Merge branch 'igor/tcp-nopush-ERL-698/OTP-15357' into maint

* igor/tcp-nopush-ERL-698/OTP-15357:
  "cork" tcp socket around file:sendfile
  Add nopush TCP socket option

6 files changed, 76 insertions, 3 deletions

diff --git a/erts/emulator/drivers/common/inet_drv.c b/erts/emulator/drivers/common/inet_drv.c
index f663ab0b05..3195ca3874 100644
--- a/erts/emulator/drivers/common/inet_drv.c
+++ b/erts/emulator/drivers/common/inet_drv.c
@@ -812,6 +812,7 @@ static size_t my_strnlen(const char *s, size_t maxlen)
 #define INET_OPT_PKTOPTIONS         45  /* IP(V6)_PKTOPTIONS get ancillary data */
 #define INET_OPT_TTL                46  /* IP_TTL */
 #define INET_OPT_RECVTTL            47  /* IP_RECVTTL ancillary data */
+#define TCP_OPT_NOPUSH              48  /* super-Nagle, aka TCP_CORK */
 /* SCTP options: a separate range, from 100: */
 #define SCTP_OPT_RTOINFO		100
 #define SCTP_OPT_ASSOCINFO		101
@@ -955,6 +956,12 @@ static size_t my_strnlen(const char *s, size_t maxlen)
 #endif
 
 
+#if defined(TCP_CORK)
+#define INET_TCP_NOPUSH TCP_CORK
+#elif defined(TCP_NOPUSH) && !defined(__DARWIN__)
+#define INET_TCP_NOPUSH TCP_NOPUSH
+#endif
+
 #define BIN_REALLOC_MARGIN(x)  ((x)/4)  /* 25% */
 
 /* The general purpose sockaddr */
@@ -6598,6 +6605,19 @@ static int inet_set_opts(inet_descriptor* desc, char* ptr, int len)
 		    (long)desc->port, desc->s, ival));
 	    break;
 
+	case TCP_OPT_NOPUSH:
+#if defined(INET_TCP_NOPUSH)
+	    proto = IPPROTO_TCP;
+	    type = INET_TCP_NOPUSH;
+	    DEBUGF(("inet_set_opts(%ld): s=%d, t=%d TCP_NOPUSH=%d\r\n",
+	            (long)desc->port, desc->s, type, ival));
+	    break;
+#else
+	    /* inet_fill_opts always returns a value for this option,
+	     * so we need to ignore it if not implemented, just in case */
+	    continue;
+#endif
+
 #if defined(HAVE_MULTICAST_SUPPORT) && defined(IPPROTO_IP)
 
 	case UDP_OPT_MULTICAST_TTL:
@@ -7759,6 +7779,16 @@ static ErlDrvSSizeT inet_fill_opts(inet_descriptor* desc,
 	    proto = IPPROTO_TCP;
 	    type = TCP_NODELAY;
 	    break;
+	case TCP_OPT_NOPUSH:
+#if defined(INET_TCP_NOPUSH)
+	    proto = IPPROTO_TCP;
+	    type = INET_TCP_NOPUSH;
+	    break;
+#else
+	    *ptr++ = opt;
+	    put_int32(0, ptr);
+	    continue;
+#endif
 
 #if defined(HAVE_MULTICAST_SUPPORT) && defined(IPPROTO_IP)
 	case UDP_OPT_MULTICAST_TTL:
diff --git a/erts/preloaded/ebin/prim_inet.beam b/erts/preloaded/ebin/prim_inet.beam
index eebfe19a11..52bab031ff 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 1da852deb2..963e8933bc 100644
--- a/erts/preloaded/src/prim_inet.erl
+++ b/erts/preloaded/src/prim_inet.erl
@@ -520,13 +520,35 @@ sendfile(S, FileHandle, Offset, Length)
 sendfile(S, FileHandle, Offset, Length) ->
     case erlang:port_info(S, connected) of
         {connected, Pid} when Pid =:= self() ->
-            sendfile_1(S, FileHandle, Offset, Length);
+            Uncork = sendfile_maybe_cork(S),
+            Result = sendfile_1(S, FileHandle, Offset, Length),
+            sendfile_maybe_uncork(S, Uncork),
+            Result;
         {connected, Pid} when Pid =/= self() ->
             {error, not_owner};
         _Other ->
             {error, einval}
     end.
 
