diff --git a/src/fscq.hs b/src/fscq.hs
index 3f9c93a9..950eb635 100644
--- a/src/fscq.hs
+++ b/src/fscq.hs
@@ -44,7 +44,8 @@ cachesize = 100000
 -- using its own threads).
 -- Disabled by default: seems to be buggy
 -- (1) FSCQ no longer terminates gracefully in response to SIGTERM
--- (2) More seriously, data doesn't seem to be persisted correctly
+-- (2) fuseDestroy (and thus AsyncFS.umount) are not called with opqueue, for
+--     unclear reasons
 useDowncalls :: Bool
 useDowncalls = False
 
@@ -492,16 +493,22 @@ fscqWrite fr m_fsxp _ inum bs offset = withMVar m_fsxp $ \fsxp -> do
   (wlen, ()) <- fr $ AsyncFS._AFS__file_get_sz fsxp inum
   len <- return $ fromIntegral $ wordToNat 64 wlen
   endpos <- return $ (fromIntegral offset) + (fromIntegral (BS.length bs))
+  -- NOTE: logged writes are disabled since they are not crash safe
+  -- (the design of DFSCQ should permit them to be crash safe, but the proofs
+  -- don't cover mixing direct and logged writes, and logged writes currently
+  -- are not synced correctly)
   (okspc, do_log) <- if len < endpos then do
     (ok, _) <- fr $ AsyncFS._AFS__file_truncate fsxp inum ((endpos + 4095) `div` 4096)
-    return (ok, True)
+    --return (ok, True)
+    return (ok, False)
   else do
     bslen <- return $ fromIntegral $ BS.length bs
     if ((fromIntegral offset) `mod` 4096 == 0) && (bslen `mod` 4096 == 0) && bslen > 4096 * 5 then
       -- block is large and aligned -> bypass write
       return $ (Errno.OK (), False)
     else
-      return $ (Errno.OK (), True)
+      -- return $ (Errno.OK (), True)
+      return $ (Errno.OK (), False)
   case okspc of
     Errno.OK _ -> do
       r <- foldM (write_piece do_log fsxp len) (WriteOK 0) (compute_range_pieces offset bs)