fix(brain): context, brain_get and brain_read read through brainfs (#157, slice 2) (#165)

readExcerpt, ReadNote and ReadScoped now read through brainfs (os.Root, regular files only, type checked on the opened handle), so a symlink swapped in after a check is not followed and a FIFO does not hang the reader. ReadRegularMax bounds the read itself with io.LimitReader(max+1). notes.Target carries the brain root.

Refs #157. Adversarial review round 2: PASS (round 1 BLOCKED: unbounded read, rule-9 test names).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Mathias <mthbqv@gmail.com>

ingestion/internal/brainfs/brainfs.go

@@ -20,6 +20,9 @@ import (
 // ErrNotRegular is returned for an entry that is not a regular file.
 var ErrNotRegular = errors.New("not a regular file")
 
+// ErrTooLarge is returned by ReadRegularMax for a file over the limit.
+var ErrTooLarge = errors.New("file too large")
+
 // Brain is an open brain directory.
 type Brain struct {
 	root *os.Root
@@ -74,7 +77,22 @@ func (b *Brain) ReadRegular(rel string) ([]byte, error) {
 // cannot create one; fs.protected_hardlinks=1 and the pod's secrets are on a
 // different filesystem from the brain volume, so a local caller cannot link
 // them in. Refusing Nlink > 1 is the next step if either changes.
-func (b *Brain) readOpened(rel string) ([]byte, error) {
+func (b *Brain) readOpened(rel string) ([]byte, error) { return b.readOpenedMax(rel, -1) }
+
+// ReadRegularMax is ReadRegular with a size limit, checked on the opened
+// handle before reading (max < 0 means no limit).
+func (b *Brain) ReadRegularMax(rel string, max int64) ([]byte, error) {
+	fi, err := b.root.Lstat(rel)
+	if err != nil {
+		return nil, err
+	}
+	if !fi.Mode().IsRegular() {
+		return nil, fmt.Errorf("%s: %w", rel, ErrNotRegular)
+	}
+	return b.readOpenedMax(rel, max)
+}
+
+func (b *Brain) readOpenedMax(rel string, max int64) ([]byte, error) {
 	f, err := b.root.OpenFile(rel, os.O_RDONLY|syscall.O_NONBLOCK, 0)
 	if err != nil {
 		return nil, err
@@ -87,9 +105,31 @@ func (b *Brain) readOpened(rel string) ([]byte, error) {
 	if !hfi.Mode().IsRegular() {
 		return nil, fmt.Errorf("%s: %w", rel, ErrNotRegular)
 	}
-	return io.ReadAll(f)
+	if max >= 0 && hfi.Size() > max {
+		return nil, fmt.Errorf("%s is %d bytes (limit %d): %w", rel, hfi.Size(), max, ErrTooLarge)
+	}
+	var r io.Reader = f
+	if wrapAfterStat != nil {
+		r = wrapAfterStat(r)
+	}
+	if max < 0 {
+		return io.ReadAll(r)
+	}
+	// The handle's size can grow after the check; bound the read itself.
+	data, err := io.ReadAll(io.LimitReader(r, max+1))
+	if err != nil {
+		return nil, err
+	}
+	if int64(len(data)) > max {
+		return nil, fmt.Errorf("%s grew past %d bytes while read: %w", rel, max, ErrTooLarge)
+	}
+	return data, nil
 }
 
+// wrapAfterStat is a test seam: it runs between the size check and the read,
+// and wraps the reader the read consumes.
+var wrapAfterStat func(io.Reader) io.Reader
+
 // WalkRegular walks a brain-relative directory through the root and calls fn
 // for each regular file (directories are descended; symlinks, FIFOs and
 // devices are skipped, never followed). An entry error is passed to onErr and

