[erts, kernel] reuseaddr/reuseport/exclusiveaddruse support/fixes

erts/preloaded/src/erts.app.src

@@ -42,7 +42,7 @@
 	{registered, []},
 	{applications, []},
 	{env, []},
-	{runtime_dependencies, ["stdlib-4.1", "kernel-8.5", "sasl-3.3"]}
+	{runtime_dependencies, ["stdlib-4.1", "kernel-@OTP-18344@", "sasl-3.3"]}
     ]}.
 
 %% vim: ft=erlang

erts/preloaded/src/prim_inet.erl

@@ -1513,6 +1513,9 @@ is_sockopt_val(Opt, Val) ->
 %% Socket options processing: Encoding option NAMES:
 %%
 enc_opt(reuseaddr)       -> ?INET_OPT_REUSEADDR;
+enc_opt(reuseport)       -> ?INET_OPT_REUSEPORT;
+enc_opt(reuseport_lb)    -> ?INET_OPT_REUSEPORT_LB;
+enc_opt(exclusiveaddruse) -> ?INET_OPT_EXCLUSIVEADDRUSE;
 enc_opt(keepalive)       -> ?INET_OPT_KEEPALIVE;
 enc_opt(dontroute)       -> ?INET_OPT_DONTROUTE;
 enc_opt(linger)          -> ?INET_OPT_LINGER;
@@ -1581,6 +1584,9 @@ enc_opt(sctp_get_peer_addr_info)   -> ?SCTP_OPT_GET_PEER_ADDR_INFO.
 %% Decoding option NAMES:
 %%
 dec_opt(?INET_OPT_REUSEADDR)      -> reuseaddr;
+dec_opt(?INET_OPT_REUSEPORT)      -> reuseport;
+dec_opt(?INET_OPT_REUSEPORT_LB)   -> reuseport_lb;
+dec_opt(?INET_OPT_EXCLUSIVEADDRUSE) -> exclusiveaddruse;
 dec_opt(?INET_OPT_KEEPALIVE)      -> keepalive;
 dec_opt(?INET_OPT_DONTROUTE)      -> dontroute;
 dec_opt(?INET_OPT_LINGER)         -> linger;
@@ -1664,6 +1670,9 @@ type_opt(_,   Opt) ->
 %% Types of option values, by option name:
 %%
 type_opt_1(reuseaddr)       -> bool;
+type_opt_1(reuseport)       -> bool;
+type_opt_1(reuseport_lb)    -> bool;
+type_opt_1(exclusiveaddruse) -> bool;
 type_opt_1(keepalive)       -> bool;
 type_opt_1(dontroute)       -> bool;
 type_opt_1(linger)          -> {bool,int};

