Merge branch 'raimo/mptcp-maint/OTP-19857' into maint

* raimo/mptcp-maint/OTP-19857:
  Update preloaded
  Implement TCP_KEEP* and TCP_USER_TIMEOUT for `gen_tcp`
  Implement TCP_USER_TIMEOUT for `socket`
  Implement `mptcp` for `gen_tcp_socket` compatibility module

Author: RaimoNiskanen

erts/emulator/drivers/common/inet_drv.c

@@ -923,6 +923,10 @@ static size_t my_strnlen(const char *s, size_t maxlen)
 #define TCP_OPT_NOPUSH              48  /* super-Nagle, aka TCP_CORK */
 #define INET_LOPT_TCP_READ_AHEAD    49  /* Read ahead of packet data */
 #define INET_LOPT_NON_BLOCK_SEND    50  /* Non-blocking send, only SCTP */
+#define TCP_OPT_KEEPCNT             51  /* TCP_KEEPCNTK */
+#define TCP_OPT_KEEPIDLE            52  /* TCP_KEEPIDLE */
+#define TCP_OPT_KEEPINTVL           53  /* TCP_KEEPINTVL */
+#define TCP_OPT_USER_TIMEOUT        54  /* TCP_USER_TIMEOUT */
 #define INET_LOPT_DEBUG             99  /* Enable/disable DEBUG for a socket */
 
 /* SCTP options: a separate range, from 100: */
@@ -7493,6 +7497,50 @@ static int inet_set_opts(inet_descriptor* desc, char* ptr, int len)
 	    continue;
 #endif
 
+#if defined(TCP_KEEPCNT)
+	case TCP_OPT_KEEPCNT:
+            DDBG(desc,
+                 ("INET-DRV-DBG[%d][" SOCKET_FSTR ",%T] "
+                  "inet_set_opts(keepcnt) -> %d\r\n",
+                  __LINE__, desc->s, driver_caller(desc->port), ival) );
+	    proto = IPPROTO_TCP;
+	    type = TCP_KEEPCNT;
+	    break;
+#endif
+
+#if defined(TCP_KEEPIDLE)
+	case TCP_OPT_KEEPIDLE:
+            DDBG(desc,
+                 ("INET-DRV-DBG[%d][" SOCKET_FSTR ",%T] "
+                  "inet_set_opts(keepidle) -> %d\r\n",
+                  __LINE__, desc->s, driver_caller(desc->port), ival) );
+	    proto = IPPROTO_TCP;
+	    type = TCP_KEEPIDLE;
+	    break;
+#endif
+
+#if defined(TCP_KEEPINTVL)
+	case TCP_OPT_KEEPINTVL:
+            DDBG(desc,
+                 ("INET-DRV-DBG[%d][" SOCKET_FSTR ",%T] "
+                  "inet_set_opts(keepintvl) -> %d\r\n",
+                  __LINE__, desc->s, driver_caller(desc->port), ival) );
+	    proto = IPPROTO_TCP;
+	    type = TCP_KEEPINTVL;
+	    break;
+#endif
+
+#if defined(TCP_USER_TIMEOUT)
+	case TCP_OPT_USER_TIMEOUT:
+            DDBG(desc,
+                 ("INET-DRV-DBG[%d][" SOCKET_FSTR ",%T] "
+                  "inet_set_opts(user_timeout) -> %d\r\n",
+                  __LINE__, desc->s, driver_caller(desc->port), ival) );
+	    proto = IPPROTO_TCP;
+	    type = TCP_USER_TIMEOUT;
+	    break;
+#endif
+
 #if defined(HAVE_MULTICAST_SUPPORT) && defined(IPPROTO_IP)
 
 	case UDP_OPT_MULTICAST_TTL:
@@ -9385,6 +9433,30 @@ static ErlDrvSSizeT inet_fill_opts(inet_descriptor* desc,
 	    put_int32(0, ptr);
 	    continue;
 #endif
+#if defined(TCP_KEEPCNT)
+	case TCP_OPT_KEEPCNT:
+	    proto = IPPROTO_TCP;
+	    type = TCP_KEEPCNT;
+	    break;
+#endif
+#if defined(TCP_KEEPIDLE)
+	case TCP_OPT_KEEPIDLE:
+	    proto = IPPROTO_TCP;
+	    type = TCP_KEEPIDLE;
+	    break;
+#endif
+#if defined(TCP_KEEPINTVL)
+	case TCP_OPT_KEEPINTVL:
+	    proto = IPPROTO_TCP;
+	    type = TCP_KEEPINTVL;
+	    break;
+#endif
+#if defined(TCP_USER_TIMEOUT)
+	case TCP_OPT_USER_TIMEOUT:
+	    proto = IPPROTO_TCP;
+	    type = TCP_USER_TIMEOUT;
+	    break;
+#endif
 
