Refresh efile port if gone

src/bitcask.erl

@@ -2865,6 +2865,25 @@ corrupt_file(Path, Offset, Data) ->
     ok = file:write(FH, Data),
     file:close(FH).
 
+% Verify that if the cached efile port goes away, we can recover
+% and not get stuck opening casks
+efile_error_test() ->
+    Dir = "/tmp/bc.efile.error",
+    B = bitcask:open(Dir, [read_write]),
+    ok = bitcask:put(B, <<"k">>, <<"v">>),
+    ok = bitcask:close(B),
+    Port = get(bitcask_efile_port),
+    % If this fails, we stopped using the efile port trick to list
+    % dir contents, so remove this test
+    ?assert(is_port(Port)),
+    true = erlang:port_close(Port),
+    case bitcask:open(Dir) of
+        {error, _} = Err ->
+            ?assertEqual(ok, Err);
+        B2 when is_reference(B2) ->
+            ok = bitcask:close(B2)
+    end.
+
 %% About leak_t0():
 %%
 %% If bitcask leaks file descriptors for the 'touch'ed files, output is:

src/bitcask_fileops.erl

@@ -60,8 +60,8 @@
 -include_lib("eqc/include/eqc_fsm.hrl").
 -endif.
 -compile(export_all).
--endif.
 -include_lib("eunit/include/eunit.hrl").
+-endif.
 
 %% @doc Open a new file for writing.
 %% Called on a Dirname, will open a fresh file in that directory.
@@ -854,9 +854,20 @@ ensure_dir(F) ->
             end
     end.
 
-list_dir(Directory) ->
+list_dir(Dir) ->
+    list_dir(Dir, 1).
+
+list_dir(_, 0) ->
+    {error, efile_driver_unavailable};
+list_dir(Directory, Retries) when is_integer(Retries), Retries > 0 ->
     Port = get_efile_port(),
-    prim_file:list_dir(Port, Directory).
+    case prim_file:list_dir(Port, Directory) of
+        {error, einval} ->
+            clear_efile_port(),
+            list_dir(Directory, Retries-1);
+        Result ->
+            Result
+    end.
 
 get_efile_port() ->
     Key = bitcask_efile_port,
@@ -875,6 +886,9 @@ get_efile_port() ->
             Port
     end.
 
+clear_efile_port() ->
+    erase(bitcask_efile_port).
+
 prim_file_drv_open(Driver, Portopts) ->
     try erlang:open_port({spawn, Driver}, Portopts) of
         Port ->