feat(ingestion): symlink tripwire for the Syncthing-shared brain dir (#157) (#159)

Co-authored-by: Mathias <mthbqv@gmail.com>

ingestion/cmd/server/main.go

@@ -33,6 +33,7 @@ import (
 	"github.com/mathiasbq/hyperguild/ingestion/internal/reranker"
 	"github.com/mathiasbq/hyperguild/ingestion/internal/search"
 	"github.com/mathiasbq/hyperguild/ingestion/internal/startupretry"
+	"github.com/mathiasbq/hyperguild/ingestion/internal/symlinkguard"
 	"github.com/mathiasbq/hyperguild/ingestion/internal/vectorstore"
 	"github.com/mathiasbq/hyperguild/ingestion/internal/watcher"
 	"github.com/mathiasbq/hyperguild/ingestion/internal/webhook"
@@ -173,7 +174,8 @@ func alertingWarnings(getenv func(string) string) []string {
 	var out []string
 	if getenv("BRAIN_NTFY_URL") == "" {
 		out = append(out, "alerting DISABLED: BRAIN_NTFY_URL unset — "+
-			"brain_edit_root governance rewrites and capture audit degradation will not notify")
+			"brain_edit_root governance rewrites, capture audit degradation and the "+
+			"brain symlink tripwire (hyperguild#157) will not notify")
 	}
 	return out
 }
@@ -239,6 +241,10 @@ func main() {
 	}
 
 	brainDir := envOr("INGEST_BRAIN_DIR", "../brain")
+	if err := startupSymlinkCheck(brainDir); err != nil {
+		logger.Error("brain symlink tripwire", "err", err)
+		os.Exit(1)
+	}
 	port := envOr("INGEST_PORT", "3300")
 
 	llmURL := envOr("INGEST_LLM_URL", "http://iguana:4000/v1")
@@ -360,6 +366,8 @@ func main() {
 	}
 
 	ctx := context.Background()
+	// hyperguild#157: report a symlink that appears in the brain while running.
+	go symlinkguard.Watch(ctx, brainDir, symlinkScanInterval(os.Getenv), buildNotifier())
 
 	// Built before the watcher, the graph backfill and session ingestion, and
 	// retried rather than fatal (#83). Two separate reasons:

ingestion/cmd/server/symlink_check.go

@@ -0,0 +1,39 @@
+package main
+
+import (
+	"errors"
+	"fmt"
+	"os"
+	"strconv"
+	"time"
+
+	"github.com/mathiasbq/hyperguild/ingestion/internal/symlinkguard"
+)
+
+// startupSymlinkCheck is the hyperguild#157 tripwire at boot: a symlink in the
+// brain dir would let brain readers serve a file from outside the brain, so
+// the server does not start until it is removed. A directory the scan cannot
+// read could hide one, so it refuses on that too. A missing brain dir (first
+// boot) is the only case it lets pass.
+func startupSymlinkCheck(brainDir string) error {
+	r, err := symlinkguard.Scan(brainDir)
+	if err != nil {
+		if errors.Is(err, os.ErrNotExist) {
+			return nil
+		}
+		return fmt.Errorf("refusing to start: cannot scan the brain dir %s (hyperguild#157): %w", brainDir, err)
+	}
+	if !r.Clean() {
+		return fmt.Errorf("refusing to start: brain dir %s has %s (hyperguild#157); fix them first", brainDir, r)
+	}
+	return nil
+}
+
+// symlinkScanInterval reads BRAIN_SYMLINK_SCAN_INTERVAL (seconds), falling
+// back to the default on an unset, non-numeric or non-positive value.
+func symlinkScanInterval(getenv func(string) string) time.Duration {
+	if s, err := strconv.Atoi(getenv("BRAIN_SYMLINK_SCAN_INTERVAL")); err == nil && s > 0 {
+		return time.Duration(s) * time.Second
+	}
+	return symlinkguard.DefaultInterval
+}