lib/kernel/doc/src/inet.xml

@@ -1092,6 +1092,22 @@ get_tcpi_sacked(Sock) ->
               <seemfa marker="gen_tcp#shutdown/2"><c>gen_tcp:shutdown/2</c></seemfa>
               to shut down the write side.</p>
           </item>
+          <tag><c>{exclusiveaddruse, Boolean}</c>
+          <marker id="option-exclusiveaddruse"/></tag>
+          <item>
+            <p>
+              Enables/disables exclusive address/port usage on Windows. That
+              is, by enabling this option you can prevent other sockets from
+              binding to the same address/port. By default this option is
+              disabled. That is, other sockets may use the same address/port
+              by setting <seeerl marker="#option-reuseaddr"><c>{reuseaddr,
+              true}</c></seeerl> in combination with
+              <seeerl marker="#option-reuseport"><c>{reuseport,
+              true}</c></seeerl> unless <c>{exclusiveaddruse, true}</c>
+              has been set on <c><anno>Socket</anno></c>. On non-Windows
+              systems this option is silently ignored.
+            </p>
+          </item>
           <tag><c>{header, Size}</c></tag>
           <item>
             <p>This option is only meaningful if option <c>binary</c>
@@ -1566,18 +1582,94 @@ setcap cap_sys_admin,cap_sys_ptrace,cap_dac_read_search+epi beam.smp</code>
 	      option.
 	    </p>
           </item>
-          <tag><c>{reuseaddr, Boolean}</c></tag>
+          <tag><c>{reuseaddr, Boolean}</c><marker id="option-reuseaddr"/></tag>
+          <item>
+            <p>
+              Allows or disallows reuse of local address. By default, reuse
+              is disallowed.
+            </p>
+            <note>
+              <p>
+                On windows <c>{reuseaddr, true}</c> will have no effect unless
+                also <seeerl marker="#option-reuseport"><c>{reuseport,
+                true}</c></seeerl> is set. If both are set, the
+                <c>SO_REUSEADDR</c> Windows socket option will be enabled.
+                This since setting <c>SO_REUSEADDR</c> on Windows more or less
+                has the same behavior as setting both <c>SO_REUSEADDR</c> and
+                <c>SO_REUSEPORT</c> on BSD. This behavior was introduced as
+                of OTP @OTP-18324@.
+              </p>
+              <p>
+                Between OTP 25.0 and up to the predecessor of OTP @OTP-18324@,
+                the <c>SO_REUSEADDR</c> option was set on Windows if
+                <c>{reuseaddr, true}</c> was set. Prior to OTP 25.0, the
+                <c>{reuseaddr, true}</c> option on Windows was silently
+                ignored.
+              </p>
+              <p>
+                See also the
+                <seeerl marker="#option-exclusiveaddruse"><c>exclusiveaddruse</c></seeerl>
+                option.
+              </p>
+            </note>
+          </item>
+          <tag><c>{reuseport, Boolean}</c><marker id="option-reuseport"/></tag>
+          <item>
+            <p>
+              Allows or disallows reuse of local port which <i>may or may not</i>
+              have load balancing depending on the underlying OS. By default,
+              reuse is disallowed. See also
+              <seeerl marker="#option-reuseport_lb"><c>reuseport_lb</c></seeerl>.
+            </p>
+            <note>
+              <p>
+                On windows <c>{reuseport, true}</c> will have no effect unless
+                also <seeerl marker="#option-reuseaddr"><c>{reuseaddr,
+                true}</c></seeerl> is set. If both are set, the
+                <c>SO_REUSEADDR</c> Windows socket option will be enabled.
+                This since setting <c>SO_REUSEADDR</c> on Windows more or less
+                has the same behavior as setting both <c>SO_REUSEADDR</c> and
+                <c>SO_REUSEPORT</c> on BSD. The <c>reuseport</c> option was
+                introduced as of OTP @OTP-18324@.
+              </p>
+              <p>
+                See also the
+                <seeerl marker="#option-exclusiveaddruse"><c>exclusiveaddruse</c></seeerl>
+                option.
+              </p>
+            </note>
+            <note>
+              <p>
+                <c>reuseport</c> <i>may or may not</i> be the same underlying
+                option as
+                <seeerl marker="#option-reuseport_lb"><c>reuseport_lb</c></seeerl>
+                depending on the underlying OS. They, for example, are on Linux.
+                When they are the same underlying option, operating on both may
+                cause them to interact in surprising ways. For example,
+                by enabling <c>reuseport</c> and then disabling
+                <c>reuseport_lb</c> both will end up being disabled.
+              </p>
+            </note>
+          </item>
+          <tag><c>{reuseport_lb, Boolean}</c><marker id="option-reuseport_lb"/></tag>
           <item>
             <p>
-              Allows or disallows local reuse of address. By
-              default, reuse is disallowed.
+              Allows or disallows reuse of local port <i>with</i> load balancing.
+              By default, reuse is disallowed. See also
+              <seeerl marker="#option-reuseport"><c>reuseport</c></seeerl>.
             </p>
-            <note><p>
-              On Windows this option will be ignored unless
-              <c><anno>Socket</anno></c> is an UDP socket. This since the
-              behavior of <c>reuseaddr</c> is very different on Windows
-              compared to other system.
-            </p></note>
+            <note>
+              <p>
+                <c>reuseport_lb</c> <i>may or may not</i> be the same underlying
+                option as
+                <seeerl marker="#option-reuseport"><c>reuseport</c></seeerl>
+                depending on the underlying OS. They, for example, are on Linux.
+                When they are the same underlying option, operating on both may
+                cause them to interact in surprising ways. For example,
+                by enabling <c>reuseport_lb</c> and then disabling
+                <c>reuseport</c> both will end up being disabled.
+              </p>
+            </note>
           </item>
           <tag><c>{send_timeout, Integer}</c></tag>
           <item>

lib/kernel/src/gen_sctp.erl

@@ -58,13 +58,16 @@
         {buffer, non_neg_integer()} |
         {debug, boolean()} |
         {dontroute, boolean()} |
+        {exclusiveaddruse, boolean()} |
         {high_msgq_watermark, pos_integer()} |
         {linger, {boolean(), non_neg_integer()}} |
         {low_msgq_watermark, pos_integer()} |
         {mode, list | binary} | list | binary |
         {priority, non_neg_integer()} |
         {recbuf, non_neg_integer()} |
         {reuseaddr, boolean()} |
+        {reuseport, boolean()} |
+        {reuseport_lb, boolean()} |
 	{ipv6_v6only, boolean()} |
         {sndbuf, non_neg_integer()} |
         {sctp_autoclose, non_neg_integer()} |
@@ -84,13 +87,16 @@
         buffer |
         debug |
         dontroute |
+        exclusiveaddruse |
         high_msgq_watermark |
         linger |
         low_msgq_watermark |
         mode |
         priority |
         recbuf |
         reuseaddr |
+        reuseport |
+        reuseport_lb |
 	ipv6_v6only |
         sctp_autoclose |
         sctp_disable_fragments |

