperBackoff holds off asking a user's keychain again after it proved nothing.
+var userHelperBackoff = newHelperBackoff()
+
+// CollectProofs answers the certificate challenges in checks from every store this Mac
+// can reach. The root daemon reads the System keychain itself, which is where MDM
+// installs device identities, and reaches the console user's login keychain only by
+// launching a helper into that user's session. A Mac sitting at the login window
+// therefore yields device proofs alone.
+func CollectProofs(ctx context.Context, checks []*proto.Checks, peerKey []byte, cfg Config) []certposture.Proof {
+	challenges := certificateChallenges(checks)
+	if len(challenges) == 0 {
+		logNoChallenges(checks)
+		return nil
+	}
+
+	// A helper already runs inside the user's session, so it reads its own keychain
+	// directly and must never launch another one.
+	if os.Geteuid() != 0 {
+		return CollectChallenges(ctx, DefaultStore(), challenges, peerKey)
+	}
+
+	proofs := CollectChallenges(ctx, DefaultStore(), challenges, peerKey)
+	if cfg.OwnerUnknown {
+		return proofs
+	}
+
+	userProofs, err := collectAsConsoleUser(ctx, cfg.ProfileOwner, challenges, peerKey)
+	if err != nil {
+		log.Debugf("certificate posture: console user keychain unavailable: %v", err)
+	}
+	return mergeProofs(proofs, userProofs)
+}
+
+// UserContext identifies the user whose keychain a collection would include: the console
+// user when it owns the active profile, or empty when no user keychain would be asked. A
+// change means a collection made earlier no longer reflects what this Mac can prove.
+func UserContext(cfg Config) string {
+	if os.Geteuid() != 0 || cfg.OwnerUnknown {
+		return ""
+	}
+	user, ok := CurrentConsoleUser()
+	if !ok || !user.isOwner(cfg.ProfileOwner) {
+		return ""
+	}
+	return strconv.FormatUint(uint64(user.UID), 10) + ":" + user.Name
+}
+
+// collectAsConsoleUser runs the helper inside the desktop session of the logged-in
+// user. Dropping to their uid is not enough: keychain access is an XPC call to a
+// per-session securityd, so the helper has to enter their Mach bootstrap namespace,
+// which is what launchctl asuser does.
+func collectAsConsoleUser(ctx context.Context, owner string, challenges []*proto.CertificateChallenge, peerKey []byte) ([]certposture.Proof, error) {
+	user, ok := CurrentConsoleUser()
+	if !ok {
+		return nil, nil
+	}
+	if !user.isOwner(owner) {
+		log.Debugf("certificate posture: console user %s does not own the active profile, no user keychain is asked", user.Name)
+		return nil, nil
+	}
+
+	uid := strconv.FormatUint(uint64(user.UID), 10)
+	backoffKey := helperBackoffKey(uid, challenges)
+	if !userHelperBackoff.allow(backoffKey, time.Now()) {
+		log.Debugf("certificate posture: the keychain of uid %s proved nothing recently, not asking again yet", uid)
+		return nil, nil
+	}
+
+	binary, err := os.Executable()
+	if err != nil {
+		return nil, fmt.Errorf("resolve own binary: %w", err)
+	}
+
+	parent := ctx
+	ctx, cancel := context.WithTimeout(ctx, helperTimeout)
+	defer cancel()
+
+	// Absolute paths, because the daemon's PATH is configurable through the service
+	// environment, and sudo selects the user by uid so the name never has to round-trip.
+	cmd := exec.CommandContext(ctx, "/bin/launchctl", "asuser", uid, "/usr/bin/sudo", "-u", "#"+uid, "-H", "--", binary, "posture", "cert-proof")
+	killHelperGroupOnCancel(cmd)
+
+	log.Debugf("certificate posture: asking the desktop session of uid %s to answer %d challenges", uid, len(challenges))
+	proofs, err := runHelperCmd(cmd, helperRequest(challenges, peerKey))
+	// A run that completed, or ran into the timeout waiting on a prompt nobody answered,
+	// tells whether the keychain proves anything. A launch or output failure, or a run the
+	// caller cut short, says nothing about it and must not hold off the next one.
+	timedOut := errors.Is(ctx.Err(), context.DeadlineExceeded) && parent.Err() == nil
+	if err == nil || timedOut {
+		userHelperBackoff.record(backoffKey, err == nil && len(proofs) > 0, time.Now())
+	}
+	if err != nil {
+		return nil, fmt.Errorf("run helper as uid %s: %w", uid, err)
+	}
+	log.Debugf("certificate posture: desktop session of uid %s returned %d proofs", uid, len(proofs))
+	return proofs, nil
+}
+
+// helperStore is the store the helper reads: the user's login keychain alone. The
+// session's search list also holds the System keychain, which the daemon reads itself,
+// and using a System keychain key from the user's session would ask for an
+// administrator's approval.
+func helperStore() Store {
+	home, err := os.UserHomeDir()
+	if err != nil {
+		log.Debugf("certificate posture: no home directory, searching the default keychain list: %v", err)
+		return NewKeychainStore()
+	}
+	login := filepath.Join(home, "Library", "Keychains", "login.keychain-db")
+	if _, err := os.Stat(login); err != nil {
+		// Keychains created before macOS 10.12 keep the old file name.
+		login = filepath.Join(home, "Library", "Keychains", "login.keychain")
+	}
+	return NewKeychainStore(login)
+}

client/internal/certproof/collect_js.go

@@ -0,0 +1,31 @@
+//go:build js
+
+package certproof
+
+import (
+	"context"
+
+	"github.com/netbirdio/netbird/shared/management/certposture"
+	"github.com/netbirdio/netbird/shared/management/proto"
+)
+
+// CollectProofs proves nothing in a browser: it has no TPM, certificate store or file
+// store to sign with, so the stores stay out of the WebAssembly build.
+func CollectProofs(context.Context, []*proto.Checks, []byte, Config) []certposture.Proof {
+	return nil
+}
+
+// UserContext identifies the user whose certificates a collection would include. A
+// browser has no per-user store, so it never changes.
+func UserContext(Config) string {
+	return ""
+}
+
+// DefaultStore is empty in a browser.
+func DefaultStore() Store {
+	return Stores{}
+}
+
+func helperStore() Store {
+	return DefaultStore()
+}

client/internal/certproof/collect_other.go

@@ -0,0 +1,31 @@
+//go:build ((!darwin && !windows) || ios) && !js
+
+package certproof
+
+import (
+	"context"
+
+	"github.com/netbirdio/netbird/shared/management/certposture"
+	"github.com/netbirdio/netbird/shared/management/proto"
+)
+
+// CollectProofs answers the certificate challenges in checks from the PEM directory cfg
+// names, joined by its PKCS#11 token when it names one. Only macOS and Windows keep
+// per-user certificates out of reach of a privileged daemon, so every other platform
+// reads its store in the daemon itself.
+func CollectProofs(ctx context.Context, checks []*proto.Checks, peerKey []byte, cfg Config) []certposture.Proof {
+	return Collect(ctx, storeWithToken(cfg), checks, peerKey)
+}
+
+// UserContext identifies the user whose certificates a collection would include. These
+// platforms have no per-user store, so it never changes.
+func UserContext(Config) string {
+	return ""
+}
+
+// helperStore is the store the helper reads. Nothing launches a helper on these
+// platforms, so it is the store the daemon reads, configured from the same environment,
+// which lets an administrator check a setup by running the helper by hand.
+func helperStore() Store {
+	return storeWithToken(Config{PKCS11: PKCS11FromEnv()})
+}

client/internal/certproof/collect_test.go

@@ -0,0 +1,258 @@
+package certproof
+
+import (
+	"context"
+	"crypto/rand"
+	"crypto/x509"
+	"math/big"
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/shared/management/certposture"
+	"github.com/netbirdio/netbird/shared/management/certposture/certtest"
+	"github.com/netbirdio/netbird/shared/management/proto"
+)
+
+var peerKey = []byte("peer-public-key-aaaaaaaaaaaaaaaa")
+
+func TestCollect_ProvesOneMatchingCertificatePerChallenge(t *testing.T) {
+	corpCA := certtest.NewCA(t, "corp-root")
+	otherCA := certtest.NewCA(t, "other-root")
+	unrelatedCA := certtest.NewCA(t, "unrelated-root")
+
+	dir := storeDir(t)
+	deviceKey := certtest.ECDSAKey(t)
+	device := corpCA.Issue(t, deviceKey, "device")
+	writeFile(t, dir, "device.pem", certtest.CertPEM(device)+certtest.KeyPEM(t, deviceKey))
+
+	otherKey := certtest.RSAKey(t)
+	writeFile(t, dir, "other.crt", certtest.CertPEM(otherCA.Issue(t, otherKey, "other")))
+	writeFile(t, dir, "other.key", certtest.KeyPEM(t, otherKey))
+
+	writeFile(t, dir, "keyless.crt", certtest.CertPEM(corpCA.Issue(t, certtest.ECDSAKey(t), "keyless")))
+	writeFile(t, dir, "notes.txt", "ignored")
+
+	challenger := certposture.NewChallenger([]byte("secret"))
+	nonce := challenger.Nonce(peerKey, time.Now())
+	challenge := func(cas ...string) *proto.Checks {
+		return &proto.Checks{CertificateChallenge: &proto.CertificateChallenge{Nonce: nonce, CaCertificates: cas}}
+	}
+	checks := []*proto.Checks{
+		{Files: []string{"/usr/bin/agent"}},
+		challenge(corpCA.PEM),
+		challenge(corpCA.PEM),
+		challenge(otherCA.PEM),
+		challenge(unrelatedCA.PEM),
+		challenge("not a pem"),
+	}
+
+	proofs := Collect(context.Background(), NewFileStore(dir), checks, peerKey)
+
+	require.Len(t, proofs, 2)
+	var subjects []string
+	for _, p := range proofs {
+		chain, err := challenger.Verify(p, peerKey, time.Now())
+		require.NoError(t, err)
+		subjects = append(subjects, chain[0].Subject.CommonName)
+	}
+	assert.ElementsMatch(t, []string{"device", "other"}, subjects)
+}
+
+func TestCollect_NothingToProve(t *testing.T) {
+	dir := storeDir(t)
+	key := certtest.ECDSAKey(t)
+	ca := certtest.NewCA(t, "root")
+	writeFile(t, dir, "device.pem", certtest.CertPEM(ca.Issue(t, key, "device"))+certtest.KeyPEM(t, key))
+
+	tests := []struct {
+		name   string
+		store  Store
+		checks []*proto.Checks
+	}{
+		{"no checks", NewFileStore(dir), nil},
+		{"files only", NewFileStore(dir), []*proto.Checks{{Files: []string{"/bin/x"}}}},
+		{"challenge without nonce", NewFileStore(dir), []*proto.Checks{{CertificateChallenge: &proto.CertificateChallenge{CaCertificates: []string{ca.PEM}}}}},
+		{"missing store dir", NewFileStore(filepath.Join(dir, "missing")), []*proto.Checks{{CertificateChallenge: &proto.CertificateChallenge{Nonce: make([]byte, certposture.NonceSize), CaCertificates: []string{ca.PEM}}}}},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			assert.Nil(t, Collect(context.Background(), tt.store, tt.checks, peerKey))
+		})
+	}
+}
+
+func TestFileStore_ChainWithIntermediate(t *testing.T) {
+	root := certtest.NewCA(t, "root")
+	intermediate := certtest.NewIntermediate(t, root, "intermediate")
+	key := certtest.ECDSAKey(t)
+	leaf := intermediate.Issue(t, key, "device")
+
+	dir := storeDir(t)
+	writeFile(t, dir, "device.pem", certtest.CertPEM(leaf)+certtest.CertPEM(intermediate.Cert)+certtest.KeyPEM(t, key))
+
+	candidates, err := NewFileStore(dir).Candidates(context.Background())
+	require.NoError(t, err)
+	require.Len(t, candidates, 1)
+	require.Len(t, candidates[0].Chain, 2)
+
+	roots, err := certposture.ParseCAs([]string{root.PEM})
+	require.NoError(t, err)
+	assert.NoError(t, certposture.VerifyChain(candidates[0].Chain, roots, time.Now()))
+}
+
+// storeDir is a PEM directory the store accepts: t.TempDir follows the umask, which
+// leaves the directory group-writable on systems with a user-private group umask.
+func storeDir(t *testing.T) string {
+	t.Helper()
+	dir := t.TempDir()
+	require.NoError(t, os.Chmod(dir, 0o700))
+	return dir
+}
+
+func writeFile(t *testing.T, dir, name, content string) {
+	t.Helper()
+	require.NoError(t, os.WriteFile(filepath.Join(dir, name), []byte(content), 0o600))
+}
+
+func TestCollectChallenges_RefusesMalformedInput(t *testing.T) {
+	ca := certtest.NewCA(t, "corp-root")
+	dir := storeDir(t)
+	key := certtest.ECDSAKey(t)
+	writeFile(t, dir, "device.pem", certtest.CertPEM(ca.Issue(t, key, "device"))+certtest.KeyPEM(t, key))
+	store := NewFileStore(dir)
+	nonce := certposture.NewChallenger([]byte("secret")).Nonce(peerKey, time.Now())
+
+	tests := []struct {
+		name    string
+		nonce   []byte
+		peerKey []byte
+		want    int
+	}{
+		{"issued nonce and peer key are signed", nonce, peerKey, 1},
+		{"short nonce is not signed", nonce[:8], peerKey, 0},
+		{"oversized nonce is not signed", append(append([]byte{}, nonce...), 0), peerKey, 0},
+		{"short peer key is not signed", nonce, peerKey[:16], 0},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			challenges := []*proto.CertificateChallenge{{Nonce: tt.nonce, CaCertificates: []string{ca.PEM}}}
+			assert.Len(t, CollectChallenges(context.Background(), store, challenges, tt.peerKey), tt.want,
+				"the device key signs only what management could have issued")
+		})
+	}
+}
+
+func TestFileStore_SkipsKeyOfAnotherCertificate(t *testing.T) {
+	ca := certtest.NewCA(t, "corp-root")
+	dir := storeDir(t)
+
+	// A stale key next to a renewed certificate, sorted before the good pair, must not
+	// produce a proof that management rejects and stop the search there.
+	writeFile(t, dir, "a-renewed.crt", certtest.CertPEM(ca.Issue(t, certtest.ECDSAKey(t), "renewed")))
+	writeFile(t, dir, "a-renewed.key", certtest.KeyPEM(t, certtest.ECDSAKey(t)))
+	goodKey := certtest.ECDSAKey(t)
+	good := ca.Issue(t, goodKey, "good")
+	writeFile(t, dir, "b-good.pem", certtest.CertPEM(good)+certtest.KeyPEM(t, goodKey))
+
+	candidates, err := NewFileStore(dir).Candidates(context.Background())
+	require.NoError(t, err)
+	require.Len(t, candidates, 1, "only the certificate whose key matches is a candidate")
+	assert.True(t, good.Equal(candidates[0].Chain[0]), "the matching pair is kept")
+
+	challenger := certposture.NewChallenger([]byte("secret"))
+	now := time.Now()
+	nonce := challenger.Nonce(peerKey, now)
+	proofs := CollectChallenges(context.Background(), NewFileStore(dir), []*proto.CertificateChallenge{{Nonce: nonce, CaCertificates: []string{ca.PEM}}}, peerKey)
+	require.Len(t, proofs, 1)
+	_, err = challenger.Verify(proofs[0], peerKey, now)
+	assert.NoError(t, err, "the proof sent is one management accepts")
+}
+
+// staticStore hands out fixed candidates, for scenarios no on-disk layout can express.
+type staticStore []Candidate
+
+func (s staticStore) Candidates(context.Context) ([]Candidate, error) { return s, nil }
+
+func TestCollectChallenges_RoutesAroundExpiredCopyOfRenewedIntermediate(t *testing.T) {
+	root := certtest.NewCA(t, "root")
+	intermediate := certtest.NewIntermediate(t, root, "issuing-ca")
+
+	// Renewing a CA with the same key pair leaves two certificates with the same
+	// subject and key in the store. The expired one sorts first here.
+	expiredTmpl := *intermediate.Cert
+	expiredTmpl.SerialNumber = big.NewInt(1)
+	expiredTmpl.NotBefore = time.Now().Add(-72 * time.Hour)
+	expiredTmpl.NotAfter = time.Now().Add(-48 * time.Hour)
+	der, err := x509.CreateCertificate(rand.Reader, &expiredTmpl, root.Cert, intermediate.Key.Public(), root.Key)
+	require.NoError(t, err)
+	expired, err := x509.ParseCertificate(der)
+	require.NoError(t, err)
+
+	key := certtest.ECDSAKey(t)
+	leaf := intermediate.Issue(t, key, "device")
+	pool := []*x509.Certificate{expired, intermediate.Cert}
+
+	chain := buildChain(leaf, pool)
+	require.Len(t, chain, 2)
+	require.True(t, expired.Equal(chain[1]), "precondition: the first-match chain runs through the expired copy")
+
+	challenger := certposture.NewChallenger([]byte("secret"))
+	now := time.Now()
+	nonce := challenger.Nonce(peerKey, now)
+	store := staticStore{{Chain: chain, Signer: key, Intermediates: pool}}
+
+	proofs := CollectChallenges(context.Background(), store, []*proto.CertificateChallenge{{Nonce: nonce, CaCertificates: []string{root.PEM}}}, peerKey)
+
+	require.Len(t, proofs, 1, "a valid path through the renewed intermediate exists, so the challenge is answered")
+	verified, err := challenger.Verify(proofs[0], peerKey, now)
+	require.NoError(t, err)
+	assert.True(t, intermediate.Cert.Equal(verified[1]), "the proof carries the valid intermediate, not the expired copy")
+	assert.True(t, certposture.ChainMatchesCAs(certposture.EncodeChainPEM(verified), []string{root.PEM}, now),
+		"management's own check accepts the chain the proof carries")
+}
+
+func TestCollectChallenges_ProvesALeafOncePerDistinctChain(t *testing.T) {
+	rootA := certtest.NewCA(t, "root-a")
+	rootB := certtest.NewCA(t, "root-b")
+	issuer := certtest.NewIntermediate(t, rootA, "issuing-ca")
+
+	// The same issuing CA cross-signed by a second root: one leaf, two valid paths.
+	crossTmpl := *issuer.Cert
+	crossTmpl.SerialNumber = big.NewInt(2)
+	der, err := x509.CreateCertificate(rand.Reader, &crossTmpl, rootB.Cert, issuer.Key.Public(), rootB.Key)
+	require.NoError(t, err)
+	cross, err := x509.ParseCertificate(der)
+	require.NoError(t, err)
+
+	key := certtest.ECDSAKey(t)
+	leaf := issuer.Issue(t, key, "device")
+	pool := []*x509.Certificate{issuer.Cert, cross}
+	store := staticStore{{Chain: buildChain(leaf, pool), Signer: key, Intermediates: pool}}
+
+	challenger := certposture.NewChallenger([]byte("secret"))
+	now := time.Now()
+	nonce := challenger.Nonce(peerKey, now)
+	challenges := []*proto.CertificateChallenge{
+		{Nonce: nonce, CaCertificates: []string{rootA.PEM}},
+		{Nonce: nonce, CaCertificates: []string{rootB.PEM}},
+		{Nonce: nonce, CaCertificates: []string{rootA.PEM}},
+	}
+
+	proofs := CollectChallenges(context.Background(), store, challenges, peerKey)
+
+	require.Len(t, proofs, 2, "one proof per distinct chain, the repeated root-a challenge reuses the first")
+	for _, root := range []*certtest.CA{rootA, rootB} {
+		matched := false
+		for _, p := range proofs {
+			chain, err := challenger.Verify(p, peerKey, now)
+			require.NoError(t, err)
+			matched = matched || certposture.ChainMatchesCAs(certposture.EncodeChainPEM(chain), []string{root.PEM}, now)
+		}
+		assert.True(t, matched, "management can match a chain for the check that trusts %s", root.Cert.Subject.CommonName)
+	}
+}

client/internal/certproof/collect_windows.go

@@ -0,0 +1,108 @@
+package certproof
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"os/exec"
+	"syscall"
+	"time"
+
+	log "github.com/sirupsen/logrus"
+	"golang.org/x/sys/windows"
+
+	"github.com/netbirdio/netbird/shared/management/certposture"
+	"github.com/netbirdio/netbird/shared/management/proto"
+)
+
+const helperTimeout = 30 * time.Second
+
+// CollectProofs answers the certificate challenges in checks from every store this
+// machine can reach. The service reads the local machine store itself, where AD and
+// Intune enrol device certificates, and reaches the signed-in user's store by launching
+// a helper with that session's token. A machine at the sign-in screen therefore proves
+// device certificates alone.
+func CollectProofs(ctx context.Context, checks []*proto.Checks, peerKey []byte, cfg Config) []certposture.Proof {
+	challenges := certificateChallenges(checks)
+	if len(challenges) == 0 {
+		logNoChallenges(checks)
+		return nil
+	}
+
+	proofs := CollectChallenges(ctx, DefaultStore(), challenges, peerKey)
+
+	// The helper already runs as the signed-in user, and an ordinary process has no
+	// right to a session token, so only the service goes looking for one.
+	if !runningAsLocalSystem() || cfg.OwnerUnknown {
+		return proofs
+	}
+
+	userProofs, err := collectAsDesktopUser(ctx, cfg.ProfileOwner, challenges, peerKey)
+	if err != nil {
+		log.Debugf("certificate posture: user certificate store unavailable: %v", err)
+	}
+	return mergeProofs(proofs, userProofs)
+}
+
+// UserContext identifies the session whose store a collection would include: a session
+// of the profile owner, or empty when no user store would be asked. A change means a
+// collection made earlier no longer reflects what this machine can prove.
+func UserContext(cfg Config) string {
+	if !runningAsLocalSystem() || cfg.OwnerUnknown {
+		return ""
+	}
+	user, ok := CurrentDesktopUser(cfg.ProfileOwner)
+	if !ok {
+		return ""
+	}
+	defer user.Close()
+	return fmt.Sprintf("%d:%s", user.Session, user.Name)
+}
+
+// helperStore is the store the helper reads. It runs as the signed-in user, so it wants
+// that user's store rather than the machine store the service already read.
+func helperStore() Store {
+	return NewUserStore()
+}
+
+// collectAsDesktopUser runs the helper inside the interactive session of the signed-in
+// user. Unlike a keychain on macOS, a Windows service can assume a user identity
+// directly, so the session token goes straight into the child process.
+func collectAsDesktopUser(ctx context.Context, owner string, challenges []*proto.CertificateChallenge, peerKey []byte) ([]certposture.Proof, error) {
+	user, ok := CurrentDesktopUser(owner)
+	if !ok {
+		return nil, nil
+	}
+	defer user.Close()
+
+	binary, err := os.Executable()
+	if err != nil {
+		return nil, fmt.Errorf("resolve own binary: %w", err)
+	}
+
+	// The user's own environment, not the service's: the service environment may carry
+	// secrets such as a setup key that the signed-in user must not be able to read.
+	env, err := user.Token.Environ(false)
+	if err != nil {
+		return nil, fmt.Errorf("build environment of %s: %w", user.Name, err)
+	}
+
+	ctx, cancel := context.WithTimeout(ctx, helperTimeout)
+	defer cancel()
+
+	cmd := exec.CommandContext(ctx, binary, "posture", "cert-proof")
+	cmd.Env = env
+	cmd.SysProcAttr = &syscall.SysProcAttr{
+		Token:         syscall.Token(user.Token),
+		HideWindow:    true,
+		CreationFlags: windows.CREATE_NO_WINDOW,
+	}
+
+	log.Debugf("certificate posture: asking session %d to answer %d challenges", user.Session, len(challenges))
+	proofs, err := runHelperCmd(cmd, helperRequest(challenges, peerKey))
+	if err != nil {
+		return nil, fmt.Errorf("run helper in session %d: %w", user.Session, err)
+	}
+	log.Debugf("certificate posture: session %d returned %d proofs", user.Session, len(proofs))
+	return proofs, nil
+}

client/internal/certproof/collector.go

@@ -0,0 +1,189 @@
+package certproof
+
+import (
+	"context"
+	"sync"
+	"time"
+
+	log "github.com/sirupsen/logrus"
+
+	"github.com/netbirdio/netbird/shared/management/certposture"
+	"github.com/netbirdio/netbird/shared/management/proto"
+)
+
+const (
+	collectTimeout = 45 * time.Second
+
+	// lostAfter is how many deadlines an abandoned collection is waited for before a new
+	// one may start beside it. A store that is merely slow finishes well within that and
+	// releases its slot; only one wedged for good is worked around.
+	lostAfter = 10
+
+	// maxInFlight bounds how many abandoned collections may pile up. A store that never
+	// answers would otherwise leak a goroutine on every collection for the life of the
+	// daemon.
+	maxInFlight = 2
+)
+
+// collectSlots tracks the collections running in the process. Collectors share one by
+// default: a collection abandoned in a token, TPM or keychain call keeps running after
+// the engine that started it has stopped, and the next engine must account for it.
+type collectSlots struct {
+	mu   sync.Mutex
+	next uint64
+	// running holds the start time of each collection still running, by slot id.
+	running map[uint64]time.Time
+}
+
+var processSlots collectSlots
+
+// Collector runs CollectProofs with a deadline and, in the normal case, one collection
+// at a time in the process. Token, TPM and keychain calls cannot be interrupted, and on
+// Linux there is not even a process to kill, so a collection that overruns is abandoned
+// rather than awaited. A single wedged call would then refuse every later collection for
+// the life of the daemon, so once an abandoned collection has been running for lostAfter
+// deadlines another may start beside it, up to maxInFlight. The zero value is ready to
+// use.
+type Collector struct {
+	// slots overrides the process-wide slots when set.
+	slots *collectSlots
+	// timeout overrides collectTimeout when set.
+	timeout time.Duration
+	// now overrides the clock when set.
+	now func() time.Time
+}
+
+// Collect answers the certificate challenges in checks, returning no proofs when there
+// are no challenges, when previous collections still hold the slots, or when this one
+// does not finish in time. Missing proofs fail the certificate check on management.
+func (c *Collector) Collect(ctx context.Context, checks []*proto.Checks, peerKey []byte, cfg Config) []certposture.Proof {
+	return c.collect(ctx, checks, func(ctx context.Context) []certposture.Proof {
+		return CollectProofs(ctx, checks, peerKey, cfg)
+	})
+}
+
+func (c *Collector) collect(ctx context.Context, checks []*proto.Checks, run func(context.Context) []certposture.Proof) []certposture.Proof {
+	if len(certificateChallenges(checks)) == 0 {
+		return nil
+	}
+	slots := c.slotsInUse()
+	slot, ok := slots.start(c.clock(), time.Duration(lostAfter)*c.deadline())
+	if !ok {
+		return nil
+	}
+
+	ctx, cancel := context.WithTimeout(ctx, c.deadline())
+	defer cancel()
+
+	done := make(chan []certposture.Proof, 1)
+	go func() {
+		// The slot is freed before the result is delivered, so a caller that starts the
+		// next collection right after this one returned is not turned away.
+		proofs := run(ctx)
+		slots.finish(slot)
+		done <- proofs
+	}()
+
+	select {
+	case proofs := <-done:
+		return proofs
+	case <-ctx.Done():
+		log.Warnf("certificate posture: proof collection did not finish within %s, sending no proofs", c.deadline())
+		return nil
+	}
+}
+
+// Stuck reports whether lost collections hold every slot. No collection starts again
+// until one of them returns, which for a store that never answers means a restart.
+func (c *Collector) Stuck() bool {
+	return c.slotsInUse().allLost(c.clock(), time.Duration(lostAfter)*c.deadline())
+}
+
+func (c *Collector) slotsInUse() *collectSlots {
+	if c.slots != nil {
+		return c.slots
+	}
+	return &processSlots
+}
+
+func (c *Collector) clock() time.Time {
+	if c.now != nil {
+		return c.now()
+	}
+	return time.Now()
+}
+
+func (c *Collector) deadline() time.Duration {
+	if c.timeout > 0 {
+		return c.timeout
+	}
+	return collectTimeout
+}
+
+// start claims a slot at now, reporting its id and whether the caller may collect. A
+// collection that has held a slot for lost is taken to be wedged.
+func (s *collectSlots) start(now time.Time, lost time.Duration) (uint64, bool) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	if len(s.running) > 0 {
+		oldest := s.oldest()
+		switch {
+		case len(s.running) >= maxInFlight:
+			log.Warnf("certificate posture: %d proof collections are wedged, sending no proofs", len(s.running))
+			return 0, false
+		case now.Sub(oldest) < lost:
+			log.Warnf("certificate posture: previous proof collection is still running, sending no proofs")
+			return 0, false
+		default:
+			log.Warnf("certificate posture: a proof collection has been stuck since %s, starting another", oldest.Format(time.RFC3339))
+		}
+	}
+	if s.running == nil {
+		s.running = make(map[uint64]time.Time, maxInFlight)
+	}
+	s.next++
+	s.running[s.next] = now
+	return s.next, true
+}
+
+// finish releases the slot with the given id.
+func (s *collectSlots) finish(slot uint64) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	delete(s.running, slot)
+}
+
+// allLost reports whether every slot is held by a collection that has run for lost.
+func (s *collectSlots) allLost(now time.Time, lost time.Duration) bool {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	if len(s.running) < maxInFlight {
+		return false
+	}
+	for _, started := range s.running {
+		if now.Sub(started) < lost {
+			return false
+		}
+	}
+	return true
+}
+
+// idle reports whether no collection is running.
+func (s *collectSlots) idle() bool {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	return len(s.running) == 0
+}
+
+// oldest returns the start time of the longest-running collection. The caller holds mu
+// and has checked that one is running.
+func (s *collectSlots) oldest() time.Time {
+	var oldest time.Time
+	for _, started := range s.running {
+		if oldest.IsZero() || started.Before(oldest) {
+			oldest = started
+		}
+	}
+	return oldest
+}

client/internal/certproof/collector_test.go

@@ -0,0 +1,207 @@
+package certproof
+
+import (
+	"context"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/shared/management/certposture"
+	"github.com/netbirdio/netbird/shared/management/proto"
+)
+
+var challengeChecks = []*proto.Checks{{CertificateChallenge: &proto.CertificateChallenge{
+	Nonce: certposture.NewChallenger([]byte("secret")).Nonce(peerKey, time.Now()),
+}}}
+
+func TestCollector_SkipsChecksWithoutChallenges(t *testing.T) {
+	var c Collector
+	called := false
+
+	proofs := c.collect(context.Background(), []*proto.Checks{{Files: []string{"/bin/agent"}}}, func(context.Context) []certposture.Proof {
+		called = true
+		return nil
+	})
+
+	assert.Nil(t, proofs)
+	assert.False(t, called, "no store is touched when no check carries a challenge")
+}
+
+func TestCollector_ReturnsProofs(t *testing.T) {
+	var c Collector
+	want := []certposture.Proof{{Nonce: []byte("nonce")}}
+
+	proofs := c.collect(context.Background(), challengeChecks, func(context.Context) []certposture.Proof { return want })
+
+	assert.Equal(t, want, proofs, "a collection that finishes in time is returned as is")
+}
+
+func TestCollector_AbandonsStuckCollection(t *testing.T) {
+	c := Collector{timeout: 50 * time.Millisecond, slots: &collectSlots{}}
+	release := make(chan struct{})
+	finished := make(chan struct{})
+
+	// A token or keychain call that ignores its context and blocks well past the deadline.
+	stuck := func(context.Context) []certposture.Proof {
+		defer close(finished)
+		<-release
+		return []certposture.Proof{{Nonce: []byte("late")}}
+	}
+
+	start := time.Now()
+	proofs := c.collect(context.Background(), challengeChecks, stuck)
+	assert.Nil(t, proofs, "an overrunning collection yields no proofs")
+	assert.Less(t, time.Since(start), time.Second, "the caller is released at the deadline, not when the call returns")
+
+	called := false
+	proofs = c.collect(context.Background(), challengeChecks, func(context.Context) []certposture.Proof {
+		called = true
+		return nil
+	})
+	assert.Nil(t, proofs)
+	assert.False(t, called, "no second collection starts while the first is still running")
+
+	close(release)
+	<-finished
+	require.Eventually(t, func() bool { return c.slots.idle() }, time.Second, 5*time.Millisecond, "the collector frees up once the stuck call returns")
+
+	want := []certposture.Proof{{Nonce: []byte("nonce")}}
+	assert.Equal(t, want, c.collect(context.Background(), challengeChecks, func(context.Context) []certposture.Proof { return want }),
+		"collection works again after the stuck call returned")
+}
+
+func TestCollector_CancelsContextAtDeadline(t *testing.T) {
+	c := Collector{timeout: 20 * time.Millisecond, slots: &collectSlots{}}
+	cancelled := make(chan struct{})
+
+	c.collect(context.Background(), challengeChecks, func(ctx context.Context) []certposture.Proof {
+		<-ctx.Done()
+		close(cancelled)
+		return nil
+	})
+
+	select {
+	case <-cancelled:
+	case <-time.After(time.Second):
+		t.Fatal("a collection that honours its context, like the helper process, must see it cancelled")
+	}
+}
+
+// TestCollector_SingleFlightAcrossCollectors: each engine has its own Collector, and a
+// collection stuck in a token call outlives the engine that started it, so the next
+// engine's Collector must not start another one until it has finished.
+func TestCollector_SingleFlightAcrossCollectors(t *testing.T) {
+	shared := &collectSlots{}
+	first := Collector{timeout: 20 * time.Millisecond, slots: shared}
+	second := Collector{timeout: time.Second, slots: shared}
+
+	release := make(chan struct{})
+	stuck := func(context.Context) []certposture.Proof {
+		<-release
+		return nil
+	}
+	assert.Nil(t, first.collect(context.Background(), challengeChecks, stuck), "the first collection is abandoned at its deadline")
+
+	called := false
+	proofs := second.collect(context.Background(), challengeChecks, func(context.Context) []certposture.Proof {
+		called = true
+		return []certposture.Proof{{}}
+	})
+	assert.Nil(t, proofs, "no proofs while the abandoned collection still runs")
+	assert.False(t, called, "a second collection does not start on top of the abandoned one")
+
+	close(release)
+	require.Eventually(t, func() bool { return shared.idle() }, time.Second, 5*time.Millisecond)
+	assert.Len(t, second.collect(context.Background(), challengeChecks, func(context.Context) []certposture.Proof { return []certposture.Proof{{}} }), 1,
+		"collections resume once the abandoned one finished")
+}
+
+// TestCollector_ZeroValuesShareTheProcessFlag: the zero value uses the process-wide flag.
+func TestCollector_ZeroValuesShareTheProcessFlag(t *testing.T) {
+	var a, b Collector
+	assert.Same(t, a.slotsInUse(), b.slotsInUse(), "separate Collectors share the process slots")
+}
+
+// TestCollector_LostCollectionDoesNotBlockForever: a store call that never answers, such
+// as a wedged token on Linux where there is no process to kill, must not refuse every
+// later collection. Once it has run for lostAfter deadlines another starts beside it,
+// and no more than maxInFlight ever run.
+func TestCollector_LostCollectionDoesNotBlockForever(t *testing.T) {
+	now := time.Now()
+	slots := &collectSlots{}
+	c := Collector{timeout: 10 * time.Millisecond, slots: slots, now: func() time.Time { return now }}
+
+	release := make(chan struct{})
+	defer close(release)
+	var started atomic.Int32
+	wedged := func(context.Context) []certposture.Proof {
+		started.Add(1)
+		<-release
+		return nil
+	}
+
+	c.collect(context.Background(), challengeChecks, wedged)
+	waitStarted(t, &started, 1)
+	c.collect(context.Background(), challengeChecks, wedged)
+	assert.Equal(t, int32(1), started.Load(), "a second collection is refused while the first may still finish")
+	assert.False(t, c.Stuck(), "one slow collection is not stuck yet")
+
+	now = now.Add(time.Duration(lostAfter) * c.deadline())
+	c.collect(context.Background(), challengeChecks, wedged)
+	waitStarted(t, &started, 2)
+	assert.False(t, c.Stuck(), "the second collection may still return, so the collector is not stuck yet")
+
+	now = now.Add(time.Duration(lostAfter) * c.deadline())
+	c.collect(context.Background(), challengeChecks, wedged)
+	assert.Equal(t, int32(2), started.Load(), "no more than maxInFlight collections run")
+	assert.True(t, c.Stuck(), "with every slot held by a lost collection the collector reports itself stuck")
+}
+
+// TestCollector_LostWindowFollowsTheOldestRunningCollection: A wedges, B starts beside it
+// once A counts as lost, then A returns. B is recent, so a third collection must wait for
+// B to be lost in its own right rather than inherit A's start time.
+func TestCollector_LostWindowFollowsTheOldestRunningCollection(t *testing.T) {
+	now := time.Now()
+	slots := &collectSlots{}
+	c := Collector{timeout: 10 * time.Millisecond, slots: slots, now: func() time.Time { return now }}
+	lost := time.Duration(lostAfter) * c.deadline()
+
+	releaseA := make(chan struct{})
+	releaseB := make(chan struct{})
+	defer close(releaseB)
+	var started atomic.Int32
+	blockOn := func(release chan struct{}) func(context.Context) []certposture.Proof {
+		return func(context.Context) []certposture.Proof {
+			started.Add(1)
+			<-release
+			return nil
+		}
+	}
+
+	c.collect(context.Background(), challengeChecks, blockOn(releaseA))
+	waitStarted(t, &started, 1)
+	now = now.Add(lost)
+	c.collect(context.Background(), challengeChecks, blockOn(releaseB))
+	waitStarted(t, &started, 2)
+
+	close(releaseA)
+	require.Eventually(t, func() bool {
+		slots.mu.Lock()
+		defer slots.mu.Unlock()
+		return len(slots.running) == 1
+	}, time.Second, 5*time.Millisecond, "A frees its slot once it returns")
+
+	c.collect(context.Background(), challengeChecks, blockOn(make(chan struct{})))
+	assert.Equal(t, int32(2), started.Load(), "a third collection waits while B is recent")
+}
+
+// waitStarted waits for the collections the test launched to have started. collect may
+// return at its deadline before the goroutine running the store call was scheduled.
+func waitStarted(t *testing.T, started *atomic.Int32, n int32) {
+	t.Helper()
+	require.Eventually(t, func() bool { return started.Load() == n }, time.Second, time.Millisecond,
+		"%d collections should have started", n)
+}

client/internal/certproof/consoleuser_darwin.go

@@ -0,0 +1,123 @@
+//go:build !ios
+
+package certproof
+
+import (
+	"bytes"
+	"fmt"
+	"strconv"
+	"sync"
+
+	"github.com/ebitengine/purego"
+	log "github.com/sirupsen/logrus"
+)
+
+const (
+	systemConfigurationFramework = "/System/Library/Frameworks/SystemConfiguration.framework/SystemConfiguration"
+
+	encodingUTF8       = 0x08000100
+	consoleNameBufSize = 256
+)
+
+var (
+	consoleOnce sync.Once
+	consoleErr  error
+
+	scDynamicStoreCopyConsoleUser func(store uintptr, uid, gid *uint32) uintptr
+	cfStringGetCString            func(str uintptr, buffer *byte, size int, encoding uint32) bool
+)
+
+// ConsoleUser is the account whose desktop session owns the display. Its login keychain
+// is the only user keychain a NetBird daemon can reach, and only while it is logged in.
+type ConsoleUser struct {
+	Name string
+	UID  uint32
+	GID  uint32
+}
+
+// CurrentConsoleUser reports the user sitting at the desktop. The second return value is
+// false when nobody is: at the login window macOS either reports no console user at all
+// or attributes the session to root, and neither has a login keychain to offer.
+func CurrentConsoleUser() (ConsoleUser, bool) {
+	if err := loadConsoleUser(); err != nil {
+		log.Debugf("console user lookup unavailable: %v", err)
+		return ConsoleUser{}, false
+	}
+
+	var uid, gid uint32
+	name := scDynamicStoreCopyConsoleUser(0, &uid, &gid)
+	if name == 0 {
+		log.Debug("no console user is logged in, no login keychain is reachable")
+		return ConsoleUser{}, false
+	}
+	defer release(name)
+
+	user := ConsoleUser{Name: cfString(name), UID: uid, GID: gid}
+	if !user.hasDesktop() {
+		log.Debugf("console session belongs to %q uid=%d, which is not a desktop login, no login keychain is reachable", user.Name, user.UID)
+		return ConsoleUser{}, false
+	}
+	return user, true
+}
+
+// hasDesktop reports whether the console session is a real user desktop. The login
+// window runs as root and some macOS releases name it "loginwindow" instead.
+func (u ConsoleUser) hasDesktop() bool {
+	switch u.Name {
+	case "", "root", "loginwindow":
+		return false
+	}
+	return u.UID != 0
+}
+
+// isOwner reports whether the console user is owner, the account of the active profile,
+// which is recorded as a short user name or, for an account without one, a numeric uid.
+// With no owner the console user counts, as macOS has a single console user.
+func (u ConsoleUser) isOwner(owner string) bool {
+	return owner == "" || owner == u.Name || owner == strconv.FormatUint(uint64(u.UID), 10)
+}
+
+func cfString(str uintptr) string {
+	buf := make([]byte, consoleNameBufSize)
+	if !cfStringGetCString(str, &buf[0], len(buf), encodingUTF8) {
+		return ""
+	}
+	if end := bytes.IndexByte(buf, 0); end >= 0 {
+		return string(buf[:end])
+	}
+	return string(buf)
+}
+
+// loadConsoleUser resolves the console user symbols. It loads the keychain bindings
+// first because CFRelease is resolved there and released strings depend on it.
+func loadConsoleUser() error {
+	if err := loadKeychain(); err != nil {
+		return err
+	}
+	consoleOnce.Do(func() { consoleErr = resolveConsoleUser() })
+	return consoleErr
+}
+
+func resolveConsoleUser() error {
+	systemConfiguration, err := purego.Dlopen(systemConfigurationFramework, purego.RTLD_LAZY|purego.RTLD_GLOBAL)
+	if err != nil {
+		return fmt.Errorf("open %s: %w", systemConfigurationFramework, err)
+	}
+	coreFoundation, err := purego.Dlopen(coreFoundationFramework, purego.RTLD_LAZY|purego.RTLD_GLOBAL)
+	if err != nil {
+		return fmt.Errorf("open %s: %w", coreFoundationFramework, err)
+	}
+
+	symbol, err := purego.Dlsym(systemConfiguration, "SCDynamicStoreCopyConsoleUser")
+	if err != nil {
+		return fmt.Errorf("resolve SCDynamicStoreCopyConsoleUser: %w", err)
+	}
+	purego.RegisterFunc(&scDynamicStoreCopyConsoleUser, symbol)
+
+	symbol, err = purego.Dlsym(coreFoundation, "CFStringGetCString")
+	if err != nil {
+		return fmt.Errorf("resolve CFStringGetCString: %w", err)
+	}
+	purego.RegisterFunc(&cfStringGetCString, symbol)
+	return nil
+}

