Merge branch 'raimo/diffserv-socket-option/OTP-13582' into maint

* raimo/diffserv-socket-option/OTP-13582:
  Tune 'tclass' semantics
  Implement IPV6_TCLASS

7 files changed, 97 insertions, 7 deletions

diff --git a/erts/emulator/drivers/common/inet_drv.c b/erts/emulator/drivers/common/inet_drv.c
index 254d3baeb1..1885338ce5 100644
--- a/erts/emulator/drivers/common/inet_drv.c
+++ b/erts/emulator/drivers/common/inet_drv.c
@@ -777,6 +777,7 @@ static int is_nonzero(const char *s, size_t n)
 #define INET_LOPT_NETNS             38  /* Network namespace pathname */
 #define INET_LOPT_TCP_SHOW_ECONNRESET 39  /* tell user about incoming RST */
 #define INET_LOPT_LINE_DELIM        40  /* Line delimiting char */
+#define INET_OPT_TCLASS             41  /* IPv6 transport class */
 /* SCTP options: a separate range, from 100: */
 #define SCTP_OPT_RTOINFO		100
 #define SCTP_OPT_ASSOCINFO		101
@@ -1330,6 +1331,7 @@ static ErlDrvTermData am_reuseaddr;
 static ErlDrvTermData am_dontroute;
 static ErlDrvTermData am_priority;
 static ErlDrvTermData am_tos;
+static ErlDrvTermData am_tclass;
 static ErlDrvTermData am_ipv6_v6only;
 static ErlDrvTermData am_netns;
 #endif
@@ -3720,6 +3722,7 @@ static void inet_init_sctp(void) {
     INIT_ATOM(dontroute);
     INIT_ATOM(priority);
     INIT_ATOM(tos);
+    INIT_ATOM(tclass);
     INIT_ATOM(ipv6_v6only);
     INIT_ATOM(netns);
     
@@ -6228,6 +6231,15 @@ static int inet_set_opts(inet_descriptor* desc, char* ptr, int len)
 #else
 	    continue;
 #endif
+#if defined(IPV6_TCLASS) && defined(SOL_IPV6)
+	case INET_OPT_TCLASS:
+	    proto = SOL_IPV6;
+	    type = IPV6_TCLASS;
+	    propagate = 1;
+	    DEBUGF(("inet_set_opts(%ld): s=%d, IPV6_TCLASS=%d\r\n",
+		    (long)desc->port, desc->s, ival));
+	    break;
+#endif
 
 	case TCP_OPT_NODELAY:
 	    proto = IPPROTO_TCP; 
@@ -6661,6 +6673,19 @@ static int sctp_set_opts(inet_descriptor* desc, char* ptr, int len)
 	    continue; /* Option not supported -- ignore it */
 #	endif
 