ingestion/internal/brainfs/brainfs_test.go

@@ -2,6 +2,7 @@ package brainfs
 
 import (
 	"errors"
+	"io"
 	"os"
 	"path/filepath"
 	"syscall"
@@ -90,3 +91,51 @@ func TestWalkRegularYieldsOnlyRegularFiles(t *testing.T) {
 	require.NoError(t, b.WalkRegular(".", func(rel string) error { got = append(got, rel); return nil }, nil))
 	assert.Equal(t, []string{"a/n.md"}, got)
 }
+
+func TestReadRegularMaxRefusesAFileOverTheLimit(t *testing.T) {
+	dir := t.TempDir()
+	require.NoError(t, os.WriteFile(filepath.Join(dir, "big.md"), make([]byte, 100), 0o644))
+	b, err := Open(dir)
+	require.NoError(t, err)
+	defer func() { _ = b.Close() }()
+	_, err = b.ReadRegularMax("big.md", 99)
+	assert.True(t, errors.Is(err, ErrTooLarge), "got %v", err)
+	got, err := b.ReadRegularMax("big.md", 100)
+	require.NoError(t, err)
+	assert.Len(t, got, 100)
+}
+
+func TestReadRegularMaxRefusesAndStopsReadingAFileThatGrowsPastTheLimit(t *testing.T) {
+	dir := t.TempDir()
+	p := filepath.Join(dir, "grow.md")
+	require.NoError(t, os.WriteFile(p, make([]byte, 10), 0o644))
+	consumed := &countingReader{}
+	wrapAfterStat = func(r io.Reader) io.Reader {
+		f, err := os.OpenFile(p, os.O_WRONLY|os.O_APPEND, 0)
+		require.NoError(t, err)
+		_, err = f.Write(make([]byte, 1000))
+		require.NoError(t, err)
+		require.NoError(t, f.Close())
+		consumed.r = r
+		return consumed
+	}
+	t.Cleanup(func() { wrapAfterStat = nil })
+	b, err := Open(dir)
+	require.NoError(t, err)
+	defer func() { _ = b.Close() }()
+	got, err := b.ReadRegularMax("grow.md", 100)
+	assert.True(t, errors.Is(err, ErrTooLarge), "got %v", err)
+	assert.Nil(t, got)
+	assert.LessOrEqual(t, consumed.n, 101, "read past max+1 bytes")
+}
+
+type countingReader struct {
+	r io.Reader
+	n int
+}
+
+func (c *countingReader) Read(p []byte) (int, error) {
+	n, err := c.r.Read(p)
+	c.n += n
+	return n, err
+}

ingestion/internal/brainfs/brainfstest/brainfstest.go

@@ -0,0 +1,103 @@
+// Package brainfstest is the shared harness for hyperguild#157's swap-race
+// tests: it keeps swapping a regular note, a symlink to an outside file, or a
+// FIFO (with a writer held open) into a path while a reader under test loops.
+// It observes two things: no sentinel byte from the outside file reaches the
+// reader's output, and the reader returns within a timeout. It does not cover
+// hard links, other paths into the reader, or races it does not generate.
+package brainfstest
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"sync"
+	"sync/atomic"
+	"syscall"
+	"testing"
+	"time"
+)
+
+// Sentinel marks the outside file's contents.
+const Sentinel = "zqbrainfssentinelonefiftyseven"
+
+// Outside writes a file outside any brain dir containing Sentinel.
+func Outside(t *testing.T) string {
+	t.Helper()
+	p := filepath.Join(t.TempDir(), "secret.md")
+	if err := os.WriteFile(p, []byte("# secret\n\n"+Sentinel), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	return p
+}
+
+// StartSwap renames, in a loop, a regular note (content inside) and either
+// a symlink to outside or a FIFO into path, until the test ends. FIFOs get a
+// writer held open until the swapper exits, so a reader that skips the
+// handle check parks forever instead of reading EOF. The swapper is stopped
+// and awaited before t.TempDir cleanup.
+func StartSwap(t *testing.T, path, inside, outside string, fifo bool) {
+	t.Helper()
+	var stop atomic.Bool
+	var wg sync.WaitGroup
+	wg.Add(1)
+	go func() {
+		defer wg.Done()
+		tmp := path + ".swap"
+		var writers []*os.File
+		defer func() {
+			for _, w := range writers {
+				_ = w.Close()
+			}
+		}()
+		for i := 0; !stop.Load(); i++ {
+			_ = os.Remove(tmp)
+			switch {
+			case i%2 == 0:
+				_ = os.WriteFile(tmp, []byte(inside), 0o644)
+			case fifo:
+				if err := syscall.Mkfifo(tmp, 0o644); err == nil {
+					if w, err := os.OpenFile(tmp, os.O_RDWR, 0); err == nil {
+						writers = append(writers, w)
+					}
+				}
+			default:
+				_ = os.Symlink(outside, tmp)
+			}
+			_ = os.Rename(tmp, path)
+		}
+	}()
+	t.Cleanup(func() { stop.Store(true); wg.Wait() })
+}
+
+// AssertNoLeak loops read for d while a symlink swapper runs on path, and
+// fails if any output contains Sentinel.
+func AssertNoLeak(t *testing.T, path, inside string, d time.Duration, read func() string) {
+	t.Helper()
+	StartSwap(t, path, inside, Outside(t), false)
+	leaks, iters := 0, 0
+	for deadline := time.Now().Add(d); time.Now().Before(deadline); iters++ {
+		leaks += strings.Count(read(), Sentinel)
+	}
+	if leaks > 0 {
+		t.Fatalf("outside file leaked %d times in %d iterations", leaks, iters)
+	}
+}
+
+// AssertNoHang loops read for d while a FIFO swapper (held writers) runs on
+// path, and fails if a read does not return within timeout.
+func AssertNoHang(t *testing.T, path, inside string, d, timeout time.Duration, read func()) {
+	t.Helper()
+	StartSwap(t, path, inside, Outside(t), true)
+	done := make(chan struct{})
+	go func() {
+		for deadline := time.Now().Add(d); time.Now().Before(deadline); {
+			read()
+		}
+		close(done)
+	}()
+	select {
+	case <-done:
+	case <-time.After(timeout):
+		t.Fatal("reader hung on a swapped-in FIFO with a writer held open")
+	}
+}