socket: distinguish eagain from closed, and retry transparently via write-select

CHANGELOG.md

@@ -26,6 +26,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Added support for configuring pins and width for sdmmc on ESP32
 - Added support for map comprehensions
 - Added USB CDC port drivers for ESP32, RP2, and STM32 platforms
+- Implemented `enif_select_write` (previously declared but not implemented), and reworked
+  `enif_select`/`enif_select_read`/`enif_select_write` so a resource can hold independent
+  pending read and write selects (each with its own ref/message/pid) on the same event at
+  the same time
 - Added a Linux `gpio` driver for the generic_unix port (in `avm_unix`) using sysfs
 
 ### Changed
@@ -69,6 +73,15 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Fixed several underallocation issues that could trigger data corruption on `binary:replace`, `zlib:compress` and bsd socket recv code.
 - Fixed a bug where `catch` would raise on regular atom results
 - Fixed ESP32 socket driver holding the global socket-list lock across blocking TCP connects, leaking the port on connect failure, losing concurrent `accept` waiters, leaking `netbuf` on receive error paths, and a recycled-`netconn` race between socket close and the event handler
+- `socket:send/2,3` now correctly distinguishes transient send backpressure (lwIP `ERR_MEM` / BSD
+  `EAGAIN`|`EWOULDBLOCK`) from a closed connection at the NIF level, and waits for the socket to
+  become writable again (using a new `nif_select_write/2`) and retries internally, so callers get
+  `ok | {error, Reason}` and no longer need to implement their own retry/backoff for partial sends
+  or backpressure; `ssl:send/2`, `ssl:recv/2` and the TLS handshake/close-notify loops likewise wait
+  for write-readiness instead of busy-looping on `want_write`
+- Fixed Erlang distribution over sockets (`socket_dist_controller`) silently ignoring
+  `socket:send/2` errors on `tick` and outgoing distribution data; the connection is now
+  terminated instead of continuing to run against a broken socket
 
 ## [0.7.0-alpha.1] - 2026-04-06
 

examples/erlang/rp2/picow_tcp_server.erl

@@ -78,8 +78,6 @@ echo(ConnectedSocket) ->
             case socket:send(ConnectedSocket, Data) of
                 ok ->
                     io:format("All data was sent~n");
-                {ok, Rest} ->
-                    io:format("Some data was sent.  Remaining: ~p~n", [Rest]);
                 {error, Reason} ->
                     io:format("An error occurred sending data: ~p~n", [Reason])
             end,

examples/erlang/tcp_socket_server.erl

@@ -61,8 +61,6 @@ echo(ConnectedSocket) ->
             case socket:send(ConnectedSocket, Data) of
                 ok ->
                     io:format("All data was sent~n");
-                {ok, Rest} ->
-                    io:format("Some data was sent.  Remaining: ~p~n", [Rest]);
                 {error, Reason} ->
                     io:format("An error occurred sending data: ~p~n", [Reason])
             end,

examples/erlang/udp_socket_client.erl

@@ -37,10 +37,6 @@ loop(Socket) ->
             io:format("Send packet ~p to ~p.~n", [
                 Data, Dest
             ]);
-        {ok, Rest} ->
-            io:format("Send packet ~p to ~p.  Remaining: ~p~n", [
-                Data, Dest, Rest
-            ]);
         {error, Reason} ->
             io:format("An error occurred sending a packet: ~p~n", [Reason])
     end,

libs/avm_network/src/epmd.erl

@@ -157,7 +157,7 @@ process_req(
 ) ->
     case socket:recv(Socket, 1, nowait) of
         {select, {select_info, recv, Ref}} ->
-            socket:send(Socket, <<?ALIVE2_X_RESP, 0, Creation:32>>),
+            ok = socket:send(Socket, <<?ALIVE2_X_RESP, 0, Creation:32>>),
             State#state{
                 recv_selects = [{Ref, client} | RecvSelects],
                 clients = [