client/internal/certproof/consoleuser_darwin_test.go

@@ -0,0 +1,51 @@
+//go:build !ios
+
+package certproof
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestConsoleUser_OnlyADesktopSessionCanBeValidated(t *testing.T) {
+	tests := []struct {
+		name    string
+		user    ConsoleUser
+		desktop bool
+	}{
+		{"logged in user", ConsoleUser{Name: "maycon", UID: 501, GID: 20}, true},
+		{"login window as root", ConsoleUser{Name: "root", UID: 0}, false},
+		{"login window by name", ConsoleUser{Name: "loginwindow", UID: 0}, false},
+		{"named user still at uid 0", ConsoleUser{Name: "admin", UID: 0}, false},
+		{"no console user", ConsoleUser{}, false},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			assert.Equal(t, tt.desktop, tt.user.hasDesktop(), "only a real desktop session offers a login keychain")
+		})
+	}
+}
+
+// CurrentConsoleUser runs against the real SystemConfiguration framework. A machine with
+// a desktop open must report a non-root user; a headless runner must report none.
+func TestCurrentConsoleUser_AgreesWithItself(t *testing.T) {
+	user, ok := CurrentConsoleUser()
+	if !ok {
+		t.Log("no console user, running headless")
+		return
+	}
+	assert.NotEmpty(t, user.Name, "a console user must have a name")
+	assert.NotZero(t, user.UID, "a desktop session never belongs to uid 0")
+	assert.True(t, user.hasDesktop(), "a reported console user must be a desktop session")
+}
+
+func TestConsoleUser_IsOwner(t *testing.T) {
+	user := ConsoleUser{Name: "maycon", UID: 501, GID: 20}
+
+	assert.True(t, user.isOwner(""), "a profile without owner accepts the single console user")
+	assert.True(t, user.isOwner("maycon"), "the owner by short name")
+	assert.True(t, user.isOwner("501"), "the owner recorded as a numeric uid")
+	assert.False(t, user.isOwner("viktor"), "another account's profile must not read this user's keychain")
+	assert.False(t, user.isOwner("502"), "another uid is another account")
+}

client/internal/certproof/desktopuser_windows.go

@@ -0,0 +1,186 @@
+package certproof
+
+import (
+	"fmt"
+	"strings"
+	"unsafe"
+
+	log "github.com/sirupsen/logrus"
+	"golang.org/x/sys/windows"
+)
+
+const (
+	noActiveSession = 0xFFFFFFFF
+	servicesSession = 0
+
+	// wtsCurrentServer is WTS_CURRENT_SERVER_HANDLE; wtsActive and wtsDisconnected are
+	// WTSActive and WTSDisconnected of WTS_CONNECTSTATE_CLASS. None is exported by
+	// x/sys/windows.
+	wtsCurrentServer = windows.Handle(0)
+	wtsActive        = 0
+	wtsDisconnected  = 4
+)
+
+// DesktopUser is an interactive session and the account signed into it. The user's
+// certificate store is readable only from a process running as that account, because
+// its private keys are protected against the user profile rather than the machine.
+type DesktopUser struct {
+	Session uint32
+	Name    string
+	Token   windows.Token
+}
+
+// Close releases the session token.
+func (u DesktopUser) Close() {
+	if err := u.Token.Close(); err != nil {
+		log.Debugf("failed closing desktop session token: %v", err)
+	}
+}
+
+// CurrentDesktopUser returns a token for the session whose certificate store should be
+// asked: a session of owner, the account the active profile belongs to, with the
+// physical console preferred over remote sessions. With no owner only the console user
+// counts. Picking any signed-in user instead would let whoever else is logged in to a
+// terminal server or VDI host decide the result. The second return is false when no
+// such session exists, and only machine certificates can then be proven.
+//
+// Obtaining the token needs SE_TCB_NAME, which the LocalSystem service has and an
+// ordinary process does not.
+func CurrentDesktopUser(owner string) (DesktopUser, bool) {
+	console := windows.WTSGetActiveConsoleSessionId()
+	if owner == "" {
+		return consoleUser(console)
+	}
+
+	sessions, err := userSessions(console)
+	if err != nil {
+		log.Debugf("cannot enumerate terminal sessions: %v", err)
+		return DesktopUser{}, false
+	}
+	var found DesktopUser
+	var ok bool
+	for _, session := range sessions {
+		user, signedIn := desktopUser(session)
+		if !signedIn {
+			continue
+		}
+		switch {
+		case !sameAccountName(user.Name, owner):
+			user.Close()
+		case !ok:
+			found, ok = user, true
+		case strings.EqualFold(user.Name, found.Name):
+			// Another session of the same account; the first one in preference order wins.
+			user.Close()
+		default:
+			// An owner recorded without a domain matches accounts of several domains here.
+			// Picking one would let another domain's user answer for the owner.
+			log.Debugf("certificate posture: profile owner %s matches both %s and %s, no user certificate store is used", owner, found.Name, user.Name)
+			user.Close()
+			found.Close()
+			return DesktopUser{}, false
+		}
+	}
+	if !ok {
+		log.Debugf("certificate posture: profile owner %s has no signed-in session, no user certificate store is reachable", owner)
+	}
+	return found, ok
+}
+
+func consoleUser(console uint32) (DesktopUser, bool) {
+	if console == noActiveSession || console == servicesSession {
+		log.Debug("no console session, no user certificate store is reachable")
+		return DesktopUser{}, false
+	}
+	user, ok := desktopUser(console)
+	if !ok {
+		log.Debugf("console session %d has nobody signed in, no user certificate store is reachable", console)
+	}
+	return user, ok
+}
+
+// sameAccountName compares DOMAIN\account names case-insensitively, as Windows does, and an
+// owner given without a domain against the account part alone. The session side comes from
+// the session token's own SID, which Windows resolves from its cache of signed-in users.
+// Resolving the owner name to a SID instead would ask the domain controller, which on a
+// laptop that cannot reach it yet blocks for tens of seconds, past the collection deadline.
+func sameAccountName(sessionName, owner string) bool {
+	if strings.EqualFold(sessionName, owner) {
+		return true
+	}
+	if strings.Contains(owner, `\`) {
+		return false
+	}
+	_, account, found := strings.Cut(sessionName, `\`)
+	return found && strings.EqualFold(account, owner)
+}
+
+func desktopUser(session uint32) (DesktopUser, bool) {
+	var token windows.Token
+	if err := windows.WTSQueryUserToken(session, &token); err != nil {
+		log.Debugf("no user token for session %d: %v", session, err)
+		return DesktopUser{}, false
+	}
+
+	name, err := tokenAccount(token)
+	if err != nil {
+		log.Debugf("session %d token has no readable account: %v", session, err)
+		if closeErr := token.Close(); closeErr != nil {
+			log.Debugf("failed closing session token: %v", closeErr)
+		}
+		return DesktopUser{}, false
+	}
+	return DesktopUser{Session: session, Name: name, Token: token}, true
+}
+
+func tokenAccount(token windows.Token) (string, error) {
+	user, err := token.GetTokenUser()
+	if err != nil {
+		return "", fmt.Errorf("read token user: %w", err)
+	}
+	account, domain, _, err := user.User.Sid.LookupAccount("")
+	if err != nil {
+		return "", fmt.Errorf("look up account: %w", err)
+	}
+	if domain == "" {
+		return account, nil
+	}
+	return domain + `\` + account, nil
+}
+
+// userSessions lists the sessions a user can be signed in to, console first, then
+// active remote sessions, then disconnected ones, whose user is still signed in. Session
+// 0 is skipped: it hosts services and never belongs to an interactive user.
+func userSessions(console uint32) ([]uint32, error) {
+	var info *windows.WTS_SESSION_INFO
+	var count uint32
+	if err := windows.WTSEnumerateSessions(wtsCurrentServer, 0, 1, &info, &count); err != nil {
+		return nil, fmt.Errorf("enumerate sessions: %w", err)
+	}
+	defer windows.WTSFreeMemory(uintptr(unsafe.Pointer(info)))
+
+	var sessions []uint32
+	if console != noActiveSession && console != servicesSession {
+		sessions = append(sessions, console)
+	}
+	for _, state := range []uint32{wtsActive, wtsDisconnected} {
+		for _, session := range unsafe.Slice(info, count) {
+			if session.SessionID == servicesSession || session.SessionID == console || session.State != state {
+				continue
+			}
+			sessions = append(sessions, session.SessionID)
+		}
+	}
+	return sessions, nil
+}
+
+// runningAsLocalSystem reports whether this process is the service. The helper runs as
+// the signed-in user and must read its own store rather than launching another helper.
+func runningAsLocalSystem() bool {
+	user, err := windows.GetCurrentProcessToken().GetTokenUser()
+	if err != nil {
+		log.Debugf("failed reading own token user: %v", err)
+		return false
+	}
+	return user.User.Sid.IsWellKnown(windows.WinLocalSystemSid)
+}

client/internal/certproof/desktopuser_windows_test.go

@@ -0,0 +1,61 @@
+package certproof
+
+import (
+	"os"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestSameAccountName(t *testing.T) {
+	tests := []struct {
+		session, owner string
+		want           bool
+	}{
+		{`CORP\alice`, `CORP\alice`, true},
+		{`CORP\alice`, `corp\ALICE`, true},
+		{`CORP\alice`, `alice`, true},
+		{`CORP\alice`, `OTHER\alice`, false},
+		{`CORP\alice`, `bob`, false},
+		{`alice`, `alice`, true},
+		{`CORP\alice`, `CORP\alic`, false},
+	}
+	for _, tt := range tests {
+		assert.Equal(t, tt.want, sameAccountName(tt.session, tt.owner), "session %q, owner %q", tt.session, tt.owner)
+	}
+}
+
+// CurrentDesktopUser against the real session manager: an owner no session belongs to
+// must never yield a session, whoever else is signed in.
+func TestCurrentDesktopUser_UnknownOwnerHasNoSession(t *testing.T) {
+	user, ok := CurrentDesktopUser(`NO-SUCH-DOMAIN\no-such-user-netbird`)
+	if ok {
+		user.Close()
+	}
+	assert.False(t, ok, "a profile owner without a session must not borrow another user's store")
+}
+
+// TestCurrentDesktopUser_FindsTheOwnersSession runs against a real machine as LocalSystem,
+// with NB_TEST_DESKTOP_OWNER naming an account that is signed in. That account's session
+// must be found, and only that account's.
+func TestCurrentDesktopUser_FindsTheOwnersSession(t *testing.T) {
+	owner := os.Getenv("NB_TEST_DESKTOP_OWNER")
+	if owner == "" {
+		t.Skip("set NB_TEST_DESKTOP_OWNER to a signed-in account and run as LocalSystem")
+	}
+	user, ok := CurrentDesktopUser(owner)
+	require.True(t, ok, "the signed-in owner %s must have a session", owner)
+	defer user.Close()
+	t.Logf("owner %s resolved to session %d as %s", owner, user.Session, user.Name)
+	assert.True(t, sameAccountName(user.Name, owner), "the session found belongs to %s, not %s", owner, user.Name)
+	assert.NotZero(t, user.Session, "session 0 hosts services and never belongs to the owner")
+
+	console, consoleOK := CurrentDesktopUser("")
+	if consoleOK {
+		defer console.Close()
+		t.Logf("without an owner the console user counts: session %d as %s", console.Session, console.Name)
+	} else {
+		t.Log("without an owner nobody counts: no user at the console")
+	}
+}

client/internal/certproof/doc.go

@@ -0,0 +1,26 @@
+// Package certproof answers certificate posture challenges: it finds the certificates a
+// peer holds a private key for and signs the challenge nonce, bound to the peer's
+// WireGuard key, with that key. Management verifies the signature and the chain against
+// the CAs of the check, so only a holder of the key passes and a proof made for one peer
+// cannot be replayed by another. Keys in the OS stores, the TPM and a PKCS#11 token are
+// used through the platform, which signs; a plain PEM key file is the exception, parsed
+// and used in the daemon's memory.
+//
+// Machine stores are read by the daemon itself: the Windows LocalMachine store, the macOS
+// System keychain, and on Linux a directory of PEM files, TSS2 key files the TPM signs
+// with, and a PKCS#11 token. On Linux the directory and the token URI come from the
+// daemon's environment (NB_CERT_STORE_DIR, NB_CERT_PKCS11_URI), as does the token PIN
+// (NB_TPM_PIN); none of them is read from the profile config.
+//
+// User stores cannot be read by a privileged daemon, so it starts this binary as
+// "netbird posture cert-proof" inside the user's session and receives only signatures
+// and chains on its stdout. On macOS uid is not what unlocks a login keychain, the
+// session's bootstrap namespace is, which is why the helper enters it through launchctl
+// asuser. On Windows a service opening CURRENT_USER silently reads its own empty hive, so
+// the helper runs with the session's token instead. Its output is untrusted: sizes are
+// capped and only proofs for requested nonces are kept.
+//
+// A user proof shows that some process in that user's session could use a key whose
+// certificate chains to the CA. It does not show which binary signed, or that the key is
+// hardware-bound; device stores are the stronger signal.
+package certproof

client/internal/certproof/helper.go

@@ -0,0 +1,66 @@
+package certproof
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+
+	log "github.com/sirupsen/logrus"
+
+	"github.com/netbirdio/netbird/shared/management/certposture"
+	"github.com/netbirdio/netbird/shared/management/proto"
+)
+
+// HelperRequest is the work the daemon hands to a helper running in a user session. The
+// peer key binds every signature to this machine, so a proof cannot be replayed onto
+// another peer.
+type HelperRequest struct {
+	PeerKey    []byte            `json:"peerKey"`
+	Challenges []HelperChallenge `json:"challenges"`
+}
+
+// HelperChallenge is one certificate challenge in the form the helper needs.
+type HelperChallenge struct {
+	Nonce          []byte   `json:"nonce"`
+	CACertificates []string `json:"caCertificates"`
+}
+
+// HelperResponse carries the proofs the helper produced from its own keychain.
+type HelperResponse struct {
+	Proofs []certposture.Proof `json:"proofs"`
+}
+
+// RunHelper answers the challenges on in from the store of the user running this
+// process and writes the proofs to out. It is the child half of the console user
+// lookup: the daemon cannot read a login keychain, so it launches this in the user's
+// session instead. Only the signature crosses back, never the private key.
+func RunHelper(ctx context.Context, in io.Reader, out io.Writer) error {
+	return runHelper(ctx, helperStore(), in, out)
+}
+
+func runHelper(ctx context.Context, store Store, in io.Reader, out io.Writer) error {
+	var req HelperRequest
+	if err := json.NewDecoder(in).Decode(&req); err != nil {
+		return fmt.Errorf("decode helper request: %w", err)
+	}
+
+	challenges := make([]*proto.CertificateChallenge, 0, len(req.Challenges))
+	for _, challenge := range req.Challenges {
+		challenges = append(challenges, &proto.CertificateChallenge{
+			Nonce:          challenge.Nonce,
+			CaCertificates: challenge.CACertificates,
+		})
+	}
+
+	var proofs []certposture.Proof
+	if len(challenges) > 0 {
+		proofs = CollectChallenges(ctx, store, challenges, req.PeerKey)
+	}
+	log.Debugf("certificate posture helper: answering %d challenges with %d proofs", len(challenges), len(proofs))
+
+	if err := json.NewEncoder(out).Encode(HelperResponse{Proofs: proofs}); err != nil {
+		return fmt.Errorf("encode helper response: %w", err)
+	}
+	return nil
+}

client/internal/certproof/helper_backoff.go

@@ -0,0 +1,78 @@
+package certproof
+
+import (
+	"crypto/sha256"
+	"encoding/hex"
+	"slices"
+	"strings"
+	"sync"
+	"time"
+
+	"github.com/netbirdio/netbird/shared/management/proto"
+)
+
+// helperQuietPeriod is how long a user whose helper proved nothing is not asked again
+// for the same CAs. On macOS each helper run may show a keychain prompt, so retrying on
+// every collection would put up a new one each time the user ignored or denied the last.
+const helperQuietPeriod = time.Hour
+
+// helperBackoff remembers, per user and set of CAs asked about, until when the helper
+// is not launched again because its last run proved nothing.
+type helperBackoff struct {
+	mu    sync.Mutex
+	until map[string]time.Time
+}
+
+func newHelperBackoff() *helperBackoff {
+	return &helperBackoff{until: map[string]time.Time{}}
+}
+
+// allow reports whether the helper may be launched for key now.
+func (b *helperBackoff) allow(key string, now time.Time) bool {
+	b.mu.Lock()
+	defer b.mu.Unlock()
+	return !now.Before(b.until[key])
+}
+
+// record stores the outcome of a helper run for key: one that proved something clears
+// the back-off, one that proved nothing starts it. Expired entries are dropped, so the
+// map holds only users currently held off.
+func (b *helperBackoff) record(key string, proven bool, now time.Time) {
+	b.mu.Lock()
+	defer b.mu.Unlock()
+	for other, until := range b.until {
+		if !now.Before(until) {
+			delete(b.until, other)
+		}
+	}
+	if proven {
+		delete(b.until, key)
+		return
+	}
+	b.until[key] = now.Add(helperQuietPeriod)
+}
+
+// helperBackoffKey identifies a user and the set of CAs the challenges accept, in any
+// order, leaving out the nonces, which rotate without changing what the user is asked to
+// prove.
+func helperBackoffKey(user string, challenges []*proto.CertificateChallenge) string {
+	sets := make([]string, 0, len(challenges))
+	for _, challenge := range challenges {
+		cas := make([]string, 0, len(challenge.GetCaCertificates()))
+		for _, ca := range challenge.GetCaCertificates() {
+			sum := sha256.Sum256([]byte(ca))
+			cas = append(cas, hex.EncodeToString(sum[:]))
+		}
+		slices.Sort(cas)
+		sets = append(sets, strings.Join(cas, ","))
+	}
+	slices.Sort(sets)
+
+	h := sha256.New()
+	h.Write([]byte(user))
+	for _, set := range sets {
+		h.Write([]byte{0})
+		h.Write([]byte(set))
+	}
+	return hex.EncodeToString(h.Sum(nil))
+}

client/internal/certproof/helper_backoff_test.go

@@ -0,0 +1,69 @@
+package certproof
+
+import (
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+
+	"github.com/netbirdio/netbird/shared/management/proto"
+)
+
+// TestHelperBackoff: a user whose helper proved nothing, for one ignoring or denying a
+// keychain prompt, is not asked again for the same CAs within the quiet period, while a
+// success or other CAs are asked at once.
+func TestHelperBackoff(t *testing.T) {
+	now := time.Now()
+	b := newHelperBackoff()
+	key := helperBackoffKey("501", []*proto.CertificateChallenge{{CaCertificates: []string{"ca-a"}}})
+
+	assert.True(t, b.allow(key, now), "a user never asked is asked")
+
+	b.record(key, false, now)
+	assert.False(t, b.allow(key, now.Add(helperQuietPeriod-time.Second)), "no new prompt within the quiet period")
+	assert.True(t, b.allow(key, now.Add(helperQuietPeriod)), "asked again once the quiet period passed")
+
+	other := helperBackoffKey("501", []*proto.CertificateChallenge{{CaCertificates: []string{"ca-b"}}})
+	assert.True(t, b.allow(other, now), "a check for other CAs is asked at once")
+	assert.True(t, b.allow(helperBackoffKey("502", []*proto.CertificateChallenge{{CaCertificates: []string{"ca-a"}}}), now), "another user is asked at once")
+
+	b.record(key, true, now)
+	assert.True(t, b.allow(key, now), "a run that proved something clears the quiet period")
+}
+
+// TestHelperBackoffKey_IgnoresNonces: nonces rotate every window without changing what
+// the user is asked to prove, so they do not reset the quiet period.
+func TestHelperBackoffKey_IgnoresNonces(t *testing.T) {
+	a := helperBackoffKey("501", []*proto.CertificateChallenge{{Nonce: []byte("one"), CaCertificates: []string{"ca"}}})
+	b := helperBackoffKey("501", []*proto.CertificateChallenge{{Nonce: []byte("two"), CaCertificates: []string{"ca"}}})
+	assert.Equal(t, a, b, "the key does not depend on the nonce")
+}
+
+// TestHelperBackoffKey_IgnoresOrder: the same CAs, listed in another order or with the
+// challenges reordered, are the same question to the user.
+func TestHelperBackoffKey_IgnoresOrder(t *testing.T) {
+	a := helperBackoffKey("501", []*proto.CertificateChallenge{
+		{CaCertificates: []string{"ca-1", "ca-2"}},
+		{CaCertificates: []string{"ca-3"}},
+	})
+	b := helperBackoffKey("501", []*proto.CertificateChallenge{
+		{CaCertificates: []string{"ca-3"}},
+		{CaCertificates: []string{"ca-2", "ca-1"}},
+	})
+	assert.Equal(t, a, b, "the key does not depend on the order of CAs or challenges")
+
+	c := helperBackoffKey("501", []*proto.CertificateChallenge{{CaCertificates: []string{"ca-1", "ca-2", "ca-3"}}})
+	assert.NotEqual(t, a, c, "the same CAs grouped into other challenges are another question")
+}
+
+// TestHelperBackoff_PrunesExpired: entries whose quiet period passed are dropped when the
+// next outcome is recorded, so the map does not grow with every user and CA set seen.
+func TestHelperBackoff_PrunesExpired(t *testing.T) {
+	now := time.Now()
+	b := newHelperBackoff()
+	b.record("old", false, now)
+	b.record("new", false, now.Add(helperQuietPeriod))
+
+	assert.NotContains(t, b.until, "old", "an expired entry is pruned")
+	assert.Contains(t, b.until, "new", "a current entry is kept")
+}

client/internal/certproof/helper_kill_other.go

@@ -0,0 +1,10 @@
+//go:build !unix && !windows
+
+package certproof
+
+import "os/exec"
+
+// startHelper starts cmd.
+func startHelper(cmd *exec.Cmd) (func(), error) {
+	return func() {}, cmd.Start()
+}

client/internal/certproof/helper_kill_unix.go

@@ -0,0 +1,35 @@
+//go:build unix
+
+package certproof
+
+import (
+	"errors"
+	"os/exec"
+	"syscall"
+)
+
+// killHelperGroupOnCancel puts cmd in a process group of its own and kills the whole
+// group when cmd's context ends. The helper may run below launchers such as launchctl
+// and sudo, which do not pass a kill on, and it keeps the output pipes open, so killing
+// only the direct child would leave the helper running and Wait blocked until the
+// helper exits on its own, for one waiting on a keychain prompt nobody answers.
+func killHelperGroupOnCancel(cmd *exec.Cmd) {
+	if cmd.SysProcAttr == nil {
+		cmd.SysProcAttr = &syscall.SysProcAttr{}
+	}
+	cmd.SysProcAttr.Setpgid = true
+	cmd.Cancel = func() error {
+		err := syscall.Kill(-cmd.Process.Pid, syscall.SIGKILL)
+		if errors.Is(err, syscall.ESRCH) {
+			return nil
+		}
+		return err
+	}
+	cmd.WaitDelay = helperWaitDelay
+}
+
+// startHelper starts cmd. Killing the helper with everything below it is arranged by
+// killHelperGroupOnCancel where the helper runs under a launcher.
+func startHelper(cmd *exec.Cmd) (func(), error) {
+	return func() {}, cmd.Start()
+}

client/internal/certproof/helper_kill_windows.go

@@ -0,0 +1,77 @@
+//go:build windows
+
+package certproof
+
+import (
+	"fmt"
+	"os/exec"
+	"unsafe"
+
+	log "github.com/sirupsen/logrus"
+	"golang.org/x/sys/windows"
+)
+
+// startHelper starts cmd inside a job object that is terminated when cmd's context ends
+// and closed, killing whatever is left in it, once the helper has been awaited. Killing
+// only the helper on Windows leaves its descendants running, and one that holds the
+// output pipe would block Wait until it exits, so WaitDelay bounds that wait as well.
+func startHelper(cmd *exec.Cmd) (func(), error) {
+	job, err := newKillOnCloseJob()
+	if err != nil {
+		return nil, err
+	}
+	closeJob := func() {
+		if err := windows.CloseHandle(job); err != nil {
+			log.Debugf("failed to close certificate proof helper job: %v", err)
+		}
+	}
+
+	cmd.Cancel = func() error {
+		return windows.TerminateJobObject(job, 1)
+	}
+	cmd.WaitDelay = helperWaitDelay
+	if err := cmd.Start(); err != nil {
+		closeJob()
+		return nil, err
+	}
+
+	// A helper outside the job is still killed directly and bounded by WaitDelay; only
+	// its descendants would outlive it.
+	if err := assignToJob(job, cmd.Process.Pid); err != nil {
+		log.Debugf("failed to put certificate proof helper %d in its job: %v", cmd.Process.Pid, err)
+		cmd.Cancel = func() error { return cmd.Process.Kill() }
+	}
+	return closeJob, nil
+}
+
+// newKillOnCloseJob creates a job object whose processes are killed when its last handle
+// is closed.
+func newKillOnCloseJob() (windows.Handle, error) {
+	job, err := windows.CreateJobObject(nil, nil)
+	if err != nil {
+		return 0, fmt.Errorf("create job object: %w", err)
+	}
+	info := windows.JOBOBJECT_EXTENDED_LIMIT_INFORMATION{
+		BasicLimitInformation: windows.JOBOBJECT_BASIC_LIMIT_INFORMATION{
+			LimitFlags: windows.JOB_OBJECT_LIMIT_KILL_ON_JOB_CLOSE,
+		},
+	}
+	if _, err := windows.SetInformationJobObject(job, windows.JobObjectExtendedLimitInformation,
+		uintptr(unsafe.Pointer(&info)), uint32(unsafe.Sizeof(info))); err != nil {
+		_ = windows.CloseHandle(job)
+		return 0, fmt.Errorf("set job object limits: %w", err)
+	}
+	return job, nil
+}
+
+// assignToJob puts the process pid into job.
+func assignToJob(job windows.Handle, pid int) error {
+	process, err := windows.OpenProcess(windows.PROCESS_SET_QUOTA|windows.PROCESS_TERMINATE, false, uint32(pid))
+	if err != nil {
+		return fmt.Errorf("open process: %w", err)
+	}
+	defer func() {
+		_ = windows.CloseHandle(process)
+	}()
+	return windows.AssignProcessToJobObject(job, process)
+}

client/internal/certproof/helper_kill_windows_test.go

@@ -0,0 +1,29 @@
+//go:build windows
+
+package certproof
+
+import (
+	"context"
+	"os/exec"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// TestRunHelperCmd_KillsHelperTree: the helper's child holds stdout and runs for a minute,
+// like a helper stuck below a launcher. Killing only the helper would leave Wait blocked
+// on the child's pipe until WaitDelay gives up; terminating the job ends both at once.
+func TestRunHelperCmd_KillsHelperTree(t *testing.T) {
+	ctx, cancel := context.WithTimeout(context.Background(), 200*time.Millisecond)
+	defer cancel()
+
+	cmd := exec.CommandContext(ctx, "cmd.exe", "/c", "ping -n 60 127.0.0.1")
+
+	start := time.Now()
+	_, err := runHelperCmd(cmd, HelperRequest{Challenges: []HelperChallenge{{Nonce: []byte("asked")}}})
+
+	require.Error(t, err)
+	assert.Less(t, time.Since(start), helperWaitDelay, "the whole tree is killed at the timeout, not left for WaitDelay")
+}

client/internal/certproof/helper_run.go

@@ -0,0 +1,126 @@
+package certproof
+
+import (
+	"bytes"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"os/exec"
+	"strings"
+	"time"
+
+	log "github.com/sirupsen/logrus"
+
+	"github.com/netbirdio/netbird/shared/management/certposture"
+)
+
+const (
+	maxHelperStdout = 1 << 20
+	maxHelperStderr = 4 << 10
+
+	// helperWaitDelay bounds how long a helper's output is awaited after it was killed.
+	helperWaitDelay = 2 * time.Second
+)
+
+var errHelperOutputTooLarge = errors.New("helper output exceeds the size limit")
+
+// runHelperCmd feeds req to the helper process cmd and returns the proofs it answered.
+// The helper runs as an unprivileged user who can control its output, so both streams
+// are capped and only proofs for a nonce req asked about are kept, one per challenge.
+func runHelperCmd(cmd *exec.Cmd, req HelperRequest) ([]certposture.Proof, error) {
+	payload, err := json.Marshal(req)
+	if err != nil {
+		return nil, fmt.Errorf("encode helper request: %w", err)
+	}
+
+	stdout := &cappedBuffer{limit: maxHelperStdout}
+	stderr := &cappedBuffer{limit: maxHelperStderr}
+	cmd.Stdin = bytes.NewReader(payload)
+	cmd.Stdout = stdout
+	cmd.Stderr = stderr
+
+	release, err := startHelper(cmd)
+	if err != nil {
+		return nil, fmt.Errorf("start helper: %w", err)
+	}
+	defer release()
+	if err := cmd.Wait(); err != nil {
+		return nil, fmt.Errorf("%w: %s", err, strings.TrimSpace(stderr.String()))
+	}
+	logHelperStderr(stderr.String())
+	if stdout.truncated {
+		return nil, errHelperOutputTooLarge
+	}
+
+	var resp HelperResponse
+	if err := json.Unmarshal(stdout.Bytes(), &resp); err != nil {
+		return nil, fmt.Errorf("decode helper response: %w", err)
+	}
+	return requestedProofs(req, resp.Proofs), nil
+}
+
+// logHelperStderr records what a helper that succeeded wrote to stderr, such as a user
+// certificate it could not sign with, which is otherwise only visible in the user's
+// session. Each line is quoted because the helper's user controls its content.
+func logHelperStderr(stderr string) {
+	for _, line := range strings.Split(strings.TrimSpace(stderr), "\n") {
+		if line = strings.TrimSpace(line); line != "" {
+			log.Debugf("certificate posture helper: %q", line)
+		}
+	}
+}
+
+// requestedProofs keeps the proofs whose nonce belongs to one of req's challenges, at
+// most as many as req has challenges.
+func requestedProofs(req HelperRequest, proofs []certposture.Proof) []certposture.Proof {
+	var kept []certposture.Proof
+	for _, proof := range proofs {
+		if len(kept) == len(req.Challenges) {
+			break
+		}
+		if !req.asked(proof.Nonce) {
+			log.Debugf("certificate posture: dropping helper proof for a nonce that was not requested")
+			continue
+		}
+		kept = append(kept, proof)
+	}
+	return kept
+}
+
+func (r HelperRequest) asked(nonce []byte) bool {
+	for _, challenge := range r.Challenges {
+		if len(challenge.Nonce) > 0 && bytes.Equal(challenge.Nonce, nonce) {
+			return true
+		}
+	}
+	return false
+}
+
+// cappedBuffer keeps the first limit bytes written to it and discards the rest, so a
+// misbehaving child cannot grow the parent's memory without bound. The buffer is a
+// named field rather than embedded: an embedded bytes.Buffer would promote ReadFrom,
+// which io.Copy prefers over Write, bypassing the cap.
+type cappedBuffer struct {
+	buf       bytes.Buffer
+	limit     int
+	truncated bool
+}
+
+func (b *cappedBuffer) Write(p []byte) (int, error) {
+	if room := b.limit - b.buf.Len(); room < len(p) {
+		b.truncated = true
+		if room > 0 {
+			b.buf.Write(p[:room])
+		}
+		return len(p), nil
+	}
+	return b.buf.Write(p)
+}
+
+func (b *cappedBuffer) Bytes() []byte {
+	return b.buf.Bytes()
+}
+
+func (b *cappedBuffer) String() string {
+	return b.buf.String()
+}

client/internal/certproof/helper_run_test.go

@@ -0,0 +1,119 @@
+//go:build !windows && !js
+
+package certproof
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"os"
+	"os/exec"
+	"strings"
+	"testing"
+	"time"
+
+	log "github.com/sirupsen/logrus"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/shared/management/certposture"
+)
+
+// fakeHelper is a child process that drains its stdin and then prints script's output,
+// standing in for a helper running in a user session the daemon cannot trust.
+func fakeHelper(t *testing.T, script string) *exec.Cmd {
+	t.Helper()
+	return exec.Command("/bin/sh", "-c", "cat >/dev/null; "+script)
+}
+
+func printJSON(t *testing.T, v any) string {
+	t.Helper()
+	out, err := json.Marshal(v)
+	require.NoError(t, err)
+	return fmt.Sprintf("printf '%%s' '%s'", out)
+}
+
+func TestRunHelperCmd_KeepsOnlyRequestedProofs(t *testing.T) {
+	req := HelperRequest{PeerKey: peerKey, Challenges: []HelperChallenge{{Nonce: []byte("asked")}}}
+	resp := HelperResponse{Proofs: []certposture.Proof{
+		{Nonce: []byte("injected"), Signature: []byte("x")},
+		{Nonce: []byte("asked"), Signature: []byte("first")},
+		{Nonce: []byte("asked"), Signature: []byte("second")},
+	}}
+
+	proofs, err := runHelperCmd(fakeHelper(t, printJSON(t, resp)), req)
+
+	require.NoError(t, err)
+	require.Len(t, proofs, 1, "a helper may not return more proofs than challenges or proofs for nonces it was not asked about")
+	assert.Equal(t, []byte("first"), proofs[0].Signature, "the first proof for the requested nonce is kept")
+}
+
+func TestRunHelperCmd_RejectsOversizedOutput(t *testing.T) {
+	req := HelperRequest{Challenges: []HelperChallenge{{Nonce: []byte("asked")}}}
+	script := fmt.Sprintf("head -c %d /dev/zero", maxHelperStdout+1)
+
+	_, err := runHelperCmd(fakeHelper(t, script), req)
+
+	assert.ErrorIs(t, err, errHelperOutputTooLarge)
+}
+
+func TestRunHelperCmd_CapsStderrInError(t *testing.T) {
+	req := HelperRequest{Challenges: []HelperChallenge{{Nonce: []byte("asked")}}}
+	script := fmt.Sprintf("head -c %d /dev/zero | tr '\\0' 'a' >&2; exit 3", 10*maxHelperStderr)
+
+	_, err := runHelperCmd(fakeHelper(t, script), req)
+
+	require.Error(t, err)
+	assert.LessOrEqual(t, len(err.Error()), maxHelperStderr+100, "a chatty helper must not blow up the daemon's error or log line")
+	assert.True(t, strings.Contains(err.Error(), "exit status 3"), "the exit status is kept: %v", err)
+}
+
+// TestRunHelperCmd_LogsStderrOfSuccessfulHelper covers a helper that answers but warns,
+// for one about a certificate it could not sign with: the warning reaches the daemon's
+// log, quoted so a helper cannot forge log lines with embedded newlines.
+func TestRunHelperCmd_LogsStderrOfSuccessfulHelper(t *testing.T) {
+	var logged bytes.Buffer
+	log.SetOutput(&logged)
+	level := log.GetLevel()
+	log.SetLevel(log.DebugLevel)
+	t.Cleanup(func() {
+		log.SetOutput(os.Stderr)
+		log.SetLevel(level)
+	})
+
+	req := HelperRequest{Challenges: []HelperChallenge{{Nonce: []byte("asked")}}}
+	script := "printf 'failed signing certificate proof for CN=user\\nforged line\\n' >&2; " + printJSON(t, HelperResponse{})
+
+	_, err := runHelperCmd(fakeHelper(t, script), req)
+
+	require.NoError(t, err)
+	assert.Contains(t, logged.String(), "failed signing certificate proof for CN=user", "the helper's warning reaches the daemon log")
+	assert.NotContains(t, logged.String(), "\nforged line", "helper output cannot start a log line of its own")
+}
+
+// TestRunHelperCmd_KillsHelperBelowLauncher stands in for launchctl and sudo starting the
+// helper: the direct child spawns a grandchild that holds stdout and never exits, like a
+// helper waiting on a keychain prompt. The timeout must end both, not wait for the
+// grandchild.
+func TestRunHelperCmd_KillsHelperBelowLauncher(t *testing.T) {
+	ctx, cancel := context.WithTimeout(context.Background(), 200*time.Millisecond)
+	defer cancel()
+
+	cmd := exec.CommandContext(ctx, "/bin/sh", "-c", "sleep 30 & cat >/dev/null; wait")
+	killHelperGroupOnCancel(cmd)
+
+	start := time.Now()
+	_, err := runHelperCmd(cmd, HelperRequest{Challenges: []HelperChallenge{{Nonce: []byte("asked")}}})
+
+	require.Error(t, err)
+	assert.Less(t, time.Since(start), 5*time.Second, "the helper is killed at the timeout, not awaited until it exits")
+}
+
+func TestRunHelperCmd_RejectsGarbage(t *testing.T) {
+	req := HelperRequest{Challenges: []HelperChallenge{{Nonce: []byte("asked")}}}
+
+	_, err := runHelperCmd(fakeHelper(t, "printf 'not json'"), req)
+
+	assert.ErrorContains(t, err, "decode helper response")
+}

client/internal/certproof/helper_spawn.go

@@ -0,0 +1,63 @@
+//go:build (darwin && !ios) || windows
+
+package certproof
+
+import (
+	"crypto/sha256"
+	"crypto/x509"
+
+	log "github.com/sirupsen/logrus"
+
+	"github.com/netbirdio/netbird/shared/management/certposture"
+	"github.com/netbirdio/netbird/shared/management/proto"
+)
+
+func helperRequest(challenges []*proto.CertificateChallenge, peerKey []byte) HelperRequest {
+	req := HelperRequest{PeerKey: peerKey, Challenges: make([]HelperChallenge, 0, len(challenges))}
+	for _, challenge := range challenges {
+		req.Challenges = append(req.Challenges, HelperChallenge{
+			Nonce:          challenge.GetNonce(),
+			CACertificates: challenge.GetCaCertificates(),
+		})
+	}
+	return req
+}
+
+// mergeProofs appends the user session proofs to the device proofs, dropping a leaf that
+// both stores hold so the same certificate is proven once.
+func mergeProofs(device, user []certposture.Proof) []certposture.Proof {
+	if len(user) == 0 {
+		return device
+	}
+
+	seen := make(map[[sha256.Size]byte]struct{}, len(device))
+	for _, proof := range device {
+		if len(proof.Chain) > 0 {
+			seen[sha256.Sum256(proof.Chain[0])] = struct{}{}
+		}
+	}
+
+	merged := device
+	for _, proof := range user {
+		if len(proof.Chain) == 0 {
+			continue
+		}
+		fingerprint := sha256.Sum256(proof.Chain[0])
+		if _, done := seen[fingerprint]; done {
+			continue
+		}
+		seen[fingerprint] = struct{}{}
+		merged = append(merged, proof)
+		logUserProof(proof)
+	}
+	return merged
+}
+
+func logUserProof(proof certposture.Proof) {
+	leaf, err := x509.ParseCertificate(proof.Chain[0])
+	if err != nil {
+		log.Debugf("certificate posture: user proof carries an unparsable leaf: %v", err)
+		return
+	}
+	log.Debugf("certificate posture: signed-in user proved %q issued by %q", leaf.Subject, leaf.Issuer)
+}

client/internal/certproof/helper_spawn_test.go

@@ -0,0 +1,56 @@
+//go:build (darwin && !ios) || windows
+
+package certproof
+
+import (
+	"crypto/sha256"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/shared/management/certposture"
+	"github.com/netbirdio/netbird/shared/management/certposture/certtest"
+	"github.com/netbirdio/netbird/shared/management/proto"
+)
+
+func TestMergeProofs_ProvesACertificateHeldByBothStoresOnce(t *testing.T) {
+	ca := certtest.NewCA(t, "corp-root")
+	shared := ca.Issue(t, certtest.ECDSAKey(t), "shared")
+	userOnly := ca.Issue(t, certtest.ECDSAKey(t), "user-only")
+
+	device := []certposture.Proof{{Chain: [][]byte{shared.Raw}}}
+	user := []certposture.Proof{{Chain: [][]byte{shared.Raw}}, {Chain: [][]byte{userOnly.Raw}}, {}}
+
+	merged := mergeProofs(device, user)
+
+	require.Len(t, merged, 2, "the shared leaf is proven once and the empty chain is dropped")
+	assert.Equal(t, shared.Raw, merged[0].Chain[0], "the device proof keeps its place")
+	assert.Equal(t, userOnly.Raw, merged[1].Chain[0], "the user-only certificate is appended")
+}
+
+func TestMergeProofs_KeepsDeviceProofsWhenNoUserSession(t *testing.T) {
+	ca := certtest.NewCA(t, "corp-root")
+	device := []certposture.Proof{{Chain: [][]byte{ca.Issue(t, certtest.ECDSAKey(t), "device").Raw}}}
+
+	merged := mergeProofs(device, nil)
+
+	require.Len(t, merged, 1, "a machine with nobody signed in still sends its device proof")
+	assert.Equal(t, sha256.Sum256(device[0].Chain[0]), sha256.Sum256(merged[0].Chain[0]), "the device proof is unchanged")
+}
+
+func TestHelperRequest_CarriesEveryChallenge(t *testing.T) {
+	ca := certtest.NewCA(t, "corp-root")
+	challenges := []*proto.CertificateChallenge{
+		{Nonce: []byte("first"), CaCertificates: []string{ca.PEM}},
+		{Nonce: []byte("second")},
+	}
+
+	req := helperRequest(challenges, peerKey)
+
+	require.Len(t, req.Challenges, 2, "every challenge must reach the helper")
+	assert.Equal(t, peerKey, req.PeerKey, "the peer key binds the signature to this machine")
+	assert.Equal(t, []byte("first"), req.Challenges[0].Nonce, "the nonce must survive unchanged")
+	assert.Equal(t, []string{ca.PEM}, req.Challenges[0].CACertificates, "the accepted CAs must survive unchanged")
+	assert.Empty(t, req.Challenges[1].CACertificates, "a challenge without CAs stays without CAs")
+}

client/internal/certproof/helper_test.go

@@ -0,0 +1,63 @@
+package certproof
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/shared/management/certposture"
+	"github.com/netbirdio/netbird/shared/management/certposture/certtest"
+)
+
+func TestRunHelper_ProofSurvivesTheProcessBoundary(t *testing.T) {
+	ca := certtest.NewCA(t, "corp-root")
+	dir := storeDir(t)
+	key := certtest.ECDSAKey(t)
+	writeFile(t, dir, "device.pem", certtest.CertPEM(ca.Issue(t, key, "device"))+certtest.KeyPEM(t, key))
+
+	challenger := certposture.NewChallenger([]byte("secret"))
+	nonce := challenger.Nonce(peerKey, time.Now())
+	request, err := json.Marshal(HelperRequest{
+		PeerKey:    peerKey,
+		Challenges: []HelperChallenge{{Nonce: nonce, CACertificates: []string{ca.PEM}}},
+	})
+	require.NoError(t, err, "request must encode")
+
+	var stdout bytes.Buffer
+	require.NoError(t, runHelper(context.Background(), NewFileStore(dir), bytes.NewReader(request), &stdout))
+
+	var resp HelperResponse
+	require.NoError(t, json.Unmarshal(stdout.Bytes(), &resp), "helper must emit decodable JSON")
+	require.Len(t, resp.Proofs, 1, "the matching certificate should produce one proof")
+
+	// Verify exactly as management does, so the proof is proven to survive the encode,
+	// the process boundary and the decode intact.
+	chain, err := challenger.Verify(resp.Proofs[0], peerKey, time.Now())
+	require.NoError(t, err, "the decoded proof must verify against the issued nonce")
+	assert.Equal(t, "device", chain[0].Subject.CommonName, "the proven leaf should be the device certificate")
+}
+
+func TestRunHelper_NoChallengesYieldsEmptyResponse(t *testing.T) {
+	request, err := json.Marshal(HelperRequest{PeerKey: peerKey})
+	require.NoError(t, err)
+
+	var stdout bytes.Buffer
+	require.NoError(t, runHelper(context.Background(), NewFileStore(storeDir(t)), bytes.NewReader(request), &stdout))
+
+	var resp HelperResponse
+	require.NoError(t, json.Unmarshal(stdout.Bytes(), &resp), "an empty request must still emit valid JSON")
+	assert.Empty(t, resp.Proofs, "no challenges should produce no proofs")
+}
+
+func TestRunHelper_RejectsMalformedRequest(t *testing.T) {
+	var stdout bytes.Buffer
+	err := runHelper(context.Background(), NewFileStore(storeDir(t)), bytes.NewReader([]byte("not json")), &stdout)
+
+	require.Error(t, err, "a malformed request must fail rather than emit an empty proof set")
+	assert.Empty(t, stdout.String(), "nothing should be written to stdout on a decode failure")
+}