ingestion/cmd/server/symlink_check_test.go

@@ -0,0 +1,65 @@
+package main
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// hyperguild#157: the server refuses to start while a symlink exists in the
+// brain dir, or while any part of it cannot be scanned (which could hide
+// one); a missing brain dir is the only thing it lets pass.
+func TestStartupSymlinkCheckRefusesABrainWithASymlink(t *testing.T) {
+	brain := t.TempDir()
+	require.NoError(t, os.Symlink(t.TempDir(), filepath.Join(brain, "planted")))
+	err := startupSymlinkCheck(brain)
+	require.Error(t, err)
+	assert.Contains(t, err.Error(), "planted")
+}
+
+func TestStartupSymlinkCheckRefusesABrainWithAnUnreadableDir(t *testing.T) {
+	if os.Geteuid() == 0 {
+		t.Skip("root reads mode-000 directories")
+	}
+	brain := t.TempDir()
+	locked := filepath.Join(brain, "aaa")
+	require.NoError(t, os.MkdirAll(locked, 0o755))
+	require.NoError(t, os.Chmod(locked, 0o000))
+	t.Cleanup(func() { _ = os.Chmod(locked, 0o755) })
+	err := startupSymlinkCheck(brain)
+	require.Error(t, err)
+	assert.Contains(t, err.Error(), "aaa")
+}
+
+func TestStartupSymlinkCheckPassesACleanOrMissingBrain(t *testing.T) {
+	brain := t.TempDir()
+	require.NoError(t, os.WriteFile(filepath.Join(brain, "n.md"), []byte("x"), 0o644))
+	assert.NoError(t, startupSymlinkCheck(brain))
+	assert.NoError(t, startupSymlinkCheck(filepath.Join(brain, "missing")))
+}
+
+func TestSymlinkScanIntervalFallsBackOnNonPositive(t *testing.T) {
+	for _, v := range []string{"0", "-5", "", "junk"} {
+		got := symlinkScanInterval(func(string) string { return v })
+		assert.Positive(t, got, "env %q", v)
+	}
+}
+
+func TestAlertingWarningsNamesTheSymlinkTripwire(t *testing.T) {
+	w := strings.Join(alertingWarnings(func(string) string { return "" }), " ")
+	assert.Contains(t, w, "symlink tripwire")
+}
+
+func TestStartupSymlinkCheckRefusesAnUnlistableBrainRoot(t *testing.T) {
+	if os.Geteuid() == 0 {
+		t.Skip("root lists mode-0100 directories")
+	}
+	brain := t.TempDir()
+	require.NoError(t, os.Chmod(brain, 0o100))
+	t.Cleanup(func() { _ = os.Chmod(brain, 0o755) })
+	assert.Error(t, startupSymlinkCheck(brain))
+}