libs/estdlib/src/socket.erl

@@ -46,6 +46,7 @@
 %% internal nifs
 -export([
     nif_select_read/2,
+    nif_select_write/2,
     nif_accept/1,
     nif_recv/2,
     nif_recvfrom/2,
@@ -571,49 +572,58 @@ recvfrom0_nowait(Socket, Length, Ref) ->
 %%-----------------------------------------------------------------------------
 %% @param   Socket the socket
 %% @param   Data the data to send
-%% @returns `{ok, Rest}' if successful; `{error, Reason}', otherwise.
+%% @returns `ok' if successful; `{error, Reason}', otherwise.
 %% @doc     Send data on the specified socket.
 %%
 %%          Note that this function will block until data is sent
 %%          on the socket.  The data may not have been received by the
 %%          intended recipient, and the data may not even have been sent
 %%          over the network.
 %%
+%%          Partial sends and transient send backpressure (e.g. a full TCP
+%%          send buffer) are retried internally, waiting for the socket to
+%%          become writable again, until all data has been accepted by the
+%%          stack or an error occurs. Callers therefore do not need to
+%%          implement their own retry logic for `eagain'-like conditions.
+%%
 %% Example:
 %%
 %%          `ok = socket:send(ConnectedSocket, Data)'
 %% @end
 %%-----------------------------------------------------------------------------
 -spec send(Socket :: socket(), Data :: iodata()) ->
-    ok | {ok, Rest :: binary()} | {error, Reason :: term()}.
+    ok | {error, Reason :: term()}.
 send(Socket, Data) when is_binary(Data) ->
-    ?MODULE:nif_send(Socket, Data);
+    send_all(Socket, Data);
 send(Socket, Data) ->
-    ?MODULE:nif_send(Socket, erlang:iolist_to_binary(Data)).
+    send_all(Socket, erlang:iolist_to_binary(Data)).
 
 %%-----------------------------------------------------------------------------
 %% @param   Socket the socket
 %% @param   Data the data to send
 %% @param   Dest the destination to which to send the data
-%% @returns `{ok, Rest}' if successful; `{error, Reason}', otherwise.
+%% @returns `ok' if successful; `{error, Reason}', otherwise.
 %% @doc     Send data on the specified socket to the specified destination.
 %%
 %%          Note that this function will block until data is sent
 %%          on the socket.  The data may not have been received by the
 %%          intended recipient, and the data may not even have been sent
 %%          over the network.
 %%
+%%          Transient send backpressure is retried internally, waiting for
+%%          the socket to become writable again, as with `send/2'.
+%%
 %% Example:
 %%
 %%          `ok = socket:sendto(ConnectedSocket, Data, Dest)'
 %% @end
 %%-----------------------------------------------------------------------------
 -spec sendto(Socket :: socket(), Data :: iodata(), Dest :: sockaddr()) ->
-    ok | {ok, Rest :: binary()} | {error, Reason :: term()}.
+    ok | {error, Reason :: term()}.
 sendto(Socket, Data, Dest) when is_binary(Data) ->
-    ?MODULE:nif_sendto(Socket, Data, Dest);
+    sendto_all(Socket, Data, Dest);
 sendto(Socket, Data, Dest) ->
-    ?MODULE:nif_sendto(Socket, erlang:iolist_to_binary(Data), Dest).
+    sendto_all(Socket, erlang:iolist_to_binary(Data), Dest).
 
 %%-----------------------------------------------------------------------------
 %% @param   Socket the socket
@@ -708,6 +718,10 @@ shutdown(_Socket, _How) ->
 nif_select_read(_Socket, _Ref) ->
     erlang:nif_error(undefined).
 
+%% @private
+nif_select_write(_Socket, _Ref) ->
+    erlang:nif_error(undefined).
+
 %% @private
 nif_accept(_Socket) ->
     erlang:nif_error(undefined).
@@ -731,3 +745,62 @@ nif_send(_Socket, _Data) ->
 %% @private
 nif_sendto(_Socket, _Data, _Dest) ->
     erlang:nif_error(undefined).
+
+%% @private
+%% Send all of Data on Socket, retrying partial sends and transient send
+%% backpressure ({error, eagain}) by waiting for the socket to become
+%% writable again, until all data has been accepted or a real error occurs.
+send_all(_Socket, <<>>) ->
+    ok;
+send_all(Socket, Data) ->
+    case ?MODULE:nif_send(Socket, Data) of
+        ok ->
+            ok;
+        {ok, Rest} ->
+            send_all(Socket, Rest);
+        {error, eagain} ->
+            case wait_write(Socket) of
+                ok ->
+                    send_all(Socket, Data);
+                {error, _Reason} = Error ->
+                    Error
+            end;
+        {error, _Reason} = Error ->
+            Error
+    end.
+
+%% @private
+sendto_all(_Socket, <<>>, _Dest) ->
+    ok;
+sendto_all(Socket, Data, Dest) ->
+    case ?MODULE:nif_sendto(Socket, Data, Dest) of
+        ok ->
+            ok;
+        {ok, Rest} ->
+            sendto_all(Socket, Rest, Dest);
+        {error, eagain} ->
+            case wait_write(Socket) of
+                ok ->
+                    sendto_all(Socket, Data, Dest);
+                {error, _Reason} = Error ->
+                    Error
+            end;
+        {error, _Reason} = Error ->
+            Error
+    end.
+
+%% @private
+%% Block the calling process until Socket is writable (or closed).
+wait_write(Socket) ->
+    Ref = erlang:make_ref(),
+    case ?MODULE:nif_select_write(Socket, Ref) of
+        ok ->
+            receive
+                {'$socket', Socket, select, Ref} ->
+                    ok;
+                {'$socket', Socket, abort, {Ref, Reason}} ->
+                    {error, Reason}
+            end;
+        {error, _Reason} = Error ->
+            Error
+    end.

