Add 'show_econnreset' TCP socket option

An ECONNRESET is a socket error which tells us that a TCP peer has sent
an RST. The RST indicates that they have aborted the connection and
that the payload we have received should not be considered complete. Up
until now, the implementation of TCP in inet_drv.c has hidden the
receipt of the RST from the user, treating it as though it was just
a FIN terminating the read side of the socket.

There are many cases where user code needs to be able to distinguish
between a socket that was closed normally and one that was aborted.
Setting the option {show_econnreset, true} enables the user to receive
ECONNRESET errors on both active and passive sockets.

A connected socket returned from gen_tcp:accept/1 will inherit the
show_econnreset setting of the listening socket.

By default this option is set to {show_econnreset, false}.

Note that this patch only enables the reporting of ECONNRESET when
the socket is being read from. It does not report ECONNRESET (or
EPIPE) when the user tries to write to a connection when an RST
has already been received. Currently the TCP implementation in
inet_drv.c hides all such send errors from the user in favour
of returning {error, close}. A separate patch will be needed to
enable the reporting of such errors.

7 files changed, 189 insertions, 5 deletions

diff --git a/erts/emulator/drivers/common/inet_drv.c b/erts/emulator/drivers/common/inet_drv.c
index e001f31932..b5903f0c07 100644
--- a/erts/emulator/drivers/common/inet_drv.c
+++ b/erts/emulator/drivers/common/inet_drv.c
@@ -835,6 +835,7 @@ static int my_strncasecmp(const char *s1, const char *s2, size_t n)
 #define TCP_ADDF_PENDING_SHUT_RDWR 32 /* Call shutdown(sock, SHUT_RDWR) when queue empties */
 #define TCP_ADDF_PENDING_SHUTDOWN \
 		(TCP_ADDF_PENDING_SHUT_WR | TCP_ADDF_PENDING_SHUT_RDWR)
+#define TCP_ADDF_SHOW_ECONNRESET   64 /* Tell user about incoming RST */
 
 /* *_REQ_* replies */
 #define INET_REP_ERROR       0
@@ -879,6 +880,7 @@ static int my_strncasecmp(const char *s1, const char *s2, size_t n)
 #define INET_LOPT_MSGQ_HIWTRMRK     36  /* set local msgq high watermark */
 #define INET_LOPT_MSGQ_LOWTRMRK     37  /* set local msgq low watermark */
 #define INET_LOPT_NETNS             38  /* Network namespace pathname */
+#define INET_LOPT_TCP_SHOW_ECONNRESET 39  /* tell user about incoming RST */
 /* SCTP options: a separate range, from 100: */
 #define SCTP_OPT_RTOINFO		100
 #define SCTP_OPT_ASSOCINFO		101
@@ -6255,6 +6257,16 @@ static int inet_set_opts(inet_descriptor* desc, char* ptr, int len)
 	    continue;
 #endif
 
+	case INET_LOPT_TCP_SHOW_ECONNRESET:
+	    if (desc->sprotocol == IPPROTO_TCP) {
+		tcp_descriptor* tdesc = (tcp_descriptor*) desc;
+		if (ival)
+		    tdesc->tcp_add_flags |= TCP_ADDF_SHOW_ECONNRESET;
+		else
+		    tdesc->tcp_add_flags &= ~TCP_ADDF_SHOW_ECONNRESET;
+	    }
+	    continue;
+
 	case INET_OPT_REUSEADDR: 
 #ifdef __WIN32__
 	    continue;  /* Bjorn says */
@@ -7234,6 +7246,17 @@ static ErlDrvSSizeT inet_fill_opts(inet_descriptor* desc,
 	    continue;
 #endif
 
