Merge branch 'max-au/erl-check-io-desync-23/OTP-16780/PR-2701' into maint-23

* max-au/erl-check-io-desync-23/OTP-16780/PR-2701:
  kernel: Fix tcp misc testcase resource cleanup
  erl_check_io: do not discard ERTS_POLL_EV_IN from active_events

2 files changed, 69 insertions, 3 deletions

diff --git a/erts/emulator/sys/common/erl_check_io.c b/erts/emulator/sys/common/erl_check_io.c
index 6305590690..fe9d6b29c0 100644
--- a/erts/emulator/sys/common/erl_check_io.c
+++ b/erts/emulator/sys/common/erl_check_io.c
@@ -1761,8 +1761,10 @@ erts_check_io(ErtsPollThread *psi, ErtsMonotonicTime timeout_time)
 
                 reactive_events = state->active_events;
 
-                if (state->flags & ERTS_EV_FLAG_IN_SCHEDULER)
+                if (state->flags & ERTS_EV_FLAG_IN_SCHEDULER) {
                     reactive_events &= ~ERTS_POLL_EV_IN;
+                    state->active_events |= ERTS_POLL_EV_IN;
+                }
 
                 /* Reactivate the poll op if there are still active events */
                 if (reactive_events) {
diff --git a/lib/kernel/test/gen_tcp_misc_SUITE.erl b/lib/kernel/test/gen_tcp_misc_SUITE.erl
index 55bf50edd9..220a5b55f1 100644
--- a/lib/kernel/test/gen_tcp_misc_SUITE.erl
+++ b/lib/kernel/test/gen_tcp_misc_SUITE.erl
@@ -55,7 +55,7 @@
 	 active_once_closed/1, send_timeout/1, send_timeout_active/1,
          otp_7731/1, zombie_sockets/1, otp_7816/1, otp_8102/1,
          wrapping_oct/0, wrapping_oct/1, otp_9389/1, otp_13939/1,
-         otp_12242/1, delay_send_error/1]).
+         otp_12242/1, delay_send_error/1, bidirectional_traffic/1]).
 
 %% Internal exports.
 -export([sender/3, not_owner/1, passive_sockets_server/2, priority_server/1, 
@@ -100,7 +100,7 @@ all() ->
      active_once_closed, send_timeout, send_timeout_active, otp_7731,
      wrapping_oct,
      zombie_sockets, otp_7816, otp_8102, otp_9389,
-     otp_12242, delay_send_error].
+     otp_12242, delay_send_error, bidirectional_traffic].
 
 groups() -> 
     [].
@@ -3574,6 +3574,7 @@ delay_send_error(_Config) ->
           [{packet, 1}, {active, false}, {delay_send, true}]),
     {ok, S} = gen_tcp:accept(L),
     %% Do a couple of sends first to see that it works
+
     ok = gen_tcp:send(C, "hello"),
     ok = gen_tcp:send(C, "hello"),
     ok = gen_tcp:send(C, "hello"),
@@ -3594,3 +3595,66 @@ delay_send_error(_Config) ->
             ok
     end,
     ok = gen_tcp:close(C).
+
+-define(ACTIVE_N, 20).
+
+%% 30-second test for gen_tcp in {active, N} mode, ensuring it does not get stuck.
+%% Verifies that erl_check_io properly handles extra EPOLLIN signals.
+bidirectional_traffic(Config) when is_list(Config) ->
+    Workers = erlang:system_info(schedulers_online) * 2,
+    Payload = crypto:strong_rand_bytes(32),
+    {ok, LSock} = gen_tcp:listen(0, [binary, {packet, 0}, {active, false}, {reuseaddr, true}]),
+    %% get all sockets to know failing ends
+    {ok, Port} = inet:port(LSock),
+    Control = self(),
+    Receivers = [spawn_link(fun () -> exchange(LSock, Port, Payload, Control) end) || _ <- lists:seq(1, Workers)],
+    Result =
+        receive
+            {timeout, Socket, Total} ->
+                {fail, {timeout, Socket, Total}};
+            {error, Socket, Reason} ->
+                {fail, {error, Socket, Reason}}
+        after 30000 ->
+            %% if it does not fail in 30 seconds, it most likely works
+            ok
+        end,
+    [begin unlink(Rec), exit(Rec, kill) end || Rec <- Receivers],
+    Result.
+
+exchange(LSock, Port, Payload, Control) ->
+    %% spin up client
+    _ClntRcv = spawn(
+        fun () ->
+            {ok, Client} = gen_tcp:connect("localhost", Port, [binary, {packet, 0}, {active, ?ACTIVE_N}]),
+            send_recv_loop(Client, Payload, Control)
+        end),
+    {ok, Socket} = gen_tcp:accept(LSock),
+    %% sending process
+    send_recv_loop(Socket, Payload, Control).
+
+send_recv_loop(Socket, Payload, Control) ->
+    %% {active, N} must be set to active > 12 to trigger the issue
+    %% {active, 30} seems to trigger it quite often & reliably
+    inet:setopts(Socket, [{active, ?ACTIVE_N}]),
+    _Snd = spawn_link(
+        fun Sender() ->
+            _ = gen_tcp:send(Socket, Payload),
+            Sender()
+        end),
+    recv(Socket, 0, Control).
+
+recv(Socket, Total, Control) ->
+    receive
+        {tcp, Socket, Data} ->
+            recv(Socket, Total + byte_size(Data), Control);
+        {tcp_passive, Socket} ->
+            inet:setopts(Socket, [{active, ?ACTIVE_N}]),
+            recv(Socket, Total, Control);
+        {tcp_closed, Socket} ->
+            exit(terminate);
+        Other->
+            Control ! {error, Socket, Other}
+    after 2000 ->
+        %% no data received in 2 seconds, test failed
+        Control ! {timeout, Socket, Total}
+    end.