kernel: Fix an application terminate deadlock possibility

Could happen if get_child was called during terminate.
io (since it is a group_leader) also causes problems after
get_child was called.

Split up and do it async.

4 files changed, 64 insertions, 21 deletions

diff --git a/lib/kernel/src/application_master.erl b/lib/kernel/src/application_master.erl
index 68f78c6eb8..bc15b5a7de 100644
--- a/lib/kernel/src/application_master.erl
+++ b/lib/kernel/src/application_master.erl
@@ -1,7 +1,7 @@
 %%
 %% %CopyrightBegin%
 %% 
-%% Copyright Ericsson AB 1996-2013. All Rights Reserved.
+%% Copyright Ericsson AB 1996-2014. All Rights Reserved.
 %% 
 %% The contents of this file are subject to the Erlang Public License,
 %% Version 1.1, (the "License"); you may not use this file except in
@@ -27,7 +27,7 @@
 
 -include("application_master.hrl").
 
--record(state, {child, appl_data, children = [], procs = 0, gleader}).
+-record(state, {child, appl_data, children = [], procs = 0, gleader, req=[]}).
 
 %%-----------------------------------------------------------------
 %% Func: start_link/1
@@ -205,22 +205,25 @@ terminate_loop(Child, State) ->
 
 %%-----------------------------------------------------------------
 %% The Application Master is linked to *all* processes in the group
-%% (application).  
+%% (application).
 %%-----------------------------------------------------------------
 handle_msg({get_child, Tag, From}, State) ->
-    From ! {Tag, get_child_i(State#state.child)},
-    State;
+    get_child_i(State, Tag, From);
 handle_msg({stop, Tag, From}, State) ->
     catch terminate(normal, State),
     From ! {Tag, ok},
     exit(normal);
+handle_msg({child, Ref, GrandChild, Mod}, #state{req=Reqs0}=State) ->
+    {value, {_, Tag, From}, Reqs} = lists:keytake(Ref, 1, Reqs0),
+    From ! {Tag, {GrandChild, Mod}},
+    State#state{req=Reqs};
 handle_msg(_, State) ->
     State.
 
-
-terminate(Reason, State) ->
-    terminate_child(State#state.child, State),
-    kill_children(State#state.children),
+terminate(Reason, State = #state{child=Child, children=Children, req=Reqs}) ->
+    _ = [From ! {Tag, error} || {_, Tag, From} <- Reqs],
+    terminate_child(Child, State),
+    kill_children(Children),
     exit(Reason).
 
 
@@ -342,8 +345,8 @@ start_supervisor(Type, M, A) ->
 
 loop_it(Parent, Child, Mod, AppState) ->
     receive
-	{Parent, get_child} ->
-	    Parent ! {self(), Child, Mod},
+	{Parent, get_child, Ref} ->
+	    Parent ! {child, Ref, Child, Mod},
 	    loop_it(Parent, Child, Mod, AppState);
 	{Parent, terminate} ->
 	    NewAppState = prep_stop(Mod, AppState),
@@ -382,10 +385,15 @@ prep_stop(Mod, AppState) ->
 	    NewAppState
     end.
 
-get_child_i(Child) ->
-    Child ! {self(), get_child},
-    receive
-	{Child, GrandChild, Mod} -> {GrandChild, Mod}
+get_child_i(#state{child=Child, req=Reqs}=State, Tag, From) ->
+    Ref = erlang:make_ref(),
+    case erlang:is_process_alive(Child) of
+	true ->
+	    Child ! {self(), get_child, Ref},
+	    State#state{req=[{Ref, Tag, From}|Reqs]};
+	false ->
+	    From ! {Tag, error},
+	    State
     end.
 
 terminate_child_i(Child, State) ->
diff --git a/lib/kernel/test/application_SUITE.erl b/lib/kernel/test/application_SUITE.erl
index ff62297f2d..ccb3760309 100644
--- a/lib/kernel/test/application_SUITE.erl
+++ b/lib/kernel/test/application_SUITE.erl
@@ -1,7 +1,7 @@
 %%
 %% %CopyrightBegin%
 %% 
-%% Copyright Ericsson AB 1996-2013. All Rights Reserved.
+%% Copyright Ericsson AB 1996-2014. All Rights Reserved.
 %% 
 %% The contents of this file are subject to the Erlang Public License,
 %% Version 1.1, (the "License"); you may not use this file except in
@@ -36,7 +36,7 @@
 -export([config_change/1, persistent_env/1,
 	 distr_changed_tc1/1, distr_changed_tc2/1,
 	 ensure_started/1, ensure_all_started/1,
-	 shutdown_func/1, do_shutdown/1, shutdown_timeout/1]).
+	 shutdown_func/1, do_shutdown/1, shutdown_timeout/1, shutdown_deadlock/1]).
 