+	case INET_LOPT_TCP_SHOW_ECONNRESET:
+	    if (desc->sprotocol == IPPROTO_TCP) {
+		tcp_descriptor* tdesc = (tcp_descriptor*) desc;
+		*ptr++ = opt;
+		ival = !!(tdesc->tcp_add_flags & TCP_ADDF_SHOW_ECONNRESET);
+		put_int32(ival, ptr);
+	    } else {
+		TRUNCATE_TO(0,ptr);
+	    }
+	    continue;
+
 	case INET_OPT_PRIORITY:
 #ifdef SO_PRIORITY
 	    type = SO_PRIORITY;
@@ -9077,6 +9100,11 @@ static tcp_descriptor* tcp_inet_copy(tcp_descriptor* desc,SOCKET s,
     copy_desc->low           = desc->low;
     copy_desc->send_timeout  = desc->send_timeout;
     copy_desc->send_timeout_close = desc->send_timeout_close;
+
+    if (desc->tcp_add_flags & TCP_ADDF_SHOW_ECONNRESET)
+	copy_desc->tcp_add_flags |= TCP_ADDF_SHOW_ECONNRESET;
+    else
+	copy_desc->tcp_add_flags &= ~TCP_ADDF_SHOW_ECONNRESET;
     
     /* The new port will be linked and connected to the original caller */
     port = driver_create_port(port, owner, "tcp_inet", (ErlDrvData) copy_desc);
@@ -10014,7 +10042,10 @@ static int tcp_recv(tcp_descriptor* desc, int request_len)
 	int err = sock_errno();
 	if (err == ECONNRESET) {
 	    DEBUGF((" => detected close (connreset)\r\n"));
-	    return tcp_recv_closed(desc);
+	    if (desc->tcp_add_flags & TCP_ADDF_SHOW_ECONNRESET)
+		return tcp_recv_error(desc, err);
+	    else
+		return tcp_recv_closed(desc);
 	}
 	if (err == ERRNO_BLOCK) {
 	    DEBUGF((" => would block\r\n"));
@@ -10226,7 +10257,19 @@ static void tcp_inet_event(ErlDrvData e, ErlDrvEvent event)
     if (netEv.lNetworkEvents & FD_CLOSE) {
 	/* error in err = netEv.iErrorCode[FD_CLOSE_BIT] */
 	DEBUGF(("Detected close in %s, line %d\r\n", __FILE__, __LINE__));
-	tcp_recv_closed(desc);
+	if (desc->tcp_add_flags & TCP_ADDF_SHOW_ECONNRESET) {
+	    err = netEv.iErrorCode[FD_CLOSE_BIT];
+	    if (err == ECONNRESET)
+		tcp_recv_error(desc, err);
+	    else if (err == ECONNABORTED && IS_CONNECTED(INETP(desc))) {
+		/* translate this error to ECONNRESET */
+		tcp_recv_error(desc, ECONNRESET);
+	    }
+	    else
+		tcp_recv_closed(desc);
+	}
+	else
+	    tcp_recv_closed(desc);
     }
     DEBUGF(("tcp_inet_event(%ld) }\r\n", (long)desc->inet.port));
     return;
diff --git a/erts/preloaded/src/prim_inet.erl b/erts/preloaded/src/prim_inet.erl
index 622e1be869..fb0f67aa8a 100644
--- a/erts/preloaded/src/prim_inet.erl
+++ b/erts/preloaded/src/prim_inet.erl
@@ -1140,6 +1140,7 @@ enc_opt(delay_send)      -> ?INET_LOPT_TCP_DELAY_SEND;
 enc_opt(packet_size)     -> ?INET_LOPT_PACKET_SIZE;
 enc_opt(read_packets)    -> ?INET_LOPT_READ_PACKETS;
 enc_opt(netns)           -> ?INET_LOPT_NETNS;
+enc_opt(show_econnreset) -> ?INET_LOPT_TCP_SHOW_ECONNRESET;
 enc_opt(raw)             -> ?INET_OPT_RAW;
 % Names of SCTP opts:
 enc_opt(sctp_rtoinfo)	 	   -> ?SCTP_OPT_RTOINFO;
@@ -1197,6 +1198,7 @@ dec_opt(?INET_LOPT_TCP_DELAY_SEND)   -> delay_send;
 dec_opt(?INET_LOPT_PACKET_SIZE)      -> packet_size;
 dec_opt(?INET_LOPT_READ_PACKETS)     -> read_packets;
 dec_opt(?INET_LOPT_NETNS)           -> netns;
+dec_opt(?INET_LOPT_TCP_SHOW_ECONNRESET) -> show_econnreset;
 dec_opt(?INET_OPT_RAW)              -> raw;
 dec_opt(I) when is_integer(I)     -> undefined.
 