ingestion/internal/symlinkguard/symlinkguard.go

@@ -0,0 +1,153 @@
+// Package symlinkguard is the hyperguild#157 tripwire. The brain directory is
+// a Syncthing sendreceive folder and Syncthing syncs symlinks; several brain
+// readers (search, vectorstore, context, graph) follow a symlink out of the
+// brain and would index or return whatever it points to. Until every reader
+// reads through an os.Root, a symlink inside the brain, or a directory the
+// scan cannot read (which could hide one), is an incident: the server refuses
+// to start on one (main), and Watch reports one that appears while it runs.
+package symlinkguard
+
+import (
+	"context"
+	"fmt"
+	"io/fs"
+	"log/slog"
+	"os"
+	"path/filepath"
+	"sort"
+	"strconv"
+	"strings"
+	"time"
+)
+
+// DefaultInterval is used when Watch is given a non-positive interval.
+const DefaultInterval = 5 * time.Minute
+
+// Notifier is the out-of-band alert sink (audit.Notifier satisfies it).
+type Notifier interface {
+	Notify(ctx context.Context, msg string) error
+}
+
+// Result is what a scan found: symlinks, and directories it could not read
+// (each of which could hide a symlink). Paths are brain-relative.
+type Result struct {
+	Symlinks   []string
+	Unreadable []string
+}
+
+// Clean reports whether the scan found nothing to act on.
+func (r Result) Clean() bool { return len(r.Symlinks) == 0 && len(r.Unreadable) == 0 }
+
+// String describes the findings for a log line or alert. Paths are quoted, so
+// a name containing a newline cannot blur two findings together.
+func (r Result) String() string {
+	q := func(xs []string) string {
+		out := make([]string, len(xs))
+		for i, x := range xs {
+			out[i] = strconv.Quote(x)
+		}
+		return strings.Join(out, ", ")
+	}
+	var parts []string
+	if len(r.Symlinks) > 0 {
+		parts = append(parts, fmt.Sprintf("%d symlink(s): %s", len(r.Symlinks), q(r.Symlinks)))
+	}
+	if len(r.Unreadable) > 0 {
+		parts = append(parts, fmt.Sprintf("%d unreadable dir(s): %s", len(r.Unreadable), q(r.Unreadable)))
+	}
+	return strings.Join(parts, "; ")
+}
+
+// Scan walks the brain dir and reports every symlink in it and every
+// directory it could not read, continuing past both. Only the ROOT .git is
+// skipped (it is brain-sync's, and no reader walks it); a nested .git is
+// scanned like any other directory. brainDir itself may be a symlink (a
+// volume mount); it is resolved first. The error is non-nil when the root
+// cannot be resolved (wrapping os.ErrNotExist when it is missing) or cannot be
+// listed.
+func Scan(brainDir string) (Result, error) {
+	root, err := filepath.EvalSymlinks(brainDir)
+	if err != nil {
+		return Result{}, fmt.Errorf("resolve brain dir: %w", err)
+	}
+	var r Result
+	rootGit := filepath.Join(root, ".git")
+	walkErr := filepath.WalkDir(root, func(p string, d fs.DirEntry, err error) error {
+		rel, _ := filepath.Rel(root, p)
+		rel = filepath.ToSlash(rel)
+		if err != nil {
+			if p == root {
+				return err
+			}
+			r.Unreadable = append(r.Unreadable, rel)
+			if d != nil && d.IsDir() {
+				return filepath.SkipDir
+			}
+			return nil
+		}
+		if d.IsDir() && p == rootGit {
+			return filepath.SkipDir
+		}
+		if d.Type()&os.ModeSymlink != 0 {
+			r.Symlinks = append(r.Symlinks, rel)
+		}
+		return nil
+	})
+	sort.Strings(r.Symlinks)
+	sort.Strings(r.Unreadable)
+	if walkErr != nil {
+		// Only the root's own error reaches here (deeper errors are recorded
+		// as Unreadable). A root that cannot be listed cannot be shown clean
+		// (PR #159 round 2: discarding this reported clean while search
+		// served a planted symlink).
+		return r, fmt.Errorf("scan brain dir: %w", walkErr)
+	}
+	return r, nil
+}
+
+// Watch rescans every interval (DefaultInterval if non-positive) until ctx
+// ends, and notifies once per distinct non-clean result. A result counts as
+// reported only after Notify succeeds, so an alert lost to a notifier outage
+// is retried on the next tick.
+func Watch(ctx context.Context, brainDir string, interval time.Duration, n Notifier) {
+	if interval <= 0 {
+		interval = DefaultInterval
+	}
+	t := time.NewTicker(interval)
+	defer t.Stop()
+	reported := ""
+	for {
+		select {
+		case <-ctx.Done():
+			return
+		case <-t.C:
+		}
+		r, err := Scan(brainDir)
+		var key, msg string
+		switch {
+		case err != nil:
+			key = "error: " + err.Error()
+			msg = "brain symlink tripwire (hyperguild#157): cannot scan the brain dir: " + err.Error()
+		case r.Clean():
+			reported = ""
+			continue
+		default:
+			key = r.String()
+			msg = "brain symlink tripwire (hyperguild#157): " + key +
+				". Readers may follow symlinks out of the brain; remove them."
+		}
+		if key == reported {
+			continue
+		}
+		slog.Error("brain symlink tripwire", "finding", key)
+		if n == nil {
+			reported = key
+			continue
+		}
+		if err := n.Notify(ctx, msg); err != nil {
+			slog.Warn("symlink tripwire notify failed; will retry", "err", err)
+			continue
+		}
+		reported = key
+	}
+}

