Fix deadlock problem in connection handling

3 files changed, 44 insertions, 14 deletions

diff --git a/lib/common_test/src/ct_gen_conn.erl b/lib/common_test/src/ct_gen_conn.erl
index 5aab4dd2dd..12d5b5b9fd 100644
--- a/lib/common_test/src/ct_gen_conn.erl
+++ b/lib/common_test/src/ct_gen_conn.erl
@@ -27,7 +27,7 @@
 -compile(export_all).
 
 -export([start/4, stop/1]).
--export([call/2, do_within_time/2]).
+-export([call/2, call/3, do_within_time/2]).
 
 -ifdef(debug).
 -define(dbg,true).
@@ -97,7 +97,7 @@ start(Name,Address,InitData,CallbackMod) ->
 %%%
 %%% @doc Close the telnet connection and stop the process managing it.
 stop(Pid) ->
-    call(Pid,stop).
+    call(Pid,stop,5000).
 
 %%%-----------------------------------------------------------------
 %%% @spec log(Heading,Format,Args) -> ok
@@ -176,7 +176,10 @@ do_within_time(Fun,Timeout) ->
 
 %%%=================================================================
 %%% Internal functions
-call(Pid,Msg) ->
+call(Pid, Msg) ->
+    call(Pid, Msg, infinity).
+
+call(Pid, Msg, Timeout) ->
     MRef = erlang:monitor(process,Pid),
     Ref = make_ref(),
     Pid ! {Msg,{self(),Ref}},
@@ -191,6 +194,12 @@ call(Pid,Msg) ->
 	    end;
 	{'DOWN',MRef,process,_,Reason}  -> 
 	    {error,{process_down,Pid,Reason}}
+    after Timeout ->
+	    log("ct_gen_conn",
+		"Connection process ~p not responding. Killing now!",
+		[Pid]),
+	    exit(Pid, kill),
+	    {error,{process_down,Pid,forced_termination}}
     end.
 
 return({To,Ref},Result) ->
diff --git a/lib/common_test/src/ct_ssh.erl b/lib/common_test/src/ct_ssh.erl
index 2e8e46d184..09cd4ef02a 100644
--- a/lib/common_test/src/ct_ssh.erl
+++ b/lib/common_test/src/ct_ssh.erl
@@ -235,7 +235,7 @@ connect(KeyOrName, ConnType, ExtraOpts) ->
 disconnect(SSH) ->
     case get_handle(SSH) of
 	{ok,Pid} ->
-	    try_log(heading(disconnect,SSH), "Handle: ~p", [Pid]),
+	    try_log(heading(disconnect,SSH), "Handle: ~p", [Pid], 5000),
 	    case ct_gen_conn:stop(Pid) of
 		{error,{process_down,Pid,noproc}} ->
 		    {error,already_closed};
@@ -1228,11 +1228,11 @@ terminate(SSHRef, State) ->
     case State#state.conn_type of
 	ssh ->
 	    try_log(heading(disconnect_ssh,State#state.target),
-		    "SSH Ref: ~p",[SSHRef]),
+		    "SSH Ref: ~p",[SSHRef], 5000),
 	    ssh:close(SSHRef);
 	sftp ->
 	    try_log(heading(disconnect_sftp,State#state.target),
-		    "SFTP Ref: ~p",[SSHRef]),
+		    "SFTP Ref: ~p",[SSHRef], 5000),
 	    ssh_sftp:stop_channel(SSHRef)
     end.
 