lib/kernel/src/gen_tcp.erl

@@ -43,6 +43,7 @@
         {deliver,         port | term} |
         {dontroute,       boolean()} |
         {exit_on_close,   boolean()} |
+        {exclusiveaddruse, boolean()} |
         {header,          non_neg_integer()} |
         {high_msgq_watermark, pos_integer()} |
         {high_watermark,  non_neg_integer()} |
@@ -63,6 +64,8 @@
          ValueBin :: binary()} |
         {recbuf,          non_neg_integer()} |
         {reuseaddr,       boolean()} |
+        {reuseport,       boolean()} |
+        {reuseport_lb,    boolean()} |
         {send_timeout,    non_neg_integer() | infinity} |
         {send_timeout_close, boolean()} |
         {show_econnreset, boolean()} |
@@ -84,6 +87,7 @@
         deliver |
         dontroute |
         exit_on_close |
+        exclusiveaddruse |
         header |
         high_msgq_watermark |
         high_watermark |
@@ -103,6 +107,8 @@
                       (ValueBin :: binary())} |
         recbuf |
         reuseaddr |
+        reuseport |
+        reuseport_lb |
         send_timeout |
         send_timeout_close |
         show_econnreset |

lib/kernel/src/gen_udp.erl

@@ -38,6 +38,7 @@
         {deliver,         port | term} |
         {dontroute,       boolean()} |
         {drop_membership, membership()} |
+        {exclusiveaddruse, boolean()} |
         {header,          non_neg_integer()} |
         {high_msgq_watermark, pos_integer()} |
         {low_msgq_watermark, pos_integer()} |
@@ -53,6 +54,8 @@
         {read_packets,    non_neg_integer()} |
         {recbuf,          non_neg_integer()} |
         {reuseaddr,       boolean()} |
+        {reuseport,       boolean()} |
+        {reuseport_lb,    boolean()} |
         {sndbuf,          non_neg_integer()} |
         {tos,             non_neg_integer()} |
         {tclass,          non_neg_integer()} |
@@ -68,6 +71,7 @@
         debug |
         deliver |
         dontroute |
+        exclusiveaddruse |
         header |
         high_msgq_watermark |
         low_msgq_watermark |
@@ -84,6 +88,8 @@
         read_packets |
         recbuf |
         reuseaddr |
+        reuseport |
+        reuseport_lb |
         sndbuf |
         tos |
         tclass |

lib/kernel/src/inet.erl

@@ -953,9 +953,9 @@ stats() ->
 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 connect_options() ->
     [debug,
-     tos, tclass, priority, reuseaddr, keepalive, linger, nodelay,
-     sndbuf, recbuf,
-     recvtos, recvtclass, ttl, recvttl,
+     tos, tclass, priority, reuseaddr, reuseport, reuseport_lb,
+     exclusiveaddruse, keepalive,
+     linger, nodelay, sndbuf, recbuf, recvtos, recvtclass, ttl, recvttl,
      header, active, packet, packet_size, buffer, mode, deliver, line_delimiter,
      exit_on_close, high_watermark, low_watermark, high_msgq_watermark,
      low_msgq_watermark, send_timeout, send_timeout_close, delay_send, raw,
@@ -1044,8 +1044,8 @@ con_add(Name, Val, #connect_opts{} = R, Opts, AllOpts) ->
 listen_options() ->
     [debug,
      tos, tclass,
-     priority, reuseaddr, keepalive, linger, sndbuf, recbuf, nodelay,
-     recvtos, recvtclass, ttl, recvttl,
+     priority, reuseaddr, reuseport, reuseport_lb, exclusiveaddruse, keepalive,
+     linger, sndbuf, recbuf, nodelay, recvtos, recvtclass, ttl, recvttl,
      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,

lib/kernel/src/inet_int.hrl

@@ -135,6 +135,9 @@
 -define(UDP_OPT_ADD_MEMBERSHIP,  14).
 -define(UDP_OPT_DROP_MEMBERSHIP, 15).
 -define(INET_OPT_IPV6_V6ONLY,    16).
+-define(INET_OPT_REUSEPORT,      17).
+-define(INET_OPT_REUSEPORT_LB,   18).
+-define(INET_OPT_EXCLUSIVEADDRUSE, 19).
 % "Local" options: codes start from 20:
 -define(INET_LOPT_BUFFER,        20).
 -define(INET_LOPT_HEADER,        21).