+#       if defined(IPV6_TCLASS) && defined(SOL_IPV6)
+	case INET_OPT_TCLASS:
+	{
+	    arg.ival= get_int32 (curr);	  curr += 4;
+	    proto   = SOL_IPV6;
+	    type    = IPV6_TCLASS;
+	    arg_ptr = (char*) (&arg.ival);
+	    arg_sz  = sizeof  ( arg.ival);
+	    break;
+	}
+#	endif
+
+
 	case INET_OPT_IPV6_V6ONLY:
 #       if HAVE_DECL_IPV6_V6ONLY
 	{
@@ -7162,6 +7187,15 @@ static ErlDrvSSizeT inet_fill_opts(inet_descriptor* desc,
 	    put_int32(0, ptr);
 	    continue;
 #endif
+	case INET_OPT_TCLASS:
+#if defined(IPV6_TCLASS) && defined(SOL_IPV6)
+	    proto = SOL_IPV6;
+	    type = IPV6_TCLASS;
+	    break;
+#else
+	    TRUNCATE_TO(0,ptr);
+	    continue;
+#endif
 	case INET_OPT_REUSEADDR: 
 	    type = SO_REUSEADDR; 
 	    break;
@@ -7556,6 +7590,7 @@ static ErlDrvSSizeT sctp_fill_opts(inet_descriptor* desc,
 	case INET_OPT_DONTROUTE:
 	case INET_OPT_PRIORITY :
 	case INET_OPT_TOS      :
+	case INET_OPT_TCLASS   :
 	case INET_OPT_IPV6_V6ONLY:
 	case SCTP_OPT_AUTOCLOSE:
 	case SCTP_OPT_MAXSEG   :
@@ -7629,6 +7664,19 @@ static ErlDrvSSizeT sctp_fill_opts(inet_descriptor* desc,
 		continue;
 #	    endif
 	    }
+	    case INET_OPT_TCLASS:
+	    {
+#           if defined(IPV6_TCLASS) && defined(SOL_IPV6)
+		proto  = SOL_IPV6;
+		type   = IPV6_TCLASS;
+		is_int = 1;
+		tag    = am_tclass;
+		break;
+#	    else
+		/* Not supported -- ignore */
+		continue;
+#	    endif
+	    }
 	    case INET_OPT_IPV6_V6ONLY:
 #           if HAVE_DECL_IPV6_V6ONLY
 	    {
diff --git a/erts/preloaded/ebin/prim_inet.beam b/erts/preloaded/ebin/prim_inet.beam
index ae4861c336..1a573ce297 100644
--- a/erts/preloaded/ebin/prim_inet.beam
+++ b/erts/preloaded/ebin/prim_inet.beam
diff --git a/erts/preloaded/src/prim_inet.erl b/erts/preloaded/src/prim_inet.erl
index bcf16402b0..61f727e8a4 100644
--- a/erts/preloaded/src/prim_inet.erl
+++ b/erts/preloaded/src/prim_inet.erl
@@ -347,7 +347,17 @@ accept_opts(L, S) ->
     case getopts(L, [active, nodelay, keepalive, delay_send, priority, tos]) of
 	{ok, Opts} ->
 	    case setopts(S, Opts) of
-		ok -> {ok, S};
+		ok ->
+		    case getopts(L, [tclass]) of
+			{ok, []} ->
+			    {ok, S};
+			{ok, TClassOpts} ->
+			    case setopts(S, TClassOpts) of
+				ok ->
+				    {ok, S};
+				Error -> close(S), Error
+			    end
+		    end;
 		Error -> close(S), Error
 	    end;
 	Error ->
@@ -1196,6 +1206,7 @@ enc_opt(sndbuf)          -> ?INET_OPT_SNDBUF;
 enc_opt(recbuf)          -> ?INET_OPT_RCVBUF;
 enc_opt(priority)        -> ?INET_OPT_PRIORITY;
 enc_opt(tos)             -> ?INET_OPT_TOS;
+enc_opt(tclass)          -> ?INET_OPT_TCLASS;
 enc_opt(nodelay)         -> ?TCP_OPT_NODELAY;
 enc_opt(multicast_if)    -> ?UDP_OPT_MULTICAST_IF;
 enc_opt(multicast_ttl)   -> ?UDP_OPT_MULTICAST_TTL;
@@ -1255,6 +1266,7 @@ dec_opt(?INET_OPT_SNDBUF)         -> sndbuf;
 dec_opt(?INET_OPT_RCVBUF)         -> recbuf;
 dec_opt(?INET_OPT_PRIORITY)       -> priority;
 dec_opt(?INET_OPT_TOS)            -> tos;
+dec_opt(?INET_OPT_TCLASS)         -> tclass;
 dec_opt(?TCP_OPT_NODELAY)         -> nodelay;
 dec_opt(?UDP_OPT_MULTICAST_IF)    -> multicast_if;
 dec_opt(?UDP_OPT_MULTICAST_TTL)   -> multicast_ttl;
@@ -1329,6 +1341,7 @@ type_opt_1(sndbuf)          -> int;
 type_opt_1(recbuf)          -> int;
 type_opt_1(priority)        -> int;
 type_opt_1(tos)             -> int;
+type_opt_1(tclass)          -> int;
 type_opt_1(nodelay)         -> bool;
 type_opt_1(ipv6_v6only)     -> bool;
 %% multicast
diff --git a/lib/kernel/doc/src/inet.xml b/lib/kernel/doc/src/inet.xml
index 3f4e3684f4..d907cef7d3 100644
--- a/lib/kernel/doc/src/inet.xml
+++ b/lib/kernel/doc/src/inet.xml
@@ -1092,6 +1092,15 @@ setcap cap_sys_admin,cap_sys_ptrace,cap_dac_read_search+epi beam.smp</code>
               The option is ignored on platforms where it is not
               implemented. Use with caution.</p>
           </item>
+          <tag><c>{tclass, Integer}</c></tag>
+          <item>
+            <p>
+	      Sets <c>IPV6_TCLASS IP</c> level options on platforms
+	      where this is implemented. The behavior and allowed range
+	      varies between different systems.
+              The option is ignored on platforms where it is not
+              implemented. Use with caution.</p>
+          </item>
         </taglist>
         <p>In addition to these options, <em>raw</em>
           option specifications can be used. The raw options are
diff --git a/lib/kernel/src/inet.erl b/lib/kernel/src/inet.erl
index 75dd800c6b..f5c13ecdd7 100644
--- a/lib/kernel/src/inet.erl
+++ b/lib/kernel/src/inet.erl
@@ -677,7 +677,7 @@ parse_strict_address(Addr) ->
 %% Return a list of available options
 options() ->
     [
-     tos, priority, reuseaddr, keepalive, dontroute, linger,
+     tos, tclass, priority, reuseaddr, keepalive, dontroute, linger,
      broadcast, sndbuf, recbuf, nodelay, ipv6_v6only,
      buffer, header, active, packet, deliver, mode,
      multicast_if, multicast_ttl, multicast_loop,
@@ -698,7 +698,7 @@ stats() ->
 %% Available options for tcp:connect
 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 connect_options() ->
-    [tos, priority, reuseaddr, keepalive, linger, sndbuf, recbuf, nodelay,
+    [tos, tclass, priority, reuseaddr, keepalive, linger, sndbuf, recbuf, nodelay,
      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,
@@ -766,7 +766,7 @@ con_add(Name, Val, #connect_opts{} = R, Opts, AllOpts) ->
 %% Available options for tcp:listen
 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 listen_options() ->
-    [tos, priority, reuseaddr, keepalive, linger, sndbuf, recbuf, nodelay,
+    [tos, tclass, priority, reuseaddr, keepalive, linger, sndbuf, recbuf, nodelay,
      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,
@@ -846,7 +846,7 @@ tcp_module_1(Opts, Address) ->
 %% Available options for udp:open
 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 udp_options() ->
-    [tos, priority, reuseaddr, sndbuf, recbuf, header, active, buffer, mode, 
+    [tos, tclass, priority, reuseaddr, sndbuf, recbuf, header, active, buffer, mode,
      deliver, ipv6_v6only,
      broadcast, dontroute, multicast_if, multicast_ttl, multicast_loop,
      add_membership, drop_membership, read_packets,raw,
@@ -917,7 +917,7 @@ udp_module(Opts) ->
 %  (*) passing of open FDs ("fdopen") is not supported.
 sctp_options() ->
 [   % The following are generic inet options supported for SCTP sockets:
-    mode, active, buffer, tos, priority, dontroute, reuseaddr, linger, sndbuf,
+    mode, active, buffer, tos, tclass, priority, dontroute, reuseaddr, linger, sndbuf,
     recbuf, ipv6_v6only, high_msgq_watermark, low_msgq_watermark,
 