libs/estdlib/src/socket_dist_controller.erl

@@ -142,8 +142,12 @@ handle_call(getstat, _From, #state{received = Received, sent = Sent} = State) ->
     {reply, {ok, Received, Sent, 0}, State}.
 
 handle_cast(tick, #state{socket = Socket, sent = Sent} = State) ->
-    socket:send(Socket, <<0:32>>),
-    {noreply, State#state{sent = Sent + 1}};
+    case socket:send(Socket, <<0:32>>) of
+        ok ->
+            {noreply, State#state{sent = Sent + 1}};
+        {error, Reason} ->
+            {stop, {send_error, Reason}, State}
+    end;
 handle_cast({handshake_complete, DHandle}, State0) ->
     ok = erlang:dist_ctrl_get_data_notification(DHandle),
     % We now need to get messages when data is coming.
@@ -154,8 +158,10 @@ handle_cast({handshake_complete, DHandle}, State0) ->
     end.
 
 handle_info(dist_data, State0) ->
-    State1 = send_data_loop(State0),
-    {noreply, State1};
+    case send_data_loop(State0) of
+        {ok, State1} -> {noreply, State1};
+        {stop, Reason, State1} -> {stop, Reason, State1}
+    end;
 handle_info(
     {'$socket', Socket, select, SelectHandle},
     #state{socket = Socket, select_handle = SelectHandle} = State0
@@ -209,10 +215,14 @@ send_data_loop(#state{dhandle = DHandle, socket = Socket, sent = Sent} = State)
     case erlang:dist_ctrl_get_data(DHandle) of
         none ->
             ok = erlang:dist_ctrl_get_data_notification(DHandle),
-            State;
+            {ok, State};
         Data ->
             DataBin = ?PRE_PROCESS(Data),
             DataSize = byte_size(DataBin),
-            socket:send(Socket, <<DataSize:32, DataBin/binary>>),
-            send_data_loop(State#state{sent = Sent + 1})
+            case socket:send(Socket, <<DataSize:32, DataBin/binary>>) of
+                ok ->
+                    send_data_loop(State#state{sent = Sent + 1});
+                {error, Reason} ->
+                    {stop, {send_error, Reason}, State}
+            end
     end.

libs/estdlib/src/ssl.erl

@@ -200,8 +200,13 @@ handshake_loop(SSLContext, Socket) ->
                     Error
             end;
         want_write ->
-            % We're currrently missing non-blocking writes
-            handshake_loop(SSLContext, Socket);
+            case wait_write(Socket) of
+                ok ->
+                    handshake_loop(SSLContext, Socket);
+                {error, _Reason} = Error ->
+                    socket:close(Socket),
+                    Error
+            end;
         {error, _Reason} = Error ->
             socket:close(Socket),
             Error
@@ -253,8 +258,13 @@ close_notify_loop(SSLContext, Socket) ->
                     Error
             end;
         want_write ->
-            % We're currrently missing non-blocking writes
-            close_notify_loop(SSLContext, Socket);
+            case wait_write(Socket) of
+                ok ->
+                    close_notify_loop(SSLContext, Socket);
+                {error, _Reason} = Error ->
+                    socket:close(Socket),
+                    Error
+            end;
         {error, _Reason} = Error ->
             socket:close(Socket),
             Error
@@ -283,8 +293,12 @@ send({SSLContext, Socket} = SSLSocket, Binary) ->
                     Error
             end;
         want_write ->
-            % We're currrently missing non-blocking writes
-            send(SSLSocket, Binary);
+            case wait_write(Socket) of
+                ok ->
+                    send(SSLSocket, Binary);
+                {error, _Reason} = Error ->
+                    Error
+            end;
         {error, _Reason} = Error ->
             Error
     end.
@@ -318,8 +332,27 @@ recv0({SSLContext, Socket} = SSLSocket, Length, Remaining, Acc) ->
                     Error
             end;
         want_write ->
-            % We're currrently missing non-blocking writes
-            recv0(SSLSocket, Length, Remaining, Acc);
+            case wait_write(Socket) of
+                ok ->
+                    recv0(SSLSocket, Length, Remaining, Acc);
+                {error, _Reason} = Error ->
+                    Error
+            end;
+        {error, _Reason} = Error ->
+            Error
+    end.
+
+%% @private
+wait_write(Socket) ->
+    Ref = erlang:make_ref(),
+    case socket:nif_select_write(Socket, Ref) of
+        ok ->
+            receive
+                {'$socket', Socket, select, Ref} ->
+                    ok;
+                {'$socket', Socket, abort, {Ref, Reason}} ->
+                    {error, Reason}
+            end;
         {error, _Reason} = Error ->
             Error
     end.