Merge pull request #280 from thalesmg/20250318-fix-ws-gun-death-race

fix(ws): avoiding hanging if gun connection process dies before upgrade

Author: thalesmg

src/emqtt_sock.erl

@@ -76,7 +76,23 @@ ssl_upgrade(Host, Sock, SslOpts0, Timeout) ->
 send(Sock, Data) when is_port(Sock) ->
     send_tcp_data(Sock, Data);
 send(#ssl_socket{ssl = SslSock}, Data) ->
-    ssl:send(SslSock, Data);
+    case ssl:send(SslSock, Data) of
+        ok ->
+            ok;
+        {error, closed}->
+            %% We attempt to grab an async exception with more information, if
+            %% available; otherwise, bail out.
+            receive
+                {ssl_error, _Sock, DetailedReason} ->
+                    {error, DetailedReason};
+                {ssl_closed, _Sock} ->
+                    {error, closed}
+            after 1 ->
+                    {error, closed}
+            end;
+        {error, Reason}->
+            {error, Reason}
+    end;
 send(QuicStream, Data) when is_reference(QuicStream) ->
     case quicer:send(QuicStream, Data) of
         {ok, _Len} ->

src/emqtt_ws.erl

@@ -51,9 +51,11 @@ connect(Host0, Port, Opts, Timeout) ->
                         {ok, StreamRef} -> {ok, {ConnPid, StreamRef}};
                         Error -> Error
                     end;
-                Error -> Error
+                Error ->
+                    Error
             end;
-        Error -> Error
+        Error ->
+            Error
     end.
 
 opts([], Acc) -> Acc;
@@ -73,6 +75,8 @@ upgrade(ConnPid, Opts, Timeout) ->
         {gun_response, ConnPid, _, _, Status, Headers} ->
             {error, {ws_upgrade_failed, Status, Headers}};
         {gun_error, ConnPid, StreamRef, Reason} ->
+            {error, {ws_upgrade_failed, Reason}};
+        {gun_down, ConnPid, _, Reason, _} ->
             {error, {ws_upgrade_failed, Reason}}
     after Timeout ->
         {error, timeout}