client/internal/certproof/keychain_darwin.go

@@ -0,0 +1,483 @@
+//go:build !ios
+
+package certproof
+
+import (
+	"bytes"
+	"context"
+	"crypto"
+	"crypto/x509"
+	"errors"
+	"fmt"
+	"io"
+	"os"
+	"sync"
+	"unsafe"
+
+	"github.com/ebitengine/purego"
+	log "github.com/sirupsen/logrus"
+)
+
+const (
+	securityFramework       = "/System/Library/Frameworks/Security.framework/Security"
+	coreFoundationFramework = "/System/Library/Frameworks/CoreFoundation.framework/CoreFoundation"
+
+	errSecItemNotFound          = -25300
+	errSecInteractionNotAllowed = -25308
+)
+
+// errKeyNeedsApproval reports a key whose access list does not include netbird, so using
+// it needs the user's approval, which a daemon has no UI to ask for.
+var errKeyNeedsApproval = errors.New("the key's access control requires user approval for netbird; " +
+	"import the identity with netbird allowed (security import <identity.p12> -k <keychain> " +
+	"-T /Applications/NetBird.app/Contents/MacOS/netbird), " +
+	"or set AllowAllAppsAccess in the MDM certificate payload, which allows every application")
+
+var (
+	keychainOnce sync.Once
+	keychainErr  error
+
+	secItemCopyMatching        func(query uintptr, result *uintptr) int32
+	secIdentityCopyCertificate func(identity uintptr, cert *uintptr) int32
+	secIdentityCopyPrivateKey  func(identity uintptr, key *uintptr) int32
+	secCertificateCopyData     func(cert uintptr) uintptr
+	secKeyCreateSignature      func(key, algorithm, data uintptr, err *uintptr) uintptr
+
+	secKeychainOpen           func(path *byte, keychain *uintptr) int32
+	secKeychainCopySearchList func(searchList *uintptr) int32
+	secKeychainGetPath        func(keychain uintptr, pathLength *uint32, path *byte) int32
+	cfArrayCreate             func(alloc uintptr, values *uintptr, count int, callBacks uintptr) uintptr
+
+	cfDictionaryCreate     func(alloc uintptr, keys, values *uintptr, count int, keyCallBacks, valueCallBacks uintptr) uintptr
+	cfArrayGetCount        func(array uintptr) int
+	cfArrayGetValueAtIndex func(array uintptr, index int) uintptr
+	cfDataCreate           func(alloc uintptr, data *byte, length int) uintptr
+	cfDataGetLength        func(data uintptr) int
+	cfDataGetBytePtr       func(data uintptr) unsafe.Pointer
+	cfErrorGetCode         func(err uintptr) int
+	cfRelease              func(ref uintptr)
+
+	kSecClass, kSecClassIdentity, kSecClassCertificate, kSecMatchLimit, kSecMatchLimitAll, kSecReturnRef  uintptr
+	kSecMatchSearchList                                                                                   uintptr
+	kSecKeyAlgorithmECDSASHA256, kSecKeyAlgorithmECDSASHA384, kSecKeyAlgorithmRSAPSSSHA256                uintptr
+	kCFBooleanTrue, kCFTypeDictionaryKeyCallBacks, kCFTypeDictionaryValueCallBacks, kCFTypeArrayCallBacks uintptr
+)
+
+// systemKeychain is where MDM installs device identities.
+const systemKeychain = "/Library/Keychains/System.keychain"
+
+// DefaultStore is the System keychain for the root daemon, so it never touches a user's
+// keychain, whose keys would ask for approval in a session the daemon has no UI in. Run
+// by an ordinary user, it is that user's keychain search list.
+func DefaultStore() Store {
+	if os.Geteuid() == 0 {
+		return NewKeychainStore(systemKeychain)
+	}
+	return NewKeychainStore()
+}
+
+// KeychainStore yields the identities of a set of keychains, or of the process's keychain
+// search list when none is named, reached through purego so the client keeps building
+// with CGO_ENABLED=0.
+type KeychainStore struct {
+	keychains []string
+}
+
+// NewKeychainStore returns a store that searches the keychain files at paths, or the
+// process's keychain search list when no path is given.
+func NewKeychainStore(paths ...string) *KeychainStore {
+	return &KeychainStore{keychains: paths}
+}
+
+func (s *KeychainStore) Candidates(_ context.Context) ([]Candidate, error) {
+	if err := loadKeychain(); err != nil {
+		return nil, err
+	}
+	searchList, done, err := searchListOf(s.keychains)
+	if err != nil {
+		return nil, err
+	}
+	defer done()
+
+	var leaves []*x509.Certificate
+	err = eachIdentity(searchList, func(_ uintptr, der []byte) (bool, error) {
+		cert, err := x509.ParseCertificate(der)
+		if err != nil {
+			log.Warnf("skipping keychain identity: %v", err)
+			return false, nil
+		}
+		log.Debugf("keychain identity: subject=%q issuer=%q serial=%s expires=%s", cert.Subject, cert.Issuer, cert.SerialNumber, cert.NotAfter)
+		leaves = append(leaves, cert)
+		return false, nil
+	})
+	if err != nil {
+		return nil, err
+	}
+	// The certificate query runs even without identities: it separates a keychain that is
+	// readable but holds no identity from one the process cannot read at all.
+	pool, err := keychainCertificates(searchList)
+	if err != nil {
+		return nil, err
+	}
+	if len(leaves) == 0 {
+		log.Debugf("keychain search list holds no identities usable for certificate posture, but %d readable certificates: an identity needs its private key in the same keychain", len(pool))
+		return nil, nil
+	}
+	log.Debugf("keychain search list holds %d identities and %d certificates for chain building", len(leaves), len(pool))
+
+	candidates := make([]Candidate, 0, len(leaves))
+	for _, leaf := range leaves {
+		chain := buildChain(leaf, pool)
+		log.Debugf("keychain candidate %q issued by %q built a chain of %d certificates", leaf.Subject, leaf.Issuer, len(chain))
+		if len(chain) == 1 && leaf.CheckSignatureFrom(leaf) != nil {
+			log.Debugf("keychain candidate %q has no issuer in the keychain, its proof carries the leaf alone and only verifies if the challenge supplies %q", leaf.Subject, leaf.Issuer)
+		}
+		candidates = append(candidates, Candidate{Chain: chain, Signer: &keychainSigner{leaf: leaf, keychains: s.keychains}, Intermediates: pool})
+	}
+	return candidates, nil
+}
+
+// keychainSigner holds only the certificate; the identity is looked up again at signing
+// time so no keychain references outlive a call.
+type keychainSigner struct {
+	leaf      *x509.Certificate
+	keychains []string
+}
+
+func (s *keychainSigner) Public() crypto.PublicKey {
+	return s.leaf.PublicKey
+}
+
+func (s *keychainSigner) Sign(_ io.Reader, digest []byte, opts crypto.SignerOpts) ([]byte, error) {
+	scheme, err := schemeFor(s.leaf.PublicKey, opts)
+	if err != nil {
+		return nil, err
+	}
+	log.Debugf("signing certificate posture challenge with keychain key of %q", s.leaf.Subject)
+
+	searchList, done, err := searchListOf(s.keychains)
+	if err != nil {
+		return nil, err
+	}
+	defer done()
+
+	algorithm := keychainAlgorithm(scheme)
+	var signature []byte
+	err = eachIdentity(searchList, func(identity uintptr, der []byte) (bool, error) {
+		if !bytes.Equal(der, s.leaf.Raw) {
+			return false, nil
+		}
+		signature, err = signWithIdentity(identity, algorithm, digest)
+		return true, err
+	})
+	if err != nil {
+		return nil, err
+	}
+	if signature == nil {
+		return nil, errors.New("certificate is no longer in the keychain")
+	}
+	log.Debugf("keychain signed certificate posture challenge for %q, %d bytes", s.leaf.Subject, len(signature))
+	return signature, nil
+}
+
+func keychainAlgorithm(scheme sigScheme) uintptr {
+	switch scheme {
+	case schemeECDSASHA384:
+		return kSecKeyAlgorithmECDSASHA384
+	case schemeRSAPSSSHA256:
+		return kSecKeyAlgorithmRSAPSSSHA256
+	default:
+		return kSecKeyAlgorithmECDSASHA256
+	}
+}
+
+func signWithIdentity(identity, algorithm uintptr, digest []byte) ([]byte, error) {
+	var key uintptr
+	if status := secIdentityCopyPrivateKey(identity, &key); status != 0 {
+		return nil, fmt.Errorf("SecIdentityCopyPrivateKey: %d", status)
+	}
+	defer release(key)
+
+	data := cfDataCreate(0, &digest[0], len(digest))
+	if data == 0 {
+		return nil, errors.New("CFDataCreate returned NULL")
+	}
+	defer release(data)
+
+	var cfErr uintptr
+	signature := secKeyCreateSignature(key, algorithm, data, &cfErr)
+	if signature == 0 {
+		if cfErr == 0 {
+			return nil, errors.New("SecKeyCreateSignature failed without a CFError")
+		}
+		defer release(cfErr)
+		code := cfErrorGetCode(cfErr)
+		if code == errSecInteractionNotAllowed {
+			return nil, fmt.Errorf("SecKeyCreateSignature: CFError %d: %w", code, errKeyNeedsApproval)
+		}
+		return nil, fmt.Errorf("SecKeyCreateSignature: CFError %d", code)
+	}
+	defer release(signature)
+	return dataBytes(signature), nil
+}
+
+// searchListOf opens the keychain files at paths as a CFArray for kSecMatchSearchList,
+// or yields 0, the process's own search list, when paths is empty. The caller calls done
+// once it no longer uses the list.
+func searchListOf(paths []string) (searchList uintptr, done func(), err error) {
+	if len(paths) == 0 {
+		return 0, func() {}, nil
+	}
+	list, err := openSearchList(paths)
+	if err != nil {
+		return 0, nil, err
+	}
+	return list, func() { release(list) }, nil
+}
+
+// openSearchList opens the keychain files at paths, at least one, as a CFArray. The
+// caller releases the array.
+func openSearchList(paths []string) (uintptr, error) {
+	refs := make([]uintptr, 0, len(paths))
+	defer func() {
+		for _, ref := range refs {
+			release(ref)
+		}
+	}()
+	for _, path := range paths {
+		cpath := append([]byte(path), 0)
+		var keychain uintptr
+		if status := secKeychainOpen(&cpath[0], &keychain); status != 0 {
+			return 0, fmt.Errorf("SecKeychainOpen %s: %d", path, status)
+		}
+		refs = append(refs, keychain)
+	}
+	// The array retains the keychains, so the references opened here are released.
+	list := cfArrayCreate(0, &refs[0], len(refs), kCFTypeArrayCallBacks)
+	if list == 0 {
+		return 0, errors.New("CFArrayCreate returned NULL")
+	}
+	return list, nil
+}
+
+// eachIdentity calls fn with every identity in searchList, or in the process's search
+// list when it is 0, and its certificate. An identity whose certificate cannot be read is
+// skipped rather than ending the walk.
+func eachIdentity(searchList uintptr, fn func(identity uintptr, der []byte) (bool, error)) error {
+	return eachMatching(searchList, kSecClassIdentity, "identity", func(identity uintptr) (bool, error) {
+		var cert uintptr
+		if status := secIdentityCopyCertificate(identity, &cert); status != 0 {
+			log.Debugf("skipping keychain identity: SecIdentityCopyCertificate: %d", status)
+			return false, nil
+		}
+		der := certificateDER(cert)
+		release(cert)
+		if der == nil {
+			log.Debug("skipping keychain identity whose certificate has no DER data")
+			return false, nil
+		}
+		return fn(identity, der)
+	})
+}
+
+func keychainCertificates(searchList uintptr) ([]*x509.Certificate, error) {
+	var certs []*x509.Certificate
+	var unparsable int
+	err := eachMatching(searchList, kSecClassCertificate, "certificate", func(item uintptr) (bool, error) {
+		if cert, err := x509.ParseCertificate(certificateDER(item)); err == nil {
+			certs = append(certs, cert)
+			return false, nil
+		}
+		unparsable++
+		return false, nil
+	})
+	log.Debugf("keychain holds %d parsable certificates, %d unparsable", len(certs), unparsable)
+	return certs, err
+}
+
+func eachMatching(searchList, class uintptr, name string, fn func(item uintptr) (bool, error)) error {
+	keys := []uintptr{kSecClass, kSecMatchLimit, kSecReturnRef}
+	values := []uintptr{class, kSecMatchLimitAll, kCFBooleanTrue}
+	if searchList != 0 {
+		keys = append(keys, kSecMatchSearchList)
+		values = append(values, searchList)
+	}
+	query := cfDictionaryCreate(0, &keys[0], &values[0], len(keys), kCFTypeDictionaryKeyCallBacks, kCFTypeDictionaryValueCallBacks)
+	defer release(query)
+
+	var items uintptr
+	switch status := secItemCopyMatching(query, &items); status {
+	case 0:
+	case errSecItemNotFound:
+		log.Debugf("keychain %s query returned errSecItemNotFound (%d): the search list holds no item of this class", name, errSecItemNotFound)
+		return nil
+	default:
+		log.Debugf("keychain %s query returned OSStatus %d", name, status)
+		return fmt.Errorf("SecItemCopyMatching: %d", status)
+	}
+	defer release(items)
+
+	n := cfArrayGetCount(items)
+	log.Debugf("keychain %s query returned %d items", name, n)
+	for i := 0; i < n; i++ {
+		if stop, err := fn(cfArrayGetValueAtIndex(items, i)); stop || err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+// certificateDER returns the DER form of cert, or nil when SecCertificateCopyData
+// returns NULL, which it does for an object that is not a valid certificate.
+func certificateDER(cert uintptr) []byte {
+	data := secCertificateCopyData(cert)
+	if data == 0 {
+		return nil
+	}
+	defer release(data)
+	return dataBytes(data)
+}
+
+func dataBytes(data uintptr) []byte {
+	n := cfDataGetLength(data)
+	if n <= 0 {
+		return nil
+	}
+	return bytes.Clone(unsafe.Slice((*byte)(cfDataGetBytePtr(data)), n))
+}
+
+// release drops a CoreFoundation reference. CFRelease crashes the process on NULL, and
+// several Security calls return NULL on failure, so every release goes through here.
+func release(ref uintptr) {
+	if ref != 0 {
+		cfRelease(ref)
+	}
+}
+
+func loadKeychain() error {
+	keychainOnce.Do(func() {
+		if keychainErr = resolveKeychain(); keychainErr != nil {
+			log.Debugf("macOS keychain unavailable for certificate posture: %v", keychainErr)
+			return
+		}
+		log.Debugf("macOS Security framework loaded for certificate posture, running as uid=%d euid=%d", os.Getuid(), os.Geteuid())
+		logSearchList()
+	})
+	return keychainErr
+}
+
+// logSearchList reports the keychains the process searches. The root daemon sees the
+// System keychain and System Roots, never a user's login keychain.
+func logSearchList() {
+	if secKeychainCopySearchList == nil || secKeychainGetPath == nil {
+		log.Debug("keychain search list diagnostics unavailable on this macOS version")
+		return
+	}
+
+	var list uintptr
+	if status := secKeychainCopySearchList(&list); status != 0 {
+		log.Debugf("SecKeychainCopySearchList returned OSStatus %d", status)
+		return
+	}
+	defer release(list)
+
+	n := cfArrayGetCount(list)
+	log.Debugf("keychain search list contains %d keychains", n)
+	for i := 0; i < n; i++ {
+		log.Debugf("keychain search list[%d]: %s", i, keychainPath(cfArrayGetValueAtIndex(list, i)))
+	}
+}
+
+func keychainPath(keychain uintptr) string {
+	path := make([]byte, 1024)
+	length := uint32(len(path))
+	if status := secKeychainGetPath(keychain, &length, &path[0]); status != 0 {
+		return fmt.Sprintf("<SecKeychainGetPath: %d>", status)
+	}
+	return string(path[:length])
+}
+
+func resolveKeychain() error {
+	security, err := purego.Dlopen(securityFramework, purego.RTLD_LAZY|purego.RTLD_GLOBAL)
+	if err != nil {
+		return fmt.Errorf("open %s: %w", securityFramework, err)
+	}
+	coreFoundation, err := purego.Dlopen(coreFoundationFramework, purego.RTLD_LAZY|purego.RTLD_GLOBAL)
+	if err != nil {
+		return fmt.Errorf("open %s: %w", coreFoundationFramework, err)
+	}
+
+	for _, fn := range []struct {
+		ptr  any
+		lib  uintptr
+		name string
+	}{
+		{&secItemCopyMatching, security, "SecItemCopyMatching"},
+		{&secIdentityCopyCertificate, security, "SecIdentityCopyCertificate"},
+		{&secIdentityCopyPrivateKey, security, "SecIdentityCopyPrivateKey"},
+		{&secCertificateCopyData, security, "SecCertificateCopyData"},
+		{&secKeyCreateSignature, security, "SecKeyCreateSignature"},
+		{&secKeychainOpen, security, "SecKeychainOpen"},
+		{&cfArrayCreate, coreFoundation, "CFArrayCreate"},
+		{&cfDictionaryCreate, coreFoundation, "CFDictionaryCreate"},
+		{&cfArrayGetCount, coreFoundation, "CFArrayGetCount"},
+		{&cfArrayGetValueAtIndex, coreFoundation, "CFArrayGetValueAtIndex"},
+		{&cfDataCreate, coreFoundation, "CFDataCreate"},
+		{&cfDataGetLength, coreFoundation, "CFDataGetLength"},
+		{&cfDataGetBytePtr, coreFoundation, "CFDataGetBytePtr"},
+		{&cfErrorGetCode, coreFoundation, "CFErrorGetCode"},
+		{&cfRelease, coreFoundation, "CFRelease"},
+	} {
+		symbol, err := purego.Dlsym(fn.lib, fn.name)
+		if err != nil {
+			return fmt.Errorf("resolve %s: %w", fn.name, err)
+		}
+		purego.RegisterFunc(fn.ptr, symbol)
+	}
+
+	for _, global := range []struct {
+		ptr   *uintptr
+		lib   uintptr
+		name  string
+		deref bool
+	}{
+		{&kSecClass, security, "kSecClass", true},
+		{&kSecClassIdentity, security, "kSecClassIdentity", true},
+		{&kSecClassCertificate, security, "kSecClassCertificate", true},
+		{&kSecMatchLimit, security, "kSecMatchLimit", true},
+		{&kSecMatchLimitAll, security, "kSecMatchLimitAll", true},
+		{&kSecReturnRef, security, "kSecReturnRef", true},
+		{&kSecMatchSearchList, security, "kSecMatchSearchList", true},
+		{&kSecKeyAlgorithmECDSASHA256, security, "kSecKeyAlgorithmECDSASignatureDigestX962SHA256", true},
+		{&kSecKeyAlgorithmECDSASHA384, security, "kSecKeyAlgorithmECDSASignatureDigestX962SHA384", true},
+		{&kSecKeyAlgorithmRSAPSSSHA256, security, "kSecKeyAlgorithmRSASignatureDigestPSSSHA256", true},
+		{&kCFBooleanTrue, coreFoundation, "kCFBooleanTrue", true},
+		{&kCFTypeDictionaryKeyCallBacks, coreFoundation, "kCFTypeDictionaryKeyCallBacks", false},
+		{&kCFTypeDictionaryValueCallBacks, coreFoundation, "kCFTypeDictionaryValueCallBacks", false},
+		{&kCFTypeArrayCallBacks, coreFoundation, "kCFTypeArrayCallBacks", false},
+	} {
+		addr, err := purego.Dlsym(global.lib, global.name)
+		if err != nil {
+			return fmt.Errorf("resolve %s: %w", global.name, err)
+		}
+		if global.deref {
+			addr = **(**uintptr)(unsafe.Pointer(&addr))
+		}
+		*global.ptr = addr
+	}
+
+	resolveOptional(security, "SecKeychainCopySearchList", &secKeychainCopySearchList)
+	resolveOptional(security, "SecKeychainGetPath", &secKeychainGetPath)
+	return nil
+}
+
+// resolveOptional binds a diagnostic-only symbol, leaving it nil when the framework no
+// longer exports it so keychain lookups keep working without it.
+func resolveOptional(lib uintptr, name string, ptr any) {
+	symbol, err := purego.Dlsym(lib, name)
+	if err != nil {
+		log.Debugf("keychain diagnostics: %s unavailable: %v", name, err)
+		return
+	}
+	purego.RegisterFunc(ptr, symbol)
+}

client/internal/certproof/keystore.go

@@ -0,0 +1,77 @@
+package certproof
+
+import (
+	"crypto"
+	"crypto/ecdsa"
+	"crypto/rsa"
+	"crypto/x509"
+	"encoding/asn1"
+	"errors"
+	"fmt"
+	"math/big"
+	"slices"
+)
+
+var errUnsupportedScheme = errors.New("unsupported signature scheme for OS keystore")
+
+type sigScheme int
+
+const (
+	schemeECDSASHA256 sigScheme = iota + 1
+	schemeECDSASHA384
+	schemeRSAPSSSHA256
+)
+
+// schemeFor maps a crypto.Signer request onto the schemes the OS keystores perform.
+func schemeFor(pub crypto.PublicKey, opts crypto.SignerOpts) (sigScheme, error) {
+	switch pub.(type) {
+	case *ecdsa.PublicKey:
+		switch opts.HashFunc() {
+		case crypto.SHA256:
+			return schemeECDSASHA256, nil
+		case crypto.SHA384:
+			return schemeECDSASHA384, nil
+		}
+	case *rsa.PublicKey:
+		if pss, ok := opts.(*rsa.PSSOptions); ok && pss.Hash == crypto.SHA256 {
+			return schemeRSAPSSSHA256, nil
+		}
+	}
+	return 0, fmt.Errorf("%w: %T with %v", errUnsupportedScheme, pub, opts.HashFunc())
+}
+
+// buildChain extends leaf with the issuers found in pool up to a self-signed certificate.
+func buildChain(leaf *x509.Certificate, pool []*x509.Certificate) []*x509.Certificate {
+	chain := []*x509.Certificate{leaf}
+	current := leaf
+	for current.CheckSignatureFrom(current) != nil {
+		issuer := issuerIn(current, pool, chain)
+		if issuer == nil {
+			break
+		}
+		chain = append(chain, issuer)
+		current = issuer
+	}
+	return chain
+}
+
+func issuerIn(cert *x509.Certificate, pool, seen []*x509.Certificate) *x509.Certificate {
+	for _, candidate := range pool {
+		if slices.ContainsFunc(seen, candidate.Equal) {
+			continue
+		}
+		if cert.CheckSignatureFrom(candidate) == nil {
+			return candidate
+		}
+	}
+	return nil
+}
+
+// ecdsaSignatureASN1 converts the fixed-width r||s form emitted by CNG into the DER form Go verifies.
+func ecdsaSignatureASN1(raw []byte) ([]byte, error) {
+	if len(raw) == 0 || len(raw)%2 != 0 {
+		return nil, errors.New("malformed raw ECDSA signature")
+	}
+	half := len(raw) / 2
+	return asn1.Marshal(struct{ R, S *big.Int }{new(big.Int).SetBytes(raw[:half]), new(big.Int).SetBytes(raw[half:])})
+}

client/internal/certproof/keystore_test.go

@@ -0,0 +1,87 @@
+package certproof
+
+import (
+	"crypto"
+	"crypto/ecdsa"
+	"crypto/rand"
+	"crypto/rsa"
+	"crypto/sha256"
+	"crypto/x509"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/shared/management/certposture"
+	"github.com/netbirdio/netbird/shared/management/certposture/certtest"
+)
+
+func TestBuildChain_FollowsIssuersThroughThePool(t *testing.T) {
+	root := certtest.NewCA(t, "root")
+	intermediate := certtest.NewIntermediate(t, root, "intermediate")
+	unrelated := certtest.NewCA(t, "unrelated")
+	leaf := intermediate.Issue(t, certtest.ECDSAKey(t), "device")
+	pool := []*x509.Certificate{unrelated.Cert, root.Cert, leaf, intermediate.Cert}
+
+	chain := buildChain(leaf, pool)
+
+	require.Equal(t, []*x509.Certificate{leaf, intermediate.Cert, root.Cert}, chain)
+	roots, err := certposture.ParseCAs([]string{root.PEM})
+	require.NoError(t, err)
+	assert.NoError(t, certposture.VerifyChain(chain, roots, time.Now()))
+}
+
+func TestBuildChain_StopsWhereThePoolEnds(t *testing.T) {
+	root := certtest.NewCA(t, "root")
+	intermediate := certtest.NewIntermediate(t, root, "intermediate")
+	leaf := intermediate.Issue(t, certtest.ECDSAKey(t), "device")
+
+	assert.Equal(t, []*x509.Certificate{leaf}, buildChain(leaf, nil))
+	assert.Equal(t, []*x509.Certificate{leaf, intermediate.Cert}, buildChain(leaf, []*x509.Certificate{intermediate.Cert}))
+}
+
+func TestSchemeFor(t *testing.T) {
+	ecKey := certtest.ECDSAKey(t)
+	rsaKey := certtest.RSAKey(t)
+	pss := &rsa.PSSOptions{SaltLength: rsa.PSSSaltLengthEqualsHash, Hash: crypto.SHA256}
+
+	tests := []struct {
+		name string
+		pub  crypto.PublicKey
+		opts crypto.SignerOpts
+		want sigScheme
+	}{
+		{"ecdsa sha256", ecKey.Public(), crypto.SHA256, schemeECDSASHA256},
+		{"ecdsa sha384", ecKey.Public(), crypto.SHA384, schemeECDSASHA384},
+		{"rsa pss sha256", rsaKey.Public(), pss, schemeRSAPSSSHA256},
+		{"rsa pkcs1v15", rsaKey.Public(), crypto.SHA256, 0},
+		{"ed25519", certtest.Ed25519Key(t).Public(), crypto.Hash(0), 0},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			got, err := schemeFor(tc.pub, tc.opts)
+			if tc.want == 0 {
+				assert.ErrorIs(t, err, errUnsupportedScheme)
+				return
+			}
+			require.NoError(t, err)
+			assert.Equal(t, tc.want, got)
+		})
+	}
+}
+
+func TestECDSASignatureASN1(t *testing.T) {
+	key := certtest.ECDSAKey(t).(*ecdsa.PrivateKey)
+	digest := sha256.Sum256([]byte("nonce"))
+	r, s, err := ecdsa.Sign(rand.Reader, key, digest[:])
+	require.NoError(t, err)
+	raw := append(r.FillBytes(make([]byte, 32)), s.FillBytes(make([]byte, 32))...)
+
+	der, err := ecdsaSignatureASN1(raw)
+	require.NoError(t, err)
+	assert.True(t, ecdsa.VerifyASN1(&key.PublicKey, digest[:], der))
+
+	_, err = ecdsaSignatureASN1(raw[:63])
+	assert.Error(t, err)
+}

client/internal/certproof/pinlatch.go

@@ -0,0 +1,54 @@
+package certproof
+
+import (
+	"crypto/sha256"
+	"encoding/binary"
+	"errors"
+	"sync"
+)
+
+// errPINRejectedBefore is returned instead of logging in with a PIN the token already
+// refused: every failed login counts towards the token's lockout, which for a TPM is
+// shared with everything else on the machine, and proofs are collected on every sync.
+var errPINRejectedBefore = errors.New("PKCS#11 token rejected this PIN before, not trying it again")
+
+// errPINNeedsToken refuses a PIN that names no token to log in to.
+var errPINNeedsToken = errors.New("a PKCS#11 PIN needs the token named in NB_CERT_PKCS11_URI, as token=<label>")
+
+// rejectedPINs outlives a single store, since a store is built for each collection.
+var rejectedPINs = &pinLatch{keys: map[[sha256.Size]byte]struct{}{}}
+
+// pinLoginMu serializes logging in to a PKCS#11 token with a PIN, from checking the latch
+// to recording a rejection.
+var pinLoginMu sync.Mutex
+
+// pinLatch remembers PINs a token rejected. Keys are hashes, so the PIN itself is not
+// kept in memory any longer than the store that read it.
+type pinLatch struct {
+	mu   sync.Mutex
+	keys map[[sha256.Size]byte]struct{}
+}
+
+func (l *pinLatch) has(key [sha256.Size]byte) bool {
+	l.mu.Lock()
+	defer l.mu.Unlock()
+	_, ok := l.keys[key]
+	return ok
+}
+
+func (l *pinLatch) add(key [sha256.Size]byte) {
+	l.mu.Lock()
+	defer l.mu.Unlock()
+	l.keys[key] = struct{}{}
+}
+
+// rejectedPINKey identifies a PIN for one token of one module, so a PIN another token
+// rejected is still tried on the token it belongs to.
+func rejectedPINKey(module, token string, pin []byte) [sha256.Size]byte {
+	var buf []byte
+	for _, part := range [][]byte{[]byte(module), []byte(token), pin} {
+		buf = binary.BigEndian.AppendUint64(buf, uint64(len(part)))
+		buf = append(buf, part...)
+	}
+	return sha256.Sum256(buf)
+}

client/internal/certproof/pinlatch_test.go

@@ -0,0 +1,29 @@
+package certproof
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestRejectedPINKey_ScopesToModuleTokenAndPIN(t *testing.T) {
+	base := rejectedPINKey("/lib/a.so", "netbird", []byte("1234"))
+
+	assert.Equal(t, base, rejectedPINKey("/lib/a.so", "netbird", []byte("1234")), "the same PIN on the same token is the same key")
+	assert.NotEqual(t, base, rejectedPINKey("/lib/a.so", "netbird", []byte("4321")), "a corrected PIN must be tried")
+	assert.NotEqual(t, base, rejectedPINKey("/lib/a.so", "piv", []byte("1234")), "a PIN rejected by one token is still tried on another")
+	assert.NotEqual(t, base, rejectedPINKey("/lib/b.so", "netbird", []byte("1234")), "a PIN rejected through one module is still tried through another")
+	assert.NotEqual(t, rejectedPINKey("ab", "", []byte("1")), rejectedPINKey("a", "b", []byte("1")),
+		"field boundaries are part of the key, so shifted fields do not collide")
+}
+
+func TestPINLatch_RemembersRejectedKeys(t *testing.T) {
+	latch := &pinLatch{keys: map[[32]byte]struct{}{}}
+	rejected := rejectedPINKey("/lib/a.so", "netbird", []byte("0000"))
+	other := rejectedPINKey("/lib/a.so", "netbird", []byte("1234"))
+
+	assert.False(t, latch.has(rejected), "nothing is rejected before a login fails")
+	latch.add(rejected)
+	assert.True(t, latch.has(rejected), "a rejected PIN is not tried again")
+	assert.False(t, latch.has(other), "other PINs are unaffected")
+}

client/internal/certproof/pkcs11store.go

@@ -0,0 +1,333 @@
+package certproof
+
+import (
+	"context"
+	"crypto"
+	"crypto/sha256"
+	"crypto/x509"
+	"errors"
+	"fmt"
+	"io"
+	"os"
+
+	log "github.com/sirupsen/logrus"
+
+	"github.com/netbirdio/netbird/client/internal/pkcs11"
+)
+
+const (
+	// PINEnv carries the user PIN of the PKCS#11 token. It is read from the daemon's
+	// environment only, so the PIN never lands in the profile config or on a command line.
+	PINEnv = "NB_TPM_PIN"
+	// PKCS11URIEnv carries the RFC 7512 URI of the token. Like the PIN it lives in the
+	// daemon's environment rather than the profile config, since it may carry pin-value.
+	PKCS11URIEnv = "NB_CERT_PKCS11_URI"
+)
+
+// PKCS11Config names the token whose certificates the store yields. URI is an RFC 7512
+// PKCS#11 URI, or empty for the first token the p11-kit proxy exposes. PIN is the user
+// PIN, and takes precedence over a pin-value or pin-source the URI carries.
+type PKCS11Config struct {
+	URI string
+	PIN string
+}
+
+// PKCS11FromEnv returns the token named by NB_CERT_PKCS11_URI with the PIN set in
+// NB_TPM_PIN, either empty when unset.
+func PKCS11FromEnv() PKCS11Config {
+	return PKCS11Config{URI: os.Getenv(PKCS11URIEnv), PIN: os.Getenv(PINEnv)}
+}
+
+// PKCS11Store yields the identities of a PKCS#11 token, which is how tpm2-pkcs11 exposes
+// TPM-held keys on Linux. Certificates on the token are paired with keys by CKA_ID, the
+// convention tpm2_ptool addcert and pkcs11-tool follow; certificate files in the PEM
+// directory by public key. Every signature happens on the token.
+type PKCS11Store struct {
+	uri     *pkcs11.URI
+	pin     string
+	certDir string
+}
+
+// NewPKCS11Store parses cfg.URI, standing in the bare defaults when it is empty. Files in
+// certDir without a key of their own are paired with the token's keys by public key.
+//
+// A PIN is only accepted together with a token label: without one the PIN would go to
+// whichever token the module lists first, which on a machine with a smartcard plugged
+// in may be the card, and every wrong PIN counts towards that card's lockout.
+func NewPKCS11Store(cfg PKCS11Config, certDir string) (*PKCS11Store, error) {
+	store := &PKCS11Store{uri: &pkcs11.URI{}, pin: cfg.PIN, certDir: certDir}
+	if cfg.URI != "" {
+		parsed, err := pkcs11.ParseURI(cfg.URI)
+		if err != nil {
+			return nil, err
+		}
+		store.uri = parsed
+	}
+	if (cfg.PIN != "" || store.uri.HasPIN()) && store.uri.Token == "" {
+		return nil, errPINNeedsToken
+	}
+	return store, nil
+}
+
+func (s *PKCS11Store) Candidates(_ context.Context) ([]Candidate, error) {
+	session, err := s.open()
+	if err != nil {
+		return nil, err
+	}
+	defer session.Close()
+
+	certs, err := tokenCertificates(session)
+	if err != nil {
+		return nil, err
+	}
+	fileChains, err := s.fileChains()
+	if err != nil {
+		return nil, err
+	}
+	log.Debugf("%s holds %d certificates, %d certificate files without a key wait for its keys", s, len(certs), len(fileChains))
+
+	pool := make([]*x509.Certificate, 0, len(certs))
+	for _, cert := range certs {
+		pool = append(pool, cert.cert)
+	}
+	for _, chain := range fileChains {
+		pool = append(pool, chain...)
+	}
+
+	var candidates []Candidate
+	for _, cert := range certs {
+		if _, err := privateKey(session, cert.id); err != nil {
+			log.Debugf("%s certificate %q has no usable private key: %v", s, cert.cert.Subject, err)
+			continue
+		}
+		candidates = append(candidates, s.candidate(cert.cert, cert.id, pool))
+	}
+	if len(fileChains) == 0 {
+		return candidates, nil
+	}
+
+	keys, err := tokenPublicKeys(session)
+	if err != nil {
+		return nil, err
+	}
+	for _, chain := range fileChains {
+		leaf := chain[0]
+		id, ok := keys.idFor(leaf.PublicKey)
+		if !ok {
+			log.Debugf("%s holds no key for certificate %q from %s", s, leaf.Subject, s.certDir)
+			continue
+		}
+		candidates = append(candidates, s.candidate(leaf, id, pool))
+	}
+	return candidates, nil
+}
+
+func (s *PKCS11Store) candidate(leaf *x509.Certificate, id []byte, pool []*x509.Certificate) Candidate {
+	chain := buildChain(leaf, pool)
+	log.Debugf("%s candidate %q issued by %q built a chain of %d certificates", s, leaf.Subject, leaf.Issuer, len(chain))
+	return Candidate{Chain: chain, Signer: &pkcs11Signer{store: s, leaf: leaf, id: id}, Intermediates: pool}
+}
+
+// fileChains reads the certificate files in the PEM directory that carry no key of their
+// own; the file store answers for the ones that do.
+func (s *PKCS11Store) fileChains() ([][]*x509.Certificate, error) {
+	if s.certDir == "" {
+		return nil, nil
+	}
+	paths, err := certFiles(s.certDir)
+	if err != nil {
+		return nil, err
+	}
+	var chains [][]*x509.Certificate
+	for _, path := range paths {
+		chain, signer, err := loadPEM(path)
+		if err != nil || signer != nil {
+			continue
+		}
+		chains = append(chains, chain)
+	}
+	return chains, nil
+}
+
+func (s *PKCS11Store) String() string {
+	if s.uri.Token == "" {
+		return "PKCS#11 token"
+	}
+	return fmt.Sprintf("PKCS#11 token %q", s.uri.Token)
+}
+
+func (s *PKCS11Store) open() (*pkcs11.Session, error) {
+	module, err := pkcs11.Load(s.uri.Module())
+	if err != nil {
+		return nil, err
+	}
+	pin, err := s.userPIN()
+	if err != nil {
+		return nil, err
+	}
+	if pin == nil {
+		return module.OpenSession(s.uri.Token, nil)
+	}
+
+	// The check, the login and recording a rejection happen under one lock, so two
+	// collections running at once cannot both send a PIN the token is about to refuse.
+	pinLoginMu.Lock()
+	defer pinLoginMu.Unlock()
+	key := rejectedPINKey(s.uri.Module(), s.uri.Token, pin)
+	if rejectedPINs.has(key) {
+		return nil, errPINRejectedBefore
+	}
+	session, err := module.OpenSession(s.uri.Token, pin)
+	if pkcs11.PINRejected(err) {
+		rejectedPINs.add(key)
+		return nil, fmt.Errorf("%s rejected the PIN, not retrying it until the daemon restarts: %w", s, err)
+	}
+	return session, err
+}
+
+func (s *PKCS11Store) userPIN() ([]byte, error) {
+	if s.pin != "" {
+		return []byte(s.pin), nil
+	}
+	return s.uri.PIN()
+}
+
+type tokenKey struct {
+	id     []byte
+	public crypto.PublicKey
+}
+
+type tokenKeys []tokenKey
+
+func tokenPublicKeys(session *pkcs11.Session) (tokenKeys, error) {
+	objects, err := session.FindObjects(pkcs11.Attribute{Type: pkcs11.AttrClass, Value: pkcs11.ULong(pkcs11.ClassPublicKey)})
+	if err != nil {
+		return nil, err
+	}
+	keys := make(tokenKeys, 0, len(objects))
+	for _, object := range objects {
+		id, err := session.Attribute(object, pkcs11.AttrID)
+		if err != nil {
+			return nil, err
+		}
+		public, err := session.PublicKey(object)
+		if err != nil {
+			log.Debugf("skipping public key on PKCS#11 token: %v", err)
+			continue
+		}
+		keys = append(keys, tokenKey{id: id, public: public})
+	}
+	return keys, nil
+}
+
+// idFor finds the token key whose public half is pub, so a certificate kept outside the
+// token is still signed for by the key inside it.
+func (k tokenKeys) idFor(pub crypto.PublicKey) ([]byte, bool) {
+	for _, key := range k {
+		equaler, ok := key.public.(interface{ Equal(crypto.PublicKey) bool })
+		if ok && len(key.id) > 0 && equaler.Equal(pub) {
+			return key.id, true
+		}
+	}
+	return nil, false
+}
+
+type tokenCertificate struct {
+	cert *x509.Certificate
+	id   []byte
+}
+
+func tokenCertificates(session *pkcs11.Session) ([]tokenCertificate, error) {
+	objects, err := session.FindObjects(
+		pkcs11.Attribute{Type: pkcs11.AttrClass, Value: pkcs11.ULong(pkcs11.ClassCertificate)},
+		pkcs11.Attribute{Type: pkcs11.AttrCertificateType, Value: pkcs11.ULong(pkcs11.CertificateX509)},
+	)
+	if err != nil {
+		return nil, err
+	}
+	certs := make([]tokenCertificate, 0, len(objects))
+	for _, object := range objects {
+		der, err := session.Attribute(object, pkcs11.AttrValue)
+		if err != nil {
+			return nil, err
+		}
+		cert, err := x509.ParseCertificate(der)
+		if err != nil {
+			log.Warnf("skipping unparsable certificate on PKCS#11 token: %v", err)
+			continue
+		}
+		id, err := session.Attribute(object, pkcs11.AttrID)
+		if err != nil {
+			return nil, err
+		}
+		certs = append(certs, tokenCertificate{cert: cert, id: id})
+	}
+	return certs, nil
+}
+
+var errNoPrivateKey = errors.New("no private key shares the certificate's CKA_ID")
+
+func privateKey(session *pkcs11.Session, id []byte) (pkcs11.Object, error) {
+	if len(id) == 0 {
+		return 0, errNoPrivateKey
+	}
+	keys, err := session.FindObjects(
+		pkcs11.Attribute{Type: pkcs11.AttrClass, Value: pkcs11.ULong(pkcs11.ClassPrivateKey)},
+		pkcs11.Attribute{Type: pkcs11.AttrID, Value: id},
+	)
+	if err != nil {
+		return 0, err
+	}
+	if len(keys) == 0 {
+		return 0, errNoPrivateKey
+	}
+	return keys[0], nil
+}
+
+// pkcs11Signer holds only the certificate and its CKA_ID; the key is looked up in a fresh
+// session at signing time so no token handle outlives a call.
+type pkcs11Signer struct {
+	store *PKCS11Store
+	leaf  *x509.Certificate
+	id    []byte
+}
+
+func (s *pkcs11Signer) Public() crypto.PublicKey {
+	return s.leaf.PublicKey
+}
+
+func (s *pkcs11Signer) Sign(_ io.Reader, digest []byte, opts crypto.SignerOpts) ([]byte, error) {
+	scheme, err := schemeFor(s.leaf.PublicKey, opts)
+	if err != nil {
+		return nil, err
+	}
+	session, err := s.store.open()
+	if err != nil {
+		return nil, err
+	}
+	defer session.Close()
+
+	key, err := privateKey(session, s.id)
+	if err != nil {
+		return nil, err
+	}
+	signature, err := session.Sign(pkcs11Mechanism(scheme), key, digest)
+	if err != nil {
+		return nil, err
+	}
+	if scheme == schemeRSAPSSSHA256 {
+		return signature, nil
+	}
+	return ecdsaSignatureASN1(signature)
+}
+
+// pkcs11Mechanism maps a signature scheme onto the token mechanism that consumes a digest.
+func pkcs11Mechanism(scheme sigScheme) pkcs11.Mechanism {
+	if scheme == schemeRSAPSSSHA256 {
+		return pkcs11.Mechanism{
+			Type: pkcs11.MechRSAPKCSPSS,
+			PSS:  &pkcs11.PSSParams{Hash: pkcs11.MechSHA256, MGF: pkcs11.MGF1SHA256, SaltLen: sha256.Size},
+		}
+	}
+	return pkcs11.Mechanism{Type: pkcs11.MechECDSA}
+}