@@ -1296,6 +1298,7 @@ type_opt_1(delay_send)      -> bool;
 type_opt_1(packet_size)     -> uint;
 type_opt_1(read_packets)    -> uint;
 type_opt_1(netns)           -> binary;
+type_opt_1(show_econnreset) -> bool;
 %% 
 %% SCTP options (to be set). If the type is a record type, the corresponding
 %% record signature is returned, otherwise, an "elementary" type tag 
diff --git a/lib/kernel/doc/src/inet.xml b/lib/kernel/doc/src/inet.xml
index 77a8caaaf6..a364f9e0c9 100644
--- a/lib/kernel/doc/src/inet.xml
+++ b/lib/kernel/doc/src/inet.xml
@@ -1037,6 +1037,33 @@ setcap cap_sys_admin,cap_sys_ptrace,cap_dac_read_search+epi beam.smp
 			<marker id="option-sndbuf"></marker>
           </item>
 
+          <tag><c>{show_econnreset, Boolean}</c>(TCP/IP sockets)</tag>
+          <item>
+            <p>When this option is set to <c>false</c>, as it is by
+              default, an RST that is received from the TCP peer is treated
+              as a normal close (as though a FIN was sent). A caller
+              to <seealso marker="gen_tcp#recv/2">gen_tcp:recv/2</seealso>
+              will get <c>{error, closed}</c>. In active
+              mode the controlling process will receive a
+              <c>{tcp_close, Socket}</c> message, indicating that the
+              peer has closed the connection.</p>
+            <p>Setting this option to <c>true</c> will allow you to
+              distinguish between a connection that was closed normally,
+              and one which was aborted (intentionally or unintentionally)
+              by the TCP peer. A call to
+              <seealso marker="gen_tcp#recv/2">gen_tcp:recv/2</seealso>
+              will return <c>{error, econnreset}</c>. In
+              active mode, the controlling process will receive a
+              <c>{tcp_error, Socket, econnreset}</c> message
+              before the usual <c>{tcp_closed, Socket}</c>, as is
+              the case for any other socket error.</p>
+            <p>A connected socket returned from
+              <seealso marker="gen_tcp#accept/1">gen_tcp:accept/1</seealso>
+              will inherit the <c>show_econnreset</c> setting from the
+              listening socket.</p>
+			<marker id="option-show_econnreset"></marker>
+          </item>
+
           <tag><c>{sndbuf, Size}</c></tag>
           <item>
             <p>The minimum size of the send buffer to use for the socket.
diff --git a/lib/kernel/src/gen_tcp.erl b/lib/kernel/src/gen_tcp.erl
index bc8ffbe5e3..86251fc8f1 100644
--- a/lib/kernel/src/gen_tcp.erl
+++ b/lib/kernel/src/gen_tcp.erl
@@ -58,6 +58,7 @@
         {reuseaddr,       boolean()} |
         {send_timeout,    non_neg_integer() | infinity} |
         {send_timeout_close, boolean()} |
+        {show_econnreset, boolean()} |
         {sndbuf,          non_neg_integer()} |
         {tos,             non_neg_integer()} |
 	{ipv6_v6only,     boolean()}.
@@ -89,6 +90,7 @@
         reuseaddr |
         send_timeout |
         send_timeout_close |
+        show_econnreset |
         sndbuf |
         tos |
 	ipv6_v6only.
diff --git a/lib/kernel/src/inet.erl b/lib/kernel/src/inet.erl
index d668738109..1ae90aaf0c 100644
--- a/lib/kernel/src/inet.erl
+++ b/lib/kernel/src/inet.erl
@@ -654,7 +654,7 @@ options() ->
      multicast_if, multicast_ttl, multicast_loop,
      exit_on_close, high_watermark, low_watermark,
      high_msgq_watermark, low_msgq_watermark,
-     send_timeout, send_timeout_close
+     send_timeout, send_timeout_close, show_econnreset
     ].
 