+sendfile_maybe_cork(S) ->
+    case getprotocol(S) of
+        tcp ->
+            case getopts(S, [nopush]) of
+                {ok, [{nopush,false}]} ->
+                    _ = setopts(S, [{nopush,true}]),
+                    true;
+                _ ->
+                    false
+            end;
+        _ -> false
+    end.
+
+sendfile_maybe_uncork(S, true) ->
+    _ = setopts(S, [{nopush,false}]),
+    ok;
+sendfile_maybe_uncork(_, false) ->
+    ok.
+
 sendfile_1(S, FileHandle, Offset, 0) ->
     sendfile_1(S, FileHandle, Offset, (1 bsl 63) - 1);
 sendfile_1(_S, _FileHandle, Offset, Length) when
@@ -1318,6 +1340,7 @@ enc_opt(pktoptions)      -> ?INET_OPT_PKTOPTIONS;
 enc_opt(ttl)             -> ?INET_OPT_TTL;
 enc_opt(recvttl)         -> ?INET_OPT_RECVTTL;
 enc_opt(nodelay)         -> ?TCP_OPT_NODELAY;
+enc_opt(nopush)          -> ?TCP_OPT_NOPUSH;
 enc_opt(multicast_if)    -> ?UDP_OPT_MULTICAST_IF;
 enc_opt(multicast_ttl)   -> ?UDP_OPT_MULTICAST_TTL;
 enc_opt(multicast_loop)  -> ?UDP_OPT_MULTICAST_LOOP;
@@ -1379,6 +1402,7 @@ 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(?TCP_OPT_NOPUSH)          -> nopush;
 dec_opt(?INET_OPT_RECVTOS)        -> recvtos;
 dec_opt(?INET_OPT_RECVTCLASS)     -> recvtclass;
 dec_opt(?INET_OPT_PKTOPTIONS)     -> pktoptions;
@@ -1465,6 +1489,7 @@ type_opt_1(pktoptions)      -> opts;
 type_opt_1(ttl)             -> int;
 type_opt_1(recvttl)         -> bool;
 type_opt_1(nodelay)         -> bool;
+type_opt_1(nopush)          -> bool;
 type_opt_1(ipv6_v6only)     -> bool;
 %% multicast
 type_opt_1(multicast_ttl)   -> int;
diff --git a/lib/kernel/doc/src/inet.xml b/lib/kernel/doc/src/inet.xml
index 127c110df4..87b08e4e36 100644
--- a/lib/kernel/doc/src/inet.xml
+++ b/lib/kernel/doc/src/inet.xml
@@ -1147,6 +1147,18 @@ setcap cap_sys_admin,cap_sys_ptrace,cap_dac_read_search+epi beam.smp</code>
               is turned on for the socket, which means that also small
               amounts of data are sent immediately.</p>
           </item>
+          <tag><c>{nopush, Boolean}</c>(TCP/IP sockets)</tag>
+          <item>
+            <p>This translates to <c>TCP_NOPUSH</c> on BSD and
+              to <c>TCP_CORK</c> on Linux.</p>
+            <p>If <c>Boolean == true</c>, the corresponding option
+              is turned on for the socket, which means that small
+              amounts of data are accumulated until a full MSS-worth
+              of data is available or this option is turned off.</p>
+            <p>Note that while <c>TCP_NOPUSH</c> socket option is available on OSX, its semantics
+              is very different (e.g., unsetting it does not cause immediate send
+              of accumulated data). Hence, <c>nopush</c> option is intentionally ignored on OSX.</p>
+          </item>
           <tag><c>{packet, PacketType}</c>(TCP/IP sockets)</tag>
           <item>
             <p><marker id="packet"/>Defines the type of packets to use for a socket.
diff --git a/lib/kernel/src/inet_int.hrl b/lib/kernel/src/inet_int.hrl
index c8e09d18ad..f6525d7261 100644
--- a/lib/kernel/src/inet_int.hrl
+++ b/lib/kernel/src/inet_int.hrl
@@ -162,6 +162,7 @@
 -define(INET_OPT_PKTOPTIONS,      45).
 -define(INET_OPT_TTL,             46).
 -define(INET_OPT_RECVTTL,         47).
+-define(TCP_OPT_NOPUSH,           48).
 % Specific SCTP options: separate range:
 -define(SCTP_OPT_RTOINFO,	 	100).
 -define(SCTP_OPT_ASSOCINFO,	 	101).
diff --git a/lib/kernel/test/inet_sockopt_SUITE.erl b/lib/kernel/test/inet_sockopt_SUITE.erl
index ada9c2689c..27ff74e309 100644
--- a/lib/kernel/test/inet_sockopt_SUITE.erl
+++ b/lib/kernel/test/inet_sockopt_SUITE.erl
@@ -110,9 +110,14 @@ simple(Config) when is_list(Config) ->
     {S1,S2} = create_socketpair(Opt, Opt),
     {ok,Opt} = inet:getopts(S1,OptTags),
     {ok,Opt} = inet:getopts(S2,OptTags),
-    COpt = [{X,case X of nodelay -> false;_ -> Y end} || {X,Y} <- Opt],
+    NoPushOpt = case os:type() of
+                    {unix, Osname} when Osname =:= linux; Osname =:= freebsd -> {nopush, true};
+                    {_,_} -> {nopush, false}
+                end,
+    COpt = [{X,case X of nodelay -> false;_ -> Y end} || {X,Y} <- [NoPushOpt|Opt]],
+    COptTags = [X || {X,_} <- COpt],
     inet:setopts(S1,COpt),
-    {ok,COpt} = inet:getopts(S1,OptTags),
+    {ok,COpt} = inet:getopts(S1,COptTags),
     {ok,Opt} = inet:getopts(S2,OptTags),
     gen_tcp:close(S1),
     gen_tcp:close(S2),