ingestion/internal/symlinkguard/symlinkguard_test.go

@@ -0,0 +1,189 @@
+package symlinkguard
+
+import (
+	"context"
+	"errors"
+	"os"
+	"path/filepath"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// hyperguild#157: the brain dir is a Syncthing sendreceive folder and
+// Syncthing syncs symlinks; several readers follow them out of the brain.
+// Until every reader goes through an os.Root, a symlink anywhere in the brain,
+// or a directory the scan cannot read (which could hide one), is an incident.
+
+func TestScanReportsSymlinksAtEveryDepthIncludingNestedGit(t *testing.T) {
+	brain := t.TempDir()
+	outside := t.TempDir()
+	require.NoError(t, os.MkdirAll(filepath.Join(brain, "wiki", "w", "facts", ".git"), 0o755))
+	require.NoError(t, os.MkdirAll(filepath.Join(brain, ".git", "objects"), 0o755))
+	require.NoError(t, os.Symlink(filepath.Join(outside, "f"), filepath.Join(brain, "wiki", "w", "facts", "leak.md")))
+	require.NoError(t, os.Symlink(outside, filepath.Join(brain, "wiki", "dirlink")))
+	// PR #159 review: only the ROOT .git is skipped; a nested .git is not
+	// special to the brain readers, which walk into it.
+	require.NoError(t, os.Symlink(filepath.Join(outside, "g"), filepath.Join(brain, "wiki", "w", "facts", ".git", "leak.md")))
+	require.NoError(t, os.Symlink(outside, filepath.Join(brain, ".git", "objects", "alt")))
+
+	r, err := Scan(brain)
+	require.NoError(t, err)
+	assert.ElementsMatch(t, []string{"wiki/w/facts/leak.md", "wiki/dirlink", "wiki/w/facts/.git/leak.md"}, r.Symlinks)
+	assert.Empty(t, r.Unreadable)
+}
+
+// PR #159 review: WalkDir stopped at the first unreadable directory and the
+// startup check passed, while search (which keeps walking) served a planted
+// symlink. Scan now records the unreadable dir and keeps going.
+func TestScanReportsUnreadableDirsAndKeepsWalkingPastThem(t *testing.T) {
+	if os.Geteuid() == 0 {
+		t.Skip("root reads mode-000 directories")
+	}
+	brain := t.TempDir()
+	locked := filepath.Join(brain, "aaa")
+	require.NoError(t, os.MkdirAll(locked, 0o755))
+	require.NoError(t, os.Chmod(locked, 0o000))
+	t.Cleanup(func() { _ = os.Chmod(locked, 0o755) })
+	require.NoError(t, os.Symlink(t.TempDir(), filepath.Join(brain, "zzz-planted")))
+
+	r, err := Scan(brain)
+	require.NoError(t, err)
+	assert.Equal(t, []string{"aaa"}, r.Unreadable)
+	assert.Equal(t, []string{"zzz-planted"}, r.Symlinks, "the walk continues past the unreadable dir")
+}
+
+func TestScanResolvesASymlinkedRootAndReportsSymlinkedEntriesInsideIt(t *testing.T) {
+	real := t.TempDir()
+	link := filepath.Join(t.TempDir(), "brain")
+	require.NoError(t, os.Symlink(real, link))
+	require.NoError(t, os.WriteFile(filepath.Join(real, "n.md"), []byte("x"), 0o644))
+	require.NoError(t, os.Symlink(t.TempDir(), filepath.Join(real, "inner")))
+	r, err := Scan(link)
+	require.NoError(t, err)
+	assert.Equal(t, []string{"inner"}, r.Symlinks)
+}
+
+func TestScanOfAMissingRootIsErrNotExist(t *testing.T) {
+	_, err := Scan(filepath.Join(t.TempDir(), "nope"))
+	assert.True(t, errors.Is(err, os.ErrNotExist), "got %v", err)
+}
+
+type recNotifier struct {
+	mu   sync.Mutex
+	msgs []string
+	fail int // fail the first N calls
+}
+
+func (r *recNotifier) Notify(_ context.Context, m string) error {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+	r.msgs = append(r.msgs, m)
+	if r.fail > 0 {
+		r.fail--
+		return errors.New("ntfy down")
+	}
+	return nil
+}
+
+func (r *recNotifier) count() int { r.mu.Lock(); defer r.mu.Unlock(); return len(r.msgs) }
+
+func TestWatchNotifiesOncePerDistinctSetAndAgainForANewSet(t *testing.T) {
+	brain := t.TempDir()
+	n := &recNotifier{}
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	go Watch(ctx, brain, 10*time.Millisecond, n)
+
+	time.Sleep(50 * time.Millisecond)