 #if defined(HAVE_MULTICAST_SUPPORT) && defined(IPPROTO_IP)
 	case UDP_OPT_MULTICAST_TTL:

erts/emulator/nifs/common/prim_socket_int.h

@@ -567,6 +567,11 @@ extern BOOLEAN_T esock_getopt_int(SOCKET sock,
                                   int    opt,
                                   int*   valP);
 
+extern BOOLEAN_T esock_getopt_uint(SOCKET        sock,
+                                   int           level,
+                                   int           opt,
+                                   unsigned int *valP);
+
 
 /* ** Socket Registry functions *** */
 extern void esock_send_reg_add_msg(ErlNifEnv*   env,

erts/emulator/nifs/common/prim_socket_nif.c

@@ -1310,6 +1310,10 @@ static ERL_NIF_TERM esock_getopt_int_opt(ErlNifEnv*       env,
                                          ESockDescriptor* descP,
                                          int              level,
                                          int              opt);
+static ERL_NIF_TERM esock_getopt_uint_opt(ErlNifEnv*       env,
+                                          ESockDescriptor* descP,
+                                          int              level,
+                                          int              opt);
 static ERL_NIF_TERM esock_getopt_size_opt(ErlNifEnv*       env,
                                           ESockDescriptor* descP,
                                           int              level,
@@ -1719,6 +1723,11 @@ static ERL_NIF_TERM esock_setopt_int_opt(ErlNifEnv*       env,
                                          int              level,
                                          int              opt,
                                          ERL_NIF_TERM     eVal);
+static ERL_NIF_TERM esock_setopt_uint_opt(ErlNifEnv*       env,
+                                          ESockDescriptor* descP,
+                                          int              level,
+                                          int              opt,
+                                          ERL_NIF_TERM     eVal);
 #if (defined(SO_RCVTIMEO) || defined(SO_SNDTIMEO))      \
     && defined(ESOCK_USE_RCVSNDTIMEO)
 static ERL_NIF_TERM esock_setopt_timeval_opt(ErlNifEnv*       env,
@@ -3762,7 +3771,14 @@ static struct ESockOpt optLevelTCP[] =
 #endif
         &esock_atom_nopush},
         {0, NULL, NULL, &esock_atom_syncnt},
-        {0, NULL, NULL, &esock_atom_user_timeout}
+        {
+#ifdef TCP_USER_TIMEOUT
+            TCP_USER_TIMEOUT,
+            esock_setopt_uint_opt, esock_getopt_uint_opt,
+#else
+            0, NULL, NULL,
+#endif
+            &esock_atom_user_timeout}
 
     };
 
@@ -8302,6 +8318,31 @@ ERL_NIF_TERM esock_setopt_int_opt(ErlNifEnv*       env,
 
 
 
+/* esock_setopt_uint_opt - set an option that has an unsigned integer value
+ */
+
+static
+ERL_NIF_TERM esock_setopt_uint_opt(ErlNifEnv*       env,
+                                   ESockDescriptor* descP,
+                                   int              level,
+                                   int              opt,
+                                   ERL_NIF_TERM     eVal)
+{
+    ERL_NIF_TERM result;
+    unsigned int val;
+
+    if (GET_UINT(env, eVal, &val)) {
+        result =
+            esock_setopt_level_opt(env, descP, level, opt,
+                                   &val, sizeof(val));
+    } else {
+        result = esock_make_invalid(env, esock_atom_value);
+    }
+    return result;
+}
+
+
+
 /* esock_setopt_str_opt - set an option that has an string value
  */
 
@@ -9908,6 +9949,24 @@ ERL_NIF_TERM esock_getopt_int_opt(ErlNifEnv*       env,
 
 
 
+/* esock_getopt_uint_opt - get an unsigned integer option
+ */
+static
+ERL_NIF_TERM esock_getopt_uint_opt(ErlNifEnv*       env,
+                                   ESockDescriptor* descP,
+                                   int              level,
+                                   int              opt)
+{
+    unsigned int val;
+
+    if (! esock_getopt_uint(descP->sock, level, opt, &val))
+        return esock_make_error_errno(env, sock_errno());
+
+    return esock_make_ok2(env, MKUI(env, val));
+}
+
+
+
 /* esock_getopt_int - get an integer option
  */
 extern
@@ -9932,6 +9991,30 @@ BOOLEAN_T esock_getopt_int(SOCKET           sock,
 
 
 