 %% Return a list of statistics options
@@ -672,7 +672,8 @@ connect_options() ->
     [tos, priority, reuseaddr, keepalive, linger, sndbuf, recbuf, nodelay,
      header, active, packet, packet_size, buffer, mode, deliver,
      exit_on_close, high_watermark, low_watermark, high_msgq_watermark,
-     low_msgq_watermark, send_timeout, send_timeout_close, delay_send, raw].
+     low_msgq_watermark, send_timeout, send_timeout_close, delay_send, raw,
+     show_econnreset].
     
 connect_options(Opts, Family) ->
     BaseOpts = 
@@ -740,7 +741,7 @@ listen_options() ->
      header, active, packet, buffer, mode, deliver, backlog, ipv6_v6only,
      exit_on_close, high_watermark, low_watermark, high_msgq_watermark,
      low_msgq_watermark, send_timeout, send_timeout_close, delay_send,
-     packet_size, raw].
+     packet_size, raw, show_econnreset].
 
 listen_options(Opts, Family) ->
     BaseOpts = 
diff --git a/lib/kernel/src/inet_int.hrl b/lib/kernel/src/inet_int.hrl
index 889b596a22..7bd973af99 100644
--- a/lib/kernel/src/inet_int.hrl
+++ b/lib/kernel/src/inet_int.hrl
@@ -147,6 +147,7 @@
 -define(INET_LOPT_MSGQ_HIWTRMRK,  36).
 -define(INET_LOPT_MSGQ_LOWTRMRK,  37).
 -define(INET_LOPT_NETNS,          38).
+-define(INET_LOPT_TCP_SHOW_ECONNRESET, 39).
 % Specific SCTP options: separate range:
 -define(SCTP_OPT_RTOINFO,	 	100).
 -define(SCTP_OPT_ASSOCINFO,	 	101).
diff --git a/lib/kernel/test/gen_tcp_misc_SUITE.erl b/lib/kernel/test/gen_tcp_misc_SUITE.erl
index 4f0d7a7d50..f08f28b650 100644
--- a/lib/kernel/test/gen_tcp_misc_SUITE.erl
+++ b/lib/kernel/test/gen_tcp_misc_SUITE.erl
@@ -31,6 +31,8 @@
 	 init_per_testcase/2, end_per_testcase/2,
 	 otp_3924/1, otp_3924_sender/4, closed_socket/1,
 	 shutdown_active/1, shutdown_passive/1, shutdown_pending/1,
+	 show_econnreset_active/1, show_econnreset_active_once/1,
+	 show_econnreset_passive/1,
 	 default_options/1, http_bad_packet/1, 
 	 busy_send/1, busy_disconnect_passive/1, busy_disconnect_active/1,
 	 fill_sendq/1, partial_recv_and_close/1, 
@@ -92,6 +94,8 @@ all() ->
      iter_max_socks, passive_sockets, active_n,
      accept_closed_by_other_process, otp_3924, closed_socket,
      shutdown_active, shutdown_passive, shutdown_pending,
+     show_econnreset_active, show_econnreset_active_once,
+     show_econnreset_passive,
      default_options, http_bad_packet, busy_send,
      busy_disconnect_passive, busy_disconnect_active,
      fill_sendq, partial_recv_and_close,
@@ -1075,6 +1079,109 @@ shutdown_pending(Config) when is_list(Config) ->
 	     gen_tcp:close(S)
      end.
 