@@ -1337,9 +1337,12 @@ get_handle(SSH) ->
 %%%-----------------------------------------------------------------
 %%% 
 call(SSH, Msg) ->
+    call(SSH, Msg, infinity).
+	
+call(SSH, Msg, Timeout) ->
     case get_handle(SSH) of
 	{ok,Pid} ->
-	    ct_gen_conn:call(Pid, Msg);
+	    ct_gen_conn:call(Pid, Msg, Timeout);
 	Error ->
 	    Error
     end.
@@ -1368,11 +1371,16 @@ log(Heading, Str, Args) ->
 %%%-----------------------------------------------------------------
 %%% 
 try_log(Heading, Str, Args) ->
-    case ct_util:is_silenced(ssh) of
+    try_log(Heading, Str, Args, infinity).
+
+try_log(Heading, Str, Args, Timeout) ->
+    case ct_util:is_silenced(ssh, Timeout) of
 	true ->
 	    ok;
 	false ->
-	    ct_gen_conn:log(Heading, Str, Args)
+	    ct_gen_conn:log(Heading, Str, Args);
+	_Error ->
+	    ok
     end.
 
 %%%-----------------------------------------------------------------
diff --git a/lib/common_test/src/ct_util.erl b/lib/common_test/src/ct_util.erl
index 541921df58..52a2765856 100644
--- a/lib/common_test/src/ct_util.erl
+++ b/lib/common_test/src/ct_util.erl
@@ -36,14 +36,15 @@
 	 save_suite_data_async/3, save_suite_data_async/2,
 	 read_suite_data/1, 
 	 delete_suite_data/0, delete_suite_data/1, match_delete_suite_data/1,
-	 delete_testdata/0, delete_testdata/1, set_testdata/1, get_testdata/1,
+	 delete_testdata/0, delete_testdata/1,
+	 set_testdata/1, get_testdata/1, get_testdata/2,
 	 set_testdata_async/1, update_testdata/2]).
 
 -export([override_silence_all_connections/0, override_silence_connections/1, 
 	 get_overridden_silenced_connections/0, 
 	 delete_overridden_silenced_connections/0, 
-	 silence_all_connections/0, silence_connections/1, is_silenced/1, 
-	 reset_silent_connections/0]).
+	 silence_all_connections/0, silence_connections/1,
+	 is_silenced/1, is_silenced/2, reset_silent_connections/0]).
 
 -export([get_mode/0, create_table/3, read_opts/0]).
 
@@ -254,6 +255,9 @@ set_testdata_async(TestData) ->
 get_testdata(Key) ->
     call({get_testdata, Key}).
 
+get_testdata(Key, Timeout) ->
+    call({get_testdata, Key}, Timeout).
+
 set_cwd(Dir) ->
     call({set_cwd,Dir}).
 
@@ -376,7 +380,6 @@ loop(Mode,TestData,StartDir) ->
 	    exit(Reason)
     end.
 
-
 close_connections([#conn{handle=Handle,callback=CB}|Conns]) ->
     CB:close(Handle),
     close_connections(Conns);
@@ -545,7 +548,10 @@ silence_connections(Conns) when is_list(Conns) ->
     set_testdata({silent_connections,Conns1}).
 
 is_silenced(Conn) ->
-    case get_testdata(silent_connections) of
+    is_silenced(Conn, infinity).
+
+is_silenced(Conn, Timeout) ->
+    case get_testdata(silent_connections, Timeout) of
 	Conns when is_list(Conns) ->
 	    case lists:keysearch(Conn,1,Conns) of
 		{value,{Conn,true}} ->
@@ -553,6 +559,8 @@ is_silenced(Conn) ->
 		_ ->
 		    false
 	    end;
+	Error = {error,_} ->
+	    Error;
 	_ ->
 	    false
     end.
@@ -827,6 +835,9 @@ get_profile_data(Profile, Key, StartDir) ->
 %%%-----------------------------------------------------------------
 %%% Internal functions
 call(Msg) ->
+    call(Msg, infinity).
+
+call(Msg, Timeout) ->
     case whereis(ct_util_server) of
 	undefined ->
 	    {error,ct_util_server_not_running};
@@ -840,6 +851,8 @@ call(Msg) ->
 		    Result;
 		{'DOWN',MRef,process,_,Reason}  -> 
 		    {error,{ct_util_server_down,Reason}}
+	    after
+		Timeout -> {error,timeout}
 	    end
     end.