+/* esock_getopt_uint - get an unsigned integer option
+ */
+extern
+BOOLEAN_T esock_getopt_uint(SOCKET           sock,
+                            int              level,
+                            int              opt,
+                            unsigned int    *valP)
+{
+    unsigned int val = 0;
+    SOCKOPTLEN_T valSz = sizeof(val);
+
+#ifdef __WIN32__
+    if (sock_getopt(sock, level, opt, (char*) &val, &valSz) != 0)
+#else
+    if (sock_getopt(sock, level, opt, &val, &valSz) != 0)
+#endif
+        return FALSE;
+
+    *valP = val;
+    return TRUE;
+}
+
+
+
 static
 ERL_NIF_TERM esock_getopt_size_opt(ErlNifEnv*       env,
                                    ESockDescriptor* descP,

erts/preloaded/ebin/prim_inet.beam

@@ -1585,7 +1585,11 @@ enc_opt(line_delimiter)  -> ?INET_LOPT_LINE_DELIM;
 enc_opt(raw)             -> ?INET_OPT_RAW;
 enc_opt(bind_to_device)  -> ?INET_OPT_BIND_TO_DEVICE;
 enc_opt(read_ahead)      -> ?INET_LOPT_TCP_READ_AHEAD;
-enc_opt(non_block_send)  -> ?INET_OPT_NON_BLOCK_SEND;
+enc_opt(non_block_send)  -> ?INET_LOPT_NON_BLOCK_SEND;
+enc_opt(keepcnt)         -> ?TCP_OPT_KEEPCNT;
+enc_opt(keepidle)        -> ?TCP_OPT_KEEPIDLE;
+enc_opt(keepintvl)       -> ?TCP_OPT_KEEPINTVL;
+enc_opt(user_timeout)    -> ?TCP_OPT_USER_TIMEOUT;
 enc_opt(debug)           -> ?INET_OPT_DEBUG;
 % Names of SCTP opts:
 enc_opt(sctp_rtoinfo)	 	   -> ?SCTP_OPT_RTOINFO;
@@ -1658,7 +1662,11 @@ dec_opt(?INET_LOPT_LINE_DELIM)      -> line_delimiter;
 dec_opt(?INET_OPT_RAW)              -> raw;
 dec_opt(?INET_OPT_BIND_TO_DEVICE) -> bind_to_device;
 dec_opt(?INET_LOPT_TCP_READ_AHEAD) -> read_ahead;
-dec_opt(?INET_OPT_NON_BLOCK_SEND) -> non_block_send;
+dec_opt(?INET_LOPT_NON_BLOCK_SEND) -> non_block_send;
+dec_opt(?TCP_OPT_KEEPCNT)         -> keepcnt;
+dec_opt(?TCP_OPT_KEEPIDLE)        -> keepidle;
+dec_opt(?TCP_OPT_KEEPINTVL)       -> keepintvl;
+dec_opt(?TCP_OPT_USER_TIMEOUT)    -> user_timeout;
 dec_opt(?INET_OPT_DEBUG)          -> debug;
 dec_opt(I) when is_integer(I)     -> undefined.
 
@@ -1773,6 +1781,10 @@ type_opt_1(show_econnreset) -> bool;
 type_opt_1(bind_to_device)  -> binary;
 type_opt_1(read_ahead)      -> bool;
 type_opt_1(non_block_send)  -> bool;
+type_opt_1(keepcnt)         -> int;
+type_opt_1(keepidle)        -> int;
+type_opt_1(keepintvl)       -> int;
+type_opt_1(user_timeout)    -> uint;
 type_opt_1(debug)           -> bool;
 %% 
 %% SCTP options (to be set). If the type is a record type, the corresponding

erts/preloaded/src/prim_inet.erl

@@ -626,16 +626,19 @@ _Table: ipv6 options_
 [](){: #socket_options_tcp }
 Options for level `tcp`:
 