client/internal/certproof/pkcs11store_test.go

@@ -0,0 +1,394 @@
+package certproof
+
+import (
+	"context"
+	"crypto"
+	"crypto/ecdsa"
+	"crypto/rand"
+	"crypto/rsa"
+	"crypto/x509"
+	"encoding/asn1"
+	"errors"
+	"math/big"
+	"os"
+	"runtime"
+	"strconv"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/client/internal/pkcs11"
+	"github.com/netbirdio/netbird/shared/management/certposture"
+	"github.com/netbirdio/netbird/shared/management/certposture/certtest"
+	"github.com/netbirdio/netbird/shared/management/proto"
+)
+
+const (
+	testPKCS11URIEnv = "NB_TEST_PKCS11_URI"
+	// testPKCS11DisposableEnv marks the token as disposable, allowing tests that spend
+	// attempts of its PIN lockout counter.
+	testPKCS11DisposableEnv = "NB_TEST_PKCS11_DISPOSABLE"
+)
+
+type failingStore struct{}
+
+func (failingStore) Candidates(context.Context) ([]Candidate, error) {
+	return nil, errors.New("token unplugged")
+}
+
+func TestStores_KeepsFileCertificatesWhenTokenFails(t *testing.T) {
+	ca := certtest.NewCA(t, "corp")
+	key := certtest.ECDSAKey(t)
+	dir := storeDir(t)
+	writeFile(t, dir, "device.pem", certtest.CertPEM(ca.Issue(t, key, "device"))+certtest.KeyPEM(t, key))
+
+	candidates, err := Stores{failingStore{}, NewFileStore(dir)}.Candidates(context.Background())
+	require.NoError(t, err)
+	assert.Len(t, candidates, 1, "the directory's certificate must survive a failing token")
+}
+
+// TestCollect_PKCS11TokenEndToEnd needs an initialised token with a user PIN, named by
+// NB_TEST_PKCS11_URI. With SoftHSM:
+//
+//	softhsm2-util --init-token --free --label netbird --pin 1234 --so-pin 1234
+//	NB_TEST_PKCS11_URI='pkcs11:token=netbird?module-path=/usr/lib/softhsm/libsofthsm2.so&pin-value=1234' \
+//	  go test -tags pkcs11 ./client/internal/certproof/ -run PKCS11 -v
+//
+// It imports a key and its certificate as token objects, then proves the certificate
+// through the store the way the daemon would. Every run adds one more identity to the token.
+func TestCollect_PKCS11TokenEndToEnd(t *testing.T) {
+	store, uri := pkcs11TestStore(t, "")
+
+	keys := map[string]crypto.Signer{"ecdsa": certtest.ECDSAKey(t), "rsa": certtest.RSAKey(t)}
+	for name, key := range keys {
+		t.Run(name, func(t *testing.T) {
+			ca := certtest.NewCA(t, "corp-"+name)
+			leaf := ca.Issue(t, key, "device-"+name)
+			importIdentity(t, uri, key, leaf)
+
+			challenger := certposture.NewChallenger([]byte("secret"))
+			now := time.Now()
+			nonce := challenger.Nonce(peerKey, now)
+			checks := []*proto.Checks{{CertificateChallenge: &proto.CertificateChallenge{Nonce: nonce, CaCertificates: []string{ca.PEM}}}}
+
+			proofs := Collect(context.Background(), store, checks, peerKey)
+			require.Len(t, proofs, 1, "the token-held key must prove exactly this run's certificate")
+			chain, err := challenger.Verify(proofs[0], peerKey, now)
+			require.NoError(t, err)
+			assert.True(t, leaf.Equal(chain[0]), "proof must carry the imported certificate")
+		})
+	}
+}
+
+// Attribute types the import needs and the store does not.
+const (
+	attrPrivate         = 0x2
+	attrIssuer          = 0x81
+	attrSerialNumber    = 0x82
+	attrSensitive       = 0x103
+	attrSign            = 0x108
+	attrVerify          = 0x10a
+	attrPrivateExponent = 0x123
+	attrPrime1          = 0x124
+	attrPrime2          = 0x125
+	attrExponent1       = 0x126
+	attrExponent2       = 0x127
+	attrCoefficient     = 0x128
+)
+
+var (
+	ckTrue  = []byte{1}
+	ckFalse = []byte{0}
+	// The P-256 named curve OID in DER, which is what CKA_EC_PARAMS carries.
+	oidP256 = []byte{0x06, 0x08, 0x2a, 0x86, 0x48, 0xce, 0x3d, 0x03, 0x01, 0x07}
+)
+
+// importIdentity stores key and leaf on the token the way tpm2_ptool import and addcert
+// do: private and public key objects plus the certificate, all under one CKA_ID.
+func importIdentity(t *testing.T, uri string, key crypto.Signer, leaf *x509.Certificate) {
+	t.Helper()
+	id := importKey(t, uri, key, leaf.Subject.CommonName)
+	importCertificate(t, uri, leaf, id)
+}
+
+func importKey(t *testing.T, uri string, key crypto.Signer, label string) []byte {
+	t.Helper()
+	session := readWriteSession(t, uri)
+	defer session.Close()
+
+	id := make([]byte, 8)
+	_, err := rand.Read(id)
+	require.NoError(t, err)
+
+	private := []pkcs11.Attribute{
+		attr(pkcs11.AttrClass, pkcs11.ULong(pkcs11.ClassPrivateKey)),
+		attr(pkcs11.AttrToken, ckTrue),
+		attr(attrPrivate, ckTrue),
+		attr(attrSensitive, ckTrue),
+		attr(attrSign, ckTrue),
+		attr(pkcs11.AttrLabel, []byte(label)),
+		attr(pkcs11.AttrID, id),
+	}
+	_, err = session.CreateObject(append(private, privateKeyAttributes(t, key)...)...)
+	require.NoError(t, err, "import private key")
+
+	public := []pkcs11.Attribute{
+		attr(pkcs11.AttrClass, pkcs11.ULong(pkcs11.ClassPublicKey)),
+		attr(pkcs11.AttrToken, ckTrue),
+		attr(attrPrivate, ckFalse),
+		attr(attrVerify, ckTrue),
+		attr(pkcs11.AttrLabel, []byte(label)),
+		attr(pkcs11.AttrID, id),
+	}
+	_, err = session.CreateObject(append(public, publicKeyAttributes(t, key)...)...)
+	require.NoError(t, err, "import public key")
+	return id
+}
+
+func importCertificate(t *testing.T, uri string, leaf *x509.Certificate, id []byte) {
+	t.Helper()
+	session := readWriteSession(t, uri)
+	defer session.Close()
+
+	serial, err := asn1.Marshal(leaf.SerialNumber)
+	require.NoError(t, err)
+	_, err = session.CreateObject(
+		attr(pkcs11.AttrClass, pkcs11.ULong(pkcs11.ClassCertificate)),
+		attr(pkcs11.AttrCertificateType, pkcs11.ULong(pkcs11.CertificateX509)),
+		attr(pkcs11.AttrToken, ckTrue),
+		attr(attrPrivate, ckFalse),
+		attr(pkcs11.AttrLabel, []byte(leaf.Subject.CommonName)),
+		attr(pkcs11.AttrID, id),
+		attr(pkcs11.AttrSubject, leaf.RawSubject),
+		attr(attrIssuer, leaf.RawIssuer),
+		attr(attrSerialNumber, serial),
+		attr(pkcs11.AttrValue, leaf.Raw),
+	)
+	require.NoError(t, err, "import certificate")
+}
+
+func readWriteSession(t *testing.T, uri string) *pkcs11.Session {
+	t.Helper()
+	parsed, err := pkcs11.ParseURI(uri)
+	require.NoError(t, err)
+	module, err := pkcs11.Load(parsed.Module())
+	require.NoError(t, err)
+	pin, err := parsed.PIN()
+	require.NoError(t, err)
+	session, err := module.OpenReadWriteSession(parsed.Token, pin)
+	require.NoError(t, err)
+	return session
+}
+
+func privateKeyAttributes(t *testing.T, key crypto.Signer) []pkcs11.Attribute {
+	t.Helper()
+	switch k := key.(type) {
+	case *ecdsa.PrivateKey:
+		return []pkcs11.Attribute{
+			attr(pkcs11.AttrKeyType, pkcs11.ULong(pkcs11.KeyEC)),
+			attr(pkcs11.AttrECParams, oidP256),
+			attr(pkcs11.AttrValue, ecPrivateScalar(t, k)),
+		}
+	case *rsa.PrivateKey:
+		k.Precompute()
+		return []pkcs11.Attribute{
+			attr(pkcs11.AttrKeyType, pkcs11.ULong(pkcs11.KeyRSA)),
+			attr(pkcs11.AttrModulus, k.N.Bytes()),
+			attr(pkcs11.AttrPublicExponent, big.NewInt(int64(k.E)).Bytes()),
+			attr(attrPrivateExponent, k.D.Bytes()),
+			attr(attrPrime1, k.Primes[0].Bytes()),
+			attr(attrPrime2, k.Primes[1].Bytes()),
+			attr(attrExponent1, k.Precomputed.Dp.Bytes()),
+			attr(attrExponent2, k.Precomputed.Dq.Bytes()),
+			attr(attrCoefficient, k.Precomputed.Qinv.Bytes()),
+		}
+	}
+	t.Fatalf("unsupported key %T", key)
+	return nil
+}
+
+// publicKeyAttributes describes the CKO_PUBLIC_KEY object tokens keep next to a private
+// key, which is what the store reads to pair a certificate file with its key.
+func publicKeyAttributes(t *testing.T, key crypto.Signer) []pkcs11.Attribute {
+	t.Helper()
+	switch k := key.(type) {
+	case *ecdsa.PrivateKey:
+		point := append([]byte{4}, k.X.FillBytes(make([]byte, 32))...)
+		point = append(point, k.Y.FillBytes(make([]byte, 32))...)
+		wrapped, err := asn1.Marshal(point)
+		require.NoError(t, err)
+		return []pkcs11.Attribute{
+			attr(pkcs11.AttrKeyType, pkcs11.ULong(pkcs11.KeyEC)),
+			attr(pkcs11.AttrECParams, oidP256),
+			attr(pkcs11.AttrECPoint, wrapped),
+		}
+	case *rsa.PrivateKey:
+		return []pkcs11.Attribute{
+			attr(pkcs11.AttrKeyType, pkcs11.ULong(pkcs11.KeyRSA)),
+			attr(pkcs11.AttrModulus, k.N.Bytes()),
+			attr(pkcs11.AttrPublicExponent, big.NewInt(int64(k.E)).Bytes()),
+		}
+	}
+	t.Fatalf("unsupported key %T", key)
+	return nil
+}
+
+func attr(typ uint, value []byte) pkcs11.Attribute {
+	return pkcs11.Attribute{Type: typ, Value: value}
+}
+
+// pkcs11TestStore builds the store for the token NB_TEST_PKCS11_URI names, skipping when
+// no token is configured or this build lacks PKCS#11 support.
+func pkcs11TestStore(t *testing.T, certDir string) (*PKCS11Store, string) {
+	t.Helper()
+	uri := os.Getenv(testPKCS11URIEnv)
+	if uri == "" {
+		t.Skipf("set %s to a PKCS#11 URI with a PIN to run", testPKCS11URIEnv)
+	}
+	store, err := NewPKCS11Store(PKCS11Config{URI: uri}, certDir)
+	require.NoError(t, err)
+	if _, err := pkcs11.Load(store.uri.Module()); errors.Is(err, pkcs11.ErrUnsupported) {
+		t.Skip(err)
+	}
+	return store, uri
+}
+
+// TestCollect_PKCS11KeyWithFileCertificate covers the split layout: the key lives on the
+// token, the certificate is a PEM file in the directory, and the two are paired by public
+// key because nothing on the token carries the certificate's CKA_ID.
+func TestCollect_PKCS11KeyWithFileCertificate(t *testing.T) {
+	dir := storeDir(t)
+	store, uri := pkcs11TestStore(t, dir)
+
+	keys := map[string]crypto.Signer{"ecdsa": certtest.ECDSAKey(t), "rsa": certtest.RSAKey(t)}
+	for name, key := range keys {
+		t.Run(name, func(t *testing.T) {
+			ca := certtest.NewCA(t, "corp-file-"+name)
+			leaf := ca.Issue(t, key, "device-file-"+name)
+			importKey(t, uri, key, "device-file-"+name)
+			writeFile(t, dir, "device-"+name+".pem", certtest.CertPEM(leaf))
+
+			challenger := certposture.NewChallenger([]byte("secret"))
+			now := time.Now()
+			nonce := challenger.Nonce(peerKey, now)
+			checks := []*proto.Checks{{CertificateChallenge: &proto.CertificateChallenge{Nonce: nonce, CaCertificates: []string{ca.PEM}}}}
+
+			proofs := Collect(context.Background(), store, checks, peerKey)
+			require.Len(t, proofs, 1, "the token key must prove the certificate kept on disk")
+			chain, err := challenger.Verify(proofs[0], peerKey, now)
+			require.NoError(t, err)
+			assert.True(t, leaf.Equal(chain[0]), "proof must carry the certificate from the directory")
+		})
+	}
+}
+
+func TestPKCS11Store_FileChains(t *testing.T) {
+	ca := certtest.NewCA(t, "corp")
+	dir := storeDir(t)
+	// Only certificate files without a key of their own belong to the token; the file
+	// store answers for the others, and non-certificate files are ignored.
+	writeFile(t, dir, "device.pem", certtest.CertPEM(ca.Issue(t, certtest.ECDSAKey(t), "device")))
+	writeFile(t, dir, "ca.crt", ca.PEM)
+	keyed := certtest.ECDSAKey(t)
+	writeFile(t, dir, "inline.pem", certtest.CertPEM(ca.Issue(t, keyed, "inline"))+certtest.KeyPEM(t, keyed))
+	writeFile(t, dir, "sibling.crt", certtest.CertPEM(ca.Issue(t, keyed, "sibling")))
+	writeFile(t, dir, "sibling.key", certtest.KeyPEM(t, keyed))
+	writeFile(t, dir, "notes.txt", "not a certificate")
+
+	chains, err := (&PKCS11Store{uri: &pkcs11.URI{}, certDir: dir}).fileChains()
+	require.NoError(t, err)
+	var subjects []string
+	for _, chain := range chains {
+		subjects = append(subjects, chain[0].Subject.CommonName)
+	}
+	assert.ElementsMatch(t, []string{"device", "corp"}, subjects, "only key-less certificate files are left to the token")
+
+	chains, err = (&PKCS11Store{uri: &pkcs11.URI{}}).fileChains()
+	require.NoError(t, err)
+	assert.Empty(t, chains, "no directory configured means no file certificates")
+}
+
+func TestNewPKCS11Store_PIN(t *testing.T) {
+	tests := []struct {
+		name       string
+		cfg        PKCS11Config
+		wantPIN    []byte
+		wantModule string
+	}{
+		{"pin with a token label opens that token through p11-kit", PKCS11Config{URI: "pkcs11:token=netbird", PIN: "1234"}, []byte("1234"), pkcs11.DefaultModule},
+		{"pin field wins over pin-value", PKCS11Config{URI: "pkcs11:token=netbird?module-path=" + absModule("x.so") + "&pin-value=0000", PIN: "1234"}, []byte("1234"), absModule("x.so")},
+		{"uri pin-value stands in for a missing field", PKCS11Config{URI: "pkcs11:token=netbird?pin-value=0000"}, []byte("0000"), pkcs11.DefaultModule},
+		{"no pin at all means no login", PKCS11Config{URI: "pkcs11:token=netbird"}, nil, pkcs11.DefaultModule},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			store, err := NewPKCS11Store(tt.cfg, "")
+			require.NoError(t, err)
+			pin, err := store.userPIN()
+			require.NoError(t, err)
+			assert.Equal(t, tt.wantPIN, pin, "PIN, nil meaning no login")
+			assert.Equal(t, tt.wantModule, store.uri.Module(), "module to load")
+		})
+	}
+
+	_, err := NewPKCS11Store(PKCS11Config{URI: "not-a-pkcs11-uri", PIN: "1234"}, "")
+	assert.Error(t, err, "a malformed URI must not be silently replaced by the defaults")
+
+	for name, cfg := range map[string]PKCS11Config{
+		"env pin without uri":      {PIN: "1234"},
+		"env pin, uri lacks token": {URI: "pkcs11:?module-path=" + absModule("x.so"), PIN: "1234"},
+		"inline pin-value only":    {URI: "pkcs11:?pin-value=0000"},
+		"pin-source only":          {URI: "pkcs11:?pin-source=file:/etc/netbird/pkcs11.pin"},
+	} {
+		_, err := NewPKCS11Store(cfg, "")
+		assert.ErrorIs(t, err, errPINNeedsToken, "%s: a PIN must not go to whichever token is listed first", name)
+	}
+}
+
+// TestPKCS11Store_WrongPINIsTriedOnce logs in to the real token with a wrong PIN: the
+// token refuses it, and the next collection refuses to send the same PIN again rather
+// than spending another attempt of the token's lockout counter. The one attempt it does
+// spend counts against a real token's lockout, so it runs only on a token marked
+// disposable through NB_TEST_PKCS11_DISPOSABLE.
+func TestPKCS11Store_WrongPINIsTriedOnce(t *testing.T) {
+	disposable, _ := strconv.ParseBool(os.Getenv(testPKCS11DisposableEnv))
+	if !disposable {
+		t.Skipf("set %s=1 to spend a wrong-PIN attempt on the token", testPKCS11DisposableEnv)
+	}
+	_, uri := pkcs11TestStore(t, "")
+
+	wrongPIN := "wrong-pin-" + t.Name()
+	store, err := NewPKCS11Store(PKCS11Config{URI: uri, PIN: wrongPIN}, "")
+	require.NoError(t, err)
+
+	_, err = store.Candidates(context.Background())
+	require.Error(t, err)
+	assert.True(t, pkcs11.PINRejected(err), "the token itself rejects the PIN: %v", err)
+
+	_, err = store.Candidates(context.Background())
+	assert.ErrorIs(t, err, errPINRejectedBefore, "the rejected PIN is not sent to the token again")
+
+	good, err := NewPKCS11Store(PKCS11Config{URI: uri}, "")
+	require.NoError(t, err)
+	_, err = good.Candidates(context.Background())
+	assert.NoError(t, err, "the correct PIN for the same token is unaffected")
+}
+
+// absModule is an absolute module path on the platform the test runs on, as module-path
+// must be absolute.
+func absModule(name string) string {
+	if runtime.GOOS == "windows" {
+		return `C:\lib\` + name
+	}
+	return "/lib/" + name
+}
+
+// ecPrivateScalar returns the raw private scalar the token stores in CKA_VALUE.
+func ecPrivateScalar(t *testing.T, k *ecdsa.PrivateKey) []byte {
+	t.Helper()
+	raw, err := k.Bytes()
+	require.NoError(t, err)
+	return raw
+}

client/internal/certproof/store.go

@@ -0,0 +1,266 @@
+package certproof
+
+import (
+	"context"
+	"crypto"
+	"crypto/x509"
+	"encoding/pem"
+	"errors"
+	"fmt"
+	"os"
+	"path/filepath"
+	"slices"
+	"strings"
+
+	log "github.com/sirupsen/logrus"
+)
+
+const (
+	StoreDirEnv     = "NB_CERT_STORE_DIR"
+	defaultStoreDir = "/etc/netbird/certs"
+
+	// maxStoreFileSize bounds a certificate or key file of the PEM directory. A chain with
+	// its key is a few kilobytes.
+	maxStoreFileSize = 1 << 20
+)
+
+// errKeyMismatch rejects a key that does not belong to the certificate it sits with: it
+// would sign a proof management can only reject, in place of a usable later candidate.
+var (
+	errKeyMismatch  = errors.New("private key does not match the certificate")
+	errNoSiblingKey = errors.New("no key file next to the certificate")
+)
+
+// Candidate is a certificate chain the peer can sign for. Signer never exposes the key.
+// Chain is leaf first. Intermediates holds every other certificate the store has, so a
+// path to a challenge's CAs can be found even where Chain followed a different issuer,
+// such as an expired copy of a renewed intermediate.
+type Candidate struct {
+	Chain         []*x509.Certificate
+	Signer        crypto.Signer
+	Intermediates []*x509.Certificate
+}
+
+// issuers is every certificate other than the leaf that a path may run through.
+func (c Candidate) issuers() []*x509.Certificate {
+	return append(slices.Clip(c.Chain[1:]), c.Intermediates...)
+}
+
+// Store yields the certificates a peer may prove possession of. FileStore is the PEM
+// directory implementation; OS keystores (CNG, Keychain, PKCS#11) slot in here.
+type Store interface {
+	Candidates(ctx context.Context) ([]Candidate, error)
+}
+
+// Config selects where the daemon looks for certificates. PKCS11 names a token whose keys
+// sign for certificates on the token or in the PEM directory, which NB_CERT_STORE_DIR
+// names on Linux, /etc/netbird/certs by default.
+//
+// ProfileOwner is the OS account the active profile belongs to. On macOS and Windows only
+// that account's certificate store is consulted for user certificates, so on a machine
+// with several people signed in the result does not depend on who else is logged in.
+// Empty means the profile has no owner, and only the user at the physical console counts.
+// OwnerUnknown means the owner could not be determined, and no user store is consulted:
+// guessing would let whoever sits at the console answer for the profile.
+type Config struct {
+	PKCS11       PKCS11Config
+	ProfileOwner string
+	OwnerUnknown bool
+}
+
+// FileStore reads PEM files from a directory. A file holds the chain (leaf first) and
+// either its private key or a sibling "<name>.key" file holds it. The key is a plain
+// PKCS#8, EC or RSA key, or a TSS2 key the TPM signs with.
+type FileStore struct {
+	dir string
+}
+
+func NewFileStore(dir string) *FileStore {
+	return &FileStore{dir: dir}
+}
+
+func StoreDir() string {
+	if dir := os.Getenv(StoreDirEnv); dir != "" {
+		return dir
+	}
+	return defaultStoreDir
+}
+
+func (s *FileStore) Candidates(_ context.Context) ([]Candidate, error) {
+	paths, err := certFiles(s.dir)
+	if err != nil {
+		return nil, err
+	}
+
+	var candidates []Candidate
+	for _, path := range paths {
+		chain, signer, err := loadPEM(path)
+		if err != nil {
+			log.Warnf("skipping certificate %s: %v", path, err)
+			continue
+		}
+		if signer == nil {
+			log.Debugf("certificate %s has no key file, only a token can sign for it", path)
+			continue
+		}
+		candidates = append(candidates, Candidate{Chain: chain, Signer: signer})
+	}
+	return candidates, nil
+}
+
+// certFiles lists the certificate files in dir, none when the directory does not exist.
+// The directory is checked before it is listed, so one that others can write to is
+// refused before its entries are read.
+func certFiles(dir string) ([]string, error) {
+	err := checkStoreDir(dir)
+	if errors.Is(err, os.ErrNotExist) {
+		return nil, nil
+	}
+	if err != nil {
+		return nil, fmt.Errorf("refusing certificate store: %w", err)
+	}
+	entries, err := os.ReadDir(dir)
+	if errors.Is(err, os.ErrNotExist) {
+		return nil, nil
+	}
+	if err != nil {
+		return nil, fmt.Errorf("read certificate store %s: %w", dir, err)
+	}
+	var paths []string
+	for _, entry := range entries {
+		if !entry.IsDir() && isCertFile(entry.Name()) {
+			paths = append(paths, filepath.Join(dir, entry.Name()))
+		}
+	}
+	return paths, nil
+}
+
+// loadPEM reads a certificate file and its private key, held in the file itself or in
+// the sibling "<name>.key" file. The signer is nil when neither holds a key.
+func loadPEM(path string) ([]*x509.Certificate, crypto.Signer, error) {
+	data, err := readStoreFile(path)
+	if err != nil {
+		return nil, nil, err
+	}
+	chain, signer, err := parsePEM(data)
+	if err != nil {
+		return nil, nil, err
+	}
+	if len(chain) == 0 {
+		return nil, nil, errors.New("no certificate")
+	}
+	if signer == nil {
+		signer, err = siblingKey(path)
+		switch {
+		case errors.Is(err, errNoSiblingKey):
+			// A certificate with no key of its own: the token store pairs it later.
+			return chain, nil, nil
+		case err != nil:
+			return nil, nil, err
+		}
+	}
+	if !samePublicKey(signer.Public(), chain[0].PublicKey) {
+		return nil, nil, errKeyMismatch
+	}
+	return chain, signer, nil
+}
+
+// siblingKey reads the private key from the "<name>.key" file next to a certificate
+// file, reporting errNoSiblingKey when there is none.
+func siblingKey(path string) (crypto.Signer, error) {
+	keyData, err := readStoreFile(strings.TrimSuffix(path, filepath.Ext(path)) + ".key")
+	if errors.Is(err, os.ErrNotExist) {
+		return nil, errNoSiblingKey
+	}
+	if err != nil {
+		return nil, fmt.Errorf("read key file: %w", err)
+	}
+	_, signer, err := parsePEM(keyData)
+	if err != nil {
+		return nil, err
+	}
+	if signer == nil {
+		return nil, errors.New("no private key in key file")
+	}
+	return signer, nil
+}
+
+func samePublicKey(a, b crypto.PublicKey) bool {
+	equaler, ok := a.(interface{ Equal(crypto.PublicKey) bool })
+	return ok && equaler.Equal(b)
+}
+
+func parsePEM(data []byte) ([]*x509.Certificate, crypto.Signer, error) {
+	var chain []*x509.Certificate
+	var signer crypto.Signer
+	for {
+		var block *pem.Block
+		block, data = pem.Decode(data)
+		if block == nil {
+			return chain, signer, nil
+		}
+		switch block.Type {
+		case "CERTIFICATE":
+			cert, err := x509.ParseCertificate(block.Bytes)
+			if err != nil {
+				return nil, nil, fmt.Errorf("parse certificate: %w", err)
+			}
+			chain = append(chain, cert)
+		case "PRIVATE KEY", "EC PRIVATE KEY", "RSA PRIVATE KEY", tss2KeyPEMType:
+			key, err := parsePrivateKey(block)
+			if err != nil {
+				return nil, nil, err
+			}
+			signer = key
+		}
+	}
+}
+
+func parsePrivateKey(block *pem.Block) (crypto.Signer, error) {
+	var key any
+	var err error
+	switch block.Type {
+	case tss2KeyPEMType:
+		return parseTSS2Key(block.Bytes)
+	case "EC PRIVATE KEY":
+		key, err = x509.ParseECPrivateKey(block.Bytes)
+	case "RSA PRIVATE KEY":
+		key, err = x509.ParsePKCS1PrivateKey(block.Bytes)
+	default:
+		key, err = x509.ParsePKCS8PrivateKey(block.Bytes)
+	}
+	if err != nil {
+		return nil, fmt.Errorf("parse private key: %w", err)
+	}
+	signer, ok := key.(crypto.Signer)
+	if !ok {
+		return nil, errors.New("private key cannot sign")
+	}
+	return signer, nil
+}
+
+func isCertFile(name string) bool {
+	switch strings.ToLower(filepath.Ext(name)) {
+	case ".pem", ".crt", ".cer":
+		return true
+	}
+	return false
+}
+
+// Stores queries several stores and carries on when one fails, so a broken token cannot
+// hide the certificates a directory holds. A failure is logged instead of returned
+// because Collect treats a store error as "no proofs at all".
+type Stores []Store
+
+func (s Stores) Candidates(ctx context.Context) ([]Candidate, error) {
+	var all []Candidate
+	for _, store := range s {
+		candidates, err := store.Candidates(ctx)
+		if err != nil {
+			log.Warnf("certificate store %T unavailable: %v", store, err)
+			continue
+		}
+		all = append(all, candidates...)
+	}
+	return all, nil
+}

client/internal/certproof/store_other.go

@@ -0,0 +1,39 @@
+//go:build ((!darwin && !windows) || ios) && !js
+
+package certproof
+
+import (
+	"sync"
+
+	log "github.com/sirupsen/logrus"
+
+	"github.com/netbirdio/netbird/client/internal/pkcs11"
+)
+
+var unsupportedTokenOnce sync.Once
+
+// DefaultStore is the PEM directory named by NB_CERT_STORE_DIR, or /etc/netbird/certs.
+func DefaultStore() Store {
+	return NewFileStore(StoreDir())
+}
+
+// storeWithToken reads the PEM directory cfg names, joined by the PKCS#11 token when cfg
+// names one. The token pairs the directory's key-less certificates with its own keys.
+func storeWithToken(cfg Config) Store {
+	files := NewFileStore(StoreDir())
+	if cfg.PKCS11.URI == "" && cfg.PKCS11.PIN == "" {
+		return files
+	}
+	if !pkcs11.Supported() {
+		unsupportedTokenOnce.Do(func() {
+			log.Warnf("ignoring the configured PKCS#11 token: %v", pkcs11.ErrUnsupported)
+		})
+		return files
+	}
+	token, err := NewPKCS11Store(cfg.PKCS11, StoreDir())
+	if err != nil {
+		log.Warnf("ignoring PKCS#11 URI: %v", err)
+		return files
+	}
+	return Stores{files, token}
+}

client/internal/certproof/store_other_test.go

@@ -0,0 +1,44 @@
+//go:build ((!darwin && !windows) || ios) && !js
+
+package certproof
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/client/internal/pkcs11"
+)
+
+func TestStoreWithToken(t *testing.T) {
+	dir := storeDir(t)
+	t.Setenv(StoreDirEnv, dir)
+
+	files, ok := storeWithToken(Config{}).(*FileStore)
+	require.True(t, ok, "no token configured reads the PEM directory alone")
+	assert.Equal(t, dir, files.dir, "the directory comes from NB_CERT_STORE_DIR")
+
+	assert.IsType(t, &FileStore{}, storeWithToken(Config{PKCS11: PKCS11Config{URI: "not-a-pkcs11-uri"}}), "an invalid URI must not hide the PEM directory")
+	assert.IsType(t, &FileStore{}, storeWithToken(Config{PKCS11: PKCS11Config{PIN: "1234"}}), "a PIN naming no token is refused and leaves the PEM directory")
+
+	configured := map[string]PKCS11Config{
+		"env pin with token uri": {URI: "pkcs11:token=netbird", PIN: "1234"},
+		"uri alone":              {URI: "pkcs11:token=netbird?pin-value=1234"},
+	}
+	if !pkcs11.Supported() {
+		for name, cfg := range configured {
+			assert.IsType(t, &FileStore{}, storeWithToken(Config{PKCS11: cfg}),
+				"%s: a build without PKCS#11 support reads the PEM directory alone", name)
+		}
+		return
+	}
+	for name, cfg := range configured {
+		store, ok := storeWithToken(Config{PKCS11: cfg}).(Stores)
+		require.True(t, ok, "%s joins the token to the PEM directory", name)
+		require.Len(t, store, 2, name)
+		token, ok := store[1].(*PKCS11Store)
+		require.True(t, ok, name)
+		assert.Equal(t, dir, token.certDir, "%s: the token pairs certificates from the same directory", name)
+	}
+}

client/internal/certproof/store_tpm_test.go

@@ -0,0 +1,96 @@
+package certproof
+
+import (
+	"context"
+	"crypto/ecdsa"
+	"crypto/elliptic"
+	"crypto/rand"
+	"encoding/pem"
+	"os"
+	"testing"
+	"time"
+
+	"github.com/google/go-tpm/legacy/tpm2"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/client/internal/tpm"
+	"github.com/netbirdio/netbird/client/internal/tpm/tpmtest"
+	"github.com/netbirdio/netbird/shared/management/certposture"
+	"github.com/netbirdio/netbird/shared/management/certposture/certtest"
+	"github.com/netbirdio/netbird/shared/management/proto"
+)
+
+func TestFileStore_TPMKeyFile(t *testing.T) {
+	ca := certtest.NewCA(t, "corp")
+	key, err := ecdsa.GenerateKey(elliptic.P256(), rand.Reader)
+	require.NoError(t, err)
+	leaf := ca.Issue(t, key, "device")
+
+	dir := storeDir(t)
+	writeFile(t, dir, "device.pem", certtest.CertPEM(leaf))
+	writeFile(t, dir, "device.key", tpmtest.KeyPEM(t, &key.PublicKey))
+
+	// A key that needs a password can never be used silently, so its certificate is skipped.
+	locked := certtest.ECDSAKey(t)
+	writeFile(t, dir, "locked.pem", certtest.CertPEM(ca.Issue(t, locked, "locked")))
+	writeFile(t, dir, "locked.key", tpmtest.KeyPEM(t, locked.Public().(*ecdsa.PublicKey), tpmtest.WithAuth()))
+
+	candidates, err := NewFileStore(dir).Candidates(context.Background())
+	require.NoError(t, err)
+	require.Len(t, candidates, 1, "only the key without an authorization value is usable")
+	assert.True(t, leaf.Equal(candidates[0].Chain[0]), "candidate must carry the device certificate")
+	assert.True(t, key.PublicKey.Equal(candidates[0].Signer.Public()), "signer must report the certificate's key")
+}
+
+// TestCollect_TPMKeyEndToEnd runs against the TPM named by NB_TPM_DEVICE, for example a
+// swtpm started with:
+//
+//	swtpm socket --tpm2 --server type=unixio,path=/tmp/swtpm.sock \
+//	  --ctrl type=unixio,path=/tmp/swtpm.ctrl --flags not-need-init,startup-clear
+//
+// It creates a key the way tpm2-openssl does, under a transient ECC primary in the owner
+// hierarchy, and proves the certificate for it through the regular file store.
+func TestCollect_TPMKeyEndToEnd(t *testing.T) {
+	if os.Getenv(tpm.DeviceEnv) == "" {
+		t.Skipf("set %s to a TPM device or swtpm socket to run", tpm.DeviceEnv)
+	}
+	public, private := createTPMKey(t)
+	keyPEM := tpmtest.EncodePEM(t, public, private)
+	block, _ := pem.Decode([]byte(keyPEM))
+	require.NotNil(t, block)
+	signer, err := tpm.ParseKey(block.Bytes)
+	require.NoError(t, err)
+
+	ca := certtest.NewCA(t, "corp")
+	leaf := ca.Issue(t, signer, "device")
+	dir := storeDir(t)
+	writeFile(t, dir, "device.pem", certtest.CertPEM(leaf))
+	writeFile(t, dir, "device.key", keyPEM)
+
+	challenger := certposture.NewChallenger([]byte("secret"))
+	now := time.Now()
+	nonce := challenger.Nonce(peerKey, now)
+	checks := []*proto.Checks{{CertificateChallenge: &proto.CertificateChallenge{Nonce: nonce, CaCertificates: []string{ca.PEM}}}}
+
+	proofs := Collect(context.Background(), NewFileStore(dir), checks, peerKey)
+	require.Len(t, proofs, 1, "the TPM-held key must prove the certificate")
+	chain, err := challenger.Verify(proofs[0], peerKey, now)
+	require.NoError(t, err)
+	assert.True(t, leaf.Equal(chain[0]), "proof must carry the device certificate")
+}
+
+func createTPMKey(t *testing.T) (public, private []byte) {
+	t.Helper()
+	rwc, err := tpm.Open()
+	require.NoError(t, err)
+	defer func() { _ = rwc.Close() }()
+
+	parent, _, err := tpm2.CreatePrimary(rwc, tpm2.HandleOwner, tpm2.PCRSelection{}, "", "", tpmtest.ECCSRKTemplate)
+	require.NoError(t, err)
+	defer func() { _ = tpm2.FlushContext(rwc, parent) }()
+
+	private, public, _, _, _, err = tpm2.CreateKey(rwc, parent, tpm2.PCRSelection{}, "", "", tpmtest.SigningTemplate())
+	require.NoError(t, err)
+	return public, private
+}

client/internal/certproof/store_tss2.go

@@ -0,0 +1,17 @@
+//go:build !js
+
+package certproof
+
+import (
+	"crypto"
+
+	"github.com/netbirdio/netbird/client/internal/tpm"
+)
+
+// tss2KeyPEMType is the PEM block type of a TPM 2.0 key file.
+const tss2KeyPEMType = tpm.KeyPEMType
+
+// parseTSS2Key returns a signer for a TPM-held key file that signs inside the TPM.
+func parseTSS2Key(der []byte) (crypto.Signer, error) {
+	return tpm.ParseKey(der)
+}

client/internal/certproof/store_tss2_js.go

@@ -0,0 +1,16 @@
+//go:build js
+
+package certproof
+
+import (
+	"crypto"
+	"errors"
+)
+
+// tss2KeyPEMType is the PEM block type of a TPM 2.0 key file. A browser has no TPM, so
+// the TPM library stays out of the WebAssembly build and such a key is refused.
+const tss2KeyPEMType = "TSS2 PRIVATE KEY"
+
+func parseTSS2Key([]byte) (crypto.Signer, error) {
+	return nil, errors.New("TPM keys are not supported in this build")
+}

client/internal/certproof/storefile_other.go

@@ -0,0 +1,34 @@
+//go:build !unix
+
+package certproof
+
+import (
+	"fmt"
+	"io"
+	"os"
+)
+
+// readStoreFile reads a certificate or key file of the PEM directory, up to the size
+// limit. Ownership is checked on Unix only; elsewhere the PEM directory is not the store
+// the daemon reads by default.
+func readStoreFile(path string) ([]byte, error) {
+	f, err := os.Open(path)
+	if err != nil {
+		return nil, err
+	}
+	defer func() { _ = f.Close() }()
+
+	data, err := io.ReadAll(io.LimitReader(f, maxStoreFileSize+1))
+	if err != nil {
+		return nil, fmt.Errorf("read %s: %w", path, err)
+	}
+	if len(data) > maxStoreFileSize {
+		return nil, fmt.Errorf("%s is over the %d byte limit", path, maxStoreFileSize)
+	}
+	return data, nil
+}
+
+// checkStoreDir accepts any directory where the PEM directory is not a default store.
+func checkStoreDir(string) error {
+	return nil
+}

client/internal/certproof/storefile_unix.go

@@ -0,0 +1,101 @@
+//go:build unix
+
+package certproof
+
+import (
+	"errors"
+	"fmt"
+	"io"
+	"os"
+	"syscall"
+)
+
+// readStoreFile reads a certificate or key file of the PEM directory as the daemon may
+// trust it: the path must not be a symlink and must be a regular file that only root or
+// this process's own user can write. Otherwise a user able to place or redirect a file
+// could make the daemon sign with a key of their choosing, or with a root-only key kept
+// elsewhere on the machine. The file is checked through the descriptor it is read from,
+// so it cannot be swapped between the check and the read.
+func readStoreFile(path string) ([]byte, error) {
+	// O_NONBLOCK keeps a FIFO planted in the directory from blocking the open; the
+	// regular-file check below then refuses it.
+	f, err := os.OpenFile(path, os.O_RDONLY|syscall.O_NOFOLLOW|syscall.O_NONBLOCK, 0)
+	if err != nil {
+		return nil, err
+	}
+	defer func() { _ = f.Close() }()
+
+	info, err := f.Stat()
+	if err != nil {
+		return nil, fmt.Errorf("stat %s: %w", path, err)
+	}
+	if !info.Mode().IsRegular() {
+		return nil, fmt.Errorf("%s is not a regular file", path)
+	}
+	if err := checkTrustedOwner(info); err != nil {
+		return nil, fmt.Errorf("%s: %w", path, err)
+	}
+	if info.Size() > maxStoreFileSize {
+		return nil, fmt.Errorf("%s is %d bytes, over the %d byte limit", path, info.Size(), maxStoreFileSize)
+	}
+	data, err := io.ReadAll(io.LimitReader(f, maxStoreFileSize+1))
+	if err != nil {
+		return nil, fmt.Errorf("read %s: %w", path, err)
+	}
+	if len(data) > maxStoreFileSize {
+		return nil, fmt.Errorf("%s grew past the %d byte limit while being read", path, maxStoreFileSize)
+	}
+	return data, nil
+}
+
+// checkStoreDir refuses a PEM directory that someone other than root or this process's
+// user could add files to or rename files in. The configured path may be a symlink, as
+// distributions place certificate directories behind them, but only one root or this
+// process's user owns: otherwise whoever owns the link could point it at any directory.
+func checkStoreDir(dir string) error {
+	link, err := os.Lstat(dir)
+	if err != nil {
+		return err
+	}
+	if link.Mode()&os.ModeSymlink != 0 {
+		if err := checkTrustedUID(link); err != nil {
+			return fmt.Errorf("symlink %s: %w", dir, err)
+		}
+	}
+	info, err := os.Stat(dir)
+	if err != nil {
+		return err
+	}
+	if !info.IsDir() {
+		return fmt.Errorf("%s is not a directory", dir)
+	}
+	if err := checkTrustedOwner(info); err != nil {
+		return fmt.Errorf("%s: %w", dir, err)
+	}
+	return nil
+}
+
+// checkTrustedOwner requires info to belong to root or to this process's user and to be
+// writable by nobody else. Group write is refused even for root's group, which ordinary
+// users may be members of. An owner that cannot be read is refused.
+func checkTrustedOwner(info os.FileInfo) error {
+	if err := checkTrustedUID(info); err != nil {
+		return err
+	}
+	if perm := info.Mode().Perm(); perm&0o022 != 0 {
+		return fmt.Errorf("writable by group or other users (mode %#o)", perm)
+	}
+	return nil
+}
+
+// checkTrustedUID requires info to belong to root or to this process's user.
+func checkTrustedUID(info os.FileInfo) error {
+	st, ok := info.Sys().(*syscall.Stat_t)
+	if !ok {
+		return errors.New("owner cannot be determined")
+	}
+	if st.Uid != 0 && int(st.Uid) != os.Geteuid() {
+		return fmt.Errorf("owned by uid %d, not by root", st.Uid)
+	}
+	return nil
+}