lib/kernel/src/kernel.app.src

@@ -158,7 +158,7 @@
          {shell_docs_ansi,auto}
         ]},
   {mod, {kernel, []}},
-  {runtime_dependencies, ["erts-@OTP-18248:OTP-18324@", "stdlib-@OTP-17932@",
+  {runtime_dependencies, ["erts-@OTP-18248:OTP-18324:OTP-18344@", "stdlib-@OTP-17932@",
                           "sasl-3.0", "crypto-5.0"]}
   ]
 }.

lib/kernel/test/inet_sockopt_SUITE.erl

@@ -863,9 +863,14 @@ make_check_fun(Type,Element) ->
 
 %% {OptionName,Value1,Value2,Mandatory,Changeable}
 all_listen_options() ->
+    OsType = os:type(),
+    OsVersion = os:version(),
     [{tos,0,1,false,true}, 
      {priority,0,1,false,true}, 
-     {reuseaddr,false,true,false,true}, 
+     {reuseaddr,false,true,mandatory_reuseaddr(OsType,OsVersion),true},
+     {reuseport,false,true,mandatory_reuseport(OsType,OsVersion),true},
+     {reuseport_lb,false,true,mandatory_reuseport_lb(OsType,OsVersion),true},
+     {exclusiveaddruse,false,true,mandatory_exclusiveaddruse(OsType,OsVersion),true},
      {keepalive,false,true,true,true}, 
      {linger, {false,10}, {true,10},true,true},
      {sndbuf,2048,4096,false,true}, 
@@ -888,9 +893,14 @@ all_listen_options() ->
      {packet_size,0,4,true,true}
     ].
 all_connect_options() ->
+    OsType = os:type(),
+    OsVersion = os:version(),
     [{tos,0,1,false,true}, 
      {priority,0,1,false,true}, 
-     {reuseaddr,false,true,false,true}, 
+     {reuseaddr,false,true,mandatory_reuseaddr(OsType,OsVersion),true},
+     {reuseport,false,true,mandatory_reuseport(OsType,OsVersion),true},
+     {reuseport_lb,false,true,mandatory_reuseport_lb(OsType,OsVersion),true},
+     {exclusiveaddruse,false,true,mandatory_exclusiveaddruse(OsType,OsVersion),true},
      {keepalive,false,true,true,true}, 
      {linger, {false,10}, {true,10},true,true},
      {sndbuf,2048,4096,false,true}, 
@@ -914,6 +924,46 @@ all_connect_options() ->
     ].
 
 
+%% Mandatory on a lot of system other than those listed below. Please add more...
+mandatory_reuseaddr({unix, linux}, _OsVersion) ->
+    true;
+mandatory_reuseaddr({unix, freebsd}, _OsVersion) ->
+    true;
+mandatory_reuseaddr({unix, darwin}, _OsVersion) ->
+    true;
+mandatory_reuseaddr({win32, _}, _OsVersion) ->
+    true; %% reuseaddr and reuseport are emulated by the inet-driver
+mandatory_reuseaddr(_OsType, _OsVersion) ->
+    false.
+
+%% Mandatory an a lot of system other than those listed below. Please add more...
+mandatory_reuseport({win32, _}, _OsVersion) ->
+    true; %% reuseaddr and reuseport are emulated by the inet-driver
+mandatory_reuseport({unix, linux}, {X,Y,_Z}) when X > 4 orelse X == 4 andalso Y >= 6 ->
+    true;
+mandatory_reuseport({unix, freebsd}, {X,Y,_Z}) when X > 11 orelse X == 11 andalso Y >= 4 ->
+    %% I know that it is available on 11.4, but it may be available earlier...
+    true;
+mandatory_reuseport({unix, darwin}, {X,Y,_Z}) when X > 26 orelse X == 26 andalso Y >= 6 ->
+    %% I know that it is available on 26.6, but it may be available earlier...
+    true;
+mandatory_reuseport(_OsType, _OsVersion) ->
+    false.
+
+%% Perhaps mandatory an system other than those listed below. Please add more...
+mandatory_reuseport_lb({unix, linux}, {X,Y,_Z}) when X > 4 orelse X == 4 andalso Y >= 6 ->
+    true;
+mandatory_reuseport_lb({unix, freebsd}, {X,Y,_Z}) when X > 13 orelse X == 13 andalso Y >= 1 ->
+    %% I know that it is available on 13.1, but it may be available earlier...
+    true;
+mandatory_reuseport_lb(_OsType, _OsVersion) ->
+    false.
+
+mandatory_exclusiveaddruse({win32, _}, {X,Y,_Z}) when X > 5 orelse X == 5 andalso Y >= 2 ->
+    true;
+mandatory_exclusiveaddruse(_OsType, _OsVersion) ->
+    false.
+
 create_socketpair(ListenOptions,ConnectOptions) ->
     {ok,LS}=gen_tcp:listen(0,ListenOptions),
     {ok,Port}=inet:port(LS),