     % Other options are SCTP-specific (though they may be similar to their
diff --git a/lib/kernel/src/inet_int.hrl b/lib/kernel/src/inet_int.hrl
index c8a8962e78..4e8f59a3b9 100644
--- a/lib/kernel/src/inet_int.hrl
+++ b/lib/kernel/src/inet_int.hrl
@@ -153,6 +153,7 @@
 -define(INET_LOPT_NETNS,          38).
 -define(INET_LOPT_TCP_SHOW_ECONNRESET, 39).
 -define(INET_LOPT_LINE_DELIM,     40).
+-define(INET_OPT_TCLASS,          41).
 % Specific SCTP options: separate range:
 -define(SCTP_OPT_RTOINFO,	 	100).
 -define(SCTP_OPT_ASSOCINFO,	 	101).
diff --git a/lib/kernel/test/gen_tcp_api_SUITE.erl b/lib/kernel/test/gen_tcp_api_SUITE.erl
index 77ec89b4f4..92a74465b7 100644
--- a/lib/kernel/test/gen_tcp_api_SUITE.erl
+++ b/lib/kernel/test/gen_tcp_api_SUITE.erl
@@ -38,7 +38,7 @@
 	 t_local_basic/1, t_local_unbound/1, t_local_fdopen/1,
 	 t_local_fdopen_listen/1, t_local_fdopen_listen_unbound/1,
 	 t_local_fdopen_connect/1, t_local_fdopen_connect_unbound/1,
-	 t_local_abstract/1]).
+	 t_local_abstract/1, t_accept_inet6_tclass/1]).
 
 -export([getsockfd/0,closesockfd/1]).
 
@@ -50,6 +50,7 @@ all() ->
     [{group, t_accept}, {group, t_connect}, {group, t_recv},
      t_shutdown_write, t_shutdown_both, t_shutdown_error,
      t_shutdown_async, t_fdopen, t_fdconnect, t_implicit_inet6,
+     t_accept_inet6_tclass,
      {group, t_local}].
 
 groups() -> 
@@ -521,6 +522,24 @@ local_handshake(S, SAddr, C, CAddr) ->
     SData = ok(gen_tcp:recv(C, length(SData))),
     ok.
 
+t_accept_inet6_tclass(Config) when is_list(Config) ->
+    TClassOpt = {tclass,8#56 bsl 2}, % Expedited forwarding
+    case gen_tcp:listen(0, [inet6,TClassOpt]) of
+	{ok,L} ->
+	    LPort = ok(inet:port(L)),
+	    Loopback = {0,0,0,0,0,0,0,1},
+	    Sa = ok(gen_tcp:connect(Loopback, LPort, [])),
+	    Sb = ok(gen_tcp:accept(L)),
+	    [TClassOpt] = ok(inet:getopts(Sb, [tclass])),
+	    ok = gen_tcp:close(Sb),
+	    ok = gen_tcp:close(Sa),
+	    ok = gen_tcp:close(L),
+	    ok;
+	{error,_} ->
+	    {skip,"IPv6 TCLASS not supported"}
+    end.
+
+
 %%% Utilities
 
 %% Calls M:F/length(A), which should return a timeout error, and complete