snapshotter: use syncfs system call (#2816)

`sync` system call triggers a full page cache sync which may not always
work, especially in kubernetes environment where it is easy to be
interfered by others. I have seen several cases where a broken nfs mount
is blocking kaniko from doing its job.

With `syncfs`, it only writes cache back to disk for the current
filesystem that is used by kaniko which is supposed to be more reliable.

pkg/snapshot/snapshot.go

@@ -22,6 +22,7 @@ import (
 	"io/ioutil"
 	"os"
 	"path/filepath"
+	"runtime"
 	"sort"
 	"syscall"
 
@@ -31,6 +32,7 @@ import (
 	"github.com/GoogleContainerTools/kaniko/pkg/util"
 
 	"github.com/sirupsen/logrus"
+	"golang.org/x/sys/unix"
 )
 
 // For testing
@@ -155,7 +157,20 @@ func (s *Snapshotter) scanFullFilesystem() ([]string, []string, error) {
 	// for example the hashing function that determines if files are equal uses the mtime of the files,
 	// which can lag if sync is not called. Unfortunately there can still be lag if too much data needs
 	// to be flushed or the disk does its own caching/buffering.
-	syscall.Sync()
+	if runtime.GOOS == "linux" {
+		dir, err := os.Open(s.directory)
+		if err != nil {
+			return nil, nil, err
+		}
+		defer dir.Close()
+		_, _, errno := syscall.Syscall(unix.SYS_SYNCFS, dir.Fd(), 0, 0)
+		if errno != 0 {
+			return nil, nil, errno
+		}
+	} else {
+		// fallback to full page cache sync
+		syscall.Sync()
+	}
 
 	s.l.Snapshot()