+	assert.Zero(t, n.count(), "a clean brain raises nothing")
+	require.NoError(t, os.Symlink(t.TempDir(), filepath.Join(brain, "first")))
+	require.Eventually(t, func() bool { return n.count() == 1 }, 2*time.Second, 10*time.Millisecond)
+	time.Sleep(60 * time.Millisecond)
+	assert.Equal(t, 1, n.count(), "the same set is reported once, not every tick")
+	require.NoError(t, os.Symlink(t.TempDir(), filepath.Join(brain, "second")))
+	require.Eventually(t, func() bool { return n.count() == 2 }, 2*time.Second, 10*time.Millisecond)
+}
+
+// PR #159 review: last was recorded before Notify, so an alert lost to an
+// ntfy outage was never retried while the symlink stayed.
+func TestWatchRetriesAnAlertThatFailedToSend(t *testing.T) {
+	brain := t.TempDir()
+	require.NoError(t, os.Symlink(t.TempDir(), filepath.Join(brain, "planted")))
+	n := &recNotifier{fail: 2}
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	go Watch(ctx, brain, 10*time.Millisecond, n)
+	require.Eventually(t, func() bool { return n.count() == 3 }, 2*time.Second, 10*time.Millisecond)
+	time.Sleep(60 * time.Millisecond)
+	assert.Equal(t, 3, n.count(), "after one successful send, no more")
+}
+
+func TestWatchAlertsOnAnUnreadableDir(t *testing.T) {
+	if os.Geteuid() == 0 {
+		t.Skip("root reads mode-000 directories")
+	}
+	brain := t.TempDir()
+	locked := filepath.Join(brain, "locked")
+	require.NoError(t, os.MkdirAll(locked, 0o755))
+	require.NoError(t, os.Chmod(locked, 0o000))
+	t.Cleanup(func() { _ = os.Chmod(locked, 0o755) })
+	n := &recNotifier{}
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	go Watch(ctx, brain, 10*time.Millisecond, n)
+	require.Eventually(t, func() bool { return n.count() == 1 }, 2*time.Second, 10*time.Millisecond)
+	assert.Contains(t, n.msgs[0], "locked")
+}
+
+// PR #159 review: a non-positive interval panicked time.NewTicker inside the
+// goroutine and crashlooped the server.
+func TestWatchToleratesANonPositiveInterval(t *testing.T) {
+	brain := t.TempDir()
+	ctx, cancel := context.WithCancel(context.Background())
+	done := make(chan struct{})
+	go func() { Watch(ctx, brain, 0, nil); close(done) }()
+	cancel()
+	select {
+	case <-done:
+	case <-time.After(2 * time.Second):
+		t.Fatal("Watch did not return on cancel")
+	}
+}
+
+// PR #159 round 2: a brain root that can be entered but not listed (mode
+// 0100) made WalkDir fail at the root; the error was discarded and Scan
+// reported clean, while search (which goes straight into wiki/) still served
+// a planted symlink. A root that cannot be listed is now an error.
+func TestScanOfAnUnlistableRootIsAnError(t *testing.T) {
+	if os.Geteuid() == 0 {
+		t.Skip("root lists mode-0100 directories")
+	}
+	brain := t.TempDir()
+	require.NoError(t, os.MkdirAll(filepath.Join(brain, "wiki"), 0o755))
+	require.NoError(t, os.Symlink(t.TempDir(), filepath.Join(brain, "wiki", "leak.md")))
+	require.NoError(t, os.Chmod(brain, 0o100))
+	t.Cleanup(func() { _ = os.Chmod(brain, 0o755) })
+	_, err := Scan(brain)
+	require.Error(t, err)
+	assert.False(t, errors.Is(err, os.ErrNotExist), "an unlistable root is not a missing root")
+}
+
+// The same set that disappears and comes back is reported again.
+func TestWatchReportsASetAgainAfterItWasCleared(t *testing.T) {
+	brain := t.TempDir()
+	link := filepath.Join(brain, "planted")
+	n := &recNotifier{}
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	go Watch(ctx, brain, 10*time.Millisecond, n)
+	require.NoError(t, os.Symlink(t.TempDir(), link))
+	require.Eventually(t, func() bool { return n.count() == 1 }, 2*time.Second, 10*time.Millisecond)
+	require.NoError(t, os.Remove(link))
+	time.Sleep(60 * time.Millisecond)
+	require.NoError(t, os.Symlink(t.TempDir(), link))
+	require.Eventually(t, func() bool { return n.count() == 2 }, 2*time.Second, 10*time.Millisecond)
+}