fusefrontend: sharedstorage: use byte-range lock on file header creation

Multiple host writing to the same empty file at the same time
could have overwritten each other's newly created file header,
leading to data corruption.

Fix the race by placing a byte-range lock on the file when
creating the file header.

2 files changed, 124 insertions, 2 deletions

diff --git a/tests/cluster/cluster_test.go b/tests/cluster/cluster_test.go
index 2e969ce..f9d0903 100644
--- a/tests/cluster/cluster_test.go
+++ b/tests/cluster/cluster_test.go
@@ -7,9 +7,12 @@ package cluster_test
 
 import (
 	"bytes"
+	"errors"
+	"io"
 	"math/rand"
 	"os"
 	"sync"
+	"syscall"
 	"testing"
 
 	"github.com/rfjakob/gocryptfs/v2/tests/test_helpers"
@@ -27,8 +30,8 @@ import (
 // > buffered read/write.
 //
 // See also:
-// * https://lore.kernel.org/linux-xfs/20190325001044.GA23020@dastard/
-//   Dave Chinner: XFS is the only linux filesystem that provides this behaviour.
+//   - https://lore.kernel.org/linux-xfs/20190325001044.GA23020@dastard/
+//     Dave Chinner: XFS is the only linux filesystem that provides this behaviour.
 func TestClusterConcurrentRW(t *testing.T) {
 	if os.Getenv("ENABLE_CLUSTER_TEST") != "1" {
 		t.Skipf("This test is disabled by default because it fails unless on XFS.\n" +
@@ -109,3 +112,75 @@ func TestClusterConcurrentRW(t *testing.T) {
 	go readThread(f2)
 	wg.Wait()
 }
+
+// Multiple hosts creating the same file at the same time could
+// overwrite each other's file header, leading to data
+// corruption. Passing "-sharedstorage" should prevent this.
+func TestConcurrentCreate(t *testing.T) {
+	cDir := test_helpers.InitFS(t)
+	mnt1 := cDir + ".mnt1"
+	mnt2 := cDir + ".mnt2"
+	test_helpers.MountOrFatal(t, cDir, mnt1, "-extpass=echo test", "-wpanic=0", "-sharedstorage")
+	defer test_helpers.UnmountPanic(mnt1)
+	test_helpers.MountOrFatal(t, cDir, mnt2, "-extpass=echo test", "-wpanic=0", "-sharedstorage")
+	defer test_helpers.UnmountPanic(mnt2)
+
+	var wg sync.WaitGroup
+	const loops = 10000
+
+	createOrOpen := func(path string) (f *os.File, err error) {
+		// Use the high-level os.Create/OpenFile instead of syscall.Open because we
+		// *want* Go's EINTR retry logic. glibc open(2) has similar logic.
+		f, err = os.OpenFile(path, os.O_CREATE|os.O_RDWR|os.O_EXCL, 0600)
+		if err == nil {
+			return
+		}
+		if !errors.Is(err, os.ErrExist) {
+			t.Logf("POSIX compliance issue: exclusive create failed with unexpected error: err=%v", errors.Unwrap(err))
+		}
+		f, err = os.OpenFile(path, os.O_CREATE|os.O_RDWR, 0600)
+		if err == nil {
+			return
+		}
+		t.Logf("POSIX compliance issue: non-exlusive create failed with err=%v", errors.Unwrap(err))
+		return
+	}
+
+	workerThread := func(path string) {
+		defer wg.Done()
+		buf := make([]byte, 10)
+		for i := 0; i < loops; i++ {
+			if t.Failed() {
+				return
+			}
+			f, err := createOrOpen(path)
+			if err != nil {
+				// retry
+				continue
+			}
+			defer f.Close()
+			_, err = f.WriteAt(buf, 0)
+			if err != nil {
+				t.Errorf("iteration %d: Pwrite: %v", i, err)
+				return
+			}
+			buf2 := make([]byte, len(buf)+1)
+			n, err := f.ReadAt(buf2, 0)
+			if err != nil && err != io.EOF {
+				t.Errorf("iteration %d: ReadAt: %v", i, err)
+				return
+			}
+			buf2 = buf2[:n]
+			if !bytes.Equal(buf, buf2) {
+				t.Errorf("iteration %d: corrupt data received: %x", i, buf2)
+				return
+			}
+			syscall.Unlink(path)
+		}
+	}
+
+	wg.Add(2)
+	go workerThread(mnt1 + "/foo")
+	go workerThread(mnt2 + "/foo")
+	wg.Wait()
+}
diff --git a/tests/cluster/poc_test.go b/tests/cluster/poc_test.go
new file mode 100644
index 0000000..4d7182a
--- /dev/null
+++ b/tests/cluster/poc_test.go
@@ -0,0 +1,47 @@
+package cluster
+
+// poc_test.go contains proof of concept tests for the byte-range locking logic.
+// This goes directly to an underlying filesystem without going through gocryptfs.
+
+import (
+	"syscall"
+	"testing"
+
+	"golang.org/x/sys/unix"
+
+	"github.com/rfjakob/gocryptfs/v2/tests/test_helpers"
+)
+
+// Check that byte-range locks work on an empty file
+func TestPoCFcntlFlock(t *testing.T) {
+	path := test_helpers.TmpDir + "/" + t.Name()
+
+	fd1, err := syscall.Open(path, syscall.O_CREAT|syscall.O_WRONLY|syscall.O_EXCL, 0600)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer syscall.Close(fd1)
+
+	// F_OFD_SETLK locks on the same fd always succeed, so we have to
+	// open a 2nd time.
+	fd2, err := syscall.Open(path, syscall.O_RDWR, 0)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer syscall.Close(fd2)
+
+	lk := unix.Flock_t{
+		Type:   unix.F_WRLCK,
+		Whence: unix.SEEK_SET,
+		Start:  0,
+		Len:    0,
+	}
+	err = unix.FcntlFlock(uintptr(fd1), unix.F_OFD_SETLK, &lk)
+	if err != nil {
+		t.Fatal(err)
+	}
+	err = unix.FcntlFlock(uintptr(fd2), unix.F_OFD_SETLK, &lk)
+	if err == nil {
+		t.Fatal("double-lock succeeded but should have failed")
+	}
+}