-| Option Name | Value Type | Set | Get | Other Requirements and comments                                                                          |
-| ----------- | ---------- | --- | --- | -------------------------------------------------------------------------------------------------------- |
-| congestion  | string()   | yes | yes | none                                                                                                     |
-| cork        | boolean()  | yes | yes | 'nopush' one some platforms (FreeBSD)                                                                    |
-| keepcnt     | integer()  | yes | yes | On Windows (at least), it is illegal to set to a value greater than 255.                                 |
-| keepidle    | integer()  | yes | yes | none                                                                                                     |
-| keepintvl   | integer()  | yes | yes | none                                                                                                     |
-| maxseg      | integer()  | yes | yes | Set not allowed on all platforms.                                                                        |
-| nodelay     | boolean()  | yes | yes | none                                                                                                     |
-| nopush      | boolean()  | yes | yes | 'cork' on some platforms (Linux). On Darwin this has a different meaning than on, for instance, FreeBSD. |
+| Option Name  | Value Type     | Set | Get | Other Requirements and comments                      |
+| ------------ | -------------- | --- | --- | ---------------------------------------------------- |
+| congestion   | string()       | yes | yes | none                                                 |
+| cork         | boolean()      | yes | yes | 'nopush' one some platforms (FreeBSD)                |
+| keepcnt      | integer()      | yes | yes | On Windows (at least), it is illegal to set to       |
+|              |                |     |     | a value greater than 255.                            |
+| keepidle     | integer()      | yes | yes | none                                                 |
+| keepintvl    | integer()      | yes | yes | none                                                 |
+| maxseg       | integer()      | yes | yes | Set not allowed on all platforms.                    |
+| nodelay      | boolean()      | yes | yes | none                                                 |
+| nopush       | boolean()      | yes | yes | 'cork' on some platforms (Linux). On Darwin this has |
+|              |                |     |     | a different meaning than on, for instance, FreeBSD.  |
+| user_timeout | integer() >= 0 | yes | yes | none                                                 |
 
 _Table: tcp options_
 

lib/kernel/doc/guides/socket_usage.md

@@ -274,6 +274,9 @@ way, option `send_timeout` comes in handy.
         {high_msgq_watermark, pos_integer()} |
         {high_watermark,  non_neg_integer()} |
         {keepalive,       boolean()} |
+        {keepcnt,         integer()} |
+        {keepidle,        integer()} |
+        {keepintvl,       integer()} |
         {linger,          {boolean(), non_neg_integer()}} |
         {low_msgq_watermark, pos_integer()} |
         {low_watermark,   non_neg_integer()} |
@@ -302,6 +305,7 @@ way, option `send_timeout` comes in handy.
 	{recvtos,         boolean()} |
 	{recvtclass,      boolean()} |
 	{recvttl,         boolean()} |
+        {user_timeout,    non_neg_integer()} |
 	{ipv6_v6only,     boolean()}.
 
 -doc """
@@ -337,6 +341,9 @@ this value is returned from `inet:getopts/2` when called with the option name
         high_msgq_watermark |
         high_watermark |
         keepalive |
+        keepcnt |
+        keepidle |
+        keepintvl |
         linger |
         low_msgq_watermark |
         low_watermark |
@@ -365,6 +372,7 @@ this value is returned from `inet:getopts/2` when called with the option name
         recvtclass |
         recvttl |
         pktoptions |
+        user_timeout |
 	ipv6_v6only.
 -type connect_option() ::
         {fd, Fd :: non_neg_integer()} |

lib/kernel/src/gen_tcp.erl

@@ -1059,7 +1059,14 @@ split_open_opts([], OpenOpts, OtherOpts) ->
 split_open_opts([{debug, _} = Opt|Opts], OpenOpts, OtherOpts) ->
     split_open_opts(Opts, [Opt|OpenOpts], OtherOpts);
 split_open_opts([Opt|Opts], OpenOpts, OtherOpts) ->
-    split_open_opts(Opts, OpenOpts, [Opt|OtherOpts]).
+    case Opt of
+        {debug, _} ->
+            split_open_opts(Opts, [Opt|OpenOpts], OtherOpts);
+        {protocol, _} ->
+            split_open_opts(Opts, [Opt|OpenOpts], OtherOpts);
+        _ ->
+            split_open_opts(Opts, OpenOpts, [Opt|OtherOpts])
+    end.
 
 
 %%
@@ -1277,7 +1284,11 @@ socket_opts() ->
 
           %%
           %% Level: tcp
-          nodelay => {tcp, nodelay},
+          keepcnt      => {tcp, keepcnt},
+          keepidle     => {tcp, keepidle},
+          keepintvl    => {tcp, keepintvl},
+          nodelay      => {tcp, nodelay},
+          user_timeout => {tcp, user_timeout},
 
           %%
           %% Level: ip
@@ -1585,9 +1596,10 @@ socket_open(Domain, #{fd := FD} = ExtraOpts, Extra) ->
     %% ?DBG([{fd, FD}, {opts, Opts}]),
     socket:open(FD, Opts);
 socket_open(Domain, ExtraOpts, Extra) ->
+    Proto = maps:get(protocol, ExtraOpts, proto(Domain)),
     Opts = maps:merge(Extra, ExtraOpts),
     %% ?DBG([{domain, Domain}, {extra_opts, ExtraOpts}, {extra, Extra}]),
-    socket:open(Domain, stream, proto(Domain), Opts).
+    socket:open(Domain, stream, Proto, Opts).
 
 proto(Domain) ->
     case Domain of