littlefs_open: sync the file only when O_TRUNC is specified

Because sync is a rather expensive operation.

While I'm here, improve the comment after the recent discussion
in https://github.com/apache/nuttx/pull/2913.

IMHO, we should not perform the sync even for O_TRUNC.
But I'm not going to propose the change for now.

Signed-off-by: YAMAMOTO Takashi <yamamoto@midokura.com>

fs/littlefs/lfs_vfs.c

@@ -424,11 +424,26 @@ static int littlefs_open(FAR struct file *filep, FAR const char *relpath,
         }
     }
 
-  /* Sync here in case of O_TRUNC haven't actually done immediately,
+  /* In case of O_TRUNC, sync to commit the truncation.
-   * e.g. total 8M, fileA 6M, O_TRUNC re-wrting fileA 6M, meet error.
+   *
+   * For example, consider a 6MB file on an 8MB filesystem.
+   * Open the file with O_TRUNC and (re)write some data until it grows
+   * to 6MB. Without this sync, it ends up with ENOSPC because
+   * littlefs doesn't commit the updates until sync/close and thus
+   * doesn't recycle the truncated space. Although it's normal for
+   * filesystems like littlefs, we automatically sync the truncation
+   * here to allow such a sequence to succeed, assuming it's what many
+   * of NuttX users expect.
+   *
+   * CAVEAT: On the other hand, this might surprise users who are familiar
+   * with the bare littlefs semantics.
    */
 
-  lfs_file_sync(&fs->lfs, &priv->file);
+  if (oflags & LFS_O_TRUNC)
+    {
+      lfs_file_sync(&fs->lfs, &priv->file);
+    }
+
   nxmutex_unlock(&fs->lock);
 
   /* Attach the private date to the struct file instance */