client/internal/certproof/storefile_unix_test.go

@@ -0,0 +1,136 @@
+//go:build unix
+
+package certproof
+
+import (
+	"context"
+	"os"
+	"path/filepath"
+	"syscall"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/shared/management/certposture/certtest"
+)
+
+// devicePair writes a certificate and its key as separate files into dir.
+func devicePair(t *testing.T, dir string) {
+	t.Helper()
+	ca := certtest.NewCA(t, "corp-root")
+	key := certtest.ECDSAKey(t)
+	writeFile(t, dir, "device.crt", certtest.CertPEM(ca.Issue(t, key, "device")))
+	writeFile(t, dir, "device.key", certtest.KeyPEM(t, key))
+}
+
+func candidates(t *testing.T, dir string) ([]Candidate, error) {
+	t.Helper()
+	return NewFileStore(dir).Candidates(context.Background())
+}
+
+func TestFileStore_AcceptsAPrivatePair(t *testing.T) {
+	dir := storeDir(t)
+	devicePair(t, dir)
+
+	got, err := candidates(t, dir)
+	require.NoError(t, err)
+	assert.Len(t, got, 1, "a certificate and key only their owner can write are used")
+}
+
+func TestFileStore_RefusesASymlinkedKey(t *testing.T) {
+	dir := storeDir(t)
+	devicePair(t, dir)
+
+	// The real key moves elsewhere and the directory only links to it, as a user able
+	// to write the directory would do to point the daemon at a key it should not use.
+	elsewhere := filepath.Join(storeDir(t), "other.key")
+	require.NoError(t, os.Rename(filepath.Join(dir, "device.key"), elsewhere))
+	require.NoError(t, os.Symlink(elsewhere, filepath.Join(dir, "device.key")))
+
+	got, err := candidates(t, dir)
+	require.NoError(t, err)
+	assert.Empty(t, got, "a key reached through a symlink is never used")
+}
+
+func TestFileStore_RefusesASymlinkedCertificateFile(t *testing.T) {
+	dir := storeDir(t)
+	other := storeDir(t)
+	devicePair(t, other)
+	require.NoError(t, os.Symlink(filepath.Join(other, "device.crt"), filepath.Join(dir, "device.crt")))
+	require.NoError(t, os.Symlink(filepath.Join(other, "device.key"), filepath.Join(dir, "device.key")))
+
+	got, err := candidates(t, dir)
+	require.NoError(t, err)
+	assert.Empty(t, got, "a certificate file that is a symlink is skipped")
+}
+
+func TestFileStore_RefusesFilesOthersCanWrite(t *testing.T) {
+	for name, mode := range map[string]os.FileMode{"world-writable": 0o602, "group-writable": 0o620} {
+		t.Run(name, func(t *testing.T) {
+			dir := storeDir(t)
+			devicePair(t, dir)
+			require.NoError(t, os.Chmod(filepath.Join(dir, "device.key"), mode))
+
+			got, err := candidates(t, dir)
+			require.NoError(t, err)
+			assert.Empty(t, got, "a key %s may have been replaced by someone else, even when the group is root's", name)
+		})
+	}
+}
+
+func TestFileStore_RefusesADirectoryOthersCanWrite(t *testing.T) {
+	dir := storeDir(t)
+	devicePair(t, dir)
+	require.NoError(t, os.Chmod(dir, 0o777))
+
+	_, err := candidates(t, dir)
+	assert.ErrorContains(t, err, "refusing certificate store", "files in a directory anyone can write are not trusted")
+}
+
+func TestFileStore_SkipsAFIFOWithoutBlocking(t *testing.T) {
+	dir := storeDir(t)
+	require.NoError(t, syscall.Mkfifo(filepath.Join(dir, "trap.pem"), 0o600))
+	devicePair(t, dir)
+
+	got, err := candidates(t, dir)
+	require.NoError(t, err)
+	assert.Len(t, got, 1, "a FIFO is skipped and the real pair still loads")
+}
+
+func TestReadStoreFile_RefusesOversizedFiles(t *testing.T) {
+	dir := storeDir(t)
+	path := filepath.Join(dir, "huge.pem")
+	require.NoError(t, os.WriteFile(path, make([]byte, maxStoreFileSize+1), 0o600))
+
+	_, err := readStoreFile(path)
+	assert.ErrorContains(t, err, "limit")
+}
+
+func TestFileStore_AcceptsASymlinkedDirectoryItsOwnerControls(t *testing.T) {
+	target := storeDir(t)
+	devicePair(t, target)
+	link := filepath.Join(storeDir(t), "certs")
+	require.NoError(t, os.Symlink(target, link))
+
+	// The link belongs to this process's user, as one root creates belongs to root.
+	got, err := candidates(t, link)
+	require.NoError(t, err)
+	assert.Len(t, got, 1, "a certificate directory behind a symlink the owner controls is read")
+}
+
+func TestFileStore_ChecksTheDirectoryBeforeListingIt(t *testing.T) {
+	dir := storeDir(t)
+	devicePair(t, dir)
+	require.NoError(t, os.Chmod(dir, 0o777))
+
+	paths, err := certFiles(dir)
+	assert.ErrorContains(t, err, "refusing certificate store")
+	assert.Nil(t, paths, "nothing in a directory anyone can write to is listed")
+}
+
+func TestFileStore_MissingDirectoryIsEmpty(t *testing.T) {
+	paths, err := certFiles(filepath.Join(storeDir(t), "missing"))
+	assert.NoError(t, err, "a store directory that does not exist yet is not an error")
+	assert.Empty(t, paths)
+}

client/internal/certproof/systemstore_windows.go

@@ -0,0 +1,270 @@
+package certproof
+
+import (
+	"bytes"
+	"context"
+	"crypto"
+	"crypto/sha256"
+	"crypto/x509"
+	"errors"
+	"fmt"
+	"io"
+	"slices"
+	"unsafe"
+
+	log "github.com/sirupsen/logrus"
+	"golang.org/x/sys/windows"
+)
+
+const (
+	personalStore     = "MY"
+	intermediateStore = "CA"
+
+	cryptAcquireSilentFlag          = 0x00000040
+	cryptAcquirePreferNCryptKeyFlag = 0x00020000
+	certNCryptKeySpec               = 0xFFFFFFFF
+	atKeyExchange                   = 1
+	atSignature                     = 2
+	bcryptPadPSS                    = 0x00000008
+)
+
+// errLegacyKey reports a private key CNG cannot open: it lives in a legacy CryptoAPI
+// provider, which only offers PKCS#1 v1.5 and no RSA-PSS or ECDSA.
+var errLegacyKey = errors.New("private key is held by a legacy CryptoAPI provider, which certificate posture cannot sign with")
+
+var (
+	crypt32 = windows.NewLazySystemDLL("crypt32.dll")
+	ncrypt  = windows.NewLazySystemDLL("ncrypt.dll")
+
+	procCryptAcquireCertificatePrivateKey = crypt32.NewProc("CryptAcquireCertificatePrivateKey")
+	procNCryptSignHash                    = ncrypt.NewProc("NCryptSignHash")
+	procNCryptFreeObject                  = ncrypt.NewProc("NCryptFreeObject")
+)
+
+type bcryptPSSPaddingInfo struct {
+	algID *uint16
+	salt  uint32
+}
+
+// DefaultStore is the local machine's personal certificate store, where device
+// certificates enrolled through AD or Intune are kept.
+func DefaultStore() Store {
+	return NewSystemStore()
+}
+
+// SystemStore yields the identities of a personal certificate store, completing their
+// chains from the matching intermediate CA store. Keys are used through CNG and never
+// exported.
+//
+// The location decides whose certificates these are. The local machine store is the one
+// a service reads; the current user store lives in the signed-in user's registry hive
+// with keys protected against their profile, so it is only readable while running as
+// that user.
+type SystemStore struct {
+	location uint32
+}
+
+// NewSystemStore reads the local machine store, which is what the daemon uses.
+func NewSystemStore() *SystemStore {
+	return &SystemStore{location: windows.CERT_SYSTEM_STORE_LOCAL_MACHINE}
+}
+
+// NewUserStore reads the calling user's personal store. It is only useful in a process
+// already running as that user, which is what the posture helper is.
+func NewUserStore() *SystemStore {
+	return &SystemStore{location: windows.CERT_SYSTEM_STORE_CURRENT_USER}
+}
+
+func (s *SystemStore) Candidates(_ context.Context) ([]Candidate, error) {
+	leaves, err := storeCertificates(s.location, personalStore)
+	if err != nil {
+		return nil, err
+	}
+	intermediates, err := storeCertificates(s.location, intermediateStore)
+	if err != nil {
+		return nil, err
+	}
+	log.Debugf("certificate store %s holds %d personal certificates and %d intermediates", s, len(leaves), len(intermediates))
+	if len(leaves) == 0 {
+		return nil, nil
+	}
+
+	pool := slices.Concat(intermediates, leaves)
+	candidates := make([]Candidate, 0, len(leaves))
+	for _, leaf := range leaves {
+		chain := buildChain(leaf, pool)
+		log.Debugf("certificate store %s candidate %q issued by %q built a chain of %d certificates", s, leaf.Subject, leaf.Issuer, len(chain))
+		candidates = append(candidates, Candidate{Chain: chain, Signer: &systemStoreSigner{leaf: leaf, location: s.location}, Intermediates: pool})
+	}
+	return candidates, nil
+}
+
+// String names the store location the way the Windows documentation does.
+func (s *SystemStore) String() string {
+	if s.location == windows.CERT_SYSTEM_STORE_CURRENT_USER {
+		return "CurrentUser"
+	}
+	return "LocalMachine"
+}
+
+// systemStoreSigner holds only the certificate; the store entry and its key are acquired
+// at signing time so no handles outlive a call.
+type systemStoreSigner struct {
+	leaf     *x509.Certificate
+	location uint32
+}
+
+func (s *systemStoreSigner) Public() crypto.PublicKey {
+	return s.leaf.PublicKey
+}
+
+func (s *systemStoreSigner) Sign(_ io.Reader, digest []byte, opts crypto.SignerOpts) ([]byte, error) {
+	scheme, err := schemeFor(s.leaf.PublicKey, opts)
+	if err != nil {
+		return nil, err
+	}
+	store, err := openStore(s.location, personalStore)
+	if err != nil {
+		return nil, err
+	}
+	defer func() { _ = windows.CertCloseStore(store, 0) }()
+
+	var signature []byte
+	err = eachCertificate(store, func(ctx *windows.CertContext) (bool, error) {
+		if !bytes.Equal(encodedCert(ctx), s.leaf.Raw) {
+			return false, nil
+		}
+		signature, err = signWithContext(ctx, scheme, digest)
+		return true, err
+	})
+	if err != nil {
+		return nil, err
+	}
+	if signature == nil {
+		return nil, errors.New("certificate is no longer in the personal store")
+	}
+	return signature, nil
+}
+
+func signWithContext(ctx *windows.CertContext, scheme sigScheme, digest []byte) ([]byte, error) {
+	var key uintptr
+	var keySpec uint32
+	var callerFree int32
+	ok, _, err := procCryptAcquireCertificatePrivateKey.Call(uintptr(unsafe.Pointer(ctx)), cryptAcquireSilentFlag|cryptAcquirePreferNCryptKeyFlag, 0,
+		uintptr(unsafe.Pointer(&key)), uintptr(unsafe.Pointer(&keySpec)), uintptr(unsafe.Pointer(&callerFree)))
+	if ok == 0 {
+		return nil, fmt.Errorf("acquire private key: %w", err)
+	}
+	if keySpec != certNCryptKeySpec {
+		if callerFree != 0 {
+			_ = windows.CryptReleaseContext(windows.Handle(key), 0)
+		}
+		return nil, legacyKeyError(keySpec)
+	}
+	if callerFree != 0 {
+		defer func() { _, _, _ = procNCryptFreeObject.Call(key) }()
+	}
+
+	var padding unsafe.Pointer
+	var flags uintptr
+	if scheme == schemeRSAPSSSHA256 {
+		algID, _ := windows.UTF16PtrFromString("SHA256")
+		info := bcryptPSSPaddingInfo{algID: algID, salt: sha256.Size}
+		padding, flags = unsafe.Pointer(&info), bcryptPadPSS
+	}
+	size, err := ncryptSignHash(key, padding, digest, nil, flags)
+	if err != nil {
+		return nil, err
+	}
+	signature := make([]byte, size)
+	if size, err = ncryptSignHash(key, padding, digest, signature, flags); err != nil {
+		return nil, err
+	}
+	signature = signature[:size]
+	if scheme == schemeRSAPSSSHA256 {
+		return signature, nil
+	}
+	return ecdsaSignatureASN1(signature)
+}
+
+// legacyKeyError describes a key acquired through CryptoAPI rather than CNG, naming its
+// key spec and the remedy, since the certificate otherwise looks usable in the store.
+func legacyKeyError(keySpec uint32) error {
+	spec := fmt.Sprintf("key spec %d", keySpec)
+	switch keySpec {
+	case atKeyExchange:
+		spec = "AT_KEYEXCHANGE"
+	case atSignature:
+		spec = "AT_SIGNATURE"
+	}
+	return fmt.Errorf("%w (%s); re-enrol it with a certificate template whose provider is a CNG key storage provider", errLegacyKey, spec)
+}
+
+func ncryptSignHash(key uintptr, padding unsafe.Pointer, digest, signature []byte, flags uintptr) (uint32, error) {
+	var result uint32
+	var signaturePtr uintptr
+	if len(signature) > 0 {
+		signaturePtr = uintptr(unsafe.Pointer(&signature[0]))
+	}
+	status, _, _ := procNCryptSignHash.Call(key, uintptr(padding), uintptr(unsafe.Pointer(&digest[0])), uintptr(len(digest)),
+		signaturePtr, uintptr(len(signature)), uintptr(unsafe.Pointer(&result)), flags)
+	if uint32(status) != 0 {
+		return 0, fmt.Errorf("NCryptSignHash: 0x%08x", uint32(status))
+	}
+	return result, nil
+}
+
+func storeCertificates(location uint32, name string) ([]*x509.Certificate, error) {
+	store, err := openStore(location, name)
+	if err != nil {
+		return nil, err
+	}
+	defer func() { _ = windows.CertCloseStore(store, 0) }()
+
+	var certs []*x509.Certificate
+	err = eachCertificate(store, func(ctx *windows.CertContext) (bool, error) {
+		cert, err := x509.ParseCertificate(bytes.Clone(encodedCert(ctx)))
+		if err != nil {
+			log.Warnf("skipping certificate in %s store: %v", name, err)
+			return false, nil
+		}
+		certs = append(certs, cert)
+		return false, nil
+	})
+	return certs, err
+}
+
+func openStore(location uint32, name string) (windows.Handle, error) {
+	namePtr, err := windows.UTF16PtrFromString(name)
+	if err != nil {
+		return 0, err
+	}
+	flags := location | uint32(windows.CERT_STORE_READONLY_FLAG|windows.CERT_STORE_OPEN_EXISTING_FLAG)
+	store, err := windows.CertOpenStore(windows.CERT_STORE_PROV_SYSTEM, 0, 0, flags, uintptr(unsafe.Pointer(namePtr)))
+	if err != nil {
+		return 0, fmt.Errorf("open %s certificate store: %w", name, err)
+	}
+	return store, nil
+}
+
+func eachCertificate(store windows.Handle, fn func(*windows.CertContext) (bool, error)) error {
+	var ctx *windows.CertContext
+	for {
+		next, err := windows.CertEnumCertificatesInStore(store, ctx)
+		if next == nil {
+			if errors.Is(err, windows.Errno(windows.CRYPT_E_NOT_FOUND)) || errors.Is(err, windows.ERROR_NO_MORE_FILES) {
+				return nil
+			}
+			return fmt.Errorf("enumerate certificates: %w", err)
+		}
+		ctx = next
+		if stop, err := fn(ctx); stop || err != nil {
+			_ = windows.CertFreeCertificateContext(ctx)
+			return err
+		}
+	}
+}
+
+func encodedCert(ctx *windows.CertContext) []byte {
+	return unsafe.Slice(ctx.EncodedCert, ctx.Length)
+}

client/internal/certproof/systemstore_windows_test.go

@@ -0,0 +1,16 @@
+package certproof
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestLegacyKeyError_NamesTheKeySpecAndTheRemedy(t *testing.T) {
+	for spec, want := range map[uint32]string{atKeyExchange: "AT_KEYEXCHANGE", atSignature: "AT_SIGNATURE", 7: "key spec 7"} {
+		err := legacyKeyError(spec)
+		assert.ErrorIs(t, err, errLegacyKey, "callers can tell a legacy key from other failures")
+		assert.ErrorContains(t, err, want, "the key spec found is named")
+		assert.ErrorContains(t, err, "key storage provider", "the error says how to fix the enrolment")
+	}
+}

client/internal/connect.go

@@ -26,6 +26,7 @@ import (
 	"github.com/netbirdio/netbird/client/iface"
 	"github.com/netbirdio/netbird/client/iface/device"
 	"github.com/netbirdio/netbird/client/iface/netstack"
+	"github.com/netbirdio/netbird/client/internal/certproof"
 	"github.com/netbirdio/netbird/client/internal/dns"
 	"github.com/netbirdio/netbird/client/internal/lazyconn"
 	"github.com/netbirdio/netbird/client/internal/listener"
@@ -75,6 +76,9 @@ type ConnectClient struct {
 	// netMgr gates every reconnection loop on OS-reported network
 	// availability and sweeps connections on network change.
 	netMgr *netevents.Manager
+
+	profileOwner        string
+	profileOwnerUnknown bool
 }
 
 // ConnectClientOption configures optional ConnectClient behavior.
@@ -85,6 +89,18 @@ func WithNetEvents(events *netevents.Manager) ConnectClientOption {
 	return func(c *ConnectClient) { c.netMgr = events }
 }
 
+// WithProfileOwner names the OS account the active profile belongs to, whose own
+// certificate store answers user certificate posture checks.
+func WithProfileOwner(username string) ConnectClientOption {
+	return func(c *ConnectClient) { c.profileOwner = username }
+}
+
+// WithUnknownProfileOwner records that the active profile's owner could not be
+// determined, so no user's certificate store answers certificate posture checks.
+func WithUnknownProfileOwner() ConnectClientOption {
+	return func(c *ConnectClient) { c.profileOwnerUnknown = true }
+}
+
 func NewConnectClient(
 	ctx context.Context,
 	config *profilemanager.Config,
@@ -420,6 +436,8 @@ func (c *ConnectClient) run(mobileDependency MobileDependency, runningChan chan
 			return wrapErr(err)
 		}
 		engineConfig.TempDir = mobileDependency.TempDir
+		engineConfig.CertStore.ProfileOwner = c.profileOwner
+		engineConfig.CertStore.OwnerUnknown = c.profileOwnerUnknown
 		// Leave StateDir empty when there is no state path so a disk-backed
 		// syncstore falls back to os.TempDir() instead of filepath.Dir("") == ".".
 		if path != "" {
@@ -675,6 +693,8 @@ func createEngineConfig(key wgtypes.Key, config *profilemanager.Config, peerConf
 
 		LazyConnection: lazyconn.ParseState(config.LazyConnection),
 
+		CertStore: certStoreConfig(config),
+
 		MTU:     selectMTU(config.MTU, peerConfig.Mtu),
 		LogPath: logPath,
 
@@ -701,6 +721,22 @@ func createEngineConfig(key wgtypes.Key, config *profilemanager.Config, peerConf
 	return engineConf, nil
 }
 
+// certStoreConfig reads where certificate posture finds certificates from the daemon's
+// environment, NB_CERT_STORE_DIR and NB_CERT_PKCS11_URI with NB_TPM_PIN. The profile
+// config fields that once held them are ignored, and a value left there is reported
+// once, so a setup relying on it does not silently stop proving.
+func certStoreConfig(config *profilemanager.Config) certproof.Config {
+	if config.CertStoreDir != "" || config.CertPKCS11URI != "" {
+		legacyCertConfigOnce.Do(func() {
+			log.Warnf("certificate posture: CertStoreDir and CertPKCS11URI in the profile config are ignored, set %s and %s in the daemon's environment instead",
+				certproof.StoreDirEnv, certproof.PKCS11URIEnv)
+		})
+	}
+	return certproof.Config{PKCS11: certproof.PKCS11FromEnv()}
+}
+
+var legacyCertConfigOnce sync.Once
+
 func selectMTU(localMTU uint16, peerMTU int32) uint16 {
 	var finalMTU uint16 = iface.DefaultMTU
 	if localMTU > 0 {

client/internal/connect_certstore_test.go

@@ -0,0 +1,26 @@
+package internal
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+
+	"github.com/netbirdio/netbird/client/internal/certproof"
+	"github.com/netbirdio/netbird/client/internal/profilemanager"
+)
+
+// TestCertStoreConfig_ReadsEnvironmentNotProfile: the token URI may carry the PIN, so it
+// and the store directory come from the daemon's environment, and values left in the
+// profile config from an older version are ignored.
+func TestCertStoreConfig_ReadsEnvironmentNotProfile(t *testing.T) {
+	t.Setenv(certproof.PKCS11URIEnv, "pkcs11:token=env")
+	t.Setenv(certproof.PINEnv, "1234")
+
+	cfg := certStoreConfig(&profilemanager.Config{
+		CertStoreDir:  "/from/profile",
+		CertPKCS11URI: "pkcs11:token=profile?pin-value=9999",
+	})
+
+	assert.Equal(t, "pkcs11:token=env", cfg.PKCS11.URI, "the URI comes from the environment")
+	assert.Equal(t, "1234", cfg.PKCS11.PIN, "the PIN comes from the environment")
+}

client/internal/debug/debug.go

@@ -647,7 +647,7 @@ const (
 	jsonKeyServiceEnv    = "service_env_vars"
 )
 
-var sensitiveEnvSubstrings = []string{"key", "token", "secret", "password", "credential"}
+var sensitiveEnvSubstrings = []string{"key", "token", "secret", "password", "credential", "pin", "pkcs11"}
 
 // addServiceParams reads the service.json file and adds a sanitized version to the bundle.
 // Non-NB_ env vars and vars with sensitive names are masked. Other NB_ values are anonymized.

client/internal/debug/debug_test.go

@@ -458,6 +458,9 @@ func TestIsSensitiveEnvVar(t *testing.T) {
 		{"NB_CLIENT_SECRET", true},
 		{"NB_PASSWORD", true},
 		{"NB_CREDENTIAL", true},
+		// The token PIN, and the token URI, which may carry the PIN as pin-value.
+		{"NB_TPM_PIN", true},
+		{"NB_CERT_PKCS11_URI", true},
 		{"NB_LOG_LEVEL", false},
 		{"NB_MANAGEMENT_URL", false},
 		{"NB_HOSTNAME", false},
@@ -848,6 +851,8 @@ func TestAddConfig_AllFieldsCovered(t *testing.T) {
 		"policy":               "non-config: in-memory MDM policy snapshot, surfaced via Config.Policy() / GetConfigResponse.MDMManagedFields",
 		"probing":              "non-config: marks a throwaway copy built to be diffed against; never set on a config anyone runs with",
 		"DebugBundleUploadURL": "sensitive: MDM-provided upload URL may carry credentials or query tokens; kept out of the shared bundle",
+		"CertPKCS11URI":        "deprecated and ignored; sensitive: the URI may carry the token PIN as pin-value",
+		"CertStoreDir":         "deprecated and ignored: the directory comes from NB_CERT_STORE_DIR",
 	}
 
 	mURL, _ := url.Parse("https://api.example.com:443")
@@ -891,6 +896,7 @@ func TestAddConfig_AllFieldsCovered(t *testing.T) {
 		ClientCertKeyPath:             "/tmp/key",
 		LazyConnection:                "on",
 		DebugBundleUploadURL:          "https://upload.example.test/bundle?token=secret",
+		CertPKCS11URI:                 "pkcs11:token=netbird?pin-value=pin-secret",
 		MTU:                           1280,
 		DisableIPv6:                   true,
 		SyncMessageVersion:            func(v int) *int { return &v }(1),
@@ -915,6 +921,10 @@ func TestAddConfig_AllFieldsCovered(t *testing.T) {
 			assert.NotContains(t, rendered, "DebugBundleUploadURL:", "MDM upload URL field must not be serialized into the debug bundle")
 			assert.NotContains(t, rendered, "token=secret", "MDM upload URL value must not leak into the debug bundle")
 
+			// A leftover CertPKCS11URI may carry the token PIN as pin-value.
+			assert.NotContains(t, rendered, "CertPKCS11URI", "PKCS#11 URI field must not be serialized into the debug bundle")
+			assert.NotContains(t, rendered, "pin-secret", "PKCS#11 PIN must not leak into the debug bundle")
+
 			val := reflect.ValueOf(cfg).Elem()
 			typ := val.Type()
 			var missing []string

client/internal/engine.go

@@ -37,6 +37,7 @@ import (
 	"github.com/netbirdio/netbird/client/iface/udpmux"
 	"github.com/netbirdio/netbird/client/iface/wgaddr"
 	"github.com/netbirdio/netbird/client/internal/acl"
+	"github.com/netbirdio/netbird/client/internal/certproof"
 	"github.com/netbirdio/netbird/client/internal/debug"
 	"github.com/netbirdio/netbird/client/internal/dns"
 	dnsconfig "github.com/netbirdio/netbird/client/internal/dns/config"
@@ -170,6 +171,8 @@ type EngineConfig struct {
 
 	MTU uint16
 
+	CertStore certproof.Config
+
 	// for debug bundle generation
 	ProfileConfig *profilemanager.Config
 
@@ -279,8 +282,26 @@ type Engine struct {
 
 	dnsServer dns.Server
 
-	// checks are the client-applied posture checks that need to be evaluated on the client
-	checks []*mgmProto.Checks
+	// checks are the client-applied posture checks that need to be evaluated on the client.
+	// Writers hold syncMsgMux and checksMu; readers hold either, see appliedChecks.
+	checks   []*mgmProto.Checks
+	checksMu sync.RWMutex
+	// pendingChecks are the newest received checks whose meta sync failed; the posture
+	// watcher retries them. Both are guarded by syncMsgMux, and
+	// hasPendingChecks lets the watcher skip the lock when nothing is pending.
+	pendingChecks    []*mgmProto.Checks
+	hasPendingChecks atomic.Bool
+	// infoTimeout overrides systemInfoTimeout when set.
+	infoTimeout time.Duration
+
+	// certProofs answers the certificate challenges in checks within a bounded time, and
+	// certState remembers what it last proved.
+	certProofs certproof.Collector
+	certState  certPostureState
+	// certWake wakes the posture watcher, which owns proof collection; certSendMu orders
+	// the meta syncs that carry proofs.
+	certWake   chan struct{}
+	certSendMu sync.Mutex
 
 	infoSource system.InfoSource
 
@@ -385,6 +406,7 @@ func NewEngine(
 		stateManager:       services.StateManager,
 		portForwardManager: portforward.NewManager(),
 		checks:             services.Checks,
+		certWake:           make(chan struct{}, 1),
 		probeStunTurn:      relay.NewStunTurnProbe(relay.DefaultCacheTTL),
 		jobExecutor:        jobexec.NewExecutor(),
 		clientMetrics:      services.ClientMetrics,
@@ -679,6 +701,12 @@ func (e *Engine) Start(netbirdConfig *mgmProto.NetbirdConfig, peerConfig *mgmPro
 		e.portForwardManager.Start(e.ctx, uint16(e.config.WgPort))
 	}()
 
+	e.shutdownWg.Add(1)
+	go func(ctx context.Context) {
+		defer e.shutdownWg.Done()
+		e.watchCertificatePosture(ctx)
+	}(e.ctx)
+
 	// Set the WireGuard interface for rosenpass after interface is up
 	if e.rpManager != nil {
 		e.rpManager.SetInterface(e.wgInterface)
@@ -1274,23 +1302,54 @@ func toFlowLoggerConfig(config *mgmProto.FlowConfig) (*nftypes.FlowConfig, error
 func (e *Engine) updateChecksIfNew(checks []*mgmProto.Checks) error {
 	// if checks are equal, we skip the update
 	if isChecksEqual(e.checks, checks) {
+		e.clearPendingChecks()
 		return nil
 	}
-	info, ok := e.infoSource.Refresh(e.ctx, systemInfoTimeout, checks, e.overlayAddresses()...)
-	if !ok {
-		// Gathering timed out; skip the meta sync this cycle rather than blocking the
-		// sync loop (and syncMsgMux) on a stuck system call. A later sync will retry.
-		return nil
+	if err := e.syncChecksMeta(checks); err != nil {
+		// The newest checks become the pending ones whatever failed, so the watcher
+		// never retries a set they superseded.
+		e.pendingChecks = checks
+		e.hasPendingChecks.Store(true)
+		if errors.Is(err, errSystemInfoTimeout) {
+			return nil
+		}
+		return err
 	}
-	e.applyInfoFlags(info)
-
-	if err := e.mgmClient.SyncMeta(info); err != nil {
-		return fmt.Errorf("could not sync meta: error %s", err)
-	}
-	e.checks = checks
+	e.setAppliedChecks(checks)
+	e.clearPendingChecks()
 	return nil
 }
 
+// appliedChecks returns the posture checks in effect, for callers that do not hold
+// syncMsgMux. The slice is replaced, never modified, so it may be read after return.
+func (e *Engine) appliedChecks() []*mgmProto.Checks {
+	e.checksMu.RLock()
+	defer e.checksMu.RUnlock()
+	return e.checks
+}
+
+// setAppliedChecks replaces the posture checks in effect. The caller holds syncMsgMux.
+func (e *Engine) setAppliedChecks(checks []*mgmProto.Checks) {
+	e.checksMu.Lock()
+	e.checks = checks
+	e.checksMu.Unlock()
+	e.wakeCertificatePosture()
+}
+
+// clearPendingChecks drops checks whose meta sync was still owed. The caller holds
+// syncMsgMux.
+func (e *Engine) clearPendingChecks() {
+	e.pendingChecks = nil
+	e.hasPendingChecks.Store(false)
+}
+
+func (e *Engine) infoGatherTimeout() time.Duration {
+	if e.infoTimeout > 0 {
+		return e.infoTimeout
+	}
+	return systemInfoTimeout
+}
+
 // applyInfoFlags sets the engine's config-derived feature flags on the gathered system info.
 func (e *Engine) applyInfoFlags(info *system.Info) {
 	info.SetFlags(
@@ -1317,6 +1376,7 @@ func (e *Engine) applyInfoFlags(info *system.Info) {
 func (e *Engine) currentSystemInfo(ctx context.Context) *system.Info {
 	info := e.infoSource.Current(ctx, e.overlayAddresses()...)
 	e.applyInfoFlags(info)
+	e.attachCertificateProofs(info, e.appliedChecks())
 	return info
 }
 
@@ -1332,6 +1392,7 @@ func (e *Engine) syncInfoFunc(refreshed *system.Info) func(ctx context.Context)
 		info := refreshed
 		refreshed = nil
 		e.applyInfoFlags(info)
+		e.attachCertificateProofs(info, e.appliedChecks())
 		return info
 	}
 }
@@ -1534,7 +1595,7 @@ func (e *Engine) receiveManagementEvents() {
 	e.shutdownWg.Add(1)
 	go func() {
 		defer e.shutdownWg.Done()
-		info, ok := e.infoSource.Refresh(e.ctx, systemInfoTimeout, e.checks, e.overlayAddresses()...)
+		info, ok := e.infoSource.Refresh(e.ctx, e.infoGatherTimeout(), e.appliedChecks(), e.overlayAddresses()...)
 		if !ok {
 			log.Warnf("posture checks not refreshed before the sync connect, sending the previous results")
 		}
@@ -2796,6 +2857,11 @@ func isChecksEqual(checks1, checks2 []*mgmProto.Checks) bool {
 			sortedFiles := slices.Clone(check.Files)
 			sort.Strings(sortedFiles)
 			normalized[i] = strings.Join(sortedFiles, "|")
+			if challenge := check.GetCertificateChallenge(); challenge != nil {
+				sortedCAs := slices.Clone(challenge.GetCaCertificates())
+				sort.Strings(sortedCAs)
+				normalized[i] += fmt.Sprintf("#%x|%s", challenge.GetNonce(), strings.Join(sortedCAs, "|"))
+			}
 		}
 
 		sort.Strings(normalized)

client/internal/engine_authsession.go

@@ -88,7 +88,7 @@ func (e *Engine) ExtendAuthSession(ctx context.Context, jwtToken string) (time.T
 		return time.Time{}, errors.New("management client is not initialised")
 	}
 
-	info, err := system.GetInfoWithChecks(ctx, e.checks)
+	info, err := system.GetInfoWithChecks(ctx, e.appliedChecks())
 	if err != nil {
 		log.Warnf("failed to collect system info for session extend: %v", err)
 		info = system.GetInfo(ctx)

client/internal/engine_certposture.go

@@ -0,0 +1,386 @@
+package internal
+
+import (
+	"context"
+	"crypto/sha256"
+	"encoding/binary"
+	"encoding/hex"
+	"errors"
+	"fmt"
+	"slices"
+	"strings"
+	"sync"
+	"time"
+
+	log "github.com/sirupsen/logrus"
+
+	"github.com/netbirdio/netbird/client/internal/certproof"
+	cProto "github.com/netbirdio/netbird/client/proto"
+	"github.com/netbirdio/netbird/client/system"
+	"github.com/netbirdio/netbird/shared/management/certposture"
+	mgmProto "github.com/netbirdio/netbird/shared/management/proto"
+)
+
+const (
+	// certContextPollInterval is how often the engine checks whether the user who can
+	// answer certificate challenges has changed, such as a login after an autostart.
+	certContextPollInterval = time.Minute
+	// certRetryInterval is how long after a collection that proved nothing it is tried
+	// again with the same user, for stores that come up late: a keychain unlocked after
+	// login, or a TPM resource manager started after the daemon.
+	certRetryInterval = 5 * time.Minute
+)
+
+// undeliveredContext never matches a real user context, so a collection whose meta sync
+// failed is seen as stale on the next poll.
+const undeliveredContext = "\x00undelivered"
+
+// errSystemInfoTimeout reports that gathering the system info for a meta sync timed out,
+// so the sync was skipped rather than holding syncMsgMux on a stuck system call.
+var errSystemInfoTimeout = errors.New("system info gathering timed out")
+
+// certPostureState remembers what the last certificate proof collection saw, so the
+// engine can collect again when it went stale and tell the user when proofs go missing.
+//
+// Collection runs only on the posture watcher, never on the sync path, because a token,
+// a TPM or a user's keychain prompt can take long; the sync path attaches the proofs the
+// last collection cached. delivered identifies the chains management last received, so
+// a collection that proves the same chains is not sent again: every meta sync makes
+// management recompute and push network maps to the peer and its neighbours.
+type certPostureState struct {
+	mu          sync.Mutex
+	attempted   bool
+	attemptedAt time.Time
+	userContext string
+	proven      bool
+	delivered   string
+	hasDelivery bool
+	cached      []certposture.Proof
+	cachedFor   string
+	// stuck is whether the collector was last seen with every slot held by a lost
+	// collection.
+	stuck bool
+}
+
+// record stores the outcome of a collection for the challenges identified by
+// challengesKey and reports whether it changed from proven to unproven or back. The first
+// collection only counts as a change when it proved nothing.
+func (s *certPostureState) record(challengesKey, userContext string, proofs []certposture.Proof, now time.Time) (changed bool) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	proven := len(proofs) > 0
+	changed = (s.attempted && s.proven != proven) || (!s.attempted && !proven)
+	s.attempted = true
+	s.attemptedAt = now
+	s.userContext = userContext
+	s.proven = proven
+	s.cached = proofs
+	s.cachedFor = challengesKey
+	return changed
+}
+
+// undelivered marks the last collection as not having reached management, so the next
+// poll collects and sends again instead of treating it as current.
+func (s *certPostureState) undelivered() {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	if s.attempted {
+		s.userContext = undeliveredContext
+	}
+	s.hasDelivery = false
+}
+
+// markDelivered records the chains of proofs as the ones management holds.
+func (s *certPostureState) markDelivered(proofs []certposture.Proof) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	s.delivered = provenChainsKey(proofs)
+	s.hasDelivery = true
+}
+
+// sameAsDelivered reports whether proofs prove exactly the chains management already
+// holds. Nonces and signatures are left out: they differ on every signing while the
+// result management stores, the verified chains, stays the same.
+func (s *certPostureState) sameAsDelivered(proofs []certposture.Proof) bool {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	return s.hasDelivery && s.delivered == provenChainsKey(proofs)
+}
+
+// needsCollection reports whether the cached proofs no longer reflect what the device
+// can prove for the challenges identified by challengesKey: they answer other
+// challenges, the user who can answer has changed, or nothing was proven and the retry
+// interval has passed.
+func (s *certPostureState) needsCollection(challengesKey, userContext string, now time.Time) bool {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	switch {
+	case !s.attempted || s.cachedFor != challengesKey:
+		return true
+	case userContext != s.userContext:
+		return true
+	default:
+		return !s.proven && now.Sub(s.attemptedAt) >= certRetryInterval
+	}
+}
+
+// setStuck records whether the collector is stuck and reports whether it just became so.
+func (s *certPostureState) setStuck(stuck bool) bool {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	became := stuck && !s.stuck
+	s.stuck = stuck
+	return became
+}
+
+// cachedProofs returns the cached proofs management still accepts for the challenges in
+// checks: those whose nonce is current, or from an adjacent window, for one of them. Proofs
+// signed for the previous nonce bridge the time until the watcher has signed the new one.
+func (s *certPostureState) cachedProofs(checks []*mgmProto.Checks) []certposture.Proof {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	var nonces [][]byte
+	for _, check := range checks {
+		if challenge := check.GetCertificateChallenge(); challenge != nil {
+			nonces = append(nonces, challenge.GetNonce())
+		}
+	}
+	var accepted []certposture.Proof
+	for _, proof := range s.cached {
+		if slices.ContainsFunc(nonces, func(nonce []byte) bool {
+			return certposture.NonceAcceptedAlongside(proof.Nonce, nonce)
+		}) {
+			accepted = append(accepted, proof)
+		}
+	}
+	return accepted
+}
+
+// provenChainsKey identifies the set of chains in proofs, independent of their order.
+func provenChainsKey(proofs []certposture.Proof) string {
+	keys := make([]string, 0, len(proofs))
+	for _, proof := range proofs {
+		h := sha256.New()
+		for _, cert := range proof.Chain {
+			writeLengthPrefixed(h, cert)
+		}
+		keys = append(keys, hex.EncodeToString(h.Sum(nil)))
+	}
+	slices.Sort(keys)
+	return strings.Join(keys, ",")
+}
+
+// challengesKey identifies the certificate challenges in checks: their nonces and CAs.
+func challengesKey(checks []*mgmProto.Checks) string {
+	h := sha256.New()
+	for _, check := range checks {
+		challenge := check.GetCertificateChallenge()
+		if challenge == nil {
+			continue
+		}
+		writeLengthPrefixed(h, challenge.GetNonce())
+		for _, ca := range challenge.GetCaCertificates() {
+			writeLengthPrefixed(h, []byte(ca))
+		}
+		writeLengthPrefixed(h, nil)
+	}
+	return hex.EncodeToString(h.Sum(nil))
+}
+
+func writeLengthPrefixed(h interface{ Write([]byte) (int, error) }, data []byte) {
+	var size [8]byte
+	binary.BigEndian.PutUint64(size[:], uint64(len(data)))
+	_, _ = h.Write(size[:])
+	_, _ = h.Write(data)
+}
+
+// attachCertificateProofs attaches the cached proofs management accepts for the
+// challenges in checks, without collecting: the caller is on the sync path. The info is
+// about to be sent on the sync stream, which resends it on every reconnect, so its proofs
+// count as delivered. The watcher is woken to collect when the cache is out of date.
+func (e *Engine) attachCertificateProofs(info *system.Info, checks []*mgmProto.Checks) {
+	if !certproof.HasChallenges(checks) {
+		info.CertificateProofs = nil
+		return
+	}
+	info.CertificateProofs = e.certState.cachedProofs(checks)
+	e.certState.markDelivered(info.CertificateProofs)
+	e.wakeCertificatePosture()
+}
+
+// wakeCertificatePosture asks the posture watcher to check its proofs now rather than at
+// its next tick. It never blocks.
+func (e *Engine) wakeCertificatePosture() {
+	if e.certWake == nil {
+		return
+	}
+	select {
+	case e.certWake <- struct{}{}:
+	default:
+	}
+}
+
+// publishCertificatePostureEvent tells the user when the device stops proving any
+// certificate, which management treats as failing every certificate posture check, and
+// when it proves one again. Without it the loss of access would have no visible cause.
+func (e *Engine) publishCertificatePostureEvent(proven bool) {
+	if e.statusRecorder == nil {
+		return
+	}
+	if proven {
+		e.statusRecorder.PublishEvent(cProto.SystemEvent_INFO, cProto.SystemEvent_SYSTEM,
+			"certificate posture: a certificate is proven again",
+			"A certificate required by your organization's device policy is available again.", nil)
+		return
+	}
+	e.statusRecorder.PublishEvent(cProto.SystemEvent_WARNING, cProto.SystemEvent_SYSTEM,
+		"certificate posture: no certificate could be proven",
+		"NetBird could not use a certificate required by your organization's device policy. "+
+			"Access to some resources may be blocked until one is available.", nil)
+}
+
+// publishCertificateStoreStuckEvent tells the user that reading the certificate store
+// stopped answering. Collection resumes only once a stuck call returns, which a store
+// that hangs for good never does short of a restart.
+func (e *Engine) publishCertificateStoreStuckEvent() {
+	if e.statusRecorder == nil {
+		return
+	}
+	e.statusRecorder.PublishEvent(cProto.SystemEvent_WARNING, cProto.SystemEvent_SYSTEM,
+		"certificate posture: the certificate store stopped responding",
+		"NetBird cannot read the certificates required by your organization's device policy because "+
+			"the certificate store stopped responding. If this persists, restart the NetBird service.", nil)
+}
+
+// watchCertificatePosture owns certificate proof collection until ctx is done. It
+// collects when woken by new checks or a sync, and on every tick when the cached proofs
+// went stale, such as after a login following an autostart.
+func (e *Engine) watchCertificatePosture(ctx context.Context) {
+	ticker := time.NewTicker(certContextPollInterval)
+	defer ticker.Stop()
+	for {
+		// A pending update that times out again stays pending; the applied checks'
+		// proofs are still refreshed so a lost certificate is reported meanwhile.
+		if err := e.retryPendingChecks(); err != nil && !errors.Is(err, errSystemInfoTimeout) {
+			log.Warnf("failed to sync posture checks that timed out before: %v", err)
+		}
+		if err := e.refreshCertificateProofs(); err != nil && !errors.Is(err, errSystemInfoTimeout) {
+			log.Warnf("failed to refresh certificate posture proofs: %v", err)
+		}
+
+		select {
+		case <-ctx.Done():
+			return
+		case <-ticker.C:
+		case <-e.certWake:
+		}
+	}
+}
+
+// retryPendingChecks sends the meta sync for checks whose earlier sync timed out
+// gathering the system info, and applies them once it succeeds. Without it they would
+// wait for the next network map that changes the checks.
+func (e *Engine) retryPendingChecks() error {
+	if !e.hasPendingChecks.Load() {
+		return nil
+	}
+
+	e.syncMsgMux.Lock()
+	defer e.syncMsgMux.Unlock()
+	if e.stopping() || !e.hasPendingChecks.Load() {
+		return nil
+	}
+	checks := e.pendingChecks
+	log.Debugf("posture checks: retrying the meta sync that timed out")
+	if err := e.syncChecksMeta(checks); err != nil {
+		return err
+	}
+	e.setAppliedChecks(checks)
+	e.clearPendingChecks()
+	return nil
+}
+
+// refreshCertificateProofs collects the proofs for the applied checks when the cached
+// ones are out of date, and sends them to management when they prove other chains than
+// it holds. It holds no engine lock while collecting.
+func (e *Engine) refreshCertificateProofs() error {
+	checks := e.appliedChecks()
+	if e.stopping() || !certproof.HasChallenges(checks) {
+		return nil
+	}
+	key := challengesKey(checks)
+	userContext := certproof.UserContext(e.config.CertStore)
+	if !e.certState.needsCollection(key, userContext, time.Now()) {
+		return nil
+	}
+
+	log.Debugf("certificate posture: collecting proofs")
+	peerKey := e.config.WgPrivateKey.PublicKey()
+	proofs := e.certProofs.Collect(e.ctx, checks, peerKey[:], e.config.CertStore)
+	// A collection cut short by the engine stopping proved nothing about the device, and
+	// must not be reported to the user as a lost certificate.
+	if e.stopping() {
+		return nil
+	}
+	if e.certState.record(key, userContext, proofs, time.Now()) {
+		e.publishCertificatePostureEvent(len(proofs) > 0)
+	}
+	if e.certState.setStuck(e.certProofs.Stuck()) {
+		e.publishCertificateStoreStuckEvent()
+	}
+
+	if e.certState.sameAsDelivered(proofs) {
+		log.Debugf("certificate posture: proofs unchanged since the last meta sync, not sending")
+		return nil
+	}
+	info, ok := e.infoSource.Refresh(e.ctx, e.infoGatherTimeout(), checks, e.overlayAddresses()...)
+	if !ok {
+		e.certState.undelivered()
+		return errSystemInfoTimeout
+	}
+	e.applyInfoFlags(info)
+	return e.sendMetaWithProofs(info, checks)
+}
+
+// stopping reports whether the engine is shutting down.
+func (e *Engine) stopping() bool {
+	select {
+	case <-e.ctx.Done():
+		return true
+	default:
+		return false
+	}
+}
+
+// syncChecksMeta gathers the system info that checks evaluate and sends it to management
+// with the cached certificate proofs. The caller holds syncMsgMux.
+func (e *Engine) syncChecksMeta(checks []*mgmProto.Checks) error {
+	info, ok := e.infoSource.Refresh(e.ctx, e.infoGatherTimeout(), checks, e.overlayAddresses()...)
+	if !ok {
+		// Gathering timed out; skip the meta sync this cycle rather than blocking the
+		// sync loop (and syncMsgMux) on a stuck system call. The posture watcher retries.
+		e.certState.undelivered()
+		return errSystemInfoTimeout
+	}
+	e.applyInfoFlags(info)
+	return e.sendMetaWithProofs(info, checks)
+}
+
+// sendMetaWithProofs sends info to management with the cached proofs for checks. Sends
+// are serialized and read the cache under that lock, so a send from the sync path cannot
+// overwrite newer proofs the watcher sent in the meantime.
+func (e *Engine) sendMetaWithProofs(info *system.Info, checks []*mgmProto.Checks) error {
+	e.certSendMu.Lock()
+	defer e.certSendMu.Unlock()
+
+	info.CertificateProofs = e.certState.cachedProofs(checks)
+	if err := e.mgmClient.SyncMeta(info); err != nil {
+		e.certState.undelivered()
+		return fmt.Errorf("sync meta: %w", err)
+	}
+	e.certState.markDelivered(info.CertificateProofs)
+	return nil
+}