+%%
+%% Test 'show_econnreset' option
+%%
+
+show_econnreset_active(Config) when is_list(Config) ->
+    %% First confirm everything works with option turned off.
+    {ok, L} = gen_tcp:listen(0, []),
+    {ok, Port} = inet:port(L),
+    {ok, Client} = gen_tcp:connect(localhost, Port, [{active, false}]),
+    {ok, S} = gen_tcp:accept(L),
+    ok = gen_tcp:close(L),
+    ok = inet:setopts(Client, [{linger, {true, 0}}]),
+    ok = gen_tcp:close(Client),
+    receive
+       {tcp_closed, S} ->
+	   ok;
+       Other ->
+	   ?t:fail({unexpected1, Other})
+    after 1000 ->
+	?t:fail({timeout, {server, no_tcp_closed}})
+    end,
+
+    %% Now test with option switched on.
+    %% Note: We are also testing that the show_econnreset option is
+    %% inherited from the listening socket by the accepting socket.
+    {ok, L1} = gen_tcp:listen(0, [{show_econnreset, true}]),
+    {ok, Port1} = inet:port(L1),
+    {ok, Client1} = gen_tcp:connect(localhost, Port1, [{active, false}]),
+    {ok, S1} = gen_tcp:accept(L1),
+    ok = gen_tcp:close(L1),
+    ok = inet:setopts(Client1, [{linger, {true, 0}}]),
+    ok = gen_tcp:close(Client1),
+    receive
+	{tcp_error, S1, econnreset} ->
+	    receive
+		{tcp_closed, S1} ->
+		    ok;
+		Other1 ->
+		    ?t:fail({unexpected2, Other1})
+	    after 1 ->
+		?t:fail({timeout, {server, no_tcp_closed}})
+	    end;
+	Other2 ->
+	    ?t:fail({unexpected3, Other2})
+    after 1000 ->
+	?t:fail({timeout, {server, no_tcp_error}})
+    end.
+
+show_econnreset_active_once(Config) when is_list(Config) ->
+    %% Now test using {active, once}
+    {ok, L} = gen_tcp:listen(0,
+			   [{active, false},
+			    {show_econnreset, true}]),
+    {ok, Port} = inet:port(L),
+    {ok, Client} = gen_tcp:connect(localhost, Port, [{active, false}]),
+    {ok, S} = gen_tcp:accept(L),
+    ok = gen_tcp:close(L),
+    ok = inet:setopts(Client, [{linger, {true, 0}}]),
+    ok = gen_tcp:close(Client),
+    ok = ?t:sleep(20),
+    ok = receive Msg -> {unexpected_msg, Msg} after 0 -> ok end,
+    ok = inet:setopts(S, [{active, once}]),
+    receive
+	{tcp_error, S, econnreset} ->
+	    receive
+		{tcp_closed, S} ->
+		    ok;
+		Other1 ->
+		    ?t:fail({unexpected1, Other1})
+	    after 1 ->
+		?t:fail({timeout, {server, no_tcp_closed}})
+	    end;
+	Other2 ->
+	    ?t:fail({unexpected2, Other2})
+    after 1000 ->
+	?t:fail({timeout, {server, no_tcp_error}})
+    end.
+
+show_econnreset_passive(Config) when is_list(Config) ->
+    %% First confirm everything works with option turned off.
+    {ok, L} = gen_tcp:listen(0, [{active, false}]),
+    {ok, Port} = inet:port(L),
+    {ok, Client} = gen_tcp:connect(localhost, Port, [{active, false}]),
+    {ok, S} = gen_tcp:accept(L),
+    ok = gen_tcp:close(L),
+    ok = inet:setopts(S, [{linger, {true, 0}}]),
+    ok = gen_tcp:close(S),
+    ok = ?t:sleep(1),
+    {error, closed} = gen_tcp:recv(Client, 0),
+
+    %% Now test with option switched on.
+    {ok, L1} = gen_tcp:listen(0, [{active, false}]),
+    {ok, Port1} = inet:port(L1),
+    {ok, Client1} = gen_tcp:connect(localhost, Port1,
+				 [{active, false},
+				  {show_econnreset, true}]),
+    {ok, S1} = gen_tcp:accept(L1),
+    ok = gen_tcp:close(L1),
+    ok = inet:setopts(S1, [{linger, {true, 0}}]),
+    ok = gen_tcp:close(S1),
+    ok = ?t:sleep(1),
+    {error, econnreset} = gen_tcp:recv(Client1, 0).
+
 
 %% Thanks to Luke Gorrie. Tests for a very specific problem with 
 %% corrupt data. The testcase will be killed by the timetrap timeout