 -define(TESTCASE, testcase_name).
 -define(testcase, ?config(?TESTCASE, Config)).
@@ -54,6 +54,7 @@ all() ->
      script_start, nodedown_start, permit_false_start_local,
      permit_false_start_dist, get_key, get_env, ensure_all_started,
      {group, distr_changed}, config_change, shutdown_func, shutdown_timeout,
+     shutdown_deadlock,
      persistent_env].
 
 groups() -> 
@@ -961,7 +962,7 @@ nodedown_start(Conf) when is_list(Conf) ->
 
 ensure_started(suite) -> [];
 ensure_started(doc) -> ["Test application:ensure_started/1."];
-ensure_started(Conf) ->
+ensure_started(_Conf) ->
 
     {ok, Fd} = file:open("app1.app", [write]),
     w_app1(Fd),
@@ -981,7 +982,7 @@ ensure_started(Conf) ->
 
 ensure_all_started(suite) -> [];
 ensure_all_started(doc) -> ["Test application:ensure_all_started/1-2."];
-ensure_all_started(Conf) ->
+ensure_all_started(_Conf) ->
 
     {ok, Fd1} = file:open("app1.app", [write]),
     w_app1(Fd1),
@@ -2096,7 +2097,31 @@ shutdown_timeout(Config) when is_list(Config) ->
     end,
     ok.
 
+%%%-----------------------------------------------------------------
+%%% Provokes a (previous) application shutdown deadlock
+%%%-----------------------------------------------------------------
+shutdown_deadlock(Config) when is_list(Config) ->
+    DataDir = ?config(data_dir,Config),
+    code:add_path(filename:join([DataDir,deadlock])),
+    %% ok = rpc:call(Cp1, application, start, [sasl]),
+    ok = application:start(deadlock),
+    Tester = self(),
+    application:set_env(deadlock, fail_stop, Tester),
+    spawn(fun() -> Tester ! {stop, application:stop(deadlock)} end),
 
+    receive
+	{deadlock, Server} ->
+	    spawn(fun() ->
+			  Master = application_controller:get_master(deadlock),
+			  Child = application_master:get_child(Master),
+			  Tester ! {child, Child}
+		  end),
+	    timer:sleep(100),
+	    erlang:display({self(), "Sending Continue", Server}),
+	    Server ! continue
+    end,
+    [_|_] = application:which_applications(),
+    ok.
 
 
 %%-----------------------------------------------------------------
diff --git a/lib/kernel/test/application_SUITE_data/deadlock/deadlock.app b/lib/kernel/test/application_SUITE_data/deadlock/deadlock.app
index 0c1001bed6..233c7a3f76 100644
--- a/lib/kernel/test/application_SUITE_data/deadlock/deadlock.app
+++ b/lib/kernel/test/application_SUITE_data/deadlock/deadlock.app
@@ -4,5 +4,5 @@
   {applications, [kernel, stdlib, sasl]},
   {modules, [deadlock]},
   {mod, {deadlock, []}},
-  {env, [{fail_start, false}]}
+  {env, [{fail_start, false}, {fail_stop, false}]}
 ]}.
diff --git a/lib/kernel/test/application_SUITE_data/deadlock/deadlock.erl b/lib/kernel/test/application_SUITE_data/deadlock/deadlock.erl
index 5f68bf9078..3ef6105371 100644
--- a/lib/kernel/test/application_SUITE_data/deadlock/deadlock.erl
+++ b/lib/kernel/test/application_SUITE_data/deadlock/deadlock.erl
@@ -21,7 +21,7 @@ init([sup]) ->
     {ok, {{one_for_one, 5, 10}, [
         {
             sasl_syslog_dm, {?MODULE, start_link, []},
-            permanent, brutal_kill, worker,
+            permanent, 25000, worker,
             [deadlock]
         }
     ]}};
@@ -32,6 +32,8 @@ init([sup]) ->
 init([child]) ->
     case application:get_env(deadlock, fail_start) of
         {ok, false} ->
+	    process_flag(trap_exit, true),
+	    io:format("~p: Traps exit~n",[?MODULE]),
             %% we must not fail on the first init, otherwise supervisor
             %% terminates immediately
             {ok, []};
@@ -50,6 +52,14 @@ handle_info(_Msg, State) ->
     {noreply, State}.
 
 terminate(_Reason, _State) ->
+    case application:get_env(deadlock, fail_stop) of
+        {ok, false} -> ok;
+        {ok, Tester}  ->
+	    Tester ! {deadlock, self()},
+	    io:format("~p: Waiting in terminate (~p)~n",[?MODULE,Tester]),
+	    receive continue -> ok end
+    end,
+    io:format("~p: terminates~n", [?MODULE]),
     ok.
 
 code_change(_OldVsn, State, _Extra) ->