client/internal/engine_certposture_test.go

@@ -0,0 +1,224 @@
+//go:build !windows && !darwin
+
+package internal
+
+import (
+	"context"
+	"errors"
+	"os"
+	"path/filepath"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
+
+	"github.com/netbirdio/netbird/client/internal/certproof"
+	"github.com/netbirdio/netbird/client/internal/peer"
+	cProto "github.com/netbirdio/netbird/client/proto"
+	"github.com/netbirdio/netbird/client/system"
+	"github.com/netbirdio/netbird/shared/management/certposture"
+	"github.com/netbirdio/netbird/shared/management/certposture/certtest"
+	mgmt "github.com/netbirdio/netbird/shared/management/client"
+	mgmProto "github.com/netbirdio/netbird/shared/management/proto"
+)
+
+func TestCertPostureState_Record(t *testing.T) {
+	now := time.Now()
+	one := []certposture.Proof{{}}
+
+	var proven certPostureState
+	assert.False(t, proven.record("k", "", one, now), "a first collection that proves something is not news")
+
+	var unproven certPostureState
+	assert.True(t, unproven.record("k", "", nil, now), "a first collection that proves nothing is reported")
+	assert.False(t, unproven.record("k", "", nil, now), "the same outcome again is not reported twice")
+	assert.True(t, unproven.record("k", "", one, now), "proving again is reported")
+	assert.True(t, unproven.record("k", "", nil, now), "losing the proof again is reported")
+}
+
+// TestCertPostureState_ReportsStuckOnce: a store that stopped answering is reported to
+// the user once when it wedges the collector, not on every collection that follows, and
+// again if it wedges anew after recovering.
+func TestCertPostureState_ReportsStuckOnce(t *testing.T) {
+	var s certPostureState
+	assert.True(t, s.setStuck(true), "becoming stuck is reported")
+	assert.False(t, s.setStuck(true), "staying stuck is not reported again")
+	assert.False(t, s.setStuck(false), "recovering is not a stuck report")
+	assert.True(t, s.setStuck(true), "getting stuck again is reported again")
+}
+
+func TestCertPostureState_NeedsCollection(t *testing.T) {
+	now := time.Now()
+	one := []certposture.Proof{{}}
+	var s certPostureState
+
+	assert.True(t, s.needsCollection("k", "", now), "nothing collected yet")
+
+	s.record("k", "501:alice", one, now)
+	assert.False(t, s.needsCollection("k", "501:alice", now.Add(time.Hour)), "a proven collection for the same challenges and user stays fresh")
+	assert.True(t, s.needsCollection("other", "501:alice", now), "new challenges, such as a rotated nonce, are answered again")
+	assert.True(t, s.needsCollection("k", "", now), "the user logging out changes what can be proven")
+	assert.True(t, s.needsCollection("k", "502:bob", now), "another owner session changes what can be proven")
+
+	s.record("k", "", nil, now)
+	assert.False(t, s.needsCollection("k", "", now.Add(certRetryInterval-time.Second)), "an unproven collection is not retried before the interval")
+	assert.True(t, s.needsCollection("k", "", now.Add(certRetryInterval)), "an unproven collection is retried once the interval passed")
+
+	s.record("k", "501:alice", one, now)
+	s.undelivered()
+	assert.True(t, s.needsCollection("k", "501:alice", now), "a proof management never received is collected and sent again")
+}
+
+// TestCertPostureState_CachedProofsAcrossNonceRotation: management rotates the nonce every
+// window and accepts the previous one, so a cached proof keeps being attached across one
+// rotation, while the watcher signs the new nonce, and not across two.
+func TestCertPostureState_CachedProofsAcrossNonceRotation(t *testing.T) {
+	challenger := certposture.NewChallenger([]byte("secret"))
+	peerKey := []byte("peer-public-key-aaaaaaaaaaaaaaaa")
+	now := time.Now()
+	checksAt := func(at time.Time) []*mgmProto.Checks {
+		return []*mgmProto.Checks{{CertificateChallenge: &mgmProto.CertificateChallenge{Nonce: challenger.Nonce(peerKey, at)}}}
+	}
+
+	var s certPostureState
+	current := checksAt(now)
+	s.record(challengesKey(current), "", []certposture.Proof{{Nonce: current[0].CertificateChallenge.Nonce}}, now)
+
+	assert.Len(t, s.cachedProofs(current), 1, "the proof for the current nonce is attached")
+	assert.Len(t, s.cachedProofs(checksAt(now.Add(certposture.Window))), 1, "after one rotation the proof is still accepted and attached")
+	assert.Empty(t, s.cachedProofs(checksAt(now.Add(2*certposture.Window))), "after two rotations management would reject it")
+	assert.Empty(t, s.cachedProofs([]*mgmProto.Checks{{Files: []string{"/bin/agent"}}}), "no challenge, no proof")
+}
+
+// newCertPostureEngine is an engine whose certificate store is a real PEM directory and
+// whose management client records the proofs of every meta sync.
+func newCertPostureEngine(t *testing.T, recorder *peer.Status, syncMeta func(*system.Info) error) (*Engine, string, *certtest.CA) {
+	t.Helper()
+	// The store refuses a group-writable directory, which t.TempDir yields under a
+	// user-private-group umask.
+	dir := t.TempDir()
+	require.NoError(t, os.Chmod(dir, 0o700))
+	t.Setenv(certproof.StoreDirEnv, dir)
+	key, err := wgtypes.GeneratePrivateKey()
+	require.NoError(t, err)
+
+	e := &Engine{
+		ctx:            context.Background(),
+		syncMsgMux:     &sync.Mutex{},
+		config:         &EngineConfig{WgPrivateKey: key},
+		statusRecorder: recorder,
+		mgmClient:      &mgmt.MockClient{SyncMetaFunc: syncMeta},
+	}
+	ca := certtest.NewCA(t, "corp-root")
+	peerKey := key.PublicKey()
+	nonce := certposture.NewChallenger([]byte("secret")).Nonce(peerKey[:], time.Now())
+	e.checks = []*mgmProto.Checks{{CertificateChallenge: &mgmProto.CertificateChallenge{Nonce: nonce, CaCertificates: []string{ca.PEM}}}}
+	return e, dir, ca
+}
+
+func writeDeviceCert(t *testing.T, dir string, ca *certtest.CA) string {
+	t.Helper()
+	deviceKey := certtest.ECDSAKey(t)
+	pem := certtest.CertPEM(ca.Issue(t, deviceKey, "device")) + certtest.KeyPEM(t, deviceKey)
+	require.NoError(t, os.WriteFile(filepath.Join(dir, "device.pem"), []byte(pem), 0o600))
+	return pem
+}
+
+// TestEngine_AttachCertificateProofsNeverCollects: the sync path only attaches what the
+// watcher collected, so a certificate that appears in the store is not proven on the sync
+// path until the watcher ran.
+func TestEngine_AttachCertificateProofsNeverCollects(t *testing.T) {
+	e, dir, ca := newCertPostureEngine(t, nil, func(*system.Info) error { return nil })
+	writeDeviceCert(t, dir, ca)
+
+	info := &system.Info{}
+	e.attachCertificateProofs(info, e.checks)
+	assert.Empty(t, info.CertificateProofs, "the sync path does not collect")
+
+	require.NoError(t, e.refreshCertificateProofs())
+	e.attachCertificateProofs(info, e.checks)
+	assert.Len(t, info.CertificateProofs, 1, "the sync path attaches what the watcher collected")
+}
+
+// TestEngine_RefreshReportsLostAndRegainedProofs: a device that cannot prove its
+// certificate gets one warning, and one notice when it can again.
+func TestEngine_RefreshReportsLostAndRegainedProofs(t *testing.T) {
+	recorder := peer.NewRecorder("")
+	e, dir, ca := newCertPostureEngine(t, recorder, func(*system.Info) error { return nil })
+	warnings := func() int {
+		n := 0
+		for _, ev := range recorder.GetEventHistory() {
+			if ev.Severity == cProto.SystemEvent_WARNING && ev.Category == cProto.SystemEvent_SYSTEM {
+				n++
+			}
+		}
+		return n
+	}
+	expire := func() { e.certState.attemptedAt = time.Now().Add(-certRetryInterval) }
+
+	require.NoError(t, e.refreshCertificateProofs())
+	assert.Equal(t, 1, warnings(), "the user is told the device proves no certificate")
+
+	expire()
+	require.NoError(t, e.refreshCertificateProofs())
+	assert.Equal(t, 1, warnings(), "an unchanged outcome is not reported again")
+
+	writeDeviceCert(t, dir, ca)
+	expire()
+	require.NoError(t, e.refreshCertificateProofs())
+	events := recorder.GetEventHistory()
+	require.NotEmpty(t, events)
+	assert.Equal(t, cProto.SystemEvent_INFO, events[len(events)-1].Severity, "regaining the proof is reported as good news")
+}
+
+// TestEngine_RefreshSendsOnlyChangedProofs: every meta sync makes management recompute
+// and push network maps, so a collection that proves the same chains as before must not
+// send, while a new certificate or a failed delivery must.
+func TestEngine_RefreshSendsOnlyChangedProofs(t *testing.T) {
+	var sent [][]certposture.Proof
+	failSync := false
+	e, dir, ca := newCertPostureEngine(t, nil, func(info *system.Info) error {
+		if failSync {
+			return errors.New("management unavailable")
+		}
+		sent = append(sent, info.CertificateProofs)
+		return nil
+	})
+	// expire makes the last collection due for the unproven retry.
+	expire := func() { e.certState.attemptedAt = time.Now().Add(-certRetryInterval) }
+	// switchUser makes the last collection stale as if another user signed in.
+	switchUser := func() { e.certState.userContext = "someone else" }
+
+	require.NoError(t, e.refreshCertificateProofs())
+	require.Len(t, sent, 1, "the first collection is sent")
+	assert.Empty(t, sent[0], "nothing to prove yet")
+
+	expire()
+	require.NoError(t, e.refreshCertificateProofs())
+	assert.Len(t, sent, 1, "still proving nothing is not sent again")
+
+	pem := writeDeviceCert(t, dir, ca)
+	expire()
+	require.NoError(t, e.refreshCertificateProofs())
+	require.Len(t, sent, 2, "a newly proven certificate is sent")
+	assert.Len(t, sent[1], 1, "the new proof is attached")
+
+	switchUser()
+	require.NoError(t, e.refreshCertificateProofs())
+	assert.Len(t, sent, 2, "the same chain proven again, with a fresh signature, is not sent")
+
+	failSync = true
+	require.NoError(t, os.Remove(filepath.Join(dir, "device.pem")))
+	switchUser()
+	require.Error(t, e.refreshCertificateProofs())
+	failSync = false
+	require.NoError(t, os.WriteFile(filepath.Join(dir, "device.pem"), []byte(pem), 0o600))
+
+	// Management missed the empty proof set, so it still holds the chain the device proves
+	// again now; after a failed delivery the state is unknown, and it is sent regardless.
+	require.NoError(t, e.refreshCertificateProofs())
+	assert.Len(t, sent, 3, "after a failed delivery the next collection is sent")
+}

client/internal/engine_test.go

@@ -359,6 +359,111 @@ func TestEngine_UpdateChecksIfNewRetriesAfterFailedSyncMeta(t *testing.T) {
 	assert.Equal(t, 2, syncMetaCalls)
 }
 
+// TestEngine_FailedUpdateReplacesOlderPendingChecks: checks A time out and stay pending,
+// then checks B fail to sync for another reason. B must replace A as pending, or the
+// watcher would later apply the superseded A.
+func TestEngine_FailedUpdateReplacesOlderPendingChecks(t *testing.T) {
+	key, err := wgtypes.GeneratePrivateKey()
+	require.NoError(t, err)
+	ctx, cancel := context.WithCancel(CtxInitState(context.Background()))
+	defer cancel()
+
+	relayMgr := relayClient.NewManager(ctx, nil, key.PublicKey().String(), iface.DefaultMTU)
+	engine := NewEngine(ctx, cancel, &EngineConfig{
+		WgIfaceName:  "utun107",
+		WgAddr:       wgaddr.MustParseWGAddress("100.64.0.1/24"),
+		WgPrivateKey: key,
+		WgPort:       33102,
+		MTU:          iface.DefaultMTU,
+	}, EngineServices{
+		SignalClient:   &signal.MockClient{},
+		MgmClient:      &mgmt.MockClient{SyncMetaFunc: func(*system.Info) error { return errors.New("management unavailable") }},
+		RelayManager:   relayMgr,
+		StatusRecorder: peer.NewRecorder("https://mgm"),
+	}, MobileDependency{})
+
+	checksA := []*mgmtProto.Checks{{Files: []string{"/checks/a"}}}
+	checksB := []*mgmtProto.Checks{{Files: []string{"/checks/b"}}}
+
+	engine.infoTimeout = time.Nanosecond
+	require.NoError(t, engine.updateChecksIfNew(checksA))
+	require.Equal(t, checksA, engine.pendingChecks, "precondition: the timed-out checks are pending")
+
+	// Let the abandoned gathering finish so B gets as far as the meta sync.
+	engine.infoTimeout = 0
+	require.Eventually(t, func() bool {
+		_, ok := engine.infoSource.Refresh(ctx, 10*time.Second, nil)
+		return ok
+	}, 10*time.Second, 10*time.Millisecond)
+
+	require.Error(t, engine.updateChecksIfNew(checksB), "the meta sync of B fails")
+	assert.Equal(t, checksB, engine.pendingChecks, "the newest checks replace the older pending ones")
+}
+
+// TestEngine_PendingChecksRetriedAfterInfoTimeout covers a check update whose system info
+// gathering times out: the update is kept pending rather than dropped, and the posture
+// watcher's retry sends it and applies the checks, without waiting for management to
+// send different checks.
+func TestEngine_PendingChecksRetriedAfterInfoTimeout(t *testing.T) {
+	key, err := wgtypes.GeneratePrivateKey()
+	require.NoError(t, err)
+
+	exe, err := os.Executable()
+	require.NoError(t, err)
+
+	ctx, cancel := context.WithCancel(CtxInitState(context.Background()))
+	defer cancel()
+
+	var synced []*system.Info
+	mgmClient := &mgmt.MockClient{
+		SyncMetaFunc: func(info *system.Info) error {
+			synced = append(synced, info)
+			return nil
+		},
+	}
+
+	relayMgr := relayClient.NewManager(ctx, nil, key.PublicKey().String(), iface.DefaultMTU)
+	engine := NewEngine(ctx, cancel, &EngineConfig{
+		WgIfaceName:  "utun106",
+		WgAddr:       wgaddr.MustParseWGAddress("100.64.0.1/24"),
+		WgPrivateKey: key,
+		WgPort:       33101,
+		MTU:          iface.DefaultMTU,
+	}, EngineServices{
+		SignalClient:   &signal.MockClient{},
+		MgmClient:      mgmClient,
+		RelayManager:   relayMgr,
+		StatusRecorder: peer.NewRecorder("https://mgm"),
+	}, MobileDependency{})
+
+	checks := []*mgmtProto.Checks{{Files: []string{exe}}}
+
+	// Gathering cannot finish within a nanosecond, so the sync times out.
+	engine.infoTimeout = time.Nanosecond
+	require.NoError(t, engine.updateChecksIfNew(checks), "a timed-out gathering is not an error")
+	assert.Empty(t, synced, "nothing is sent when gathering timed out")
+	assert.Nil(t, engine.checks, "timed-out checks are not applied")
+	require.True(t, engine.hasPendingChecks.Load(), "timed-out checks are kept pending")
+
+	// The timed-out gathering keeps running in the background, and no new one starts
+	// until it exits, so the retry may time out again before it goes through.
+	engine.infoTimeout = 0
+	var retryErr error
+	require.Eventually(t, func() bool {
+		retryErr = engine.retryPendingChecks()
+		return retryErr == nil || !errors.Is(retryErr, errSystemInfoTimeout)
+	}, 10*time.Second, 10*time.Millisecond, "the pending checks are sent once the earlier gathering exits")
+	require.NoError(t, retryErr)
+	require.Len(t, synced, 1, "the retry sends the meta sync")
+	assert.Len(t, synced[0].Files, 1, "the retry evaluates the pending checks")
+	assert.Equal(t, checks, engine.checks, "the retried checks are applied")
+	assert.False(t, engine.hasPendingChecks.Load(), "nothing is pending after the retry")
+
+	require.NoError(t, engine.retryPendingChecks())
+	require.NoError(t, engine.updateChecksIfNew(checks))
+	assert.Len(t, synced, 1, "applied checks are not sent again")
+}
+
 func TestEngine_UpdateNetworkMap(t *testing.T) {
 	// test setup
 	key, err := wgtypes.GeneratePrivateKey()
@@ -1179,6 +1284,32 @@ func Test_CheckFilesEqual(t *testing.T) {
 			},
 			expectedBool: true,
 		},
+		{
+			name: "Same files with rotated certificate challenge nonce should return false",
+			inputChecks1: []*mgmtProto.Checks{
+				{
+					Files:                []string{"testfile1"},
+					CertificateChallenge: &mgmtProto.CertificateChallenge{Nonce: []byte{1}, CaCertificates: []string{"ca-a"}},
+				},
+			},
+			inputChecks2: []*mgmtProto.Checks{
+				{
+					Files:                []string{"testfile1"},
+					CertificateChallenge: &mgmtProto.CertificateChallenge{Nonce: []byte{2}, CaCertificates: []string{"ca-a"}},
+				},
+			},
+			expectedBool: false,
+		},
+		{
+			name: "Same certificate challenge with CA certificates in different order should return true",
+			inputChecks1: []*mgmtProto.Checks{
+				{CertificateChallenge: &mgmtProto.CertificateChallenge{Nonce: []byte{1}, CaCertificates: []string{"ca-a", "ca-b"}}},
+			},
+			inputChecks2: []*mgmtProto.Checks{
+				{CertificateChallenge: &mgmtProto.CertificateChallenge{Nonce: []byte{1}, CaCertificates: []string{"ca-b", "ca-a"}}},
+			},
+			expectedBool: true,
+		},
 	}
 	for _, testCase := range testCases {
 		t.Run(testCase.name, func(t *testing.T) {

client/internal/pkcs11/driver_linux.go

@@ -0,0 +1,372 @@
+//go:build pkcs11 && linux && !android && (amd64 || arm64)
+
+package pkcs11
+
+import (
+	"errors"
+	"fmt"
+	"runtime"
+	"strings"
+	"unsafe"
+
+	"github.com/ebitengine/purego"
+	log "github.com/sirupsen/logrus"
+)
+
+// ulong is CK_ULONG, an unsigned long, which is pointer-sized on the 64-bit Linux ABIs
+// this file builds for. The struct layouts below assume that width and natural alignment.
+type ulong = uintptr
+
+const (
+	unavailableInformation = ^ulong(0)
+
+	flagOSLockingOK   = 0x2
+	flagRWSession     = 0x2
+	flagSerialSession = 0x4
+	userTypeUser      = 0x1
+
+	findBatch = 32
+
+	// Upper bounds on sizes a module reports, so a buggy module cannot make the daemon
+	// allocate without limit. Certificates and keys are far smaller; an RSA-16384
+	// signature is 2 KiB.
+	maxAttributeLen = 1 << 20
+	maxSignatureLen = 1 << 16
+)
+
+type version struct {
+	major byte
+	minor byte
+}
+
+type attribute struct {
+	typ   ulong
+	value unsafe.Pointer
+	len   ulong
+}
+
+type mechanism struct {
+	typ       ulong
+	parameter unsafe.Pointer
+	len       ulong
+}
+
+type pssParams struct {
+	hashAlg ulong
+	mgf     ulong
+	saltLen ulong
+}
+
+type tokenInfo struct {
+	label          [32]byte
+	manufacturerID [32]byte
+	model          [16]byte
+	serialNumber   [16]byte
+	flags          ulong
+	counters       [10]ulong
+	hardware       version
+	firmware       version
+	utcTime        [16]byte
+}
+
+type initializeArgs struct {
+	createMutex  uintptr
+	destroyMutex uintptr
+	lockMutex    uintptr
+	unlockMutex  uintptr
+	flags        ulong
+	reserved     unsafe.Pointer
+}
+
+// functionList mirrors CK_FUNCTION_LIST: a CK_VERSION padded to pointer alignment, then
+// the PKCS#11 v2.40 entry points in specification order.
+type functionList struct {
+	version version
+	_       [6]byte
+	fn      [68]uintptr
+}
+
+const (
+	fnInitialize        = 0
+	fnGetSlotList       = 4
+	fnGetTokenInfo      = 6
+	fnOpenSession       = 12
+	fnCloseSession      = 13
+	fnLogin             = 18
+	fnLogout            = 19
+	fnCreateObject      = 20
+	fnGetAttributeValue = 24
+	fnFindObjectsInit   = 26
+	fnFindObjects       = 27
+	fnFindObjectsFinal  = 28
+	fnSignInit          = 42
+	fnSign              = 43
+)
+
+// module holds the entry points of one loaded library, bound straight from its
+// CK_FUNCTION_LIST.
+type module struct {
+	cInitialize        func(args *initializeArgs) ulong
+	cGetSlotList       func(tokenPresent byte, slots *ulong, count *ulong) ulong
+	cGetTokenInfo      func(slot ulong, info *tokenInfo) ulong
+	cOpenSession       func(slot ulong, flags ulong, application unsafe.Pointer, notify uintptr, session *ulong) ulong
+	cCloseSession      func(session ulong) ulong
+	cLogin             func(session ulong, userType ulong, pin *byte, pinLen ulong) ulong
+	cLogout            func(session ulong) ulong
+	cCreateObject      func(session ulong, template *attribute, count ulong, object *ulong) ulong
+	cGetAttributeValue func(session ulong, object ulong, template *attribute, count ulong) ulong
+	cFindObjectsInit   func(session ulong, template *attribute, count ulong) ulong
+	cFindObjects       func(session ulong, objects *ulong, max ulong, count *ulong) ulong
+	cFindObjectsFinal  func(session ulong) ulong
+	cSignInit          func(session ulong, mech *mechanism, key ulong) ulong
+	cSign              func(session ulong, data *byte, dataLen ulong, signature *byte, signatureLen *ulong) ulong
+}
+
+// Supported reports whether this build can load PKCS#11 modules.
+func Supported() bool {
+	return true
+}
+
+func load(path string) (_ driver, err error) {
+	lib, err := purego.Dlopen(path, purego.RTLD_NOW|purego.RTLD_LOCAL)
+	if err != nil {
+		return nil, fmt.Errorf("open PKCS#11 module %s: %w", path, err)
+	}
+	// A module that loads but cannot be used is not cached, so each later attempt opens
+	// it again; close it here or every attempt keeps another reference to the library.
+	defer func() {
+		if err != nil {
+			if closeErr := purego.Dlclose(lib); closeErr != nil {
+				log.Debugf("failed closing PKCS#11 module %s: %v", path, closeErr)
+			}
+		}
+	}()
+	symbol, err := purego.Dlsym(lib, "C_GetFunctionList")
+	if err != nil {
+		return nil, fmt.Errorf("%s is not a PKCS#11 module: %w", path, err)
+	}
+	var getFunctionList func(list **functionList) ulong
+	purego.RegisterFunc(&getFunctionList, symbol)
+	var list *functionList
+	if rv := getFunctionList(&list); rv != rvOK || list == nil {
+		return nil, Error{Op: "C_GetFunctionList", Code: uint(rv)}
+	}
+
+	m := &module{}
+	for _, entry := range []struct {
+		fn    any
+		index int
+	}{
+		{&m.cInitialize, fnInitialize},
+		{&m.cGetSlotList, fnGetSlotList},
+		{&m.cGetTokenInfo, fnGetTokenInfo},
+		{&m.cOpenSession, fnOpenSession},
+		{&m.cCloseSession, fnCloseSession},
+		{&m.cLogin, fnLogin},
+		{&m.cLogout, fnLogout},
+		{&m.cCreateObject, fnCreateObject},
+		{&m.cGetAttributeValue, fnGetAttributeValue},
+		{&m.cFindObjectsInit, fnFindObjectsInit},
+		{&m.cFindObjects, fnFindObjects},
+		{&m.cFindObjectsFinal, fnFindObjectsFinal},
+		{&m.cSignInit, fnSignInit},
+		{&m.cSign, fnSign},
+	} {
+		if list.fn[entry.index] == 0 {
+			return nil, fmt.Errorf("%s lacks PKCS#11 entry point %d", path, entry.index)
+		}
+		purego.RegisterFunc(entry.fn, list.fn[entry.index])
+	}
+
+	args := &initializeArgs{flags: flagOSLockingOK}
+	if rv := m.cInitialize(args); rv != rvOK && rv != rvAlreadyInitialized {
+		return nil, Error{Op: "C_Initialize", Code: uint(rv)}
+	}
+	return m, nil
+}
+
+func (m *module) tokens() ([]Token, error) {
+	var count ulong
+	if rv := m.cGetSlotList(1, nil, &count); rv != rvOK {
+		return nil, Error{Op: "C_GetSlotList", Code: uint(rv)}
+	}
+	if count == 0 {
+		return nil, nil
+	}
+	slots := make([]ulong, count)
+	if rv := m.cGetSlotList(1, &slots[0], &count); rv != rvOK {
+		return nil, Error{Op: "C_GetSlotList", Code: uint(rv)}
+	}
+
+	tokens := make([]Token, 0, count)
+	for _, slot := range slots[:count] {
+		var info tokenInfo
+		if rv := m.cGetTokenInfo(slot, &info); rv != rvOK {
+			continue
+		}
+		tokens = append(tokens, Token{Slot: uint(slot), Label: strings.TrimRight(string(info.label[:]), " \x00")})
+	}
+	return tokens, nil
+}
+
+func (m *module) openSession(slot uint, readWrite bool) (uint, error) {
+	flags := ulong(flagSerialSession)
+	if readWrite {
+		flags |= flagRWSession
+	}
+	var session ulong
+	if rv := m.cOpenSession(ulong(slot), flags, nil, 0, &session); rv != rvOK {
+		return 0, Error{Op: "C_OpenSession", Code: uint(rv)}
+	}
+	return uint(session), nil
+}
+
+func (m *module) closeSession(session uint) {
+	m.cCloseSession(ulong(session))
+}
+
+func (m *module) login(session uint, pin []byte) error {
+	var pinPtr *byte
+	if len(pin) > 0 {
+		pinPtr = &pin[0]
+	}
+	rv := m.cLogin(ulong(session), userTypeUser, pinPtr, ulong(len(pin)))
+	runtime.KeepAlive(pin)
+	if rv != rvOK {
+		return Error{Op: "C_Login", Code: uint(rv)}
+	}
+	return nil
+}
+
+func (m *module) logout(session uint) {
+	m.cLogout(ulong(session))
+}
+
+func (m *module) findObjects(session uint, template []Attribute) ([]Object, error) {
+	var pinner runtime.Pinner
+	attrs := toAttributes(template, &pinner)
+	rv := m.cFindObjectsInit(ulong(session), first(attrs), ulong(len(attrs)))
+	pinner.Unpin()
+	if rv != rvOK {
+		return nil, Error{Op: "C_FindObjectsInit", Code: uint(rv)}
+	}
+	defer m.cFindObjectsFinal(ulong(session))
+
+	var objects []Object
+	for {
+		var batch [findBatch]ulong
+		var count ulong
+		if rv := m.cFindObjects(ulong(session), &batch[0], findBatch, &count); rv != rvOK {
+			return nil, Error{Op: "C_FindObjects", Code: uint(rv)}
+		}
+		if count > findBatch {
+			return nil, fmt.Errorf("C_FindObjects reported %d handles for a batch of %d", count, findBatch)
+		}
+		for _, handle := range batch[:count] {
+			objects = append(objects, Object(handle))
+		}
+		if count < findBatch {
+			return objects, nil
+		}
+	}
+}
+
+func (m *module) attribute(session uint, obj Object, typ uint) ([]byte, error) {
+	attr := attribute{typ: ulong(typ)}
+	if rv := m.cGetAttributeValue(ulong(session), ulong(obj), &attr, 1); rv != rvOK {
+		return nil, Error{Op: "C_GetAttributeValue", Code: uint(rv)}
+	}
+	if attr.len == unavailableInformation {
+		return nil, fmt.Errorf("attribute 0x%x is unavailable", typ)
+	}
+	if attr.len == 0 {
+		return nil, nil
+	}
+	if attr.len > maxAttributeLen {
+		return nil, fmt.Errorf("attribute 0x%x reports %d bytes, over the %d byte limit", typ, attr.len, maxAttributeLen)
+	}
+	value := make([]byte, attr.len)
+	var pinner runtime.Pinner
+	pinner.Pin(&value[0])
+	attr.value = unsafe.Pointer(&value[0])
+	rv := m.cGetAttributeValue(ulong(session), ulong(obj), &attr, 1)
+	pinner.Unpin()
+	if rv != rvOK {
+		return nil, Error{Op: "C_GetAttributeValue", Code: uint(rv)}
+	}
+	if attr.len > ulong(len(value)) {
+		return nil, fmt.Errorf("attribute 0x%x grew from %d to %d bytes between calls", typ, len(value), attr.len)
+	}
+	return value[:attr.len], nil
+}
+
+func (m *module) sign(session uint, mech Mechanism, key Object, data []byte) ([]byte, error) {
+	if len(data) == 0 {
+		return nil, errors.New("nothing to sign")
+	}
+	native := mechanism{typ: ulong(mech.Type)}
+	var pinner runtime.Pinner
+	if mech.PSS != nil {
+		params := &pssParams{hashAlg: ulong(mech.PSS.Hash), mgf: ulong(mech.PSS.MGF), saltLen: ulong(mech.PSS.SaltLen)}
+		pinner.Pin(params)
+		native.parameter = unsafe.Pointer(params)
+		native.len = ulong(unsafe.Sizeof(*params))
+	}
+	rv := m.cSignInit(ulong(session), &native, ulong(key))
+	pinner.Unpin()
+	if rv != rvOK {
+		return nil, Error{Op: "C_SignInit", Code: uint(rv)}
+	}
+
+	var size ulong
+	if rv := m.cSign(ulong(session), &data[0], ulong(len(data)), nil, &size); rv != rvOK {
+		return nil, Error{Op: "C_Sign", Code: uint(rv)}
+	}
+	if size == 0 || size > maxSignatureLen {
+		return nil, fmt.Errorf("C_Sign reports a %d byte signature", size)
+	}
+	signature := make([]byte, size)
+	rv = m.cSign(ulong(session), &data[0], ulong(len(data)), &signature[0], &size)
+	runtime.KeepAlive(data)
+	if rv != rvOK {
+		return nil, Error{Op: "C_Sign", Code: uint(rv)}
+	}
+	if size > ulong(len(signature)) {
+		return nil, fmt.Errorf("C_Sign wrote %d bytes into a %d byte buffer", size, len(signature))
+	}
+	return signature[:size], nil
+}
+
+func (m *module) createObject(session uint, template []Attribute) (Object, error) {
+	var pinner runtime.Pinner
+	attrs := toAttributes(template, &pinner)
+	var object ulong
+	rv := m.cCreateObject(ulong(session), first(attrs), ulong(len(attrs)), &object)
+	pinner.Unpin()
+	if rv != rvOK {
+		return 0, Error{Op: "C_CreateObject", Code: uint(rv)}
+	}
+	return Object(object), nil
+}
+
+// toAttributes builds the C template for template. The values stay Go memory referenced
+// from Go memory passed to C, so each is pinned; the caller unpins after the call.
+func toAttributes(template []Attribute, pinner *runtime.Pinner) []attribute {
+	attrs := make([]attribute, len(template))
+	for i, a := range template {
+		attrs[i].typ = ulong(a.Type)
+		if len(a.Value) > 0 {
+			pinner.Pin(&a.Value[0])
+			attrs[i].value = unsafe.Pointer(&a.Value[0])
+			attrs[i].len = ulong(len(a.Value))
+		}
+	}
+	return attrs
+}
+
+func first(attrs []attribute) *attribute {
+	if len(attrs) == 0 {
+		return nil
+	}
+	return &attrs[0]
+}

client/internal/pkcs11/driver_linux_test.go

@@ -0,0 +1,72 @@
+//go:build pkcs11 && linux && !android && (amd64 || arm64)
+
+package pkcs11
+
+import (
+	"crypto/x509"
+	"os"
+	"testing"
+	"unsafe"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestStructLayoutsMatchTheCABI(t *testing.T) {
+	assert.Equal(t, uintptr(24), unsafe.Sizeof(attribute{}), "CK_ATTRIBUTE")
+	assert.Equal(t, uintptr(24), unsafe.Sizeof(mechanism{}), "CK_MECHANISM")
+	assert.Equal(t, uintptr(24), unsafe.Sizeof(pssParams{}), "CK_RSA_PKCS_PSS_PARAMS")
+	assert.Equal(t, uintptr(208), unsafe.Sizeof(tokenInfo{}), "CK_TOKEN_INFO")
+	assert.Equal(t, uintptr(48), unsafe.Sizeof(initializeArgs{}), "CK_C_INITIALIZE_ARGS")
+	assert.Equal(t, uintptr(8), unsafe.Offsetof(functionList{}.fn), "entry points follow the padded CK_VERSION")
+	assert.Equal(t, uintptr(8+68*8), unsafe.Sizeof(functionList{}), "CK_FUNCTION_LIST v2.40")
+}
+
+// TestTrustModule_ListsSystemCertificates drives a real module through the binding:
+// p11-kit's trust module exposes the system CA store as certificate objects with no login.
+func TestTrustModule_ListsSystemCertificates(t *testing.T) {
+	module := loadFirst(t,
+		"/usr/lib/pkcs11/p11-kit-trust.so",
+		"/usr/lib/x86_64-linux-gnu/pkcs11/p11-kit-trust.so",
+		"/usr/lib/aarch64-linux-gnu/pkcs11/p11-kit-trust.so",
+		"/usr/lib64/pkcs11/p11-kit-trust.so",
+	)
+
+	tokens, err := module.Tokens()
+	require.NoError(t, err)
+	require.NotEmpty(t, tokens, "the trust module must present at least one token")
+
+	parsed := 0
+	for _, token := range tokens {
+		session, err := module.OpenSession(token.Label, nil)
+		require.NoError(t, err, token.Label)
+		objects, err := session.FindObjects(
+			Attribute{Type: AttrClass, Value: ULong(ClassCertificate)},
+			Attribute{Type: AttrCertificateType, Value: ULong(CertificateX509)},
+		)
+		require.NoError(t, err, token.Label)
+		for _, object := range objects {
+			der, err := session.Attribute(object, AttrValue)
+			require.NoError(t, err)
+			_, err = x509.ParseCertificate(der)
+			require.NoError(t, err, "CKA_VALUE must be a DER certificate")
+			parsed++
+		}
+		session.Close()
+	}
+	assert.Positive(t, parsed, "system trust anchors must be readable through the binding")
+}
+
+func loadFirst(t *testing.T, paths ...string) *Module {
+	t.Helper()
+	for _, path := range paths {
+		if _, err := os.Stat(path); err != nil {
+			continue
+		}
+		module, err := Load(path)
+		require.NoError(t, err, path)
+		return module
+	}
+	t.Skip("p11-kit trust module not installed")
+	return nil
+}

client/internal/pkcs11/driver_stub.go

@@ -0,0 +1,12 @@
+//go:build !(pkcs11 && linux && !android && (amd64 || arm64))
+
+package pkcs11
+
+// Supported reports whether this build can load PKCS#11 modules.
+func Supported() bool {
+	return false
+}
+
+func load(string) (driver, error) {
+	return nil, ErrUnsupported
+}

client/internal/pkcs11/pinrejected_test.go

@@ -0,0 +1,19 @@
+package pkcs11
+
+import (
+	"errors"
+	"fmt"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestPINRejected(t *testing.T) {
+	for _, code := range []uint{rvPINIncorrect, rvPINInvalid, rvPINLenRange, rvPINExpired, rvPINLocked} {
+		err := fmt.Errorf("open session: %w", Error{Op: "C_Login", Code: code})
+		assert.True(t, PINRejected(err), "CKR 0x%x refuses the PIN, also when wrapped", code)
+	}
+	assert.False(t, PINRejected(Error{Op: "C_Login", Code: 0x30}), "a device error says nothing about the PIN")
+	assert.False(t, PINRejected(errors.New("CKR_PIN_INCORRECT")), "only a PKCS#11 return value counts")
+	assert.False(t, PINRejected(nil))
+}

client/internal/pkcs11/pkcs11.go

@@ -0,0 +1,328 @@
+// Package pkcs11 is a minimal PKCS#11 client. It loads a module at runtime without cgo,
+// opens a token session, lists objects and signs with keys the token holds. It exists so
+// certificates whose keys live in a TPM behind tpm2-pkcs11 can be proven; whatever a
+// certificate store does not need is left out.
+package pkcs11
+
+import (
+	"encoding/binary"
+	"errors"
+	"fmt"
+	"sync"
+)
+
+// Object classes, attribute types, mechanisms and generators from PKCS#11 v2.40.
+const (
+	ClassCertificate = 0x1
+	ClassPublicKey   = 0x2
+	ClassPrivateKey  = 0x3
+
+	CertificateX509 = 0x0
+
+	AttrClass           = 0x0
+	AttrToken           = 0x1
+	AttrLabel           = 0x3
+	AttrValue           = 0x11
+	AttrCertificateType = 0x80
+	AttrKeyType         = 0x100
+	AttrSubject         = 0x101
+	AttrID              = 0x102
+	AttrModulus         = 0x120
+	AttrPublicExponent  = 0x122
+	AttrECParams        = 0x180
+	AttrECPoint         = 0x181
+
+	KeyRSA = 0x0
+	KeyEC  = 0x3
+
+	MechRSAPKCSPSS = 0xd
+	MechSHA256     = 0x250
+	MechSHA384     = 0x260
+	MechECDSA      = 0x1041
+
+	MGF1SHA256 = 0x2
+	MGF1SHA384 = 0x3
+
+	rvOK                  = 0x0
+	rvPINIncorrect        = 0xa0
+	rvPINInvalid          = 0xa1
+	rvPINLenRange         = 0xa2
+	rvPINExpired          = 0xa3
+	rvPINLocked           = 0xa4
+	rvUserAlreadyLoggedIn = 0x100
+	rvAlreadyInitialized  = 0x191
+)
+
+var ErrUnsupported = errors.New("PKCS#11 modules need a build with the pkcs11 tag on linux/amd64 or linux/arm64")
+
+// Error is a PKCS#11 return value other than CKR_OK.
+type Error struct {
+	Op   string
+	Code uint
+}
+
+func (e Error) Error() string {
+	if name, ok := returnValueNames[e.Code]; ok {
+		return fmt.Sprintf("%s: %s", e.Op, name)
+	}
+	return fmt.Sprintf("%s: CKR 0x%x", e.Op, e.Code)
+}
+
+// PINRejected reports whether err is the token refusing the user PIN. Retrying the same
+// PIN cannot succeed, and each attempt counts towards the token's lockout.
+func PINRejected(err error) bool {
+	var e Error
+	if !errors.As(err, &e) {
+		return false
+	}
+	switch e.Code {
+	case rvPINIncorrect, rvPINInvalid, rvPINLenRange, rvPINExpired, rvPINLocked:
+		return true
+	}
+	return false
+}
+
+func isCode(err error, code uint) bool {
+	var e Error
+	return errors.As(err, &e) && e.Code == code
+}
+
+var returnValueNames = map[uint]string{
+	0x2:   "CKR_HOST_MEMORY",
+	0x3:   "CKR_SLOT_ID_INVALID",
+	0x5:   "CKR_GENERAL_ERROR",
+	0x7:   "CKR_ARGUMENTS_BAD",
+	0x12:  "CKR_ATTRIBUTE_TYPE_INVALID",
+	0x13:  "CKR_ATTRIBUTE_VALUE_INVALID",
+	0x30:  "CKR_DEVICE_ERROR",
+	0x54:  "CKR_FUNCTION_NOT_SUPPORTED",
+	0x68:  "CKR_KEY_FUNCTION_NOT_PERMITTED",
+	0x70:  "CKR_MECHANISM_INVALID",
+	0x71:  "CKR_MECHANISM_PARAM_INVALID",
+	0x82:  "CKR_OBJECT_HANDLE_INVALID",
+	0xa0:  "CKR_PIN_INCORRECT",
+	0xa1:  "CKR_PIN_INVALID",
+	0xa2:  "CKR_PIN_LEN_RANGE",
+	0xa3:  "CKR_PIN_EXPIRED",
+	0xa4:  "CKR_PIN_LOCKED",
+	0xb3:  "CKR_SESSION_HANDLE_INVALID",
+	0xd0:  "CKR_TEMPLATE_INCOMPLETE",
+	0xd1:  "CKR_TEMPLATE_INCONSISTENT",
+	0xe0:  "CKR_TOKEN_NOT_PRESENT",
+	0x101: "CKR_USER_NOT_LOGGED_IN",
+	0x150: "CKR_BUFFER_TOO_SMALL",
+	0x190: "CKR_CRYPTOKI_NOT_INITIALIZED",
+}
+
+// Attribute is one entry of a PKCS#11 template. Integer-valued attributes such as the
+// object class are encoded with ULong.
+type Attribute struct {
+	Type  uint
+	Value []byte
+}
+
+// Mechanism selects a signing algorithm. PSS carries the parameters CKM_RSA_PKCS_PSS needs.
+type Mechanism struct {
+	Type uint
+	PSS  *PSSParams
+}
+
+type PSSParams struct {
+	Hash    uint
+	MGF     uint
+	SaltLen uint
+}
+
+// Object is a handle the token issued for one of its objects.
+type Object uint
+
+// Token is a slot with a token present.
+type Token struct {
+	Slot  uint
+	Label string
+}
+
+// Module is a loaded and initialised PKCS#11 library. A module is loaded once per path
+// and never finalised: tokens such as tpm2-pkcs11 do real work in C_Initialize, and the
+// process exit releases everything anyway.
+type Module struct {
+	d driver
+
+	// loginMu guards logins, the number of open sessions relying on the user login of
+	// each slot, and serializes logging in so a PIN is never sent twice at once.
+	loginMu sync.Mutex
+	logins  map[uint]int
+}
+
+var (
+	modulesMu sync.Mutex
+	modules   = map[string]*Module{}
+)
+
+// Load opens the shared library at path and initialises it, or returns the module already
+// loaded from that path.
+func Load(path string) (*Module, error) {
+	modulesMu.Lock()
+	defer modulesMu.Unlock()
+	if m, ok := modules[path]; ok {
+		return m, nil
+	}
+	d, err := load(path)
+	if err != nil {
+		return nil, err
+	}
+	m := &Module{d: d}
+	modules[path] = m
+	return m, nil
+}
+
+func (m *Module) Tokens() ([]Token, error) {
+	return m.d.tokens()
+}
+
+// OpenSession opens a read-only session with the token carrying label, or with the first
+// token when label is empty, and logs in as the user when pin is not nil. An empty,
+// non-nil pin still logs in.
+func (m *Module) OpenSession(label string, pin []byte) (*Session, error) {
+	return m.openSession(label, pin, false)
+}
+
+// OpenReadWriteSession is OpenSession for callers that create objects on the token.
+func (m *Module) OpenReadWriteSession(label string, pin []byte) (*Session, error) {
+	return m.openSession(label, pin, true)
+}
+
+func (m *Module) openSession(label string, pin []byte, readWrite bool) (*Session, error) {
+	token, err := m.token(label)
+	if err != nil {
+		return nil, err
+	}
+	handle, err := m.d.openSession(token.Slot, readWrite)
+	if err != nil {
+		return nil, err
+	}
+	s := &Session{d: m.d, handle: handle}
+	if pin == nil {
+		return s, nil
+	}
+	if err := m.acquireLogin(token.Slot, handle, pin); err != nil {
+		s.Close()
+		return nil, err
+	}
+	s.module, s.slot = m, token.Slot
+	return s, nil
+}
+
+// acquireLogin makes sure the user is logged in to the token in slot and counts the
+// session as one relying on it. PKCS#11 login state belongs to the application, not to
+// a session: every session with the token shares it, and logging out from any of them
+// ends it for all. So the PIN is sent only when no session holds the login yet, and the
+// last session to close logs out.
+func (m *Module) acquireLogin(slot, handle uint, pin []byte) error {
+	m.loginMu.Lock()
+	defer m.loginMu.Unlock()
+	if m.logins[slot] > 0 {
+		m.logins[slot]++
+		return nil
+	}
+	if err := m.d.login(handle, pin); err != nil && !isCode(err, rvUserAlreadyLoggedIn) {
+		return err
+	}
+	if m.logins == nil {
+		m.logins = make(map[uint]int)
+	}
+	m.logins[slot] = 1
+	return nil
+}
+
+// releaseLogin ends the session's share of the login, logging out when it is the last.
+func (m *Module) releaseLogin(slot, handle uint) {
+	m.loginMu.Lock()
+	defer m.loginMu.Unlock()
+	m.logins[slot]--
+	if m.logins[slot] > 0 {
+		return
+	}
+	delete(m.logins, slot)
+	m.d.logout(handle)
+}
+
+func (m *Module) token(label string) (Token, error) {
+	tokens, err := m.Tokens()
+	if err != nil {
+		return Token{}, err
+	}
+	for _, token := range tokens {
+		if label == "" || token.Label == label {
+			return token, nil
+		}
+	}
+	if label == "" {
+		return Token{}, errors.New("no token present")
+	}
+	return Token{}, fmt.Errorf("no token labelled %q among %d tokens", label, len(tokens))
+}
+
+// Session is an open session with one token. Close releases the session's share of the
+// login, logging out when no other session relies on it.
+type Session struct {
+	d      driver
+	handle uint
+	// module is set when the session relies on the token's login.
+	module *Module
+	slot   uint
+}
+
+func (s *Session) Close() {
+	if s.module != nil {
+		s.module.releaseLogin(s.slot, s.handle)
+	}
+	s.d.closeSession(s.handle)
+}
+
+// FindObjects returns the handles of every object matching all attributes of template.
+func (s *Session) FindObjects(template ...Attribute) ([]Object, error) {
+	return s.d.findObjects(s.handle, template)
+}
+
+// Attribute reads one attribute of an object.
+func (s *Session) Attribute(obj Object, typ uint) ([]byte, error) {
+	return s.d.attribute(s.handle, obj, typ)
+}
+
+// Sign signs data, normally a digest, with the token-held key in a single operation.
+func (s *Session) Sign(mech Mechanism, key Object, data []byte) ([]byte, error) {
+	return s.d.sign(s.handle, mech, key, data)
+}
+
+// CreateObject stores a new object described by template on the token.
+func (s *Session) CreateObject(template ...Attribute) (Object, error) {
+	return s.d.createObject(s.handle, template)
+}
+
+type driver interface {
+	tokens() ([]Token, error)
+	openSession(slot uint, readWrite bool) (uint, error)
+	closeSession(session uint)
+	login(session uint, pin []byte) error
+	logout(session uint)
+	findObjects(session uint, template []Attribute) ([]Object, error)
+	attribute(session uint, obj Object, typ uint) ([]byte, error)
+	sign(session uint, mech Mechanism, key Object, data []byte) ([]byte, error)
+	createObject(session uint, template []Attribute) (Object, error)
+}
+
+// ulongSize is the width of CK_ULONG on the 64-bit platforms the driver builds for.
+const ulongSize = 8
+
+// ULong encodes an integer attribute value the way the module reads a CK_ULONG.
+func ULong(v uint) []byte {
+	return binary.NativeEndian.AppendUint64(nil, uint64(v))
+}
+
+func ulongValue(b []byte) (uint, error) {
+	if len(b) != ulongSize {
+		return 0, fmt.Errorf("CK_ULONG value has %d bytes", len(b))
+	}
+	return uint(binary.NativeEndian.Uint64(b)), nil
+}

client/internal/pkcs11/publickey.go

@@ -0,0 +1,95 @@
+package pkcs11
+
+import (
+	"crypto"
+	"crypto/ecdsa"
+	"crypto/elliptic"
+	"crypto/rsa"
+	"encoding/asn1"
+	"errors"
+	"fmt"
+	"math"
+	"math/big"
+)
+
+var curvesByOID = map[string]elliptic.Curve{
+	"1.2.840.10045.3.1.7": elliptic.P256(),
+	"1.3.132.0.34":        elliptic.P384(),
+	"1.3.132.0.35":        elliptic.P521(),
+}
+
+// PublicKey reads a CKO_PUBLIC_KEY object as a Go public key. RSA and EC keys are
+// supported, the two kinds a certificate posture proof can be signed with.
+func (s *Session) PublicKey(obj Object) (crypto.PublicKey, error) {
+	raw, err := s.Attribute(obj, AttrKeyType)
+	if err != nil {
+		return nil, err
+	}
+	keyType, err := ulongValue(raw)
+	if err != nil {
+		return nil, fmt.Errorf("CKA_KEY_TYPE: %w", err)
+	}
+	switch keyType {
+	case KeyRSA:
+		modulus, exponent, err := s.attributes(obj, AttrModulus, AttrPublicExponent)
+		if err != nil {
+			return nil, err
+		}
+		return rsaPublicKey(modulus, exponent)
+	case KeyEC:
+		params, point, err := s.attributes(obj, AttrECParams, AttrECPoint)
+		if err != nil {
+			return nil, err
+		}
+		return ecPublicKey(params, point)
+	}
+	return nil, fmt.Errorf("unsupported key type 0x%x", keyType)
+}
+
+func (s *Session) attributes(obj Object, first, second uint) ([]byte, []byte, error) {
+	a, err := s.Attribute(obj, first)
+	if err != nil {
+		return nil, nil, err
+	}
+	b, err := s.Attribute(obj, second)
+	if err != nil {
+		return nil, nil, err
+	}
+	return a, b, nil
+}
+
+func rsaPublicKey(modulus, exponent []byte) (*rsa.PublicKey, error) {
+	e := new(big.Int).SetBytes(exponent)
+	if e.Sign() <= 0 || e.Cmp(big.NewInt(math.MaxInt32)) > 0 {
+		return nil, errors.New("CKA_PUBLIC_EXPONENT is out of range")
+	}
+	return &rsa.PublicKey{N: new(big.Int).SetBytes(modulus), E: int(e.Int64())}, nil
+}
+
+// ecPublicKey decodes CKA_EC_PARAMS, the named curve OID, and CKA_EC_POINT, the
+// uncompressed point wrapped in a DER OCTET STRING, which some modules hand out bare.
+func ecPublicKey(params, point []byte) (*ecdsa.PublicKey, error) {
+	var oid asn1.ObjectIdentifier
+	if _, err := asn1.Unmarshal(params, &oid); err != nil {
+		return nil, fmt.Errorf("CKA_EC_PARAMS: %w", err)
+	}
+	curve, ok := curvesByOID[oid.String()]
+	if !ok {
+		return nil, fmt.Errorf("unsupported curve %s", oid)
+	}
+	size := (curve.Params().BitSize + 7) / 8
+	raw := point
+	if len(raw) != 1+2*size {
+		if _, err := asn1.Unmarshal(point, &raw); err != nil {
+			return nil, fmt.Errorf("CKA_EC_POINT: %w", err)
+		}
+	}
+	if len(raw) != 1+2*size || raw[0] != 4 {
+		return nil, errors.New("CKA_EC_POINT is not an uncompressed point")
+	}
+	key, err := ecdsa.ParseUncompressedPublicKey(curve, raw)
+	if err != nil {
+		return nil, fmt.Errorf("CKA_EC_POINT: %w", err)
+	}
+	return key, nil
+}

client/internal/pkcs11/publickey_test.go

@@ -0,0 +1,87 @@
+package pkcs11
+
+import (
+	"crypto/ecdsa"
+	"crypto/elliptic"
+	"crypto/rand"
+	"crypto/rsa"
+	"encoding/asn1"
+	"math/big"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestECPublicKey(t *testing.T) {
+	curves := []struct {
+		name  string
+		curve elliptic.Curve
+		oid   asn1.ObjectIdentifier
+	}{
+		{"P-256", elliptic.P256(), asn1.ObjectIdentifier{1, 2, 840, 10045, 3, 1, 7}},
+		{"P-384", elliptic.P384(), asn1.ObjectIdentifier{1, 3, 132, 0, 34}},
+	}
+	for _, tt := range curves {
+		t.Run(tt.name, func(t *testing.T) {
+			key, err := ecdsa.GenerateKey(tt.curve, rand.Reader)
+			require.NoError(t, err)
+			params, err := asn1.Marshal(tt.oid)
+			require.NoError(t, err)
+			point := uncompressedPoint(key)
+			wrapped, err := asn1.Marshal(point)
+			require.NoError(t, err)
+
+			// PKCS#11 wraps the point in an OCTET STRING, but some modules return it bare.
+			for form, encoded := range map[string][]byte{"DER octet string": wrapped, "bare point": point} {
+				pub, err := ecPublicKey(params, encoded)
+				require.NoError(t, err, form)
+				assert.True(t, key.PublicKey.Equal(pub), "%s must decode to the generated key", form)
+			}
+		})
+	}
+}
+
+func TestECPublicKey_Rejections(t *testing.T) {
+	key, err := ecdsa.GenerateKey(elliptic.P256(), rand.Reader)
+	require.NoError(t, err)
+	p256, err := asn1.Marshal(asn1.ObjectIdentifier{1, 2, 840, 10045, 3, 1, 7})
+	require.NoError(t, err)
+	brainpool, err := asn1.Marshal(asn1.ObjectIdentifier{1, 3, 36, 3, 3, 2, 8, 1, 1, 7})
+	require.NoError(t, err)
+	point := uncompressedPoint(key)
+
+	_, err = ecPublicKey(brainpool, point)
+	assert.Error(t, err, "curves the proof cannot use must be rejected")
+	_, err = ecPublicKey(p256, point[:len(point)-1])
+	assert.Error(t, err, "a truncated point must be rejected")
+	_, err = ecPublicKey([]byte("junk"), point)
+	assert.Error(t, err, "malformed parameters must be rejected")
+}
+
+func TestRSAPublicKey(t *testing.T) {
+	key, err := rsa.GenerateKey(rand.Reader, 2048)
+	require.NoError(t, err)
+
+	pub, err := rsaPublicKey(key.N.Bytes(), big.NewInt(int64(key.E)).Bytes())
+	require.NoError(t, err)
+	assert.True(t, key.PublicKey.Equal(pub), "modulus and exponent must decode to the generated key")
+
+	_, err = rsaPublicKey(key.N.Bytes(), nil)
+	assert.Error(t, err, "a missing exponent must be rejected")
+}
+
+func TestULongRoundTrip(t *testing.T) {
+	v, err := ulongValue(ULong(ClassPrivateKey))
+	require.NoError(t, err)
+	assert.Equal(t, uint(ClassPrivateKey), v)
+
+	_, err = ulongValue([]byte{1, 2, 3})
+	assert.Error(t, err, "a value of the wrong width must be rejected")
+}
+
+func uncompressedPoint(key *ecdsa.PrivateKey) []byte {
+	size := (key.Curve.Params().BitSize + 7) / 8
+	point := append([]byte{4}, key.X.FillBytes(make([]byte, size))...)
+	return append(point, key.Y.FillBytes(make([]byte, size))...)
+}

client/internal/pkcs11/session_test.go

@@ -0,0 +1,72 @@
+package pkcs11
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// fakeDriver stands in for a loaded module and records the calls a session makes.
+type fakeDriver struct {
+	loginErr error
+	logins   int
+	logouts  int
+	closes   int
+}
+
+func (f *fakeDriver) tokens() ([]Token, error)             { return []Token{{Slot: 1, Label: "netbird"}}, nil }
+func (f *fakeDriver) openSession(uint, bool) (uint, error) { return 7, nil }
+func (f *fakeDriver) closeSession(uint)                    { f.closes++ }
+func (f *fakeDriver) login(uint, []byte) error             { f.logins++; return f.loginErr }
+func (f *fakeDriver) logout(uint)                          { f.logouts++ }
+func (f *fakeDriver) findObjects(uint, []Attribute) ([]Object, error) {
+	return nil, nil
+}
+func (f *fakeDriver) attribute(uint, Object, uint) ([]byte, error) { return nil, nil }
+func (f *fakeDriver) sign(uint, Mechanism, Object, []byte) ([]byte, error) {
+	return nil, nil
+}
+func (f *fakeDriver) createObject(uint, []Attribute) (Object, error) { return 0, nil }
+
+func TestOpenSession_LogsOutOnlyALoginItOwns(t *testing.T) {
+	t.Run("own login is logged out on close", func(t *testing.T) {
+		d := &fakeDriver{}
+		s, err := (&Module{d: d}).OpenSession("netbird", []byte("1234"))
+		require.NoError(t, err)
+		s.Close()
+		assert.Equal(t, 1, d.logouts, "the session that logged in logs out again")
+	})
+
+	t.Run("a login shared by sessions ends with the last one", func(t *testing.T) {
+		d := &fakeDriver{}
+		m := &Module{d: d}
+		first, err := m.OpenSession("netbird", []byte("1234"))
+		require.NoError(t, err)
+		second, err := m.OpenSession("netbird", []byte("1234"))
+		require.NoError(t, err)
+		assert.Equal(t, 1, d.logins, "the second session reuses the login instead of sending the PIN again")
+
+		first.Close()
+		assert.Zero(t, d.logouts, "logging out now would end the login the second session still uses")
+		second.Close()
+		assert.Equal(t, 1, d.logouts, "the last session to close logs out")
+		assert.Equal(t, 2, d.closes, "both sessions are closed")
+	})
+
+	t.Run("an existing login is taken over", func(t *testing.T) {
+		d := &fakeDriver{loginErr: Error{Op: "C_Login", Code: rvUserAlreadyLoggedIn}}
+		s, err := (&Module{d: d}).OpenSession("netbird", []byte("1234"))
+		require.NoError(t, err, "an existing login is good enough to use the token")
+		s.Close()
+		assert.Equal(t, 1, d.logouts, "no session of this client holds the login any more, so it ends")
+	})
+
+	t.Run("rejected pin closes the session", func(t *testing.T) {
+		d := &fakeDriver{loginErr: Error{Op: "C_Login", Code: rvPINIncorrect}}
+		_, err := (&Module{d: d}).OpenSession("netbird", []byte("0000"))
+		assert.True(t, PINRejected(err), "the PIN error reaches the caller")
+		assert.Zero(t, d.logouts, "nothing to log out after a failed login")
+		assert.Equal(t, 1, d.closes, "the session opened for the login is closed")
+	})
+}

client/internal/pkcs11/uri.go

@@ -0,0 +1,135 @@
+package pkcs11
+
+import (
+	"errors"
+	"fmt"
+	"net/url"
+	"os"
+	"path/filepath"
+	"strings"
+)
+
+// DefaultModule is p11-kit's proxy, which exposes every module the system has registered,
+// tpm2-pkcs11 included, so a URI without module-path works on a stock p11-kit setup.
+const DefaultModule = "p11-kit-proxy.so"
+
+// URI is the subset of an RFC 7512 PKCS#11 URI this client understands: the token label,
+// the module to load and where the user PIN comes from.
+type URI struct {
+	Token      string
+	ModulePath string
+	pinValue   *string
+	pinSource  string
+}
+
+// ParseURI parses raw following RFC 7512 section 2.3. A path attribute other than token
+// is refused rather than ignored: the path narrows which token is used, and ignoring a
+// constraint such as serial would widen the match to whichever token is listed first,
+// where the RFC calls for no match at all. Duplicate attributes are refused, a
+// module-path must be absolute and a module-name must be a bare name. Unknown query
+// attributes are ignored, as the RFC asks.
+func ParseURI(raw string) (*URI, error) {
+	rest, ok := strings.CutPrefix(raw, "pkcs11:")
+	if !ok {
+		return nil, errors.New("PKCS#11 URI does not start with the pkcs11 scheme")
+	}
+	path, query, _ := strings.Cut(rest, "?")
+
+	u := &URI{}
+	if err := eachAttribute(path, ";", u.setPathAttribute); err != nil {
+		return nil, err
+	}
+	if err := eachAttribute(query, "&", u.setQueryAttribute); err != nil {
+		return nil, err
+	}
+	// RFC 7512 section 2.4: a URI with both pin-value and pin-source should be refused.
+	if u.pinValue != nil && u.pinSource != "" {
+		return nil, errors.New("PKCS#11 URI carries both pin-value and pin-source")
+	}
+	return u, nil
+}
+
+func (u *URI) setPathAttribute(name, value string) error {
+	if name != "token" {
+		return fmt.Errorf("PKCS#11 URI path attribute %q is not supported, only token selects a token", name)
+	}
+	u.Token = value
+	return nil
+}
+
+func (u *URI) setQueryAttribute(name, value string) error {
+	switch name {
+	case "module-path":
+		if !filepath.IsAbs(value) {
+			return fmt.Errorf("PKCS#11 URI module-path %q must be absolute", value)
+		}
+		u.ModulePath = value
+	case "module-name":
+		if value == "" || strings.ContainsAny(value, `/\`) || strings.Contains(value, "..") {
+			return fmt.Errorf("PKCS#11 URI module-name %q must be a module name, not a path", value)
+		}
+		u.ModulePath = "lib" + value + ".so"
+	case "pin-value":
+		u.pinValue = &value
+	case "pin-source":
+		u.pinSource = value
+	}
+	return nil
+}
+
+// Module is the library to load, DefaultModule when the URI names none.
+func (u *URI) Module() string {
+	if u.ModulePath == "" {
+		return DefaultModule
+	}
+	return u.ModulePath
+}
+
+// HasPIN reports whether the URI carries a PIN, inline or as a pin-source.
+func (u *URI) HasPIN() bool {
+	return u.pinValue != nil || u.pinSource != ""
+}
+
+// PIN returns the user PIN, or nil when the URI carries none and no login should happen.
+// A pin-source names a file whose single line is the PIN.
+func (u *URI) PIN() ([]byte, error) {
+	if u.pinValue != nil {
+		return []byte(*u.pinValue), nil
+	}
+	if u.pinSource == "" {
+		return nil, nil
+	}
+	path := strings.TrimPrefix(strings.TrimPrefix(u.pinSource, "file://"), "file:")
+	pin, err := os.ReadFile(path)
+	if err != nil {
+		return nil, fmt.Errorf("read PIN: %w", err)
+	}
+	return []byte(strings.TrimRight(string(pin), "\r\n")), nil
+}
+
+// eachAttribute splits list on sep and calls fn for every name=value pair, refusing a
+// name that appears twice.
+func eachAttribute(list, sep string, fn func(name, value string) error) error {
+	if list == "" {
+		return nil
+	}
+	seen := make(map[string]struct{})
+	for _, pair := range strings.Split(list, sep) {
+		name, value, ok := strings.Cut(pair, "=")
+		if !ok {
+			return fmt.Errorf("PKCS#11 URI attribute %q has no value", pair)
+		}
+		if _, dup := seen[name]; dup {
+			return fmt.Errorf("PKCS#11 URI attribute %s appears more than once", name)
+		}
+		seen[name] = struct{}{}
+		value, err := url.PathUnescape(value)
+		if err != nil {
+			return fmt.Errorf("PKCS#11 URI attribute %s: %w", name, err)
+		}
+		if err := fn(name, value); err != nil {
+			return err
+		}
+	}
+	return nil
+}

client/internal/pkcs11/uri_test.go

@@ -0,0 +1,119 @@
+package pkcs11
+
+import (
+	"os"
+	"path/filepath"
+	"runtime"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestParseURI(t *testing.T) {
+	tests := []struct {
+		name       string
+		raw        string
+		wantToken  string
+		wantModule string
+		wantPIN    []byte
+	}{
+		{
+			name:       "token with module path and pin value",
+			raw:        "pkcs11:token=netbird?module-path=" + absModule("libtpm2_pkcs11.so") + "&pin-value=1234",
+			wantToken:  "netbird",
+			wantModule: absModule("libtpm2_pkcs11.so"),
+			wantPIN:    []byte("1234"),
+		},
+		{
+			name:       "module name becomes a library file",
+			raw:        "pkcs11:token=netbird?module-name=tpm2_pkcs11",
+			wantToken:  "netbird",
+			wantModule: "libtpm2_pkcs11.so",
+		},
+		{
+			name:       "percent encoding, unknown query attributes are ignored",
+			raw:        "pkcs11:token=my%20token?max-sessions=1&vendor-flag=on",
+			wantToken:  "my token",
+			wantModule: DefaultModule,
+		},
+		{
+			name:       "bare scheme uses the p11-kit proxy and no login",
+			raw:        "pkcs11:",
+			wantModule: DefaultModule,
+		},
+		{
+			name:       "empty pin value still logs in",
+			raw:        "pkcs11:?pin-value=",
+			wantModule: DefaultModule,
+			wantPIN:    []byte{},
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			uri, err := ParseURI(tt.raw)
+			require.NoError(t, err)
+			assert.Equal(t, tt.wantToken, uri.Token, "token label")
+			assert.Equal(t, tt.wantModule, uri.Module(), "module to load")
+			pin, err := uri.PIN()
+			require.NoError(t, err)
+			assert.Equal(t, tt.wantPIN, pin, "PIN, nil meaning no login")
+		})
+	}
+}
+
+func TestParseURI_Rejections(t *testing.T) {
+	tests := map[string]string{
+		"no scheme":            "pkcs11",
+		"other scheme":         "https://example.com",
+		"attribute no value":   "pkcs11:token",
+		"bad percent encoding": "pkcs11:token=%zz",
+		// RFC 7512 2.3: an unrecognized path attribute matches nothing. Ignoring it
+		// instead would let the token listed first answer for the one serial names.
+		"unsupported serial":       "pkcs11:token=netbird;serial=1234",
+		"unsupported model":        "pkcs11:model=SoftHSM%20v2;token=netbird",
+		"object selector":          "pkcs11:token=netbird;object=device",
+		"vendor path attribute":    "pkcs11:token=netbird;vendor-slot=2",
+		"duplicate token":          "pkcs11:token=a;token=b",
+		"duplicate module-path":    "pkcs11:token=a?module-path=" + absModule("a.so") + "&module-path=" + absModule("b.so"),
+		"duplicate pin-value":      "pkcs11:token=a?pin-value=1&pin-value=2",
+		"relative module-path":     "pkcs11:token=a?module-path=lib/x.so",
+		"bare module-path":         "pkcs11:token=a?module-path=libtpm2_pkcs11.so",
+		"module-name with path":    "pkcs11:token=a?module-name=../../tmp/evil",
+		"module-name with slash":   "pkcs11:token=a?module-name=tmp/evil",
+		"module-name with dotdot":  "pkcs11:token=a?module-name=..",
+		"empty module-name":        "pkcs11:token=a?module-name=",
+		"pin-value and pin-source": "pkcs11:token=a?pin-value=1&pin-source=file:/etc/pin",
+	}
+	for name, raw := range tests {
+		_, err := ParseURI(raw)
+		assert.Error(t, err, "%s: %s must be refused", name, raw)
+	}
+}
+
+func TestURI_PINFromFile(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "pin")
+	require.NoError(t, os.WriteFile(path, []byte("secret\n"), 0o600))
+
+	for _, source := range []string{path, "file:" + path, "file://" + path} {
+		uri, err := ParseURI("pkcs11:token=netbird?pin-source=" + source)
+		require.NoError(t, err)
+		pin, err := uri.PIN()
+		require.NoError(t, err)
+		assert.Equal(t, []byte("secret"), pin, "PIN from %s must drop the trailing newline", source)
+	}
+
+	uri, err := ParseURI("pkcs11:?pin-source=" + filepath.Join(t.TempDir(), "missing"))
+	require.NoError(t, err)
+	_, err = uri.PIN()
+	assert.Error(t, err, "a missing PIN file must fail loudly instead of logging in without a PIN")
+}
+
+// absModule is an absolute module path on the platform the test runs on, as module-path
+// must be absolute.
+func absModule(name string) string {
+	if runtime.GOOS == "windows" {
+		return `C:\lib\` + name
+	}
+	return "/lib/" + name
+}

client/internal/profilemanager/config.go

@@ -185,6 +185,16 @@ type Config struct {
 
 	ClientCertKeyPair *tls.Certificate `json:"-"`
 
+	// CertStoreDir is no longer read: certificate posture takes the directory from
+	// NB_CERT_STORE_DIR in the daemon's environment. The field is kept only to report a
+	// value left from an earlier version.
+	CertStoreDir string `json:",omitempty"`
+
+	// CertPKCS11URI is no longer read, as the URI may carry the token PIN: certificate
+	// posture takes it from NB_CERT_PKCS11_URI in the daemon's environment. The field is
+	// kept only to report a value left from an earlier version.
+	CertPKCS11URI string `json:",omitempty"`
+
 	// LazyConnection is the MDM-managed lazy-connection override ("on"/"off"/"").
 	// Runtime-only: re-derived from MDM policy on each load, never persisted.
 	LazyConnection string `json:"-"`

client/internal/tpm/open_linux.go

@@ -0,0 +1,29 @@
+package tpm
+
+import (
+	"errors"
+	"fmt"
+	"io"
+	"os"
+
+	"github.com/google/go-tpm/tpmutil"
+)
+
+// The kernel resource manager multiplexes clients and flushes what they leave behind,
+// so it is tried before the raw device.
+var devicePaths = []string{"/dev/tpmrm0", "/dev/tpm0"}
+
+func open() (io.ReadWriteCloser, error) {
+	if path := os.Getenv(DeviceEnv); path != "" {
+		return tpmutil.OpenTPM(path)
+	}
+	var errs error
+	for _, path := range devicePaths {
+		rwc, err := tpmutil.OpenTPM(path)
+		if err == nil {
+			return rwc, nil
+		}
+		errs = errors.Join(errs, err)
+	}
+	return nil, fmt.Errorf("open TPM: %w", errs)
+}

client/internal/tpm/open_other.go

@@ -0,0 +1,9 @@
+//go:build !linux
+
+package tpm
+
+import "io"
+
+func open() (io.ReadWriteCloser, error) {
+	return nil, ErrUnsupported
+}

client/internal/tpm/signer.go

@@ -0,0 +1,163 @@
+package tpm
+
+import (
+	"crypto"
+	"crypto/ecdsa"
+	"crypto/elliptic"
+	"crypto/rsa"
+	"encoding/asn1"
+	"errors"
+	"fmt"
+	"io"
+	"math/big"
+
+	legacy "github.com/google/go-tpm/legacy/tpm2"
+	"github.com/google/go-tpm/tpmutil"
+)
+
+// KeyPEMType is the PEM block type of a TPM 2.0 key file as defined by
+// draft-bottomley-tpm2-keys and written by tpm2-openssl and tpm2-tss-engine.
+const KeyPEMType = "TSS2 PRIVATE KEY"
+
+var ErrKeyNeedsAuth = errors.New("TPM key requires an authorization value")
+
+// eccSRKTemplate is the TCG reference ECC-P256 storage root key. A key whose parent is
+// a hierarchy rather than a persistent handle was wrapped by the primary this template
+// derives, and tpm2-openssl and tpm2-tss-engine derive the same one, so the TPM
+// reproduces the identical parent from the hierarchy seed without anything being stored.
+var eccSRKTemplate = legacy.Public{
+	Type:       legacy.AlgECC,
+	NameAlg:    legacy.AlgSHA256,
+	Attributes: legacy.FlagStorageDefault | legacy.FlagNoDA,
+	ECCParameters: &legacy.ECCParams{
+		Symmetric: &legacy.SymScheme{Alg: legacy.AlgAES, KeyBits: 128, Mode: legacy.AlgCFB},
+		Sign:      &legacy.SigScheme{Alg: legacy.AlgNull},
+		CurveID:   legacy.CurveNISTP256,
+	},
+}
+
+// ParseKey reads a TSS2 key file and returns a signer that produces every signature
+// inside the TPM; only the digest goes in and only the signature comes out. Keys guarded
+// by an authorization value are rejected, since nothing can supply it without prompting.
+func ParseKey(der []byte) (crypto.Signer, error) {
+	key, err := parseTSS2(der)
+	if err != nil {
+		return nil, err
+	}
+	if !key.EmptyAuth {
+		return nil, ErrKeyNeedsAuth
+	}
+
+	public, err := legacy.DecodePublic(key.PublicKey[2:])
+	if err != nil {
+		return nil, fmt.Errorf("decode TSS2 public area: %w", err)
+	}
+	pub, err := public.Key()
+	if err != nil {
+		return nil, fmt.Errorf("decode TSS2 public key: %w", err)
+	}
+	return &keySigner{key: key, public: pub}, nil
+}
+
+type keySigner struct {
+	key    *tss2Key
+	public crypto.PublicKey
+}
+
+func (s *keySigner) Public() crypto.PublicKey {
+	return s.public
+}
+
+// Sign loads the key under its parent, signs, and releases both handles. The TPM is
+// opened per signature so no handle outlives the call, which matters on a device whose
+// transient object slots are few and shared with everything else on the host.
+func (s *keySigner) Sign(_ io.Reader, digest []byte, opts crypto.SignerOpts) ([]byte, error) {
+	rwc, err := Open()
+	if err != nil {
+		return nil, err
+	}
+	defer func() { _ = rwc.Close() }()
+
+	parent := tpmutil.Handle(s.key.Parent)
+	if !persistentHandle(s.key.Parent) {
+		parent, _, err = legacy.CreatePrimary(rwc, parent, legacy.PCRSelection{}, "", "", eccSRKTemplate)
+		if err != nil {
+			return nil, fmt.Errorf("create TPM primary: %w", err)
+		}
+		defer func() { _ = legacy.FlushContext(rwc, parent) }()
+	}
+
+	public, private := s.key.blobs()
+	handle, _, err := legacy.Load(rwc, parent, "", public, private)
+	if err != nil {
+		return nil, fmt.Errorf("load TPM key: %w", err)
+	}
+	defer func() { _ = legacy.FlushContext(rwc, handle) }()
+
+	switch pub := s.public.(type) {
+	case *ecdsa.PublicKey:
+		return signECDSA(rwc, handle, digest, pub.Curve)
+	case *rsa.PublicKey:
+		return signRSA(rwc, handle, digest, opts)
+	default:
+		return nil, fmt.Errorf("unsupported TPM key type %T", s.public)
+	}
+}
+
+// signECDSA returns the signature as the ASN.1 sequence crypto.Signer is defined to
+// return; the TPM hands back the two integers on their own.
+func signECDSA(rw io.ReadWriter, handle tpmutil.Handle, digest []byte, curve elliptic.Curve) ([]byte, error) {
+	hash, err := eccHash(curve)
+	if err != nil {
+		return nil, err
+	}
+	sig, err := legacy.Sign(rw, handle, "", digest, nil, &legacy.SigScheme{Alg: legacy.AlgECDSA, Hash: hash})
+	if err != nil {
+		return nil, fmt.Errorf("TPM ECDSA signature: %w", err)
+	}
+	if sig.ECC == nil {
+		return nil, fmt.Errorf("TPM returned a %v signature for an ECDSA key", sig.Alg)
+	}
+	return asn1.Marshal(struct{ R, S *big.Int }{sig.ECC.R, sig.ECC.S})
+}
+
+func signRSA(rw io.ReadWriter, handle tpmutil.Handle, digest []byte, opts crypto.SignerOpts) ([]byte, error) {
+	hash, err := legacy.HashToAlgorithm(opts.HashFunc())
+	if err != nil {
+		return nil, fmt.Errorf("TPM hash algorithm: %w", err)
+	}
+
+	scheme := &legacy.SigScheme{Alg: legacy.AlgRSASSA, Hash: hash}
+	if pss, ok := opts.(*rsa.PSSOptions); ok {
+		// The TPM chooses the salt length itself, the digest length on most chips, so
+		// only a request for that length is taken. PSSSaltLengthAuto asks for the
+		// largest salt the key allows and is refused. Verify with PSSSaltLengthAuto.
+		if pss.SaltLength != rsa.PSSSaltLengthEqualsHash &&
+			pss.SaltLength != len(digest) {
+			return nil, fmt.Errorf("TPM cannot produce a PSS signature with salt length %d", pss.SaltLength)
+		}
+		scheme.Alg = legacy.AlgRSAPSS
+	}
+
+	sig, err := legacy.Sign(rw, handle, "", digest, nil, scheme)
+	if err != nil {
+		return nil, fmt.Errorf("TPM RSA signature: %w", err)
+	}
+	if sig.RSA == nil {
+		return nil, fmt.Errorf("TPM returned a %v signature for an RSA key", sig.Alg)
+	}
+	return sig.RSA.Signature, nil
+}
+
+func eccHash(curve elliptic.Curve) (legacy.Algorithm, error) {
+	switch curve {
+	case elliptic.P256():
+		return legacy.AlgSHA256, nil
+	case elliptic.P384():
+		return legacy.AlgSHA384, nil
+	case elliptic.P521():
+		return legacy.AlgSHA512, nil
+	default:
+		return 0, fmt.Errorf("unsupported curve %s", curve.Params().Name)
+	}
+}

client/internal/tpm/signer_test.go

@@ -0,0 +1,73 @@
+package tpm
+
+import (
+	"crypto"
+	"crypto/ecdsa"
+	"crypto/elliptic"
+	"crypto/rand"
+	"crypto/rsa"
+	"crypto/sha256"
+	"encoding/pem"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/client/internal/tpm/tpmtest"
+)
+
+func TestParseKey_ReportsPublicKeyWithoutTouchingTPM(t *testing.T) {
+	key := newP256Key(t)
+
+	signer, err := ParseKey(decodePEM(t, tpmtest.KeyPEM(t, &key.PublicKey)))
+	require.NoError(t, err)
+	assert.True(t, key.PublicKey.Equal(signer.Public()), "signer must expose the key the TPM holds")
+}
+
+func TestParseKey_RejectsKeyWithAuthorization(t *testing.T) {
+	key := newP256Key(t)
+	_, err := ParseKey(decodePEM(t, tpmtest.KeyPEM(t, &key.PublicKey, tpmtest.WithAuth())))
+	assert.ErrorIs(t, err, ErrKeyNeedsAuth)
+}
+
+func TestParseKey_RejectsMalformedKey(t *testing.T) {
+	_, err := ParseKey([]byte("not a TSS2 key"))
+	assert.Error(t, err)
+}
+
+func TestSign_FailsWhenTPMIsUnreachable(t *testing.T) {
+	t.Setenv(DeviceEnv, filepath.Join(t.TempDir(), "missing"))
+	signer, err := ParseKey(decodePEM(t, tpmtest.KeyPEM(t, &newP256Key(t).PublicKey)))
+	require.NoError(t, err)
+
+	digest := sha256.Sum256([]byte("challenge"))
+	_, err = signer.Sign(rand.Reader, digest[:], crypto.SHA256)
+	assert.Error(t, err, "signing must not fall back to software when the TPM is missing")
+}
+
+// TestSignRSA_RefusesSaltLengthsTheTPMDoesNotChoose checks the salt-length gate, which
+// runs before the TPM is touched: the TPM picks the salt itself, so a request for the
+// maximum salt (PSSSaltLengthAuto) or any other explicit length is refused.
+func TestSignRSA_RefusesSaltLengthsTheTPMDoesNotChoose(t *testing.T) {
+	digest := sha256.Sum256([]byte("challenge"))
+	for _, salt := range []int{rsa.PSSSaltLengthAuto, 20, 222} {
+		_, err := signRSA(nil, 0, digest[:], &rsa.PSSOptions{SaltLength: salt, Hash: crypto.SHA256})
+		assert.ErrorContains(t, err, "salt length", "salt length %d must be refused", salt)
+	}
+}
+
+func newP256Key(t *testing.T) *ecdsa.PrivateKey {
+	t.Helper()
+	key, err := ecdsa.GenerateKey(elliptic.P256(), rand.Reader)
+	require.NoError(t, err)
+	return key
+}
+
+func decodePEM(t *testing.T, pemData string) []byte {
+	t.Helper()
+	block, _ := pem.Decode([]byte(pemData))
+	require.NotNil(t, block)
+	require.Equal(t, KeyPEMType, block.Type)
+	return block.Bytes
+}

client/internal/tpm/tpm.go

@@ -0,0 +1,19 @@
+// Package tpm is the client's one door to the platform TPM 2.0. It opens the device
+// and turns TPM-held key files into signers; every operation opens the TPM, runs and
+// closes it, so no handle outlives a call.
+package tpm
+
+import (
+	"errors"
+	"io"
+)
+
+// DeviceEnv overrides the TPM device path, which also lets tests point at a swtpm socket.
+const DeviceEnv = "NB_TPM_DEVICE"
+
+var ErrUnsupported = errors.New("TPM is not supported on this platform")
+
+// Open connects to the platform TPM 2.0. The caller closes it after one operation.
+func Open() (io.ReadWriteCloser, error) {
+	return open()
+}

client/internal/tpm/tpmtest/tpmtest.go

@@ -0,0 +1,121 @@
+// Package tpmtest builds TSS2 key files for tests, with or without a TPM behind them.
+package tpmtest
+
+import (
+	"crypto/ecdsa"
+	"crypto/elliptic"
+	"encoding/asn1"
+	"encoding/pem"
+	"testing"
+
+	"github.com/google/go-tpm/legacy/tpm2"
+	"github.com/stretchr/testify/require"
+)
+
+const (
+	p256Bytes = 32
+
+	// KeyPEMType is the PEM block type of a TPM 2.0 key file.
+	KeyPEMType = "TSS2 PRIVATE KEY"
+)
+
+// oidLoadableKey is the key type of draft-bottomley-tpm2-keys that a parent wraps.
+var oidLoadableKey = asn1.ObjectIdentifier{2, 23, 133, 10, 1, 3}
+
+// tss2KeyDER is the ASN.1 container, written out independently of the parser under test
+// so that an encoder bug and a decoder bug cannot cancel out. Policy, secret and auth
+// policy are left out entirely: they are optional, and a key that carries them is
+// refused anyway.
+type tss2KeyDER struct {
+	Type       asn1.ObjectIdentifier
+	EmptyAuth  bool `asn1:"optional,explicit,tag:0"`
+	Parent     int64
+	PublicKey  []byte
+	PrivateKey []byte
+}
+
+// Option adjusts a key before it is encoded.
+type Option func(*tss2KeyDER)
+
+// WithParent names the handle the key is wrapped by, instead of the owner hierarchy.
+func WithParent(handle int64) Option {
+	return func(k *tss2KeyDER) { k.Parent = handle }
+}
+
+// WithAuth marks the key as guarded by an authorization value, which this client refuses
+// because nothing can supply one without prompting.
+func WithAuth() Option {
+	return func(k *tss2KeyDER) { k.EmptyAuth = false }
+}
+
+// SigningTemplate is the public area of an unrestricted P-256 signing key with no fixed
+// scheme, the shape tpm2-openssl creates certificate keys in.
+func SigningTemplate() tpm2.Public {
+	return tpm2.Public{
+		Type:          tpm2.AlgECC,
+		NameAlg:       tpm2.AlgSHA256,
+		Attributes:    tpm2.FlagSign | tpm2.FlagFixedTPM | tpm2.FlagFixedParent | tpm2.FlagSensitiveDataOrigin | tpm2.FlagUserWithAuth | tpm2.FlagNoDA,
+		ECCParameters: &tpm2.ECCParams{CurveID: tpm2.CurveNISTP256},
+	}
+}
+
+// KeyPEM encodes pub as a TSS2 PRIVATE KEY over a placeholder private blob: it parses
+// and reports pub, but no TPM can load it.
+func KeyPEM(t *testing.T, pub *ecdsa.PublicKey, opts ...Option) string {
+	t.Helper()
+	require.Equal(t, elliptic.P256(), pub.Curve, "fixture keys must be P-256")
+	area := SigningTemplate()
+	// The uncompressed encoding is 0x04 followed by the two coordinates, which is
+	// what the TPM wants them as.
+	point, err := pub.Bytes()
+	require.NoError(t, err)
+	require.Len(t, point, 1+2*p256Bytes)
+	area.ECCParameters.Point = tpm2.ECPoint{
+		XRaw: point[1 : 1+p256Bytes],
+		YRaw: point[1+p256Bytes:],
+	}
+	encoded, encErr := area.Encode()
+	require.NoError(t, encErr)
+	return EncodePEM(t, encoded, []byte("placeholder"), opts...)
+}
+
+// EncodePEM wraps the public and private blobs TPM2_Create returned into a TSS2 PRIVATE
+// KEY, giving each the TPM2B length prefix the format carries them with.
+func EncodePEM(t *testing.T, public, private []byte, opts ...Option) string {
+	t.Helper()
+
+	key := tss2KeyDER{
+		Type:       oidLoadableKey,
+		EmptyAuth:  true,
+		Parent:     int64(tpm2.HandleOwner),
+		PublicKey:  prefixTPM2B(public),
+		PrivateKey: prefixTPM2B(private),
+	}
+	for _, opt := range opts {
+		opt(&key)
+	}
+
+	der, err := asn1.Marshal(key)
+	require.NoError(t, err)
+	return string(pem.EncodeToMemory(&pem.Block{Type: KeyPEMType, Bytes: der}))
+}
+
+func prefixTPM2B(b []byte) []byte {
+	out := make([]byte, 0, len(b)+2)
+	out = append(out, byte(len(b)>>8), byte(len(b)))
+	return append(out, b...)
+}
+
+// ECCSRKTemplate is the TCG reference ECC-P256 storage root key, the parent a key under
+// a hierarchy is wrapped by. Tests that create a key in a real TPM have to use the same
+// template the signer re-derives it with.
+var ECCSRKTemplate = tpm2.Public{
+	Type:       tpm2.AlgECC,
+	NameAlg:    tpm2.AlgSHA256,
+	Attributes: tpm2.FlagStorageDefault | tpm2.FlagNoDA,
+	ECCParameters: &tpm2.ECCParams{
+		Symmetric: &tpm2.SymScheme{Alg: tpm2.AlgAES, KeyBits: 128, Mode: tpm2.AlgCFB},
+		Sign:      &tpm2.SigScheme{Alg: tpm2.AlgNull},
+		CurveID:   tpm2.CurveNISTP256,
+	},
+}

client/internal/tpm/tss2.go

@@ -0,0 +1,133 @@
+package tpm
+
+import (
+	"encoding/asn1"
+	"errors"
+	"fmt"
+	"math"
+
+	legacy "github.com/google/go-tpm/legacy/tpm2"
+)
+
+// oidLoadableKey marks a key wrapped by a parent inside the TPM, which is the only kind
+// enrollment tooling writes for a signing key and the only kind that can be loaded under
+// an SRK. The sealed and importable variants carry different material and are refused.
+var oidLoadableKey = asn1.ObjectIdentifier{2, 23, 133, 10, 1, 3}
+
+var (
+	errNotLoadable  = errors.New("TSS2 key is not a loadable key")
+	errKeyHasPolicy = errors.New("TSS2 key carries a policy, which is not supported")
+	errKeyHasSecret = errors.New("TSS2 key carries a secret, which is not supported")
+	errBadParent    = errors.New("TSS2 key names a parent that is neither persistent nor a hierarchy")
+	errBadBlob      = errors.New("TSS2 key blob is malformed")
+)
+
+// tss2KeyASN1 is the ASN.1 container of draft-bottomley-tpm2-keys, the format
+// tpm2-openssl, tpm2-tss-engine and tpm2_encodeobject write. PublicKey and PrivateKey
+// hold TPM2B structures, so each is its own two-byte length followed by that many bytes.
+//
+// EmptyAuth is decoded raw: OpenSSL-based tools write BOOLEAN TRUE as 0x01, which BER
+// allows but encoding/asn1 rejects, as it accepts only the DER form 0xff. Parent is an
+// int64 so a persistent handle such as 0x81000001 still fits on 32-bit platforms.
+type tss2KeyASN1 struct {
+	Type       asn1.ObjectIdentifier
+	EmptyAuth  asn1.RawValue   `asn1:"optional,explicit,tag:0"`
+	Policy     []asn1.RawValue `asn1:"optional,explicit,tag:1"`
+	Secret     []byte          `asn1:"optional,explicit,tag:2"`
+	AuthPolicy []asn1.RawValue `asn1:"optional,explicit,tag:3"`
+	Parent     int64
+	PublicKey  []byte
+	PrivateKey []byte
+}
+
+// tss2Key is a decoded key file the client can ask the TPM to load.
+type tss2Key struct {
+	EmptyAuth  bool
+	Parent     uint32
+	PublicKey  []byte
+	PrivateKey []byte
+}
+
+// parseTSS2 decodes a TSS2 key file and rejects everything this client cannot honour,
+// so a key that parses here is one the TPM can be asked to load.
+func parseTSS2(der []byte) (*tss2Key, error) {
+	raw := new(tss2KeyASN1)
+	rest, err := asn1.Unmarshal(der, raw)
+	if err != nil {
+		return nil, fmt.Errorf("parse TSS2 key: %w", err)
+	}
+	if len(rest) > 0 {
+		return nil, fmt.Errorf("parse TSS2 key: %d trailing bytes", len(rest))
+	}
+	emptyAuth, err := berBoolean(raw.EmptyAuth)
+	if err != nil {
+		return nil, fmt.Errorf("parse TSS2 key emptyAuth: %w", err)
+	}
+
+	switch {
+	case !raw.Type.Equal(oidLoadableKey):
+		return nil, fmt.Errorf("%w: %s", errNotLoadable, raw.Type)
+	case len(raw.Policy) > 0 || len(raw.AuthPolicy) > 0:
+		return nil, errKeyHasPolicy
+	case len(raw.Secret) > 0:
+		return nil, errKeyHasSecret
+	case raw.Parent < 0 || raw.Parent > math.MaxUint32 || !validParent(uint32(raw.Parent)):
+		return nil, fmt.Errorf("%w: %d", errBadParent, raw.Parent)
+	case !validTPM2B(raw.PublicKey) || !validTPM2B(raw.PrivateKey):
+		return nil, errBadBlob
+	}
+	return &tss2Key{
+		EmptyAuth:  emptyAuth,
+		Parent:     uint32(raw.Parent),
+		PublicKey:  raw.PublicKey,
+		PrivateKey: raw.PrivateKey,
+	}, nil
+}
+
+// berBoolean decodes an optional BOOLEAN, absent meaning false. Any non-zero content
+// byte is true, as BER (X.690 section 8.2.2) allows and OpenSSL writes.
+func berBoolean(v asn1.RawValue) (bool, error) {
+	if len(v.FullBytes) == 0 {
+		return false, nil
+	}
+	// The field is explicitly tagged, so v is the [0] wrapper and the BOOLEAN is inside it.
+	if v.Class == asn1.ClassContextSpecific && v.IsCompound {
+		var inner asn1.RawValue
+		rest, err := asn1.Unmarshal(v.Bytes, &inner)
+		if err != nil || len(rest) > 0 {
+			return false, errors.New("malformed explicit tag")
+		}
+		v = inner
+	}
+	if v.Class != asn1.ClassUniversal || v.Tag != asn1.TagBoolean || v.IsCompound || len(v.Bytes) != 1 {
+		return false, errors.New("not a BOOLEAN")
+	}
+	return v.Bytes[0] != 0, nil
+}
+
+// blobs returns the public and private areas with their TPM2B length prefix removed,
+// which is the form the load command takes them in.
+func (k *tss2Key) blobs() (public, private []byte) {
+	return k.PublicKey[2:], k.PrivateKey[2:]
+}
+
+// validParent accepts a persistent handle, under which the key was wrapped directly, or
+// one of the four hierarchies, under which the key is wrapped by a primary the TPM
+// re-derives from the hierarchy seed.
+func validParent(parent uint32) bool {
+	return persistentHandle(parent) ||
+		parent == uint32(legacy.HandleOwner) ||
+		parent == uint32(legacy.HandleNull) ||
+		parent == uint32(legacy.HandleEndorsement) ||
+		parent == uint32(legacy.HandlePlatform)
+}
+
+func persistentHandle(h uint32) bool {
+	return h>>24 == uint32(legacy.HandleTypePersistent)
+}
+
+// validTPM2B reports whether b is a TPM2B structure: a two-byte big-endian length
+// followed by exactly that many bytes.
+func validTPM2B(b []byte) bool {
+	return len(b) >= 2 && len(b)-2 == int(b[0])<<8+int(b[1])
+}

client/internal/tpm/tss2_test.go

@@ -0,0 +1,78 @@
+package tpm
+
+import (
+	"bytes"
+	"crypto/ecdsa"
+	"crypto/elliptic"
+	"crypto/rand"
+	"encoding/asn1"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/client/internal/tpm/tpmtest"
+)
+
+// derEmptyAuthTrue is the explicit [0] tag around a DER BOOLEAN TRUE, as encoding/asn1
+// writes emptyAuth; OpenSSL writes the same element with 0x01 as the content byte.
+var (
+	derEmptyAuthTrue     = []byte{0xa0, 0x03, 0x01, 0x01, 0xff}
+	opensslEmptyAuthTrue = []byte{0xa0, 0x03, 0x01, 0x01, 0x01}
+)
+
+func p256(t *testing.T) *ecdsa.PrivateKey {
+	t.Helper()
+	key, err := ecdsa.GenerateKey(elliptic.P256(), rand.Reader)
+	require.NoError(t, err)
+	return key
+}
+
+func TestParseKey_AcceptsOpenSSLBoolean(t *testing.T) {
+	key := p256(t)
+	der := decodePEM(t, tpmtest.KeyPEM(t, &key.PublicKey))
+	require.Equal(t, 1, bytes.Count(der, derEmptyAuthTrue), "fixture carries one DER emptyAuth TRUE")
+
+	// The same key as tpm2-openssl and tpm2-tss-engine write it: identical apart from
+	// the BOOLEAN content byte.
+	openssl := bytes.Replace(der, derEmptyAuthTrue, opensslEmptyAuthTrue, 1)
+
+	signer, err := ParseKey(openssl)
+	require.NoError(t, err, "BER allows any non-zero byte for TRUE, and OpenSSL writes 0x01")
+	assert.True(t, key.PublicKey.Equal(signer.Public()), "the key is the same either way")
+}
+
+func TestParseKey_AcceptsPersistentParent(t *testing.T) {
+	key := p256(t)
+	// 0x81000001 overflows a 32-bit int, the width the parent used to be decoded into.
+	der := decodePEM(t, tpmtest.KeyPEM(t, &key.PublicKey, tpmtest.WithParent(0x81000001)))
+
+	parsed, err := parseTSS2(der)
+	require.NoError(t, err)
+	assert.Equal(t, uint32(0x81000001), parsed.Parent, "persistent parent handle")
+	assert.True(t, persistentHandle(parsed.Parent), "the key is loaded under the persistent parent directly")
+}
+
+func TestParseKey_RejectsParentOutOfRange(t *testing.T) {
+	key := p256(t)
+	for _, parent := range []int64{-1, 0x1_0000_0000, 0x02000000} {
+		_, err := parseTSS2(decodePEM(t, tpmtest.KeyPEM(t, &key.PublicKey, tpmtest.WithParent(parent))))
+		assert.ErrorIs(t, err, errBadParent, "parent %#x is neither persistent nor a hierarchy", parent)
+	}
+}
+
+func TestBERBoolean(t *testing.T) {
+	absent, err := berBoolean(asn1.RawValue{})
+	require.NoError(t, err)
+	assert.False(t, absent, "an absent emptyAuth means the key needs authorization")
+
+	for content, want := range map[byte]bool{0x00: false, 0x01: true, 0xff: true} {
+		v := asn1.RawValue{Class: asn1.ClassUniversal, Tag: asn1.TagBoolean, Bytes: []byte{content}, FullBytes: []byte{0x01, 0x01, content}}
+		got, err := berBoolean(v)
+		require.NoError(t, err)
+		assert.Equal(t, want, got, "content byte %#x", content)
+	}
+
+	_, err = berBoolean(asn1.RawValue{Class: asn1.ClassUniversal, Tag: asn1.TagInteger, Bytes: []byte{1}, FullBytes: []byte{0x02, 0x01, 0x01}})
+	assert.Error(t, err, "an INTEGER is not a BOOLEAN")
+}

client/server/login_account_test.go

@@ -2,6 +2,8 @@ package server
 
 import (
 	"context"
+	"errors"
+	"path/filepath"
 	"testing"
 	"time"
 
@@ -9,9 +11,24 @@ import (
 
 	"github.com/netbirdio/netbird/client/internal"
 	"github.com/netbirdio/netbird/client/internal/auth"
+	"github.com/netbirdio/netbird/client/internal/ipcauth"
+	"github.com/netbirdio/netbird/client/internal/profilemanager"
+	"github.com/netbirdio/netbird/client/mdm"
 	"github.com/netbirdio/netbird/client/proto"
 )
 
+type jwtStoringMDMFetcher struct {
+	cache      *jwtCache
+	caller     ipcauth.Identity
+	generation uint64
+}
+
+func (f *jwtStoringMDMFetcher) Fetch() map[string]any {
+	f.generation = f.cache.currentGeneration()
+	f.cache.store("previous-profile-token", f.caller, time.Minute, f.generation)
+	return nil
+}
+
 type stubOAuthFlow struct {
 	token  auth.TokenInfo
 	onWait func()
@@ -123,6 +140,110 @@ func TestSwitchProfile_DropsAccountPromptAndPendingFlow(t *testing.T) {
 	require.False(t, pending, "the previous profile's extend flow leaked across a profile switch")
 }
 
+func TestLogin_ProfileSwitchDropsAccountPromptAndPendingFlow(t *testing.T) {
+	s, _, _, username, cfgPath := setupServerWithProfile(t)
+	s.rootCtx = internal.CtxInitState(context.Background())
+	s.isLoginRequiredFn = func(context.Context) (bool, error) {
+		return true, nil
+	}
+
+	other := "other-profile"
+	otherPath := filepath.Join(filepath.Dir(cfgPath), other+".json")
+	_, err := profilemanager.UpdateOrCreateConfig(profilemanager.ConfigInput{
+		ConfigPath:    otherPath,
+		ManagementURL: "https://api.netbird.io:443",
+	})
+	require.NoError(t, err)
+	breakProfilePrivateKey(t, otherPath)
+
+	s.forceAccountPrompt = true
+	cancelled := false
+	s.oauthAuthFlow = oauthAuthFlow{
+		flow:       &stubOAuthFlow{},
+		hint:       "user@example.com",
+		waitCancel: func() { cancelled = true },
+	}
+
+	extendCancelled := false
+	s.extendAuthSessionFlow.Set(&stubOAuthFlow{}, auth.AuthFlowInfo{DeviceCode: "device"})
+	s.extendAuthSessionFlow.SetWaitCancel(func() { extendCancelled = true })
+
+	generation := s.jwtCache.currentGeneration()
+
+	_, err = s.Login(userCtx(), &proto.LoginRequest{ProfileName: &other, Username: &username})
+	require.Error(t, err, "the broken key must stop the login before a flow is built")
+
+	active, err := s.profileManager.GetActiveProfileState()
+	require.NoError(t, err)
+	require.Equal(t, profilemanager.ID(other), active.ID, "the login did not switch the profile")
+
+	require.False(t, s.forceAccountPrompt, "the prompt flag leaked across a login-driven profile switch")
+	require.Nil(t, s.oauthAuthFlow.flow, "the previous profile's flow leaked across a login-driven profile switch")
+	require.Empty(t, s.oauthAuthFlow.hint)
+	require.True(t, cancelled, "the pending wait was not cancelled")
+
+	require.True(t, extendCancelled, "the pending extend wait was not cancelled")
+	_, _, pending := s.extendAuthSessionFlow.Get()
+	require.False(t, pending, "the previous profile's extend flow leaked across a login-driven profile switch")
+
+	require.Greater(t, s.jwtCache.currentGeneration(), generation, "the previous profile's JWT cache survived a login-driven profile switch")
+}
+
+func TestLogin_ProfileSwitchRejectsJWTObtainedUnderPreviousConfig(t *testing.T) {
+	s, _, _, username, cfgPath := setupServerWithProfile(t)
+	s.rootCtx = internal.CtxInitState(context.Background())
+	s.isLoginRequiredFn = func(context.Context) (bool, error) {
+		return false, errors.New("stop once the config is swapped")
+	}
+
+	other := "other-profile"
+	_, err := profilemanager.UpdateOrCreateConfig(profilemanager.ConfigInput{
+		ConfigPath:    filepath.Join(filepath.Dir(cfgPath), other+".json"),
+		ManagementURL: "https://api.netbird.io:443",
+	})
+	require.NoError(t, err)
+
+	// getConfig loads the MDM policy right before Login swaps s.config, so the
+	// fetcher runs where a RequestJWTAuth racing the login would land: after the
+	// switch dropped the previous profile's state, while s.config still belongs
+	// to the previous profile. It caches a token under the generation current at
+	// that point, as WaitJWTToken would.
+	generation := s.jwtCache.currentGeneration()
+	fetcher := &jwtStoringMDMFetcher{cache: s.jwtCache, caller: unprivilegedIdentity()}
+	s.mdmLoader = mdm.NewLoader(fetcher)
+
+	_, err = s.Login(userCtx(), &proto.LoginRequest{ProfileName: &other, Username: &username})
+	require.Error(t, err)
+
+	require.Greater(t, fetcher.generation, generation, "the token was not cached after the switch dropped the previous profile's state")
+	_, found := s.jwtCache.get(fetcher.caller)
+	require.False(t, found, "a JWT obtained under the previous profile's config survived the login-driven switch")
+}
+
+func TestLogin_SameProfileKeepsPendingFlow(t *testing.T) {
+	s, _, profName, username, cfgPath := setupServerWithProfile(t)
+	s.rootCtx = internal.CtxInitState(context.Background())
+	s.isLoginRequiredFn = func(context.Context) (bool, error) {
+		return true, nil
+	}
+	breakProfilePrivateKey(t, cfgPath)
+
+	cancelled := false
+	flow := &stubOAuthFlow{}
+	s.oauthAuthFlow = oauthAuthFlow{
+		flow:       flow,
+		hint:       "user@example.com",
+		waitCancel: func() { cancelled = true },
+	}
+
+	_, err := s.Login(userCtx(), &proto.LoginRequest{ProfileName: &profName, Username: &username})
+	require.Error(t, err, "the broken key must stop the login before a flow is built")
+
+	require.Equal(t, flow, s.oauthAuthFlow.flow, "a login on the same profile dropped the flow a second client could join")
+	require.Equal(t, "user@example.com", s.oauthAuthFlow.hint)
+	require.False(t, cancelled, "a login on the same profile cancelled the pending wait")
+}
+
 func TestWaitSSOLogin_JudgesTheFlowThatProducedTheToken(t *testing.T) {
 	s := newSSOTestServer(t, "user@example.com", false, "user@example.com")
 	attempts := 0

client/server/server.go

@@ -722,7 +722,7 @@ func (s *Server) Login(callerCtx context.Context, msg *proto.LoginRequest) (*pro
 		}
 	}()
 
-	ctx, activeProf, err := s.authorizeAndPrepareLogin(callerCtx, msg, activeProf)
+	ctx, activeProf, switched, err := s.authorizeAndPrepareLogin(callerCtx, msg, activeProf)
 	if err != nil {
 		// The RPC boundary is where this gets recorded: nothing logs handler
 		// errors for us, and a caller that retries would otherwise leave no
@@ -752,6 +752,9 @@ func (s *Server) Login(callerCtx context.Context, msg *proto.LoginRequest) (*pro
 	}
 	s.mutex.Lock()
 	s.config = config
+	if switched {
+		s.jwtCache.clear()
+	}
 	s.mutex.Unlock()
 
 	s.localMetrics.Reconcile(config.LocalMetricsEnabled, config.LocalMetricsAddress)
@@ -1192,11 +1195,15 @@ func (s *Server) Up(callerCtx context.Context, msg *proto.UpRequest) (*proto.UpR
 	}
 
 	if msg != nil && msg.ProfileName != nil {
-		if _, err := s.switchProfileIfNeeded(*msg.ProfileName, msg.Username, activeProf); err != nil {
+		switched, err := s.switchProfileIfNeeded(*msg.ProfileName, msg.Username, activeProf)
+		if err != nil {
 			s.mutex.Unlock()
 			log.Errorf("failed to switch profile: %v", err)
 			return nil, err
 		}
+		if switched {
+			s.dropPendingAuthFlows()
+		}
 	}
 
 	activeProf, err = s.profileManager.GetActiveProfileState()
@@ -1334,12 +1341,12 @@ func (s *Server) resolveProfileHandle(handle, username string) (*profilemanager.
 }
 
 // switchProfileIfNeeded resolves the user-supplied handle, updates the
-// active profile state if it differs from the current one, and returns
-// the resolved profile so callers can include its ID in RPC responses.
-func (s *Server) switchProfileIfNeeded(handle string, userName *string, activeProf *profilemanager.ActiveProfileState) (*profilemanager.Profile, error) {
+// active profile state if it differs from the current one, and reports
+// whether the active profile changed.
+func (s *Server) switchProfileIfNeeded(handle string, userName *string, activeProf *profilemanager.ActiveProfileState) (bool, error) {
 	if handle != profilemanager.DefaultProfileName && (userName == nil || *userName == "") {
 		log.Errorf("profile name is set to %s, but username is not provided", handle)
-		return nil, fmt.Errorf("profile name is set to %s, but username is not provided", handle)
+		return false, fmt.Errorf("profile name is set to %s, but username is not provided", handle)
 	}
 
 	var username string
@@ -1349,26 +1356,48 @@ func (s *Server) switchProfileIfNeeded(handle string, userName *string, activePr
 
 	resolved, err := s.resolveProfileHandle(handle, username)
 	if err != nil {
-		return nil, err
+		return false, err
 	}
 
-	if resolved.ID != activeProf.ID || username != activeProf.Username {
-		if s.checkProfilesDisabled() {
-			log.Errorf("profiles are disabled, you cannot use this feature without profiles enabled")
-			return nil, gstatus.Errorf(codes.Unavailable, errProfilesDisabled)
-		}
-
-		log.Infof("switching to profile %s (%s) for user %s", resolved.Name, resolved.ID, username)
-		if err := s.profileManager.SetActiveProfileState(&profilemanager.ActiveProfileState{
-			ID:       resolved.ID,
-			Username: username,
-		}); err != nil {
-			log.Errorf("failed to set active profile state: %v", err)
-			return nil, fmt.Errorf("failed to set active profile state: %w", err)
-		}
+	if resolved.ID == activeProf.ID && username == activeProf.Username {
+		return false, nil
 	}
 
-	return resolved, nil
+	if s.checkProfilesDisabled() {
+		log.Errorf("profiles are disabled, you cannot use this feature without profiles enabled")
+		return false, gstatus.Errorf(codes.Unavailable, errProfilesDisabled)
+	}
+
+	log.Infof("switching to profile %s (%s) for user %s", resolved.Name, resolved.ID, username)
+	if err := s.profileManager.SetActiveProfileState(&profilemanager.ActiveProfileState{
+		ID:       resolved.ID,
+		Username: username,
+	}); err != nil {
+		log.Errorf("failed to set active profile state: %v", err)
+		return false, fmt.Errorf("failed to set active profile state: %w", err)
+	}
+
+	return true, nil
+}
+
+func (s *Server) dropPendingAuthFlows() {
+	// A pending login flow and the account-prompt flag describe the previous
+	// profile's login; carried across a switch they would judge the new
+	// profile's token against the old profile's account. CancelFunc is
+	// non-blocking, so calling it under the mutex is safe.
+	if cancel := s.oauthAuthFlow.waitCancel; cancel != nil {
+		cancel()
+	}
+	s.oauthAuthFlow = oauthAuthFlow{}
+	s.forceAccountPrompt = false
+
+	// A pending session extend belongs to the previous profile too: its device
+	// code was issued by that profile's IdP client, and WaitExtendAuthSession
+	// would submit the resulting token against the new profile's engine.
+	s.extendAuthSessionFlow.CancelWait()
+	s.extendAuthSessionFlow.Clear()
+
+	s.jwtCache.clear()
 }
 
 // SwitchProfile switches the active profile in the daemon.
@@ -1402,23 +1431,7 @@ func (s *Server) SwitchProfile(callerCtx context.Context, msg *proto.SwitchProfi
 	s.config = config
 	s.localMetrics.Reconcile(config.LocalMetricsEnabled, config.LocalMetricsAddress)
 
-	s.jwtCache.clear()
-
-	// A pending login flow and the account-prompt flag describe the previous
-	// profile's login; carried across a switch they would judge the new
-	// profile's token against the old profile's account. CancelFunc is
-	// non-blocking, so calling it under the mutex is safe.
-	if cancel := s.oauthAuthFlow.waitCancel; cancel != nil {
-		cancel()
-	}
-	s.oauthAuthFlow = oauthAuthFlow{}
-	s.forceAccountPrompt = false
-
-	// A pending session extend belongs to the previous profile too: its device
-	// code was issued by that profile's IdP client, and WaitExtendAuthSession
-	// would submit the resulting token against the new profile's engine.
-	s.extendAuthSessionFlow.CancelWait()
-	s.extendAuthSessionFlow.Clear()
+	s.dropPendingAuthFlows()
 
 	if msg != nil && msg.ProfileName != nil {
 		s.publishProfileListChanged(*msg.ProfileName)
@@ -2704,9 +2717,23 @@ func (s *Server) checkDisableAdvancedView() *bool {
 	return nil
 }
 
+// profileOwnerOption passes the OS account of the active profile to the connect client,
+// which reads that account's certificate store for user certificate posture checks.
+func (s *Server) profileOwnerOption() []internal.ConnectClientOption {
+	if s.profileManager == nil {
+		return nil
+	}
+	activeProf, err := s.profileManager.GetActiveProfileState()
+	if err != nil {
+		log.Warnf("failed to read the active profile owner, no user certificate store is used for certificate posture: %v", err)
+		return []internal.ConnectClientOption{internal.WithUnknownProfileOwner()}
+	}
+	return []internal.ConnectClientOption{internal.WithProfileOwner(activeProf.Username)}
+}
+
 func (s *Server) connect(ctx context.Context, config *profilemanager.Config, statusRecorder *peer.Status, runningChan chan struct{}) error {
 	log.Tracef("running client connection")
-	client := internal.NewConnectClient(ctx, config, statusRecorder)
+	client := internal.NewConnectClient(ctx, config, statusRecorder, s.profileOwnerOption()...)
 	client.SetUpdateManager(s.updateManager)
 	client.SetSyncResponsePersistence(s.persistSyncResponse)
 
@@ -2862,7 +2889,7 @@ var afterLoginPreCheck func()
 // of this is reached; this one exists because that check is not synchronized
 // against a concurrent privileged request that enables the SSH server, and a
 // caller refused here must not have cancelled or switched anything either.
-func (s *Server) authorizeAndPrepareLogin(callerCtx context.Context, msg *proto.LoginRequest, activeProf *profilemanager.ActiveProfileState) (context.Context, *profilemanager.ActiveProfileState, error) {
+func (s *Server) authorizeAndPrepareLogin(callerCtx context.Context, msg *proto.LoginRequest, activeProf *profilemanager.ActiveProfileState) (context.Context, *profilemanager.ActiveProfileState, bool, error) {
 	if afterLoginPreCheck != nil {
 		afterLoginPreCheck()
 	}
@@ -2872,10 +2899,10 @@ func (s *Server) authorizeAndPrepareLogin(callerCtx context.Context, msg *proto.
 
 	stored, err := s.storedLoginConfig(activeProf, msg)
 	if err != nil {
-		return nil, nil, err
+		return nil, nil, false, err
 	}
 	if err := requirePrivilegeForConfigChange(callerCtx, stored, privilegedChangeFromLogin(msg)); err != nil {
-		return nil, nil, err
+		return nil, nil, false, err
 	}
 
 	// The update-settings decision is re-taken here for the same reason as the
@@ -2884,7 +2911,7 @@ func (s *Server) authorizeAndPrepareLogin(callerCtx context.Context, msg *proto.
 	// authoritative check, and it is the last read before persistLoginOverrides
 	// writes.
 	if s.checkUpdateSettingsDisabled() && configChangeRequested(stored, loginOverridesInput(msg)) {
-		return nil, nil, gstatus.Errorf(codes.FailedPrecondition, errUpdateSettingsDisabled)
+		return nil, nil, false, gstatus.Errorf(codes.FailedPrecondition, errUpdateSettingsDisabled)
 	}
 
 	s.mutex.Lock()
@@ -2902,19 +2929,26 @@ func (s *Server) authorizeAndPrepareLogin(callerCtx context.Context, msg *proto.
 		log.Warnf(errRestoreResidualState, err)
 	}
 
+	switched := false
 	if msg.ProfileName != nil {
-		if _, err := s.switchProfileIfNeeded(*msg.ProfileName, msg.Username, activeProf); err != nil {
-			return nil, nil, fmt.Errorf("switch profile: %w", err)
+		switched, err = s.switchProfileIfNeeded(*msg.ProfileName, msg.Username, activeProf)
+		if err != nil {
+			return nil, nil, false, fmt.Errorf("switch profile: %w", err)
+		}
+		if switched {
+			s.mutex.Lock()
+			s.dropPendingAuthFlows()
+			s.mutex.Unlock()
 		}
 	}
 
 	activeProf, err = s.profileManager.GetActiveProfileState()
 	if err != nil {
-		return nil, nil, fmt.Errorf("active profile state: %w", err)
+		return nil, nil, false, fmt.Errorf("active profile state: %w", err)
 	}
 
 	if err := persistLoginOverrides(activeProf, msg); err != nil {
-		return nil, nil, fmt.Errorf("persist login overrides: %w", err)
+		return nil, nil, false, fmt.Errorf("persist login overrides: %w", err)
 	}
 
 	// Provisioning under the same lock as the decision above, and next to the
@@ -2923,10 +2957,10 @@ func (s *Server) authorizeAndPrepareLogin(callerCtx context.Context, msg *proto.
 	// had already answered its caller would be overwritten by the config this
 	// login read before it landed.
 	if _, _, err := provisionProfileIdentity(activeProf); err != nil {
-		return nil, nil, err
+		return nil, nil, false, err
 	}
 
-	return ctx, activeProf, nil
+	return ctx, activeProf, switched, nil
 }
 
 // persistLoginOverrides writes the config fields a login request is allowed to

client/system/info.go

@@ -11,6 +11,7 @@ import (
 	log "github.com/sirupsen/logrus"
 	"google.golang.org/grpc/metadata"
 
+	"github.com/netbirdio/netbird/shared/management/certposture"
 	"github.com/netbirdio/netbird/shared/management/proto"
 )
 
@@ -61,6 +62,7 @@ type Info struct {
 	SystemManufacturer string
 	Environment        Environment
 	Files              []File // for posture checks
+	CertificateProofs  []certposture.Proof
 
 	RosenpassEnabled    bool
 	RosenpassPermissive bool
@@ -204,12 +206,27 @@ func GetInfoWithChecks(ctx context.Context, checks []*proto.Checks, excludeIPs .
 // The buffered channel lets the abandoned goroutine finish and exit once its blocking call
 // returns, so it does not leak beyond the duration of that call.
 func GetInfoWithChecksTimeout(ctx context.Context, timeout time.Duration, checks []*proto.Checks, excludeIPs ...netip.Addr) (*Info, bool) {
+	return getInfoWithChecksTimeout(ctx, timeout, checks, nil, excludeIPs...)
+}
+
+// gatherInfoWithChecks is the gathering getInfoWithChecksTimeout bounds. Tests replace it
+// to control when a gathering finishes.
+var gatherInfoWithChecks = GetInfoWithChecks
+
+// getInfoWithChecksTimeout is GetInfoWithChecksTimeout that calls done, when not nil, once
+// the gathering is over: on return when it finished in time, else when the goroutine
+// exits, which may be well after the timeout. done may be called twice and must be
+// idempotent.
+func getInfoWithChecksTimeout(ctx context.Context, timeout time.Duration, checks []*proto.Checks, done func(), excludeIPs ...netip.Addr) (*Info, bool) {
 	ctx, cancel := context.WithTimeout(ctx, timeout)
 	defer cancel()
 
 	infoCh := make(chan *Info, 1)
 	go func() {
-		info, err := GetInfoWithChecks(ctx, checks, excludeIPs...)
+		if done != nil {
+			defer done()
+		}
+		info, err := gatherInfoWithChecks(ctx, checks, excludeIPs...)
 		if err != nil {
 			if ctx.Err() != nil {
 				return
@@ -223,6 +240,9 @@ func GetInfoWithChecksTimeout(ctx context.Context, timeout time.Duration, checks
 
 	select {
 	case info := <-infoCh:
+		if done != nil {
+			done()
+		}
 		return info, true
 	case <-ctx.Done():
 		if errors.Is(ctx.Err(), context.DeadlineExceeded) {

client/system/info_source.go

@@ -4,35 +4,154 @@ import (
 	"context"
 	"net/netip"
 	"slices"
+	"sync"
 	"sync/atomic"
 	"time"
 
+	log "github.com/sirupsen/logrus"
+
 	"github.com/netbirdio/netbird/shared/management/proto"
 )
 
+const (
+	// infoLostAfter is how many timeouts an abandoned gathering is waited for before
+	// another may start beside it. A system call that is merely slow returns well within
+	// that; only one blocked for good, such as a stat on a dead network mount, is worked
+	// around.
+	infoLostAfter = 10
+
+	// maxAbandoned bounds how many gatherings may be left blocked, so a call that never
+	// returns cannot leak a goroutine on every refresh for the life of the daemon.
+	maxAbandoned = 2
+)
+
 // InfoSource gathers the system info sent to management, keeping the posture
-// check results from the last Refresh for the cheap Current snapshots.
+// check results from the last Refresh for the cheap Current snapshots. The zero value is
+// ready to use.
 type InfoSource struct {
-	files atomic.Pointer[[]File]
+	// started numbers Refresh calls in the order they began.
+	started atomic.Uint64
+	// now overrides the clock when set.
+	now func() time.Time
+
+	mu sync.Mutex
+	// abandoned holds the start of each gathering that timed out and is still blocked in
+	// a system call, by Refresh number.
+	abandoned map[uint64]time.Time
+	// files holds the file check results of the latest-started Refresh that succeeded,
+	// filesFrom its number.
+	files     []File
+	filesFrom uint64
 }
 
-// Refresh gathers the info with the posture checks evaluated, bounded by timeout.
+// Refresh gathers the info with the posture checks evaluated, bounded by timeout. It may
+// run concurrently with other calls; the results Current reuses are those of the call
+// that started last, so an older call finishing late cannot replace them. It reports
+// false on a timeout, and also while a gathering that timed out earlier is still blocked
+// in a system call: starting another would only add one more goroutine stuck on it.
+// That one is given up on once it has run for infoLostAfter timeouts, up to maxAbandoned.
 func (s *InfoSource) Refresh(ctx context.Context, timeout time.Duration, checks []*proto.Checks, excludeIPs ...netip.Addr) (*Info, bool) {
-	info, ok := GetInfoWithChecksTimeout(ctx, timeout, checks, excludeIPs...)
-	if !ok {
+	started := s.clock()
+	if !s.mayStart(started, time.Duration(infoLostAfter)*timeout) {
 		return nil, false
 	}
-	files := slices.Clone(info.Files)
-	s.files.Store(&files)
+	seq := s.started.Add(1)
+
+	var mu sync.Mutex
+	var finished, abandoned bool
+	done := func() {
+		mu.Lock()
+		defer mu.Unlock()
+		if finished {
+			return
+		}
+		finished = true
+		if abandoned {
+			s.release(seq)
+		}
+	}
+	info, ok := getInfoWithChecksTimeout(ctx, timeout, checks, done, excludeIPs...)
+	if !ok {
+		mu.Lock()
+		if !finished {
+			abandoned = true
+			s.abandon(seq, started)
+		}
+		mu.Unlock()
+		return nil, false
+	}
+	s.publishFiles(seq, info.Files)
 	return info, true
 }
 
+// mayStart reports whether a gathering may start at now. Every abandoned gathering must
+// have run for lost, and fewer than maxAbandoned may be left running.
+func (s *InfoSource) mayStart(now time.Time, lost time.Duration) bool {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	if len(s.abandoned) == 0 {
+		return true
+	}
+	if len(s.abandoned) >= maxAbandoned {
+		log.Warnf("%d system info gatherings are stuck, skipping this one", len(s.abandoned))
+		return false
+	}
+	for _, started := range s.abandoned {
+		if now.Sub(started) < lost {
+			log.Warnf("system info gathering that timed out earlier is still running, skipping this one")
+			return false
+		}
+	}
+	log.Warnf("system info gathering has been stuck for over %s, starting another", lost)
+	return true
+}
+
+// abandon records that gathering seq, started at started, timed out but is still running.
+func (s *InfoSource) abandon(seq uint64, started time.Time) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	if s.abandoned == nil {
+		s.abandoned = map[uint64]time.Time{}
+	}
+	s.abandoned[seq] = started
+}
+
+// release records that the abandoned gathering seq has returned.
+func (s *InfoSource) release(seq uint64) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	delete(s.abandoned, seq)
+}
+
+func (s *InfoSource) abandonedCount() int {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	return len(s.abandoned)
+}
+
+func (s *InfoSource) clock() time.Time {
+	if s.now != nil {
+		return s.now()
+	}
+	return time.Now()
+}
+
+func (s *InfoSource) publishFiles(seq uint64, files []File) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	if seq < s.filesFrom {
+		return
+	}
+	s.files = slices.Clone(files)
+	s.filesFrom = seq
+}
+
 // Current gathers the info without evaluating the checks, reusing the last Refresh results.
 func (s *InfoSource) Current(ctx context.Context, excludeIPs ...netip.Addr) *Info {
 	info := GetInfo(ctx)
 	info.removeAddresses(excludeIPs...)
-	if files := s.files.Load(); files != nil {
-		info.Files = *files
-	}
+	s.mu.Lock()
+	info.Files = slices.Clone(s.files)
+	s.mu.Unlock()
 	return info
 }

client/system/info_source_test.go

@@ -2,6 +2,7 @@ package system
 
 import (
 	"context"
+	"net/netip"
 	"os"
 	"path/filepath"
 	"testing"
@@ -36,6 +37,106 @@ func TestInfoSource_CurrentReusesRefreshedFiles(t *testing.T) {
 	assert.Equal(t, refreshed.Files, info.Files)
 }
 
+// TestInfoSource_RefreshSkipsWhileEarlierGatheringRuns stands in for a gathering that
+// timed out and is still blocked in a system call: no second one starts on top of it,
+// and gathering works again once it exits.
+func TestInfoSource_RefreshSkipsWhileEarlierGatheringRuns(t *testing.T) {
+	var src InfoSource
+	src.abandoned = map[uint64]time.Time{1: time.Now()}
+
+	_, ok := src.Refresh(context.Background(), 15*time.Second, nil)
+	assert.False(t, ok, "no gathering starts while an earlier one is still running")
+
+	src.abandoned = nil
+	_, ok = src.Refresh(context.Background(), 15*time.Second, nil)
+	require.True(t, ok, "gathering runs once the earlier one exited")
+
+	_, ok = src.Refresh(context.Background(), 15*time.Second, nil)
+	assert.True(t, ok, "a gathering that finished in time does not block the next one")
+}
+
+// TestInfoSource_RefreshReleasesAfterTimedOutGatheringExits checks that a gathering that
+// timed out releases the source once its goroutine finishes, not before.
+func TestInfoSource_RefreshReleasesAfterTimedOutGatheringExits(t *testing.T) {
+	var src InfoSource
+
+	_, ok := src.Refresh(context.Background(), time.Nanosecond, nil)
+	require.False(t, ok, "gathering cannot finish within a nanosecond")
+
+	require.Eventually(t, func() bool { return src.abandonedCount() == 0 }, 10*time.Second, 10*time.Millisecond,
+		"the source is released when the abandoned gathering exits")
+	_, ok = src.Refresh(context.Background(), 15*time.Second, nil)
+	assert.True(t, ok, "gathering works again after the abandoned one exited")
+}
+
+// stubGathering replaces the gathering with one that reports a single file check for the
+// path the checks name, and blocks a gathering for blockedPath until release is closed.
+// entered receives once that gathering has started.
+func stubGathering(t *testing.T, blockedPath string) (entered chan struct{}, release chan struct{}) {
+	t.Helper()
+	entered = make(chan struct{}, 1)
+	release = make(chan struct{})
+	original := gatherInfoWithChecks
+	gatherInfoWithChecks = func(_ context.Context, checks []*proto.Checks, _ ...netip.Addr) (*Info, error) {
+		path := checks[0].Files[0]
+		if path == blockedPath {
+			entered <- struct{}{}
+			<-release
+		}
+		return &Info{Files: []File{{Path: path, Exist: true}}}, nil
+	}
+	t.Cleanup(func() { gatherInfoWithChecks = original })
+	return entered, release
+}
+
+func filesCheck(path string) []*proto.Checks {
+	return []*proto.Checks{{Files: []string{path}}}
+}
+
+// TestInfoSource_RefreshRunsConcurrently: only a gathering that timed out holds off new
+// ones; a caller gathering while another gathering is in progress is served.
+func TestInfoSource_RefreshRunsConcurrently(t *testing.T) {
+	entered, release := stubGathering(t, "/slow")
+	var src InfoSource
+
+	slowDone := make(chan bool, 1)
+	go func() {
+		_, ok := src.Refresh(context.Background(), 15*time.Second, filesCheck("/slow"))
+		slowDone <- ok
+	}()
+	<-entered
+
+	_, ok := src.Refresh(context.Background(), 15*time.Second, filesCheck("/fast"))
+	assert.True(t, ok, "a gathering runs while another one is still in progress")
+
+	close(release)
+	assert.True(t, <-slowDone, "the slower gathering completes too")
+}
+
+// TestInfoSource_LateOlderRefreshDoesNotReplaceNewerResults: a Refresh for the previous
+// checks that finishes after one for the current checks must not bring back the old
+// results Current reports.
+func TestInfoSource_LateOlderRefreshDoesNotReplaceNewerResults(t *testing.T) {
+	entered, release := stubGathering(t, "/old")
+	var src InfoSource
+
+	oldDone := make(chan struct{})
+	go func() {
+		defer close(oldDone)
+		_, _ = src.Refresh(context.Background(), 15*time.Second, filesCheck("/old"))
+	}()
+	<-entered
+
+	_, ok := src.Refresh(context.Background(), 15*time.Second, filesCheck("/new"))
+	require.True(t, ok)
+	close(release)
+	<-oldDone
+
+	files := src.Current(context.Background()).Files
+	require.Len(t, files, 1)
+	assert.Equal(t, "/new", files[0].Path, "the results of the refresh that started last are kept")
+}
+
 func TestInfoSource_CurrentExcludesAddresses(t *testing.T) {
 	addrs := GetInfo(context.Background()).NetworkAddresses
 	if len(addrs) == 0 {
@@ -57,3 +158,41 @@ func TestInfoSource_CurrentExcludesAddresses(t *testing.T) {
 		assert.NotEqual(t, excluded, addr.NetIP.Addr())
 	}
 }
+
+// TestInfoSource_LostGatheringDoesNotBlockForever: a gathering blocked for good, such as
+// an os.Stat on a dead network mount, must not stop every later meta sync. Once it has
+// run for lostAfter timeouts another starts beside it, and no more than maxAbandoned
+// are ever left running.
+func TestInfoSource_LostGatheringDoesNotBlockForever(t *testing.T) {
+	entered, release := stubGathering(t, "/wedged")
+	t.Cleanup(func() { close(release) })
+	now := time.Now()
+	src := InfoSource{now: func() time.Time { return now }}
+	const timeout = 10 * time.Millisecond
+	lost := time.Duration(infoLostAfter) * timeout
+
+	_, ok := src.Refresh(context.Background(), timeout, filesCheck("/wedged"))
+	require.False(t, ok, "the wedged gathering times out")
+	<-entered
+
+	_, ok = src.Refresh(context.Background(), timeout, filesCheck("/wedged"))
+	assert.False(t, ok, "a gathering that may still finish holds off the next one")
+
+	now = now.Add(lost)
+	_, ok = src.Refresh(context.Background(), timeout, filesCheck("/wedged"))
+	require.False(t, ok, "the second wedged gathering times out too")
+	select {
+	case <-entered:
+	case <-time.After(time.Second):
+		t.Fatal("once the first gathering is lost another one starts")
+	}
+
+	now = now.Add(lost)
+	_, ok = src.Refresh(context.Background(), timeout, filesCheck("/wedged"))
+	assert.False(t, ok, "no more than maxAbandoned gatherings are left running")
+	select {
+	case <-entered:
+		t.Fatal("a third gathering started beside two lost ones")
+	case <-time.After(50 * time.Millisecond):
+	}
+}